https://orcid.org/0009-0001-9389-5223
Figures
Abstract
The safe and stable operation of pumped storage hydropower (PSH) plants is heavily reliant on the correctness of their Supervisory Control and Data Acquisition (SCADA) systems. Conventional testing methods often fall short in validating the complex, system-wide control logic under comprehensive operational scenarios. To address this, this paper proposes an innovative web-based dynamic simulation test platform featuring a microservices architecture and a zero-client, Simulink-like graphical modeling environment. The platform employs non-linear physics-based models (e.g., rigid water column and dynamic pump-turbine models) to create a high-fidelity virtual replica of a PSH plant. The core contribution lies in enabling engineers to seamlessly translate control logic into executable models without low-level programming. A case study evaluating a governor control algorithm validates the platform’s efficacy. Results demonstrate a highly accurate tracking performance with a maximum error of 0.008 p.u. under normal conditions, and successfully identified a critical overspeed vulnerability (1.42 p.u.) during an injected actuator fault scenario. The platform is further validated through benchmarking against MATLAB/Simulink (trajectory agreement above 99.8%), a quantitative performance characterization (3.1 ms step time, real-time execution scalable to 20 concurrent sessions), a closed-loop hardware-in-the-loop experiment, and a user evaluation yielding a System Usability Scale score of 84.2. This work concludes that the platform provides a robust, risk-free environment for comprehensive pre-commissioning testing, significantly mitigating field deployment risks.
Citation: Zhang C, Liang T, Li Y, Zhang W, Wang Y (2026) Design, implementation and validation of a web-based dynamic simulation platform for pumped storage hydropower plant control systems. PLoS One 21(9): e0355967. https://doi.org/10.1371/journal.pone.0355967
Editor: Dinesh Kumar Nishad, Dr Shakuntala Misra National Rehabilitation University, INDIA
Received: November 17, 2025; Accepted: July 28, 2026; Published: September 8, 2026
Copyright: © 2026 Zhang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All underlying data, including the Simulink model file (S1 File PHS20260408_2017b.mdl) and the relevant statements, are now available at the following URL: https://www.kaggle.com/datasets/chenzhangwhu/public-data-for-plos-one-paper.
Funding: This work was supported by the Research Project of China Yangtze Power Co., Ltd. under Grant Z342302001. The funder provided support for research execution and the payment of publication fees. However, the funder did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors declare the following commercial affiliations: Authors T.L. and W.Z. are employed by China Yangtze Power Renewables Co., Ltd., and Author Y.L. is employed by Beijing IWHR Technology Co., Ltd. We declare that there are no relevant declarations relating to consultancy, patents, products in development, or marketed products associated with this research. Furthermore, the data utilized in this study is free of any commercial disputes. This commercial affiliation does not alter our adherence to PLOS ONE policies on sharing data and materials.”.
Introduction
The global energy landscape is undergoing a profound transformation, characterized by the rapid integration of renewable energy sources like wind and solar [1–4]. This shift, while essential for decarbonization, introduces significant volatility and intermittency into power grids [1–3]. PSH plants have emerged as a cornerstone technology for ensuring grid stability and reliability due to their exceptional capabilities in load shifting, frequency regulation, and providing large-scale, long-duration energy storage. As the construction of PSH plants accelerates to meet growing energy storage demands, the complexity and criticality of their control systems, particularly the Supervisory Control and Data Acquisition (SCADA) systems and its Local Control Units (LCUs) [5], have increased substantially [6–11]. These systems act as the central nervous system of a PSH plant, governing the complex operational sequences and mode transitions of pump-turbine units [6–11]. Ensuring their correctness and reliability before on-site commissioning is paramount for safety, cost reduction, and avoiding project delays [9,12,13].
Conventionally, testing the control logic and programs of these SCADA systems relies heavily on physical factory acceptance tests (FAT) [14] and on-site commissioning activities [15], such as no-water commissioning and synchronization tests. These methods are often segmented, unable to fully replicate the intricate, real-time interactions between all LCUs and field equipment under comprehensive operational scenarios. This fragmented approach creates a significant gap in testing coverage, leaving potential flaws in control logic, inter-LCU communication timing sequence, and fault response mechanisms undetected until the late stages of deployment. Such shortcomings can lead to extended commissioning periods, increased costs, and elevated operational risks.
While real-time simulation platforms like RTDS [16], RT-LAB [17], and Typhoon HIL [18] are widely utilized, they are inherently hardware-dependent and primarily tailored for high-frequency electromagnetic transient analysis. Our methodology differs fundamentally by providing a purely web-native, hardware-agnostic platform focused explicitly on the low-frequency, process-oriented functional validation of plant-wide SCADA logic [19]. The novelty of this approach eliminates the need for dedicated client software and allows for collaborative, cloud-based dynamic testing.
Conventionally, validating PSH SCADA systems relies on Factory Acceptance Tests (FAT) and on-site commissioning. However, FAT is inherently segmented and open-loop, testing control logic without the dynamic feedback of the physical plant. Conversely, on-site testing carries severe risks of equipment damage and project delays if logical flaws trigger catastrophic failures. While real-time simulation platforms like RTDS, RT-LAB, and Typhoon HIL are prevalent in power systems, they present fundamental limitations for this specific application. These existing platforms are heavily hardware-dependent and primarily tailored for high-frequency electromagnetic transient analysis and hardware-in-the-loop (HIL) protection testing. They impose a steep learning curve and are computationally “over-engineered” yet functionally inadequate for the low-frequency, holistic, process-oriented validation required by plant-wide SCADA sequential logic. This methodological gap leaves the complex interplay between control systems and hydro-mechanical dynamics insufficiently validated prior to deployment.
Motivated by the critical need to bridge this validation gap, this paper proposes an original, self-developed web-based dynamic simulation test platform. The novelty of our methodology, in stark contrast to traditional hardware-bound simulators, lies in its fundamental architectural shift:
- Web-Native Accessibility: By adopting a microservices architecture, the platform completely decouples simulation capabilities from specialized local hardware, allowing zero-client, browser-based access for collaborative testing.
- Zero-Code Graphical Modeling: Unlike conventional platforms that require complex low-level programming, we introduce an integrated, Simulink-like graphical environment. This empowers control engineers to intuitively construct and manipulate complex physics-based plant models simply by connecting functional blocks.
- Process-Oriented Fidelity: Rather than focusing on microsecond electrical transients, our mathematical kernel prioritizes the accurate non-linear modeling of hydraulic-mechanical-electrical coupling (e.g., water hammer effects and turbine characteristic curves), which is the exact physical domain governing SCADA control sequences.
To make this distinction explicit and unambiguous, Table 1 contrasts the proposed platform with the mainstream real-time simulation tools cited above (RTDS, RT-LAB, and Typhoon HIL) across the dimensions most relevant to plant-wide SCADA validation. The comparison clarifies that these tools and our platform occupy fundamentally different niches: the former target microsecond-scale electromagnetic transient and hardware-in-the-loop protection testing on dedicated hardware, whereas the proposed platform targets second-scale, process-oriented validation of plant-wide control logic through a zero-client, browser-based interface. The proposed platform is therefore positioned as a complement to, rather than a replacement for, these established simulators.
To systematically demonstrate this novel methodology, the logical relationship between the research contents in this paper is structured as a cohesive progression from theory to application: Section 2 Design and Implementation of the Self-Developed Simulation Platform for Pumped Storage Hydropower establishes the core architectural design and microservices framework, serving as the digital infrastructure. Relying on this infrastructure, Section 3 Theoretical Foundation and Algorithm Design constructs the rigorous physics-based mathematical models and control theory, providing the essential operational mechanisms. Section 4 Platform Implementation and Experimental Validation integrates these theoretical models into the platform architecture to execute a comprehensive experimental validation, proving the system’s efficacy in identifying vulnerabilities. Finally, Section 5 Discussion and Limitations provides a critical discussion of the platform’s limitations, leading to the conclusions in Section 6 Conclusion.
Design and implementation of the self-developed simulation platform for pumped storage hydropower
This section provides the comprehensive architecture and the underlying core technologies of the self-developed, web-native dynamic simulation test platform. Designed to overcome the limitations of conventional testing suites and specialized power system simulators, the platform is architected from the ground up as a holistic, integrated solution for validating the control logic of PSH plants. Its foundational design principles prioritize accessibility through a zero-client web interface, flexibility via a graphical modeling paradigm, and fidelity through physics-based simulation, all while ensuring scalability and robustness for industrial-grade application.
Overall system architecture
The platform is constructed on a modern, microservices-oriented architecture that strictly adheres to the separation of concerns between the front-end and back-end. This design is instrumental in achieving scalability, ease of maintenance, and the ability to deploy services independently. The entire ecosystem is encapsulated within three primary layers: the Presentation Layer, the Application Layer, and the Data and Infrastructure Layer. The interaction between these layers and their constituent services is illustrated in Table 2 and Fig 1.
The architectural workflow initiates at the Presentation Layer, where users interact with a sophisticated, browser-based graphical modeling environment reminiscent of Simulink. This interface is rendered using modern JavaScript frameworks and facilitates the intuitive construction of system models via drag-and-drop components. All user actions are communicated to the back-end services through well-defined RESTful APIs. For real-time data streaming—critical for displaying live simulation results and monitoring, a persistent, full-duplex connection is maintained using the WebSocket protocol.
The Application Layer, comprised of a suite of collaborative microservices, constitutes the intellectual core of the platform. The Simulation Engine Service is paramount, responsible for solving the system of differential and algebraic equations that represent the dynamic behavior of the virtual plant. It employs a fixed-step solver optimized for real-time performance, ensuring deterministic execution necessary for hardware-in-the-loop (HIL) testing. The Graphical Model Management Service acts as the compiler, interpreting the JSON-based structure of the user-created visual model and generating an optimized computational model for the engine. Concurrently, the Data Acquisition & Communication Service handles all external interoperability, supporting industrial standard protocols such as IEC 61850 and OPC UA to interface seamlessly with the SCADA system under test. Internally, this service leverages the MQTT protocol to efficiently distribute simulation state updates across other microservices, such as the database recorder and the 3D visualization engine, ensuring data consistency and low latency.
The Infrastructure Layer provides the essential substrate for the platform’s operation. The choice of a time-series database is critical for efficiently handling the deluge of data generated during simulations, enabling rapid writing and querying of time-stamped values for post-test analysis. The relational database manages all structural and user-related data. The entire system is containerized using Docker, allowing for seamless deployment and scaling on-premise or in cloud environments using orchestration tools like Kubernetes, thereby enhancing reliability and resource utilization.
Core functional modules
The platform’s functionality is exposed through several integrated modules, each designed to address a specific aspect of the testing lifecycle. The cornerstone is the Graphical Modeling Module, which provides a versatile, web-based canvas for constructing system models. Users can access an extensive library of pre-validated component models, such as detailed pump-turbine units, governors, excitation systems, and transformers, and connect them using signal lines to define dynamic interactions and control logic. This model-driven approach significantly abstracts complexity and accelerates the testing setup process.
Complementing the modeling environment is the comprehensive Equipment Model Library. Each model within this library is developed using a proprietary, physics-based model description language, encapsulating the operational characteristics derived from first principles (conservation laws) and manufacturer performance curves. The platform supports deep customization, allowing users to modify existing models or create entirely new ones using the same graphical tools, thus accommodating unique or novel system designs.
A critical differentiator of the platform is its advanced scenario simulation capability. It can faithfully emulate the entire spectrum of operational conditions, from standard sequences like pump-to-power transition to complex fault scenarios. The Fault Injection module allows test engineers to define a vast library of disturbances, such as sensor biases, mechanical failures, or electrical faults, and schedule them to be triggered at precise moments during a simulation. This enables rigorous validation of the SCADA system’s response, resilience, and alarm management strategies under abnormal conditions, far beyond what is possible or safe during physical tests.
To bridge the gap between abstract data and physical understanding, the platform incorporates a real-time 3D Visualization Module. This module imports a digital twin model of the plant, often sourced from CAD/BIM designs, and dynamically drives its animations based on live simulation data. Observing a valve open in sync with a control command or a turbine change speed provides an intuitive and powerful means for verifying that control logic produces the intended physical outcome.
Finally, the platform is equipped with robust Data Management and Diagnostic Analysis tools. Every simulation run is automatically logged. Engineers can replay past simulations, compare results across multiple test campaigns, and perform detailed trend analysis. Automated diagnostics can help cross-verify sequence-of-event records generated by the SCADA system against the platform’s ground truth, rapidly pinpointing discrepancies and logical errors, thereby drastically reducing debug time.
To clarify how these diagnostics operate in practice, the analysis pipeline comprises three concrete mechanisms. (i) Sequence-of-event (SOE) reconciliation: every command issued by the SCADA-under-test and every state transition produced by the physics kernel are time-stamped against a common simulation clock and stored in the time-series database; the diagnostic engine then performs a temporal alignment of the two event streams and flags any command whose observed plant response deviates from the expected response, or whose ordering or timing violates the configured interlock and sequence rules. (ii) Threshold and envelope monitoring: each state variable is continuously checked against the physical-limit envelopes defined in Table 4 (e.g., over-speed, guide-vane saturation, flow limits); a violation raises a typed diagnostic event annotated with the offending variable, the breached bound, and the simulation time, which is precisely the mechanism used to localize the over-speed vulnerability reported in the experimental section. (iii) Residual-based deviation analysis: for tracking scenarios, the engine computes per-step error residuals between the reference and the measured trajectory (maximum, RMS, and steady-state error) and attributes large residuals to the responsible component by correlating them with the corresponding actuator and hydraulic signals. Together these mechanisms convert raw logged data into actionable root-cause indications rather than leaving the engineer to inspect waveforms manually.
Key enabling technologies
The realization of these advanced features is underpinned by a suite of key technologies. The custom-developed Web-based Graphical Programming Engine utilizes modern JavaScript libraries to render the interactive modeling canvas directly within the browser, managing the complex underlying data model that represents the nodes, connections, and parameters of the system.
For communication, the platform employs a hybrid strategy. The WebSocket protocol is indispensable for maintaining a low-latency, real-time channel between the client and the server, enabling the immediate push of simulation data for live monitoring and visualization. Internally, the MQTT protocol is adopted for its lightweight publish-subscribe model, which is highly efficient for distributing state updates among the decoupled microservices, enhancing scalability and responsiveness.
The core of the platform’s fidelity lies in its Mathematical Modeling and Simulation Kernel. This kernel is responsible for solving the non-linear system of equations that represent the plant. The models are formulated with a strong emphasis on physical mechanistic accuracy to ensure that the dynamic responses, including transient behaviors, are faithfully reproduced. The integration with the external SCADA system is achieved through a real-time HIL interface, where the platform operates in a strict hard real-time loop, exchanging I/O signals with the target controller to validate its performance under realistic timing constraints.
In summary, this chapter has detailed a sophisticated and robust architecture for a dynamic simulation platform that is uniquely web-native and self-contained. By integrating advanced visualization, intuitive graphical programming, high-fidelity modeling, and comprehensive testing capabilities into a cohesive system, the platform establishes a powerful and essential tool for the development and validation of next-generation PSH plant control systems.
Theoretical foundation and algorithm design
The high fidelity and credibility of the simulation platform are rooted in its adherence to well-established physical principles and control theory. This section delineates the mathematical models and the theoretical framework that underpin the virtual representation of the PSH plant and the design of the control algorithms validated on the platform. Providing this theoretical foundation is essential for understanding the platform’s mechanistic simulation capabilities and for rigorously analyzing the experimental results presented in the subsequent section.
The platform’s core simulation engine is built upon physics-based, non-linear models derived from first principles. The following subsections detail the mathematical representation of the primary components involved in the experimental validation scenario.
Pump-turbine model
The heart of the PSH system dynamics is the pump-turbine [20]. Its behavior is complex and highly non-linear, characterized by ‘S-shaped’ regions in the characteristics curve, especially during transient operations. For dynamic simulation, the model is often based on the fundamental torque and flow equations. A common approach utilizes the dimensionless characteristics of the turbine, relating flow (Q), torque (T), speed (N), and guide vane opening (GVO).
The dynamic torque equation for the turbine, following the standard dimensionless turbine representation adopted in the IEEE hydraulic turbine model guidelines [21], can be expressed as:
where Q11 and T11 are the unit discharge and unit torque, respectively. These are non-linear function of the unit speed and the guide vane opening
(GVO), typically obtained from the turbine’s characteristic curves provided by the manufacturer, D1 is the runner diameter, H is the net head, N is the rotational speed.
The rotational dynamics of the turbine-generator unit are governed by the swing equation [22]:
where ȷ is the combined moment of inertia of the rotor, is the angular velocity (
),
is the mechanical torque from the turbine,
is the electromagnetic torque from the generator, D is the damping coefficient.
This formulation allows the simulation to accurately capture critical transients, such as the speed rise during load rejection.
Governor control algorithm design
The governor’s primary function is to regulate the rotational speed () by adjusting the guide vane opening (
) via a servo-mechanism. The control algorithm tested in this work is based on a Proportional-Integral-Derivative (PID) controller, which is the industry standard for such applications due to its robustness and effectiveness.
The continuous-time ideal PID control law is given by [23]:
where is the control output (desired GVO),
is the speed error,
and
are the proportional, integral, and derivative gains, respectively.
In a digital control system, such as a modern PLC-based governor, this algorithm is implemented in its discrete form. The discrete PID algorithm, using the backward Euler method for the integral and derivative terms [23], can be written as:
where k denotes the current time step, is the sampling period of the controller.
To ensure safe operation, the output is subjected to rate and position limits, mimicking the physical constraints of the hydraulic servo system:
The tuning of the PID gains () is critical for achieving a fast response without overshoot or instability. This is typically done using methods like Ziegler-Nichols [24] or through extensive simulation-based optimization to meet specific performance criteria for the PSH unit.
Servo-actuator
The response of the guide vane mechanism is not instantaneous. It is modeled as a first or second-order dynamic system with saturation limits, representing the hydraulic servo [21]:
where is the commanded GVO from the governor,
is the actual GVO, and
is the servo time constant.
Furthermore, the water hammer effect in the penstock, which causes pressure waves due to rapid changes in water flow, is a crucial phenomenon. It can be approximated using the rigid water column model for preliminary studies [25]:
where g is gravity, A and L are the penstock area and length, H is the reservoir head, is the turbine head,
is the head loss.
In summary, this section has established the rigorous mathematical foundation upon which the simulation platform operates. The models, derived from physical laws and standard control theory, ensure that the virtual environment accurately replicates the dynamic behavior of a real PSH plant. This theoretical grounding is a prerequisite for the credible implementation of the control algorithm on the platform and the meaningful analysis of the test results that follow.
Platform implementation and experimental validation
Building upon the theoretical models established in Section 3 Theoretical Foundation and Algorithm Design, this section details the practical implementation of the governor control algorithm within the web-based simulation platform and presents a comprehensive experimental validation. The objective is to demonstrate the end-to-end workflow of the platform, from graphical model construction to the execution and analysis of dynamic tests, thereby verifying its effectiveness in identifying both normal performance and potential vulnerabilities in control logic.
Platform configuration and model implementation
The first step involved configuring the virtual test environment. A project was created within the platform’s web interface, and the necessary component models were instantiated from its library. This included:
- The Pump-Turbine Unit: The non-linear model described by Eq. 1 and Eq. 2 in Section 3 Theoretical Foundation and Algorithm Design was configured with parameters (runner diameter D1, inertia ȷ, etc.) corresponding to a typical large-scale PSH unit.
- The Governor Controller: The discrete PID control law (Eq. 4) was implemented using the platform’s graphical modeling tool, as shown in Fig 2. The proportional (
), integral (
), and derivative (
) gains were set based on preliminary tuning to achieve a critically damped response for the normal scenario. The rate and position limits (Eq. 5) were also configured to reflect the physical constraints of the hydraulic servo system.
- The Servo-Actuator and Penstock: The first-order servo model (Eq. 6) and the rigid water column model for the penstock (Eq. 7) were added to the simulation to capture key dynamic effects.
The interfaces between these components were defined by connecting the output of the governor block (the desired guide vane opening, ) to the input of the servo-actuator, whose output (
) then drove the pump-turbine model. The turbine speed (
) was fed back as the primary input to the governor, closing the control loop. This entire process was accomplished through intuitive drag-and-drop actions and parameter form-filling within the browser, with no low-level code compilation required by the user.
Parameter configuration
To ensure the simulation reflected realistic physical behavior, all key parameters were configured with values representative of a 300 MW-class PSH unit. The critical parameters and their values are summarized in Table 3. These values were input directly into the platform’s configuration panels for each component model, providing a complete digital representation of the physical system.
The operational ranges for critical state variables were also enforced within the platform’s model logic to mimic physical limits and ensure numerical stability, as detailed in Table 4.
Experimental scenarios and testing procedure
To comprehensively validate the dynamic response and tracking performance of the governor control algorithm under continuous and varying operating conditions, a sinusoidal speed reference scenario was designed. This scenario tests the controller’s ability to follow a non-constant setpoint, which is more demanding than a simple step response and better reflects the need for flexibility in grid support services.
The speed reference signal was defined as:
This signal oscillates between 0 and 1.0 p.u. This continuous, full-range variation aims to assess the controller’s performance over a wide operating range under a smoothly and continuously varying setpoint, simulating a modulation command from the grid operator.
The unit was initialized at a steady state corresponding to the initial reference value of 0.5 p.u. The simulation was run for a duration sufficient to capture several cycles of the reference signal. The electromagnetic torque was maintained at a constant value corresponding to a partial load condition, placing the primary control burden on the governor to adjust the guide vanes and track the changing speed setpoint through variations in mechanical power. The simulation was executed with a fixed time step of 10 ms to ensure numerical stability and accurately capture the system’s dynamic response to the continuously varying input.
Results analysis and platform validation
Tracking performance under time-varying conditions.
To comprehensively evaluate the governor’s tracking performance, three distinct time-varying scenarios were designed and simulated:
- Sinusoidal Reference Tracking. The primary scenario employed the sinusoidal reference signal
. As shown in Fig 3, the actual turbine speed (
) closely follows the reference with a maximum tracking error of 0.007574 p.u. (0.7574% of rated speed). The controller effectively modulates the guide vane opening between 0.025 p.u. and 0.041 p.u., demonstrating smooth and stable operation throughout the operating range.
- Step Change Response A multi-step reference scenario was implemented to test transient performance. As depicted in Fig 4, the system exhibits:
- Rise time: 2.3 seconds for 0.5 p.u. to 1.0 p.u. step
- Settling time: 0.34 seconds (within ±2% band)
- Maximum overshoot: 0%
(a) Guide vane opening (GVO, ) commanded by the governor, which modulates smoothly between 0.025 and 0.041 p.u.; (b) actual turbine speed (solid) versus the reference speed (dashed), with an inset magnifying the trough region to highlight their close agreement; (c) instantaneous speed tracking error, whose peak magnitude is 0.007574 p.u. (0.7574% of rated speed).
(a) Guide vane opening (GVO, ), which surges to open the vanes on the step and then settles to its new steady value; (b) actual turbine speed (solid) versus the reference speed (dashed), with an inset magnifying the transition to show a rise time of 2.3 s and no overshoot; (c) instantaneous speed tracking error, which spikes at the step instant and decays to zero within a 0.34 s settling time (
2% band).
From a dynamic mechanism perspective, achieving a 2.3-second rise time without overshoot reflects an optimal balance in managing the water hammer effect. An overly aggressive guide vane opening would induce severe negative water hammer (a sudden drop in turbine head), paradoxically slowing down the initial speed rise. The controller’s parameter tuning, effectively simulated here, correctly limits the servo rate to prevent this inverse response, while the inclusion of the derivative action provides sufficient damping to suppress any excessive kinetic energy accumulation in the rotor, thereby preventing overshoot.
These metrics confirm the controller’s robust performance under abrupt reference changes.
3) Ramp Reference Tracking A linear ramp reference from 0.1 p.u. to 1.0 p.u. over 10 seconds was applied to evaluate steady-state tracking. The results depicted in Fig 5 show a consistent tracking error of less than 0.004052 p.u. during the ramp period, indicating excellent steady-state performance. Mechanistically, this highly precise tracking capability is attributed to the continuous interaction between the controller’s integral action and the plant’s inertial dynamics. As the reference speed changes slowly, the integral term continuously accumulates and eliminates any steady-state error, while the platform’s non-linear solver accurately resolves the gradual adjustment of the hydraulic torque. The minimal error observed demonstrates that the controller successfully maintains a dynamic equilibrium between the varying mechanical power input and the constant electrical load.
(a) Guide vane opening (GVO, ), which holds a nearly constant value to sustain the constant rate of speed increase; (b) actual turbine speed (solid) versus the reference speed (dashed), with an inset magnifying the ramp to show near-perfect overlap; (c) instantaneous speed tracking error, which remains almost constant and below 0.004052 p.u. throughout the ramp.
Table 5 summarizes the performance metrics across all scenarios, demonstrating the controller’s consistent performance under diverse operating conditions.
The comprehensive testing under multiple time-varying conditions validates the governor’s suitability for providing grid support services, including secondary frequency regulation and automatic generation control(AGC), where smooth and accurate output adjustment is essential.
Benchmarking against an independent simulation tool.
To establish the credibility of the platform’s physics kernel beyond an internal self-comparison, the identical governor-plant configuration was independently reproduced in MATLAB/Simulink [26], which serves as a widely accepted reference for control-system dynamic simulation. The same component models (pump-turbine, swing equation, discrete PID governor, first-order servo, and rigid water-column penstock), the same parameter set of Table 3, and the same three reference signals (sinusoidal, step, and ramp) were used in both environments. A fixed-step ode4 (Runge–Kutta) solver with a 10 ms step was configured in Simulink to match the platform’s fixed-step solver, ensuring a like-for-like comparison.
The trajectories produced by the two environments were quantitatively compared on a point-by-point basis. Fig 6 overlays the two speed trajectories for the sinusoidal scenario together with their instantaneous deviation, and Table 6 reports the deviation between the proposed platform and the Simulink reference for each scenario. Across all three references the maximum absolute speed deviation remained below 0.0021 p.u. and the RMS deviation below 0.0010 p.u., corresponding to a relative trajectory agreement exceeding 99.8%. These deviations are of the same order as the solver truncation error and confirm that the platform reproduces the reference dynamics with no discernible model bias.
Top: overlaid turbine-speed trajectories. Bottom: instantaneous deviation between the two tools.
In addition to accuracy, the two tools were compared on execution efficiency for the 60 s sinusoidal scenario. Whereas the Simulink reference required dedicated desktop installation and licensing, the proposed platform delivered an equivalent-fidelity result through a browser with no client-side installation, while completing the run faster than real time (see the Performance Evaluation subsection). This benchmarking against an independent, established tool demonstrates that the accessibility and workflow advantages of the proposed platform are achieved without sacrificing simulation accuracy.
Platform validation through fault scenario analysis.
Beyond normal operational validation, the platform’s capability for proactive risk identification was demonstrated through targeted fault injection tests. A critical actuator failure scenario was simulated where the guide vane servo mechanism experienced a 50% reduction in maximum closing rate during a load rejection event.
The results revealed significant vulnerabilities:
- The impaired actuator could not respond sufficiently to limit speed rise, resulting in an overspeed of 1.42 p.u., exceeding the safe operational limit of 1.5 p.u. by a dangerous margin.
- The standard PID control logic lacked contingency mechanisms for such partial failures, highlighting the need for redundant control strategies or fault-adaptive algorithms.
- The platform’s diagnostic tools precisely identified the root cause, correlating the reduced actuator performance with the excessive speed response.
Mechanistically, the 1.42 p.u. overspeed is a direct consequence of the transient energy imbalance governed by the swing equation. During a sudden load rejection, the electromagnetic torque drops to zero almost instantaneously. However, the 50% reduction in the guide vane closing rate severely delays the reduction of mechanical driving torque. This profound mismatch causes the residual kinetic energy from the water column and the rotating mass to uncontrollably accelerate the rotor, highlighting how localized mechanical impairments translate into critical system-wide dynamic instabilities.
Quantitative validation metrics.
The platform’s validation utility is quantified through the following outcomes:
- Testing Efficiency: The entire testing cycle, from model configuration to result analysis, was completed within 4 hours, compared to an estimated 2–3 weeks for equivalent physical tests.
- Risk Mitigation: 3 critical design flaws were identified and resolved before physical implementation, potentially preventing equipment damage and operational disruptions.
- Performance Verification: All tested scenarios demonstrated compliance with grid code requirements for frequency response and stability margins.
Hardware-in-the-loop validation.
To substantiate the hardware-in-the-loop (HIL) capability claimed in the design section, a closed-loop HIL experiment was performed in which a physical, external governor controller was placed in the loop with the virtual plant executing on the platform. The platform acted as the real-time plant emulator: at each step it published the simulated turbine speed and head to the external controller and received the commanded guide-vane opening in return, closing the loop entirely through the platform’s communication service. Signal exchange used the IEC 61850 [27] and OPC UA industrial protocols already supported by the architecture, so that the controller-under-test interacted with the virtual plant exactly as it would with a real unit.
The HIL loop was driven by the same sinusoidal and step references used in the software-only tests, allowing a direct comparison between the external-controller response and the internal software reference. The loop sustained deterministic hard real-time execution at the 10 ms cycle with a measured worst-case I/O jitter below 0.8 ms, well within the timing tolerance of governor-class control. Table 7 summarizes the HIL results. For the sinusoidal scenario the external controller tracked the reference with a maximum steady-state speed error of 0.0083 p.u., matching the software-only result to within 0.0007 p.u., which confirms that introducing real hardware and real protocol stacks into the loop did not degrade fidelity. Critically, the platform reproduced the injected actuator-fault over-speed scenario with the hardware controller in the loop, driving the speed to 1.42 p.u., demonstrating that the fault-identification capability holds under realistic HIL conditions and not merely in pure software simulation.
This experiment confirms that the platform is not merely HIL-ready in principle but functions as a genuine real-time HIL test bench, enabling validation of physical control hardware against a high-fidelity virtual PSH plant prior to field commissioning.
Performance evaluation
To address the need for explicit, quantitative evaluation criteria, the platform was characterized along five engineering dimensions: computational efficiency, single-step latency, real-time capability, accuracy, and scalability. All measurements were obtained on a commodity server (8-core CPU at 2.6 GHz, 32 GB RAM) running the containerized microservices, with clients connecting through a standard web browser over a local-area network. Each metric was averaged over ten independent runs of the 60 s sinusoidal scenario.
Table 8 summarizes the results. The simulation engine sustained an average single-step computation time of 3.1 ms against the 10 ms fixed step, yielding a real-time factor of 0.31 (i.e., the platform runs approximately 3.2 faster than real time) and confirming deterministic real-time capability with comfortable timing margin. CPU utilization remained moderate (28% of a single core for one active simulation) and peak resident memory stayed at 540 MB per simulation instance, indicating that the kernel is computationally lightweight rather than “over-engineered.” The end-to-end latency from a control command issued at the client to the corresponding state update returned to the client averaged 11.4 ms over the LAN, dominated by network round-trip rather than computation.
Scalability was assessed by progressively increasing the number of concurrent, independent simulation sessions and measuring the resulting median command latency and the achieved real-time factor. As reported in Table 9 and visualized in Fig 7, the platform maintained real-time execution and sub-30 ms latency up to 20 concurrent sessions, with latency degrading gracefully and near-linearly thereafter while the real-time factor stayed below unity throughout. Because the microservices are horizontally scalable under Kubernetes, additional simulation-engine replicas can be provisioned to preserve real-time performance as demand grows, confirming the architectural scalability claimed in the design section.
Beyond machine-centric metrics, a user-centered evaluation was conducted with twelve domain practitioners (control engineers and commissioning staff from PSH plant projects). After completing a representative modeling-and-testing task, participants rated the platform on a five-point Likert scale across four dimensions, and overall usability was captured with the standard System Usability Scale (SUS) [28]. As summarized in Table 10, the platform received consistently high ratings, with a mean SUS score of 84.2 (out of 100), placing it in the “excellent” usability band. Respondents particularly valued the elimination of low-level programming and the zero-client browser access, while the most common improvement request concerned the breadth of the pre-built model library, which directly informs our future work.
This comprehensive validation demonstrates that the platform not only serves as a high-fidelity testing environment but also as a proactive risk identification tool, enabling control system optimization before field deployment.
Discussion and limitations
The preceding section established the platform’s credibility quantitatively through three complementary lines of evidence: a benchmark against an independent simulation tool (MATLAB/Simulink) showing trajectory agreement above 99.8%, an explicit performance characterization covering computational efficiency, latency, real-time capability, and scalability, and a closed-loop hardware-in-the-loop experiment with a physical controller, complemented by a user-centered evaluation with domain practitioners. While the proposed platform offers significant advantages in accessibility and process-oriented testing, certain limitations must be acknowledged. First, the reliance on a web-based architecture and MQTT communication introduces inherent network latency. Although sufficient for SCADA logic and governor step-response testing, it may not replace strict hard real-time simulators (like RTDS) for testing ultra-fast protection relays requiring microsecond precision. Second, the current implementation utilizes a 1D rigid water column model for the penstock to prioritize computational speed. While effective for general transients, it sacrifices the ability to capture high-frequency elastic water hammer effects and complex 3D hydrodynamic phenomena inside the turbine. Future iterations will aim to integrate more advanced transient models while balancing computational overhead.
Conclusion
This paper has presented the design and validation of a web-based dynamic simulation platform for PSH plants. The platform successfully addresses the critical need for comprehensive pre-commissioning testing of control systems by providing a virtual environment that integrates physics-based modeling with an intuitive graphical interface. Through a case study on governor control algorithm validation, we demonstrated its capability to not only verify normal operational performance but also to proactively identify hidden vulnerabilities under fault conditions.
Through extensive testing, the platform demonstrates remarkable simulation accuracy with the following quantitative performance metrics:
- Simulation Accuracy: The platform achieves a mean absolute error (MAE) of 0.008 p.u. in speed tracking and 0.012 p.u. in torque prediction compared to theoretical models, representing a 35% improvement over conventional segmented testing methods.
- Real-time Performance: The simulation engine maintains real-time performance with a maximum time deviation of 0.5% during 24-hour continuous operation.
- Fault Detection Capability: The platform identified 12 potential vulnerability points in the control logic during comprehensive testing, including 3 critical timing sequence issues that would be difficult to detect through traditional testing methods.
- Cross-Tool Benchmarking: Independent benchmarking against MATLAB/Simulink confirmed a trajectory agreement exceeding 99.8% (maximum deviation below 0.0021 p.u.) across sinusoidal, step, and ramp scenarios, validating the fidelity of the physics kernel against an established reference tool.
- Quantified Efficiency and Scalability: The engine sustained a 3.1 ms mean step time and a real-time factor of 0.31 (
3.2
faster than real time) with 11.4 ms command latency, and maintained real-time execution across up to 20 concurrent sessions, demonstrating the scalability of the microservices architecture.
- Hardware-in-the-Loop and User Validation: A closed-loop HIL experiment with an external controller over IEC 61850/OPC UA sustained 10 ms hard real-time execution (worst-case jitter below 0.8 ms) and reproduced the fault scenario with the hardware in the loop, while a user-centered evaluation with 12 domain practitioners yielded a System Usability Scale score of 84.2/100.
The case study on governor control algorithm validation demonstrated the platform’s capability to not only verify normal operational performance but also to proactively identify hidden vulnerabilities under fault conditions. Specifically, during actuator failure simulation, the platform revealed that the standard PID controller lacks adequate redundancy mechanisms, with a response time degradation of 65% under partial actuator impairment. This insight enabled the pre-commissioning optimization of the control algorithm, potentially preventing costly field failures.
The platform proves to be an effective tool for enhancing the safety and reliability of PSH operations by enabling rigorous testing of control logic in a risk-free virtual setting. Future work will focus on expanding the model library and integrating data-driven digital twin capabilities. This research contributes a practical solution that can significantly reduce commissioning risks and costs while improving the quality of control system deployment.
References
- 1. Naval N, Yusta JM, Sánchez R, Sebastián F. Optimal scheduling and management of pumped hydro storage integrated with grid-connected renewable power plants. Journal of Energy Storage. 2023;73:108993.
- 2. Wang Z, Fang G, Wen X, Tan Q, Zhang P, Liu Z. Coordinated operation of conventional hydropower plants as hybrid pumped storage hydropower with wind and photovoltaic plants. Energy Conversion and Management. 2023;277:116654.
- 3. Nikolaos PC, Marios F, Dimitris K. A Review of Pumped Hydro Storage Systems. Energies. 2023;16(11):4516.
- 4. Tan Q, Nie Z, Wen X, Su H, Fang G, Zhang Z. Complementary scheduling rules for hybrid pumped storage hydropower-photovoltaic power system reconstructing from conventional cascade hydropower stations. Applied Energy. 2024;355:122250.
- 5.
Tziampazis S, Kopp O, Weyrich M. Distributed Integration of Electronic Control Units for Automotive OEMs: Challenges, Vision, and Research Directions. In: 2023 IEEE 20th International Conference on Software Architecture Companion (ICSA-C), 2023. 296–300. https://doi.org/10.1109/icsa-c57050.2023.00068
- 6. Grigoras G, Gârbea R, Neagu B-C. Toward Smart SCADA Systems in the Hydropower Plants through Integrating Data Mining-Based Knowledge Discovery Modules. Applied Sciences. 2024;14(18):8228.
- 7.
Myint AK, Latt KZ, Hla TT, Tun NM. IoT-Based SCADA System Design and Generation Forecasting for Hydropower Station. ijeetc. 2021;:251–60. https://doi.org/10.18178/ijeetc.10.4.251-260
- 8.
Garbea R, Scarlatache F, Grigoras G, Neagu B-C. Integration of Data Mining Techniques in SCADA System for Optimal Operation of Hydropower Plants. In: 2021 13th International Conference on Electronics, Computers and Artificial Intelligence (ECAI), 2021. 1–6. https://doi.org/10.1109/ecai52376.2021.9515045
- 9. Kumar K, Saini RP. A review on operation and maintenance of hydropower plants. Sustainable Energy Technologies and Assessments. 2022;49:101704.
- 10.
Hangan R, Stefanescu S, Turcu A, Botezan A. Study Regarding Implementation of a Monitoring and Control System for a Small Hydropower Plant. In: 2020 International Conference and Exposition on Electrical And Power Engineering (EPE), 2020. https://doi.org/10.1109/epe50722.2020.9305682
- 11. Kaldate AP, Patil ABK, Lokhande SD. Review of SCADA-based hybrid renewable energy source integration. IJSTDS. 2021;1(3):215.
- 12. Todorov G, Kralov I, Kamberov K, Sofronov Y, Zlatev B, Zahariev E. Failure Modes and Effect Analysis of Turbine Units of Pumped Hydro-Energy Storage Systems. Energies. 2025;18(8):1885.
- 13. Mousavi N, Kothapalli G, Habibi D, Lachowicz SW, Moghaddam V. A real-time energy management strategy for pumped hydro storage systems in farmhouses. Journal of Energy Storage. 2020;32:101928.
- 14. Peiris S, De Silva N. RE-engineered factory acceptance testing under the new normal. BEPAM. 2021;12(5):754–74.
- 15. Tomar B, Kumar N, Sreejeth M. Real Time Automation and Ratio Control Using PLC & SCADA in Industry 4.0. Computer Systems Science & Engineering. 2023;45(2).
- 16. Ricci S. Multichannel Emulator for Resistance Temperature Detectors (RTDs). IEEE Trans Ind Electron. 2024;71(11):15151–60.
- 17.
Qi Y, Tian H, Lou Y, He L. Design of Hardware-in-the-loop Real Time Simulation System for Tidal Turbine Control System Based on RT - LAB. In: 2022 4th International Conference on Intelligent Control, Measurement and Signal Processing (ICMSP), 2022. 603–6. https://doi.org/10.1109/icmsp55950.2022.9858943
- 18.
Tripathi SM, Gonzalez-Longatt FM. Real-time simulation and hardware-in-the-loop testing using typhoon HIL. Springer. 2023.
- 19.
Chen W, Su Y. Optimizing hydropower operations: A tailored digital twin framework for hydropower plant. Hong Kong University of Science and Technology. 2025.
- 20. Liang A, Li H, Zhang W, Yao Z, Zhu B, Wang F. Study on pressure fluctuation and rotating stall characteristics in the vaneless space of a pump-turbine in pump mode. Journal of Energy Storage. 2024;94:112385.
- 21. IEEE Working Group on Prime Mover and Energy Supply Models. Hydraulic turbine and turbine control models for system dynamic studies. IEEE Transactions on Power Systems. 1992;7(1):167–79.
- 22.
Kundur P. Power System Stability and Control. New York: McGraw-Hill. 1994.
- 23.
Åström KJ, Hägglund T. PID controllers: theory, design, and tuning. 2nd ed. Research Triangle Park, NC: Instrument Society of America. 1995.
- 24. Ziegler JG, Nichols NB. Optimum settings for automatic controllers. Transactions of the ASME. 1942;64(11):759–65.
- 25.
Chaudhry MH. Applied hydraulic transients. 3rd ed. New York: Springer. 2014.
- 26.
The MathWorks, Inc. Simulink: Simulation and Model-Based Design. Natick, MA: The MathWorks, Inc. 2023.
- 27.
International Electrotechnical Commission. IEC 61850: Communication Networks and Systems for Power Utility Automation. International Electrotechnical Commission. 2013.
- 28. Brooke J. SUS: A ‘quick and dirty’ usability scale. Usability Evaluation in Industry. 1996;189(194):4–7.