Bus-type multi-CPU dynamic cooperative scheduling method and device for water-wind-light micro-grid

Abstract

The invention discloses a bus-type multi-CPU dynamic collaborative scheduling method and device for a water-wind-light micro-grid, and belongs to the technical field of hydropower station control. The method comprises the following steps: acquiring operation data of the water-wind-light micro-grid by adopting a prediction CPU in each period, according to the operation data, an ultra-short-term prediction algorithm is adopted to obtain and publish prediction results of wind and light power and load; a dispatching CPU is adopted to obtain the start-stop time, the optimal operation mode and the power instruction of the pumped storage unit through rolling optimization and mathematical planning according to the prediction result and the state of the pumped storage unit to serve as dispatching instructions; the control CPU is adopted to execute mode conversion and power instructions of the pumped storage unit according to the dispatching instruction and the state of the pumped storage unit, the final state of the executed pumped storage unit is fed back to the dispatching CPU, and closed-loop control is achieved; through a multi-CPU parallel architecture, the computing power and the real-time performance are improved, and on-demand expansion or upgrading is facilitated.

Description

Technical Field

[0001] This application belongs to the field of hydropower station control technology, and specifically relates to a bus-type multi-CPU dynamic collaborative scheduling method and device for hydropower, wind power and solar power microgrids. Background Technology

[0002] With the rapid growth of installed capacity of renewable energy sources such as wind power, solar power, and pumped storage, their share in the power grid is constantly increasing. However, compared with traditional thermal power and hydropower, wind power and solar power have obvious intermittency and volatility. In large-scale grid-connected or off-grid microgrids, this volatility can cause severe disturbances in frequency and voltage, seriously affecting the stability of the power grid.

[0003] Existing microgrid energy management systems mostly employ predictive scheduling strategies with granular time intervals of minutes or longer, and serially execute tasks such as power prediction, scheduling optimization, and execution on a single CPU or PLC controller. This centralized, serial implementation is limited by computing power and communication bandwidth, making it difficult to achieve high-precision power prediction and optimized scheduling on a second-level timescale, easily leading to delays in control command issuance or outdated prediction results. Furthermore, multi-module systems often use multiple communication protocols (such as Modbus). The data exchange methods (such as TCP, CAN, and fieldbus) are mostly timed pull-based, lacking a unified event-driven and priority arbitration mechanism. This results in loose coupling between prediction, decision-making, and execution, making it impossible to achieve true real-time closed-loop control. At the same time, the system lacks effective fault tolerance and degradation strategies when communication failures or single-point device anomalies occur, affecting the reliable operation of the microgrid. Summary of the Invention

[0004] To address the issues of limited data processing capabilities, high network latency, and poor system scalability resulting from reliance on a single centralized system for data acquisition, this application provides a bus-based multi-CPU dynamic cooperative scheduling method for hydro-wind-solar microgrids. Its improvements include: The system uses a predictive CPU to collect operational data from the hydro-wind-solar microgrid in each cycle, and then uses an ultra-short-term prediction algorithm to obtain and publish the predicted results of wind and solar power and load based on the operational data. The scheduling CPU uses rolling optimization and mathematical programming to obtain the start-up and shutdown times, optimal operating modes, and power commands of the pumped storage units based on the prediction results and the status of the pumped storage units, which are then used as scheduling commands. The control CPU executes the pumped storage unit mode conversion and power command according to the scheduling instructions and the status of the pumped storage unit, and feeds back the final status of the pumped storage unit after execution to the scheduling CPU to realize closed-loop control. The predictive CPU, scheduling CPU, and control CPU are connected via a bus, which adopts a publish / subscribe mode and employs priority arbitration and real-time interrupt mechanisms in the event of an emergency.

[0005] Optionally, the scheduling CPU uses rolling optimization and mathematical programming to obtain the start-up and shutdown times, optimal operating modes, and power commands of the pumped storage units based on the prediction results and the status of the pumped storage units, which are then used as scheduling commands. The scheduling CPU calculates the net power surplus based on the subscribed forecast results, and reads the current available regulation margin of the pumped storage and the time required for start-up and shutdown. Based on the net power surplus, it is determined whether there will be a sustained positive surplus or negative deficit in the future, and the pumped storage is planned to switch to charging or releasing mode. Based on the adjustment margin, start-up and shutdown time, and pumped storage unit status, the optimal operating mode and power command of the pumped storage unit are obtained through rolling optimization and mathematical programming. The start-up and shutdown times of pumped storage units are dynamically determined based on the forecast results.

[0006] Optionally, the step of determining whether a sustained positive surplus or negative deficit will occur in the future based on the net power surplus, and planning for pumped storage to enter charging or releasing mode, includes: If the net power surplus exceeds the pumped storage absorption threshold, the pumped storage is planned to enter the charging mode. The starting time of the surplus is selected as the target, and the pumping command is issued in advance of the pumping start time. If the unit is currently in the power generation mode, the shutdown switching time must be included in advance. If the net power surplus is negative for a preset period, the pumped storage is scheduled to enter the energy release mode. The power generation start command is issued in advance of the pumped storage start time from the moment the negative value appears. If the unit is pumping water at this time, the shutdown and reversal time needs to be included in advance. If the predicted fluctuation range is within the allowable range or the pumping capacity is limited and cannot fully cover the fluctuation, then the current state will be maintained.

[0007] Optionally, the step of obtaining the optimal operating mode and power command of the pumped storage unit through rolling optimization and mathematical programming based on the adjustment margin, start-up and shutdown time, and pumped storage unit status includes: The state of pumped storage units and their power output / absorption values ​​in each scheduling cycle are used as decision variables. The objective functions are to reduce the curtailment of new energy, reduce the number of unit start-ups and shutdowns, and reduce frequency fluctuations. The main constraints are power state constraints, power upper limit constraints, charge and discharge energy balance constraints, energy storage capacity upper and lower limit constraints, power change slope constraints, minimum start-up and shutdown duration constraints, system power balance constraints, and frequency support constraints. An objective function is established with the goals of minimizing wind and solar curtailment and minimizing frequency deviation. The minimum value of the objective function is solved under the main constraints to obtain the optimal operating mode and power command of the pumped storage unit for each time period.

[0008] Optionally, the step of dynamically deciding the start-up and shutdown time of the pumped storage unit based on the prediction results includes: When the wind and solar power is in net surplus and reaches the surplus power threshold for more than the first time, the pumping mode is activated. When the wind and solar power is in negative deficit and reaches the deficit power threshold for more than the second time, the power generation mode is activated. When the wind and solar power fluctuates between the surplus power threshold and the deficit power threshold, it remains unchanged. By scheduling the CPU to send scheduling instructions a preparation time in advance, and combining this with a sliding time window to dynamically adjust the start-up and shutdown time of the pumped storage unit using a second-level rolling mechanism.

[0009] Optionally, the control CPU executes pumped storage unit mode switching and power commands according to scheduling instructions and pumped storage unit status, and feeds back the final status of the pumped storage unit after execution to the scheduling CPU to achieve closed-loop control, including: The control CPU receives the scheduling instructions from the scheduling CPU and executes corresponding control according to the status of the pumped storage unit. If the scheduling instruction is a mode switch, the switch is completed according to the start-stop sequence. If the scheduling instruction is a fine-tuning, the set power output / absorption is achieved through the scheduler or converter control. During execution, the control CPU feeds back the final state, power and frequency of the pumped storage unit to the scheduling CPU via the bus to achieve closed-loop control.

[0010] Optionally, the bus adopts a publish / subscribe model and employs priority arbitration and real-time interruption mechanisms in the event of an emergency, including: The prediction CPU publishes new prediction results on the bus, and the scheduling CPU and control CPU subscribe to relevant topics to obtain updates asynchronously; If the frequency of the wind-solar microgrid deviates from the frequency deviation threshold, the highest priority emergency processing will be handled by the control CPU through priority arbitration / non-destructive arbitration. The bus will automatically allow high-priority instructions to preempt bus bandwidth, while low-priority messages will give way and wait. The emergency frequency adjustment strategy is executed in coordination between the scheduling CPU and the control CPU. Once the emergency event is handled, the normal process will be restored. If the CPU is predicted to acquire an urgent event between two cycles, the scheduling CPU will recalculate based on the new data. If a bus communication failure occurs, the system enters a safe mode, without scheduling or issuing mode switching commands.

[0011] Based on the same inventive concept, this application also provides a bus-type multi-CPU dynamic collaborative scheduling device for hydro-wind-solar microgrids, the improvement of which is that the device includes: The prediction unit is used to collect the operation data of the hydro-wind-solar microgrid in each cycle of the prediction CPU, and to obtain and publish the prediction results of wind and solar power and load based on the operation data using an ultra-short-term prediction algorithm. The scheduling unit is used by the scheduling CPU to obtain the start-up and shutdown time, optimal operating mode and power command of the pumped storage unit through rolling optimization and mathematical programming based on the prediction results and the status of the pumped storage unit, and then use these as scheduling commands. The control unit is used by the control CPU to execute the pumped storage unit mode conversion and power command according to the scheduling instructions and the status of the pumped storage unit, and to feed back the final status of the pumped storage unit after execution to the scheduling CPU to realize closed-loop control. The predictive CPU, scheduling CPU, and control CPU are connected via a bus, which adopts a publish / subscribe mode and employs priority arbitration and real-time interrupt mechanisms in the event of an emergency.

[0012] Optionally, the scheduling CPU uses rolling optimization and mathematical programming to obtain the start-up and shutdown times, optimal operating modes, and power commands of the pumped storage units based on the prediction results and the status of the pumped storage units, which are then used as scheduling commands. The scheduling CPU calculates the net power surplus based on the subscribed forecast results, and reads the current available regulation margin of the pumped storage and the time required for start-up and shutdown. Based on the net power surplus, it is determined whether there will be a sustained positive surplus or negative deficit in the future, and the pumped storage is planned to switch to charging or releasing mode. Based on the adjustment margin, start-up and shutdown time, and pumped storage unit status, the optimal operating mode and power command of the pumped storage unit are obtained through rolling optimization and mathematical programming. The start-up and shutdown times of pumped storage units are dynamically determined based on the forecast results.

[0013] Optionally, the step of determining whether a sustained positive surplus or negative deficit will occur in the future based on the net power surplus, and planning for pumped storage to enter charging or releasing mode, includes: If the net power surplus exceeds the pumped storage absorption threshold, the pumped storage is planned to enter the charging mode. The starting time of the surplus is selected as the target, and the pumping command is issued in advance of the pumping start time. If the unit is currently in the power generation mode, the shutdown switching time must be included in advance. If the net power surplus is negative for a preset period, the pumped storage is scheduled to enter the energy release mode. The power generation start command is issued in advance of the pumped storage start time from the moment the negative value appears. If the unit is pumping water at this time, the shutdown and reversal time needs to be included in advance. If the predicted fluctuation range is within the allowable range or the pumping capacity is limited and cannot fully cover the fluctuation, then the current state will be maintained.

[0014] Optionally, the step of obtaining the optimal operating mode and power command of the pumped storage unit through rolling optimization and mathematical programming based on the adjustment margin, start-up and shutdown time, and pumped storage unit status includes: The state of pumped storage units and their power output / absorption values ​​in each scheduling cycle are used as decision variables. The objective functions are to reduce the curtailment of new energy, reduce the number of unit start-ups and shutdowns, and reduce frequency fluctuations. The main constraints are power state constraints, power upper limit constraints, charge and discharge energy balance constraints, energy storage capacity upper and lower limit constraints, power change slope constraints, minimum start-up and shutdown duration constraints, system power balance constraints, and frequency support constraints. An objective function is established with the goals of minimizing wind and solar curtailment and minimizing frequency deviation. The minimum value of the objective function is solved under the main constraints to obtain the optimal operating mode and power command of the pumped storage unit for each time period.

[0015] Optionally, the step of dynamically deciding the start-up and shutdown time of the pumped storage unit based on the prediction results includes: When the wind and solar power is in net surplus and reaches the surplus power threshold for more than the first time, the pumping mode is activated. When the wind and solar power is in negative deficit and reaches the deficit power threshold for more than the second time, the power generation mode is activated. When the wind and solar power fluctuates between the surplus power threshold and the deficit power threshold, it remains unchanged. By scheduling the CPU to send scheduling instructions a preparation time in advance, and combining this with a sliding time window to dynamically adjust the start-up and shutdown time of the pumped storage unit using a second-level rolling mechanism.

[0016] Optionally, the control CPU executes pumped storage unit mode switching and power commands according to scheduling instructions and pumped storage unit status, and feeds back the final status of the pumped storage unit after execution to the scheduling CPU to achieve closed-loop control, including: The control CPU receives the scheduling instructions from the scheduling CPU and executes corresponding control according to the status of the pumped storage unit. If the scheduling instruction is a mode switch, the switch is completed according to the start-stop sequence. If the scheduling instruction is a fine-tuning, the set power output / absorption is achieved through the scheduler or converter control. During execution, the control CPU feeds back the final state, power and frequency of the pumped storage unit to the scheduling CPU via the bus to achieve closed-loop control.

[0017] Optionally, the bus adopts a publish / subscribe model and employs priority arbitration and real-time interruption mechanisms in the event of an emergency, including: The prediction CPU publishes new prediction results on the bus, and the scheduling CPU and control CPU subscribe to relevant topics to obtain updates asynchronously; If the frequency of the wind-solar microgrid deviates from the frequency deviation threshold, the highest priority emergency processing will be handled by the control CPU through priority arbitration / non-destructive arbitration. The bus will automatically allow high-priority instructions to preempt bus bandwidth, while low-priority messages will give way and wait. The emergency frequency adjustment strategy is executed in coordination between the scheduling CPU and the control CPU. Once the emergency event is handled, the normal process will be restored. If the CPU is predicted to acquire an urgent event between two cycles, the scheduling CPU will recalculate based on the new data. If a bus communication failure occurs, the system enters a safe mode, without scheduling or issuing mode switching commands.

[0018] Furthermore, this application also provides a computing device, comprising: at least one processor and a memory; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, a fault alarm method for a heat exchange station controller as described above is implemented.

[0019] In another aspect, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements a fault alarm method for a heat exchange station controller as described above.

[0020] Compared with the prior art, this application has the following advantages: This application discloses a bus-based multi-CPU dynamic collaborative scheduling method and device for hydro-wind-solar microgrids, comprising: using a prediction CPU to collect operating data of the hydro-wind-solar microgrid in each cycle, and using an ultra-short-term prediction algorithm to obtain and publish prediction results of wind and solar power and load based on the operating data; using a scheduling CPU to obtain the start-up and shutdown time, optimal operating mode, and power command of the pumped storage unit as scheduling instructions through rolling optimization and mathematical programming based on the prediction results and the status of the pumped storage unit; and using a control CPU to execute the switching of the pumped storage unit mode and the power command according to the scheduling instructions and the status of the pumped storage unit, and feeding back the final status of the pumped storage unit after execution to the scheduling CPU to achieve closed-loop control. This application utilizes a multi-CPU parallel architecture to deeply couple the ultra-short-term forecasting of wind and solar power with the optimized scheduling of pumped storage units, achieving a second-level rolling forecast-scheduling-control closed loop. The scheduling CPU pre-quantifies the start-up and shutdown timings of the units and makes rolling corrections, enabling the pumped storage units to complete mode switching before the arrival of renewable energy output fluctuations, maximizing the absorption of excess energy, compensating for power gaps, significantly reducing the curtailment rate, and effectively suppressing frequency and voltage fluctuations. This application employs a bus-based publish / subscribe mechanism, combined with message priority arbitration and redundant links, a watchdog timer, and hot backup. This ensures millisecond-level synchronization and delivery of prediction, scheduling, and control commands, and also allows for rapid switching to a preset degradation mode in the event of communication interruption or module failure, maintaining stable system operation. The modular multi-CPU architecture not only improves computing power and real-time performance but also facilitates on-demand expansion or upgrades (such as adding battery energy storage units or replacing prediction modules with higher computing power), thus meeting the engineering application requirements of high performance, high reliability, and ease of maintenance.

[0021] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 The flowchart of the implementation of a bus-based multi-CPU dynamic collaborative scheduling method for microgrids based on hydropower, wind power, and solar power provided in this application is shown. Figure 2 This application illustrates a collaborative scheduling architecture for a microgrid energy management system that includes pumped storage power stations, wind farms, and photovoltaic fields. Figure 3 The diagram shows the system configuration of a bus-type multi-CPU dynamic collaborative scheduling device for hydro, wind and solar microgrids provided in this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Example 1 This application provides a bus-based multi-CPU dynamic cooperative scheduling method for microgrids based on hydropower, wind power, and solar power, such as... Figure 1 ,include: The system uses a predictive CPU to collect operational data from the hydro-wind-solar microgrid in each cycle, and then uses an ultra-short-term prediction algorithm to obtain and publish the predicted results of wind and solar power and load based on the operational data. The scheduling CPU uses rolling optimization and mathematical programming to obtain the start-up and shutdown times, optimal operating modes, and power commands of the pumped storage units based on the prediction results and the status of the pumped storage units, which are then used as scheduling commands. The control CPU executes the pumped storage unit mode conversion and power command according to the scheduling instructions and the status of the pumped storage unit, and feeds back the final status of the pumped storage unit after execution to the scheduling CPU to realize closed-loop control. The predictive CPU, scheduling CPU, and control CPU are connected via a bus, which adopts a publish / subscribe mode and employs priority arbitration and real-time interrupt mechanisms in the event of an emergency.

[0026] This embodiment targets a microgrid energy management system that includes pumped-storage hydroelectric power stations, wind farms, and photovoltaic power stations. It is particularly suitable for off-grid operation or industrial park microgrids to smooth out fluctuations in renewable energy output through fine-grained energy storage scheduling. In this microgrid, wind and photovoltaic power generation are intermittent and fluctuating, while pumped-storage hydroelectric power stations, as large-scale adjustable energy storage units, can pump water for energy storage when there is a power surplus and generate electricity to supplement power when there is a power shortage.

[0027] This embodiment aims to provide a bus-based multi-CPU collaborative control dynamic energy storage scheduling system architecture to optimize the coordinated operation of pumped storage units with wind and photovoltaic power generation. This allows surplus or shortage of wind and solar power to be absorbed or supplied by pumped storage devices in real time, thereby improving the utilization rate of renewable energy in microgrids, reducing wind and solar power curtailment, and enhancing the stability of microgrids.

[0028] like Figure 2 Each CPU has a clearly defined role: the pumped storage control CPU directly interfaces with the pumped storage unit equipment, responsible for real-time control of unit start-up, shutdown, and power regulation; the AI ​​prediction CPU collects environmental and load data, runs short-term power prediction models based on wind speed and solar radiation, and implements load prediction algorithms; the scheduling CPU integrates information from all modules to calculate the system's current safety margin and optimize pumped storage start-up and shutdown times. This multi-CPU architecture achieves physical isolation and parallel operation of functional modules: prediction calculations and control execution are independent of each other and do not block each other, thus significantly improving system response speed and reliability. Each CPU exchanges information, such as current pumped storage capacity, water level, wind and solar power output prediction curves, and load levels, through shared memory or bus protocols, achieving a consistent understanding of the overall status.

[0029] The system bus employs a multi-master mechanism, supporting bus arbitration and real-time interrupts, thus ensuring that different CPU units can simultaneously access and update shared data without conflict. Especially during the transmission of critical control commands, the bus can arbitrate based on priority, ensuring that high-priority messages, such as emergency frequency modulation commands, are prioritized for bus scheduling and transmission. The overall architecture guarantees efficient, real-time, and reliable information flow between pumped-storage units and wind and solar power generation devices, laying the foundation for second-level collaborative control.

[0030] CPU Unit Configuration: In this embodiment, each functional module is deployed on an independent embedded processing unit, capable of meeting their respective real-time and computing power requirements. For example, the pumped storage control CPU can utilize a high-real-time industrial controller or microcontroller, such as the Texas Instruments TMS320 series digital signal processor (DSP) or ARM Cortex-M microcontroller, to quickly execute tasks such as unit speed regulation and switching control. This control CPU typically runs a real-time operating system or firmware, with cycle times reaching milliseconds or even faster, to match the needs of power electronics and unit mechanical control. The AI ​​prediction CPU can employ a more powerful processor, such as an ARM Cortex-A series SoC with a neural network acceleration unit, or an x86 architecture industrial PC / single-board computer, supplemented by a GPU coprocessor when necessary, to run complex algorithm models such as wind speed prediction, illumination prediction, and load prediction. The scheduling decision CPU can use a real-time processor similar to the control CPU, or be integrated with the prediction CPU on the same hardware platform (with dedicated cores partitioned on a multi-core processor), specifically for executing scheduling optimization algorithms and bus communication management. This hardware allocation allows each module to perform its specific function: the control unit reliably executes real-time control, the prediction unit fully utilizes computing power for AI calculations, and the scheduling unit makes real-time decisions between the two. To ensure the feasibility of the solution, common industrial PCs, programmable logic controllers (PLCs), or embedded control boards can be used to build the aforementioned CPU unit. For example, a PLC system with a multi-core processor can be used to distribute different tasks to different cores or modules for execution.

[0031] Bus communication and shared memory: CPUs are interconnected via a high-speed industrial bus, enabling real-time data sharing and command transmission. In practice, the appropriate bus type can be selected based on the application scenario. Controller Area Network Bus (CAN Bus): In distributed deployments, the CAN bus can serve as the communication backbone between CPU units. CAN is known for its broadcast messaging mechanism and priority arbitration. Its non-destructive bit arbitration ensures that the highest priority message on the bus is transmitted first without conflicting with lower priority messages. For example, frequency / voltage anomaly alarms or pumped storage emergency shutdown commands can be set as the highest priority message ID. If multiple CPU units send data simultaneously, the bus will automatically allow high-priority commands to preempt bus bandwidth, while lower-priority messages will wait. The CAN bus also boasts high reliability, employing differential signaling to enhance anti-interference capabilities, and achieving speeds exceeding 1 Mbps, sufficient for handling the control commands and status data transmission required for second-level scheduling. In the experimental verification system of this solution, the Controller Area Network (CAN) has been successfully used to issue power commands and collect status data between the Energy Management System (EMS) controller and distributed units. For example, the EMS uses CAN to send start / stop commands and power setpoints for pumped storage units, and receives feedback from each unit regarding voltage, current, power, and operating status.

[0032] High-speed Ethernet buses (such as EtherCAT): For systems requiring higher data refresh rates and synchronization accuracy, industrial Ethernet buses such as EtherCAT can be selected. EtherCAT has a bandwidth of 100Mbps, supports distributed clock synchronization, and can achieve data transmission and control synchronization of multiple nodes in the sub-millisecond range. For example, EtherCAT can be used to connect wind power and photovoltaic inverter control units and scheduling CPUs, reducing the cycle of power command issuance and measurement value reporting to the millisecond level, which helps to more precisely smooth power fluctuations. EtherCAT also supports flexible expansion of bus topology and master-slave communication, suitable for multiple CPUs concentrated in the same chassis or rack, achieving high-speed interconnection through a switching bus. Compared with traditional fieldbuses, EtherCAT has stronger determinism and minimal jitter, making it very suitable for real-time control.

[0033] High-speed backplane bus / shared memory: When multiple CPUs are integrated into the same device (such as a control cabinet or a single-board computer), a board-level high-speed bus or shared memory mechanism can be used for data exchange. For example, a PCI-Express (PCIe) bus or a dedicated backplane bus (such as the backplane bus of some PLC systems) can be used to connect multiple CPUs. These buses have bandwidths in the Gbps range; for example, the backplane bus rate of a certain industrial control system reaches 5Gbps. With the help of a high-speed backplane, critical information can be mapped to the shared memory area, allowing each CPU to read and write directly, achieving microsecond-level data synchronization. The shared memory uses a dual-port RAM or bus main memory arbitration mechanism to ensure consistency of parallel access. When the scheduling CPU calculates a new power setting, it can be quickly written to the shared memory, and the control CPU can read the updated value almost in real time; similarly, the wind and solar power output trend curve predicted by the CPU can also be published to the shared memory for the scheduling module to use. This memory-level data exchange latency is extremely low; for example, a multi-CPU system achieved an access performance of only 0.15 microseconds to write one word of data. It is important to note that the use of shared memory must be combined with locking mechanisms or bus arbitration to prevent concurrent write conflicts. The bus control logic will ensure that only one CPU occupies the shared resources at any given time, thereby avoiding data inconsistency.

[0034] In summary, this embodiment preferably uses a high-speed, real-time, and reliable bus architecture to connect multiple CPUs. For small- to medium-scale microgrid control, the CAN bus is a reliable choice due to its practicality and priority characteristics; for higher performance requirements, it can be upgraded to real-time Ethernet such as EtherCAT; on centralized hardware platforms, shared memory / backplane bus provides the fastest communication method. All of the above buses support interrupt notification mechanisms and bus error checking functions. When a critical event occurs (such as a predicted data update or an emergency fault signal), the bus will immediately notify the relevant CPUs via interrupts or event messages, greatly reducing information transmission latency. Multiple CPUs form a tightly coupled collaborative control network through the bus, ensuring that each module can cooperate synchronously and efficiently during second-level scheduling.

[0035] Pumped Storage Control Module: The pumped storage control CPU runs a state control algorithm for the pumped storage unit, the core of which is intelligent switching control between three operating modes: pumping, generating, and standby. This module directly interfaces with the unit's electrical and hydraulic actuators, such as the pump motor drive, electric valve and guide vane opening adjustment, and grid connection switch. The control CPU periodically collects real-time signals such as unit speed, power, water level / pressure, and guide vane opening, and performs closed-loop control based on dispatch instructions and the current status.

[0036] The software implementation of the control CPU includes start-stop sequence control, state machine management, and safety logic. For example, when a "start pumping" command is received, the control CPU determines whether the unit is currently in a switchable state (e.g., the unit is idle and the water level permits), and then executes the sequence of starting the pumps (closing the gate, adjusting the speed to the rated speed, etc.). When it is necessary to shut down or switch to power generation mode, the CPU reduces power, closes relevant valves, and switches the synchronizer according to preset logic to complete the mode transition. The control module also continuously monitors the unit's operating limits (e.g., maximum / minimum water level, unit temperature, etc.), and corrects or alarms if the scheduling command may cause exceeding the limits. Through a dedicated control CPU, the unit's start-up, shutdown, and power adjustment can be performed under hard real-time conditions, with a typical control cycle of 20ms or less, ensuring rapid support for frequency and voltage. A high real-time microcontroller or digital signal processor (DSP), such as the TI C2000 series DSP or ARM Cortex-M microcontroller, is preferably used to handle the state control and interface driving of the pumped storage unit. The CPU is directly connected to the actuators of the pumped-storage unit (such as the start-stop control of the pumps and turbines, and the speed regulation of the motors), enabling the switching logic control of the three operating modes: pumping (storage), power generation (release), and standby, as well as the rapid execution of power regulation commands. Since pumped-storage units have a significant impact on the grid frequency and voltage, the control CPU must have millisecond-level response capabilities to ensure immediate output adjustment upon receiving scheduling commands.

[0037] Wind, Solar, and Load Forecasting Module: The AI ​​prediction CPU handles the forecasting of uncertainties in the microgrid. The software integrates algorithms for wind speed forecasting, solar intensity forecasting, photovoltaic power output models, and load forecasting. Utilizing on-site sensors and historical data, the forecasting module periodically runs machine learning or physical models: for example, based on recent wind speed measurements and weather forecasts, it outputs a predicted curve for wind power output in the next few seconds to minutes; based on solar intensity sensors and time and weather conditions, it predicts short-term power changes in the photovoltaic array; and it combines historical load curves and current conditions to predict second-level load fluctuation trends, etc.

[0038] To improve prediction accuracy, the prediction CPU can employ advanced methods such as neural networks, time series analysis, or digital twin models to learn and update environmental data in real time. Specifically, considering the need for very short and precise predictions for second-level scheduling, the prediction CPU runs an ultra-short-term prediction algorithm every short interval (e.g., every second or every few seconds) and publishes the latest results to the system bus. Using a publish / subscribe software architecture, the prediction CPU broadcasts key events (such as "sudden increase in wind power forecast" or "sudden load drop forecast") as messages. Other CPUs (such as the scheduling CPU) receive these events as subscribers and immediately begin response calculations. This event-driven approach avoids repeated polling and waiting by the scheduling module, reducing latency and achieving module decoupling.

[0039] For example, when a wind speed sensor detects a sudden increase in wind speed, the prediction CPU can instantly adjust the wind power forecast upwards and issue an event message stating "Wind power output increases by X MW (predicted to start in 5 seconds)". The bus then transmits this event to the scheduling CPU for timely plan adjustments. Processors with AI acceleration capabilities are preferred, such as ARM Cortex-A series processors with GPUs or NPUs, or embedded industrial computers (IPCs). The prediction CPU runs wind speed forecasting, solar irradiance forecasting models, and load forecasting algorithms in real time. It acquires environmental and load information from sensors and historical data, using machine learning or physical models to generate real-time wind and solar power forecasts for the next few seconds to minutes, as well as load demand forecasts. For example, it continuously forecasts the trends in wind power generation and solar output for the next few minutes every second, providing ultra-short-term forecast support for scheduling decisions.

[0040] Scheduling Decision Module: The scheduling CPU is the brain of the entire system, running a dynamic energy storage scheduling algorithm to determine in real time when and at what power the pumped-storage units charge (pump) or discharge (generate) power. Its software logic takes data from the prediction module and status information from the control module as input, and performs comprehensive optimization calculations. The scheduling algorithm first calculates the current system's safety margin, i.e., the upward and downward regulation capacity that pumped-storage can provide: for example, calculating the maximum pumpable power and duration based on the remaining reservoir capacity and the unit's rated power, and calculating the maximum generating output and duration based on the water level and unit operating conditions. The scheduling CPU compares the safety margin with the predicted wind and solar power surplus / shortage to determine the next operational strategy.

[0041] If forecasts indicate that wind and solar power generation will have surplus capacity and the load will not be fully utilized in the near future (e.g., within tens of seconds), the dispatch module determines that pumped storage should be used to absorb this surplus energy. Conversely, if a situation of insufficient power generation and load exceeding wind and solar output is predicted, the pumped storage units are considered to switch to power generation mode to fill the gap. The key lies in optimizing the calculation of start-up and shutdown times: the dispatch algorithm uses rolling optimization to find the optimal time for pumped storage units to start or stop. For example, the forecast curve is used to estimate when significant surplus wind and solar power will appear, and the pumped storage units are triggered to start a few seconds in advance so that they enter pumping mode just as the surplus appears; or the pumping power is reduced tens of seconds before the predicted decline in wind and solar output to prepare for switching to power generation mode, so that the units have reached rated output when the load gap appears. This lead time is calculated by the algorithm based on the start-up and shutdown response time of the units and the slope of power change, minimizing the actual response lag as much as possible.

[0042] The scheduling module can be implemented using heuristic algorithms or mathematical optimization methods, such as short-cycle rolling optimization based on model predictive control (MPC), resolving the scheduling plan for the next half minute every second, or rapid decision-making using simple rules and threshold methods. Once a new scheduling decision is made, the scheduling CPU sends instructions to the pumped storage control CPU via the bus (e.g., "plan to enter pumping mode in t + 10 seconds, target power P = 5MW"), and updates the power setpoint curves for the future period in shared memory for reference by other modules. A high-performance industrial control computer or PLC (Programmable Logic Controller) can be used as the processor for the scheduling unit, such as an ARM Cortex-A multi-core processor or an x86 architecture industrial control computer. The scheduling CPU or control CPU is responsible for integrating information provided by each module (current pumped storage status, predicted wind and solar power output, and load curves, etc.), running optimization algorithms to calculate the system's dynamic scheduling strategy, including a safe power regulation margin and the optimal start-up and shutdown times for the pumped storage units. The scheduling CPU can read prediction data and device status through shared memory or high-speed bus, and on this basis, solve optimization problems or execute pre-set rules and strategies to generate a charging and discharging energy storage plan for the next few seconds.

[0043] Communication and Coordination Mechanisms: The entire software system employs a combination of event-driven and periodic computation. The prediction CPU is primarily event-driven, immediately notifying users when significant changes occur; it can also periodically broadcast key prediction data (e.g., updating the output prediction curve for each second of the next minute every second). The scheduling CPU can be triggered by events for immediate computation, and also performs rolling optimization at fixed intervals (e.g., once per second) to ensure continuous optimization of control accuracy even when predictions do not change drastically. The control CPU constantly listens for bus instructions and arbitrates based on its real-time status: upon receiving a new scheduling command, if there are no higher-priority tasks currently running, it executes immediately; if an urgent control task is in progress, it can decide whether to interrupt the current task or execute the scheduling command later based on the instruction priority.

[0044] To achieve this flexible scheduling, the software task queues of each CPU are designed with priority queues and real-time interrupt mechanisms. For example, on the control CPU, tasks such as unit protection and primary frequency regulation are given the highest priority, power adjustment commands given by the scheduler are given the second highest priority, and normal state updates are given low priority. When an emergency command appears on the bus (such as the "immediately increase pumped storage output" command issued when the microgrid frequency deviates significantly), the control CPU captures the command through an interrupt service routine and processes it with priority, ensuring that critical control actions are not delayed. In this embodiment, the software modules are clearly divided and loosely coupled through publish / subscribe communication, allowing each module to operate independently and complement each other, thus improving the robustness and scalability of the system.

[0045] The CPUs mentioned above are connected via a high-speed communication bus, enabling real-time data sharing and rapid instruction transmission. The choice of bus type can be determined based on the application scale and real-time requirements. In a centralized architecture, the PCI-Express (PCIe) backplane bus or other high-speed board-level buses can be used as the data channel between CPUs, providing high-speed, low-latency memory read / write and interrupt mechanism support. When all CPUs are integrated on the same control chassis or board, the PCIe bus allows them to share the same physical memory address space, achieving communication similar to shared memory, ensuring that simultaneous access and updates of critical scheduling information by multiple CPUs do not conflict.

[0046] In distributed architectures or applications requiring electrical isolation, industrial fieldbuses or industrial Ethernet, such as CAN bus or EtherCAT bus, can be used as communication media. CAN bus offers advantages such as strong anti-interference capabilities and simple, reliable connections, making it ideal for device communication in noisy environments. For example, in microgrid environments with numerous power electronic devices, CAN's reliability helps ensure the stable transmission of critical commands. In one design of this embodiment, the control modules are interconnected via a high-speed CAN-FD bus, utilizing its message priority arbitration mechanism to ensure that important control messages are sent first (the CAN protocol accesses the bus via message ID arbitration; the smaller the ID value, the higher the priority).

[0047] In another preferred embodiment, the EtherCAT industrial Ethernet bus is used to construct the real-time communication network. EtherCAT is known for its high bandwidth and distributed clock synchronization mechanism, capable of meeting sub-millisecond data refresh cycles and ensuring strict real-time synchronization control of multi-CPU systems. EtherCAT is also an open industrial Ethernet protocol; the master station cyclically sends data via standard Ethernet frames, achieving high-speed communication between master and slave nodes. In this system, the scheduling CPU can act as the EtherCAT master station, while the pumped control CPU and prediction CPU act as slave stations. Each control cycle, the master station issues scheduling command frames and collects the status of each slave station, achieving synchronization and coordination.

[0048] The bus architecture design ensures that multiple CPUs can access and update shared data simultaneously without conflicts. On one hand, by defining shared memory regions or common data structures (such as current energy storage status, real-time wind and solar power output values, and forecast curves) and ensuring atomic reads and writes through a bus arbitration mechanism, each module can read and update data in parallel. On the other hand, the bus supports real-time interrupts and arbitration, allowing emergency control commands to preempt bus transmission. For example, when a severe deviation in grid frequency is detected requiring immediate command of the pumped-storage unit, the control CPU can trigger a high-priority interrupt, quickly sending commands to the pumped-storage CPU via the bus, while non-critical data transmission from the scheduling CPU will be temporarily deferred.

[0049] The bus communication protocol preferably adopts a publish / subscribe model, where the prediction CPU publishes new prediction results on the bus, and the scheduling CPU and control CPU subscribe to relevant topics to obtain updates asynchronously. This loosely coupled communication allows information transfer between modules to be triggered in an event-driven manner, reducing direct dependencies between modules and improving system scalability and reliability. To ensure that the system's hardware and software implementation meets real-time and reliability requirements, this embodiment uses a real-time operating system (RTOS) or real-time kernel patch in the software. For example, Linux or VXWorks systems with real-time patches run on the prediction CPU and scheduling CPU, and lightweight real-time kernels such as TI-RTOS or FreeRTOS run on the control CPU to ensure deterministic task scheduling. In terms of hardware communication interfaces, each CPU is configured with redundant communication links and a watchdog timer: the watchdog is used to detect whether each module is stuck and timed out, and the redundant links (e.g., dual-bus redundancy) are used to switch to the backup bus to continue communication when the main communication bus fails, improving the system's fault tolerance.

[0050] To coordinate the aforementioned functional modules and achieve second-level energy storage collaborative scheduling, this embodiment designs a fine-grained scheduling algorithm. This algorithm uses a basic cycle time of one second and also allows for event-driven asynchronous triggering. Its main process is described below: Data Acquisition and Forecast Update (Start of Each Period): Collect key data at the current time t: Wind power output P w (t), Photovoltaic output P s Wind power prediction CPUs (P) generate wind power prediction sequences for the next ΔT (e.g., the next 30 seconds, one step per second), based on the latest data, including load power L(t) and pumped storage level / state of charge (SOC(t)). The CPU runs an ultra-short-term prediction model based on the latest data to obtain the wind power prediction sequence {P} for the next ΔT (e.g., the next 30 seconds, one step per second). w (t+1),...,P w (t+△T)}、Photovoltaic power prediction sequence {P s The load forecast sequence {L(t+1),...} is used. Updated forecast results are broadcast to the bus, triggering the scheduling calculation module. (Note: If there is no significant change within a certain period, the forecast CPU can reuse the results from the previous period, only requiring periodic correction.) Calculate the net load surplus: After the scheduling CPU subscribes to and obtains the latest forecast data, it calculates the net power surplus for each future second. (Positive values ​​indicate a power generation surplus, negative values ​​indicate a power generation deficit), where k represents the look-ahead step index in the prediction time domain: the kth discrete time step relative to the current time t. Each step length is the control / prediction period Δt (e.g., 1 second), therefore "t+k" actually refers to the time t + k·Δt; typically k=1…N (N is the number of prediction steps, e.g., N=30 represents the next 30 seconds), and k=0 can be used to represent the current time. Simultaneously, the current available regulation margin of the pumped storage system is read, including: maximum pumping power. (Depending on remaining reservoir capacity and pump power limit), maximum power generation capacity (Depending on the current water level and unit capacity), and the time required for unit start-up and shutdown. (The time required for the unit to reach rated power from standby startup or from operation shutdown).

[0051] Decision-making energy storage mode: based on The sequence determines whether a sustained positive surplus or negative deficit will appear within the future ΔT: If a period of net power surplus exists and its value exceeds the pumped storage absorption threshold, the planned pumped storage system will enter pumping (charging) mode to absorb the excess energy. The starting time t when the surplus occurs will be selected. e (For example The target is the moment when the pumping unit turns from negative to positive or exceeds a certain threshold, and the pumping start-up time T is also taken into account. start in advance Issue a pumping command. If the unit is currently in generating mode, a shutdown switchover time must be included in advance.

[0052] If forecasts indicate a prolonged power shortage, the plan is to switch from pumped storage to power generation (release) mode to supplement the supply. Similarly, the start time t of the shortage will be determined. d T in advance start The power generation start-up command is issued at the designated time. If the unit is pumping water at this time, the shutdown and reversal time must be included.

[0053] If the predicted fluctuation range is not large or the pumped storage capacity is limited and cannot fully cover the fluctuation, it may maintain the current state and stand by, only participating in frequency regulation by fine-tuning the power, without performing start-stop switching.

[0054] Dispatch decisions also take into account safety margins: ensuring that overcharging or discharging will not cause pumped-storage water levels to exceed upper or lower limits. For example, even with surplus power, it cannot exceed [the limit]. Pumping water; if the reservoir is near full capacity, the pumping power or duration is limited. Similarly, power generation is limited when the water level is low, with a certain reserve in place.

[0055] Optimize start / stop timing: For scenarios where start / stop is determined, further refine the specific timing and power curves. Use a rolling window to evaluate the plan every second, which can be achieved through the following methods: Calculate the benefits of starting early: If in Starting pumping at the right time ensures full-power absorption at the start of a surplus, maximizing the utilization of renewable energy. Starting too early increases losses from pumping under no-load conditions, while starting too late may cause missed opportunities to capitalize on the surplus. Therefore, the startup time should be chosen so that the pumping power curve covers the peak of the predicted surplus curve as much as possible.

[0056] Calculating the duration: Based on the predicted surplus and remaining reservoir capacity, determine the number of seconds for pumping to continue, avoiding excessive pumping and overflow. Similarly, for power generation, determine the duration and power output based on the gap size and water volume.

[0057] If the prediction curve has multiple peaks and troughs, the optimization algorithm needs to decide whether to frequently start and stop the generator to handle each fluctuation, or to smooth out the fluctuations over a period of time after startup. This involves a trade-off between generator start-up and shutdown losses and response speed. A start-up and shutdown penalty threshold can be set to avoid excessively frequent start-up and shutdown operations. When fluctuations are frequent but not large in magnitude, it is preferable to adjust the pumped storage power level rather than completely shutting down the generator to reduce mechanical shock.

[0058] Optimization algorithms can utilize simple rules or mathematical programming. For example, an optimization function can be established with the objectives of minimizing wind and solar power curtailment and minimizing frequency deviation, solving for the optimal start-up and shutdown sequence in each cycle. Due to the small scale of the problem (which operates on a second-level scale), the optimal strategy for the next tens of seconds can be calculated in real time. Constraints such as minimum unit operating time and minimum downtime are also considered during the optimization process to prevent excessively frequent commands.

[0059] Issuing scheduling instructions: Once a new scheduling plan is calculated, the scheduling CPU immediately sends an instruction set to the control CPU via shared memory or the bus, including: the target mode (pumping or generating), the planned switching time, and the planned power value curve. For example: "Gradually reduce the pumped-storage unit from generating power to 0 within the next 5 seconds, and switch to pumping mode and increase the pumping power to 5MW at the 6th second." For gradual power adjustments, continuous instructions or direct power curve issuance can be used, and the control CPU executes closed-loop tracking control. Instructions are sent through a publish / subscribe model to ensure that the control CPU receives them in a timely manner. The scheduling CPU also broadcasts its plan to the prediction module and other monitoring units on the bus for global synchronization (e.g., the human-machine interface can obtain the scheduling plan for display).

[0060] Control Execution and Feedback: After receiving instructions, the pumped storage control CPU executes corresponding controls based on its internal state machine. For mode switching, the switching is completed according to the start-stop sequence; for power fine-tuning, the set power output / absorption is achieved through speed governor or converter control. During execution, the control CPU feeds back key states via the bus, such as "unit started, current power 2MW, frequency steadily increasing," for dispatch decision-making reference. Simultaneously, the dispatch CPU continuously monitors the deviation between feedback and prediction: if the unit fails to reach the target within the expected time (e.g., start-up delay), subsequent instructions are adjusted in the next cycle (e.g., slightly advancing the next start-stop time or temporarily increasing power generation compensation). The entire closed-loop process is continuously executed in every second-level cycle, correcting deviations in real time to ensure consistency between the dispatch plan and actual operation.

[0061] Asynchronous event handling: If a special event occurs during the above loop, the normal process will be immediately interrupted. Emergency Frequency Deviation: If the microgrid frequency deviates from the allowable range (e.g., above the rated 0.5Hz), regardless of the current stage, the control CPU triggers the highest priority emergency handling. The scheduling CPU and control CPU work together to execute emergency frequency regulation strategies, such as immediately ordering pumped-storage units to increase output or reduce pumping power to curb the frequency deviation, while temporarily suspending the original scheduling plan. The scheduling algorithm cycle resumes after the emergency event is handled or the frequency returns to normal.

[0062] Prediction Correction: If the AI ​​prediction module acquires new critical information between two cycles (e.g., a sudden weather change rendering the previous prediction invalid), the event it publishes will cause the scheduling CPU to immediately discard the old plan and recalculate based on the new data. At this point, the new instructions may override older instructions that have not yet been executed to ensure the plan matches the latest situation.

[0063] Communication Anomaly: If a momentary interruption or data inconsistency occurs in bus communication, the system will enter a safe mode (see the security mechanism section below), attempting to restore communication while maintaining stability or awaiting manual intervention. During this period, the scheduling algorithm may be paused or run solely based on locally cached data, but no new start / stop commands will be issued.

[0064] Through the above process, the system achieves dynamic collaborative scheduling at the second-level granularity—adjusting the energy storage output plan every second based on the latest information, enabling pumped storage units to detect and respond to rapid fluctuations in wind and solar power in advance. This significantly improves the accuracy compared to traditional scheduling with a cycle of minutes or hours, and can more effectively reduce the impact of random fluctuations in renewable energy output on the microgrid's frequency and voltage. Simultaneously, multi-CPU parallel processing ensures that each step is performed almost simultaneously: predictive calculations and optimized scheduling do not need to be completed serially, but rather run concurrently using independent processing units, greatly reducing the latency from detecting changes to executing a response.

[0065] The key to this system lies in introducing a fine-grained dynamic collaborative scheduling algorithm, refining the traditional hourly energy storage scheduling to a second-level or even smaller timescale. Through multi-CPU parallel processing capabilities, the collaborative efforts of each module significantly shorten the scheduling cycle, enabling real-time responses to rapid fluctuations in wind and solar power output. The following describes the second-level scheduling algorithm in conjunction with the operational flow of this embodiment: 1. Data Acquisition and Forecasting (Cycle: 1 second): Every second (or shorter, dynamically adjustable as needed), the AI ​​prediction CPU collects the latest environmental and load data from various sensors and data sources, including current wind speed, wind direction, solar intensity, solar panel temperature, and microgrid load power. It then runs a trained short-term power prediction model to generate wind power output forecasts, solar power output forecasts, and load power forecast curves for the next few seconds to minutes. These prediction results are immediately broadcast via the bus once calculated. For example, the prediction CPU packages new prediction data into a message and sends it to the bus shared memory, marking it with a prediction sequence number.

[0066] 2. Status Awareness and Data Synchronization: The pumped storage control CPU continuously monitors the operating status of the pumped storage units, including the current mode (pumping / generating / standby), unit power output or absorption value, remaining reservoir capacity or energy storage margin (State of Charge, SOC), unit start-up / shutdown status, and required preparation time, and writes this status data to the shared storage area in real time for the scheduling CPU to read. Simultaneously, the scheduling CPU also subscribes to forecast messages and pumped storage status information on the bus. When new forecast results are released or the unit status changes, the scheduling CPU promptly obtains the latest data through interrupts or polling, achieving awareness of the overall situation.

[0067] 3. Scheduling Decision Calculation (Parallel Trigger): Once the scheduling CPU receives a new wind and solar power output forecast or load data update event, it immediately starts the dynamic scheduling algorithm in parallel to calculate a new energy storage scheduling plan. The scheduling algorithm can be based on model predictive control (MPC) or pre-set start-stop optimization rules, comprehensively considering the current energy storage capacity, the changing trend of wind and solar power forecast curves, and load demand.

[0068] The steps for scheduling CPU computation include, for example: a. Power Surplus / Deficit Calculation: Calculate the real-time power balance difference. Where L(t) represents the system load power at the current moment (the electricity demand to be met / baseline absorption power, usually in MW). When off-grid, it equals the actual load within the microgrid (equivalent value including necessary losses); when connected to the grid, it can also be taken as "local load or planned absorption / export target to the main grid" according to the operation strategy. If ΔP>0, it indicates that there is a current or imminent renewable energy power surplus; if ΔP<0, it indicates that the power gap needs to be made up by energy storage or other sources.

[0069] in, This represents the predicted available wind power generation capacity for time period t. This represents the predicted value of available photovoltaic power generation for the time period t.

[0070] b. Adjustment margin assessment: Based on the current operating status and technical constraints of the pumped-storage unit, determine the available charging and discharging power margin. For example, how much surplus power is still available for pumping (charging) in pumping mode, how much energy storage capacity is still available for release in generation mode (SOC constraint), and the unit's start-up and shutdown times and power ramp-up rate limits, etc.

[0071] c. Strategy Optimization: Based on the predicted ΔP change curve over time, a fine-grained energy storage regulation plan is formulated. For example, if a large power surplus is predicted to persist for tens of seconds in the future, the scheduling CPU calculates that pumped-storage units should enter pumping mode in advance and gradually increase power to absorb excess energy; conversely, if a power shortage is predicted, the pumped-storage units are planned to start generating power in advance. This optimization process can employ a rolling time window, looking forward several seconds to minutes in each cycle to select the start-up and shutdown times and power curve schemes that minimize wind and solar power curtailment and frequency deviation. Due to the use of precise predictions, this algorithm can anticipate power imbalances several seconds to tens of seconds in advance and make adjustment decisions, thereby reducing the start-up and shutdown delays of pumped-storage units and improving response speed.

[0072] Scheduling decisions also avoid unnecessary frequent start-ups and shutdowns: for example, the algorithm sets up start-up and shutdown post-start strategies, which only trigger unit state switching when the predicted power surplus or deficit will continue to exceed a certain time threshold, in order to prevent frequent start-ups and shutdowns of units due to temporary fluctuations.

[0073] d. Output Plan Release: The scheduling CPU writes the calculated new pumping station operation plan (such as the sequence of charging and discharging power instructions per second for the next 30 seconds, or, for simplicity, the start / stop instructions and target power in the near future) to shared memory or sends it to the control CPU and relevant modules via the bus. The plan includes actions to be performed in the near future (e.g., "start pumping at t+5 seconds, increasing power to 5MW" or "maintain standby") and the next action predicted a few seconds later. In this way, each module can obtain a consistent and up-to-date scheduling plan within its respective cycle.

[0074] The scheduling CPU is configured with a linear optimization scheduling model to minimize the curtailment of wind and solar renewable energy and suppress system frequency fluctuations. The model's decision variables include the operating state (pumping, generating, or standby) of the pumped storage units in each scheduling cycle, their power output / absorption values, and the relevant energy storage status. The model's decision variables, objective function, and constraints are given below, along with a table defining the symbols of the main variables to facilitate engineering replication.

[0075] Decision variables: Let the scheduling period be a discrete-time scale. (Period length is in seconds). In each period t, the pumped-storage unit may be in one of three states: pumping (charging), generating (releasing), or standby (idle). Therefore, the following decision variables are introduced: Continuous variables Pumped storage unit's pumping (charging) power during time period t (i.e., the power absorbed from the grid for pumped storage, in MW).

[0076] Continuous variables : Power generation (discharge) of pumped storage units during time period t (i.e., power output to the grid, in MW).

[0077] binary variables : Pumping status of the pumped storage unit during time period t (binary variable). This indicates that the unit was pumping water to recharge its power during that period. This indicates a non-pumping state.

[0078] binary variables : Power generation status of the pumped storage unit during time period t (binary variable). This indicates that the generating unit was generating electricity and releasing energy during that period. This indicates a non-power generation state.

[0079] Continuous variable E(t): Energy storage state of pumped storage unit in time period t (reservoir water volume / SOC) (equivalent to "state of charge", SOC), which can be represented by reservoir water volume or equivalent stored energy, reflecting the amount of energy (water volume) stored up to time t.

[0080] Among the variables mentioned above, binary variables and The power variable determines the unit's operating mode. , And the stored energy E(t) is a continuous decision quantity.

[0081] Objective function: The optimization model aims to reduce renewable energy curtailment and frequency fluctuations. The objective function F (expressed in single-objective weighted linear form) can be defined as follows:

[0082] in, This represents the amount of wind and solar power that is wasted during time period t (i.e., the power that is not utilized and is discarded). This refers to the change or rate of change in the output power of a pumped storage unit (which can be expressed as the absolute value of the power difference between adjacent time periods, i.e.) This indicates the range of variation in the unit's output. This can represent the number of start-stop transitions that occur within a time period t (e.g., using binary variable changes to represent start-stop events). F represents the objective function. This is the output power weighting coefficient. This refers to the start-stop frequency weighting coefficient, used to balance the trade-off between minimizing power curtailment and the frequency of start-stop operations and drastic power fluctuations. By appropriately selecting the weights, it is possible to maximize the absorption of wind and solar energy while minimizing the start-stop frequency and large power fluctuations of pumped-storage units, thereby extending equipment life and improving economic efficiency. In special cases, the objective function may also take only a single objective (e.g., Take zero, and purely minimize the total amount of wind and solar power curtailment. Alternatively, it can be solved separately as a multi-objective optimization and then weighted. The above objective function reflects the scheduling intention of minimizing renewable energy curtailment and smoothing frequency fluctuations (achieved by reducing power surges and frequent start-stops).

[0083] Constraints: The optimization model must satisfy a series of constraints on the operation of the pumped storage unit and system, including charging and discharging physical constraints and grid operation constraints. The main constraints are as follows: Power state constraint (pumping / generating mutual exclusion constraint): Pumped storage units cannot pump water and generate electricity simultaneously within the same time period. This can be represented by binary variables as a mutual exclusion constraint.

[0084] This ensures that the unit is in at most one active operating mode (pumping or generating) at any given time. When both variables are 0, it indicates that the unit is in standby mode.

[0085] Power cap constraint: Both pumping and power generation power of the unit are limited by the rated capacity of the equipment. Power should be zero when the unit is not in the corresponding state. This can be expressed in the model using the following linear inequality:

[0086]

[0087] In the formula This is the upper limit of the rated pumping power of the pumping unit. This represents the upper limit of the unit's rated generating capacity. Based on the above constraints, when... Automatic limit during (non-pumping state) ;when (Non-power generation state) time limit Meanwhile, when the state is 1, the power does not exceed the rated value.

[0088] Charge / discharge energy balance constraint (SOC update constraint): The energy state E(t) of the pumped storage unit at time t is affected by the state at the previous time and the charging / discharging power during this period, and can be expressed by the energy conservation relationship:

[0089] in Pumping efficiency (the efficiency coefficient of converting electrical energy into water potential energy during pumping). This refers to the power generation efficiency (the efficiency of hydroelectric power generation). This is the scheduling cycle time length (seconds). This constraint means that if water is pumped during this period, the stored energy (SOC) will increase; if power is generated, the water volume will decrease. The model assumes that the initial stored energy E(0) is known at t=0.

[0090] Energy storage capacity upper and lower limits constraints (SOC upper and lower limits constraints): The reservoir capacity of pumped storage units is limited, and the energy storage state must be limited within a safe range:

[0091] In the formula E max and E min These are the upper and lower limits of the reservoir's volume (equivalent energy storage), respectively. This constraint ensures that the water level remains within the permissible safe range, preventing overflow or bottoming out.

[0092] Power change slope constraint (climbing rate constraint): Due to electromechanical inertia and equipment characteristics, the power of pumped storage units cannot change drastically without constraints in adjacent moments; therefore, the maximum climbing rate needs to be limited. Upward and downward climbing constraints can be applied to both pumping and power generation, for example: Power generation ramping constraints:

[0093] Pumping power ramp-up constraint:

[0094] in, These represent the maximum power gradient (MW per second) under the unit's power generation mode. These are the upper and lower limits for the rate of power change in pumping mode. These constraints ensure smooth changes in unit output, do not exceed the technically permissible slope, and reduce frequency surges.

[0095] Minimum Start-Stop Duration Constraints: To prevent equipment damage caused by frequent start-stop cycles, once a pumped-storage unit starts pumping or generating electricity, it needs to maintain this state for a minimum duration of several seconds. Similarly, a minimum interval must be met after a shutdown before restarting. These constraints are typically implemented through logical relationships between 0-1 variables, for example: if... For the change from 0 to 1 (the unit starts generating electricity at time t), it is required that after t... Maintain the power generation state unchanged within seconds, that is:

[0096] Similarly, the minimum duration of the pumping mode This constraint must also be met. This avoids excessively frequent pumping / power generation switching, improving system stability and economy.

[0097] System power balance constraints: Within each dispatch cycle, wind power, solar power, load demand, and pumped storage output must maintain a power balance. Considering that when renewable energy output exceeds load, a surplus occurs (requiring pumping or curtailment), and when insufficient, energy storage generation is needed to supplement it, the following balance equation is established:

[0098] in, It represents the system net load / baseline absorption capacity (usually in megawatts) that needs to be met by "wind power + photovoltaic power + pumped storage power" at time t. These are the predicted wind power and solar power values ​​for time period t, respectively. This represents the load power demand during that period (or the baseline power that the system can absorb). This refers to the variable for wind and solar power curtailment, used to absorb unusable wind and solar surplus.

[0099] Pgen(t) refers to the power output of the pumped storage unit at time t, that is, the power output (usually in megawatts) of the unit to the microgrid when the unit is in the "power release / power generation" mode. The value is positive when the unit is generating power; it is zero when the unit is in standby or pumping; it is mutually exclusive with Ppump(t) and is subject to constraints such as rated upper limit, ramp rate and minimum duration in the model.

[0100] The above constraints indicate that when wind and solar power output exceeds load demand and energy storage pumped hydro power cannot be fully absorbed, the excess power is counted as... Abandoned; when Fengguang's contribution is insufficient, through Water release for power generation compensates for this, balancing both sides of the equation (ideally). At this point, setting it to 0 results in zero frequency deviation. This power balance constraint ensures energy conservation and provides a foundation for frequency stability: through optimization... The goal is to keep the sum of power generation on the left side as close as possible to the sum of power consumption on the right side, thereby minimizing the system frequency deviation.

[0101] Frequency support constraints: To provide primary frequency regulation support for the system, pumped storage units need to reserve a certain margin for both upward and downward regulation to ensure that the units can respond rapidly within seconds when the grid frequency deviation exceeds the limit. This can be achieved by restricting the units from exhausting their full capacity during normal dispatch. For example, requiring the units to reserve capacity during generation mode. Increased power is reserved for use in pumping mode. The load is reduced for standby. The constraint is as follows:

[0102] in, This indicates an upward adjustment of reserve capacity (the amount of power "up" reserved, in megawatts). In the unit's generating mode, the output is intentionally not pushed to the rated upper limit, leaving this margin for rapid power increase within seconds, used for primary frequency regulation (to push up the frequency when it drops). This indicates a reduction in reserve (power "downward" reserve, in megawatts). In the unit's pumping mode, the pumping power is intentionally not increased to the upper limit, leaving this margin to quickly reduce the pumping load within seconds (equivalent to "increasing power supply" to the system), which is used for primary frequency regulation (rapid recovery when the frequency drops or exceeds the limit).

[0103] At the same time, when the power grid frequency deviation Exceeding the threshold In such cases, dispatching should ensure sufficient backup power. The system quickly pulls the load back within limits (e.g., by increasing output through energy storage or reducing load shedding). This constraint, combined with the rapid response characteristics of energy storage, ensures that the scheduling plan has flexibility to meet real-time frequency control requirements. In other words, the scheduling model must consider frequency safety constraints during optimization, and cannot be arranged in a way that leaves no margin to correct frequency deviations in the event of power disturbances.

[0104] The main constraints are listed above. In addition, other constraints can be added as needed, such as time constraints for pumped-storage unit switching modes (a transition time of several seconds required to switch from pumping to power generation), robust constraints for wind and solar power output prediction errors, etc. By solving for the minimum value of the objective function (1) under all constraints, the optimal operating mode and power command of the pumped-storage units for each time period can be obtained, thereby maximizing wind and solar power absorption and maintaining stable power balance at the second-level scheduling level.

[0105] In the second-level real-time control process, the scheduling CPU needs to dynamically decide the start-up and shutdown timing of the pumped storage unit based on the wind and solar power prediction curves. Let... This represents the predicted net surplus / deficit of wind and solar power at time t (e.g., Its positive or negative sign reflects the surplus or deficiency of renewable energy output relative to the load. The scheduling algorithm analyzes... The changing trend is used to determine when to start or stop the pumped-storage unit: based on Start-stop criteria: When a sustained surplus in wind and solar power output is predicted ( When the output of wind and solar power is predicted to be >0 and exceeds a certain threshold, it indicates that there is excess renewable energy that cannot be absorbed by the load. In this case, the pumping mode should be activated as soon as possible to absorb the excess energy and avoid wind and solar curtailment. Conversely, when a wind and solar power deficit is predicted ( If the power generation is less than 0, meaning the power generation is insufficient to meet the load and the frequency may drop, then the pumped-storage units should be activated to supplement the power grid. To avoid misjudging temporary fluctuations, a duration criterion is often used: for example, setting a "when..." "If the power exceeds 0 continuously and reaches the surplus power threshold for more than x seconds, then the pumping unit will be started." Similarly, "When..." "If the deficit remains negative for more than y seconds and exceeds the threshold continuously, then the unit will be triggered to generate electricity." This rule can filter out high-frequency noise and guide the start-up and shutdown decisions to be made only when the trend is clear. Furthermore, in When the unit is oscillating near zero, it should remain in standby mode and not switch frequently to avoid losses caused by frequent start-stop cycles.

[0106] Consideration of the lead time Δt for start / stop commands: Since pumped-storage units require a preparation time Δt (including pump or turbine startup, grid synchronization, etc., typically ranging from several seconds to several minutes depending on unit performance) to switch from standby to actual output, the dispatching system needs to issue start / stop commands in advance to compensate for this time lag. Specifically, when the forecast curve indicates a need to start at time t+Δt (e.g., a sustained surplus is expected from t+Δt, requiring pumping), the dispatching CPU will send a "start pumping" command at time t, ensuring the unit reaches the required power around t+Δt. Similarly, for shutdowns or mode switching, a command to stop the current mode needs to be issued Δt in advance so the unit can timely switch to standby or the new mode. By considering start / stop preparation time and anticipating wind / solar power changes, the algorithm can pre-start units before severe wind / solar power fluctuations, achieving "advanced start-up and delayed shutdown," effectively supporting system balance.

[0107] Sliding forecast window and second-level rolling adjustment: This start-stop decision-making process is not completed all at once, but is continuously performed in conjunction with the rolling forecast of the sliding time window. The scheduling CPU refreshes the wind and solar power output forecast curve at fixed short intervals (e.g., every second or every few seconds), sliding the forecast window forward (e.g., always considering future T). pred (Prediction curve in seconds), and based on the latest The sequence reassesses start-up and shutdown conditions. This second-level rolling judgment mechanism ensures that scheduling can correct deviations in a timely manner if actual conditions deviate from previous forecasts. For example, if a previous forecast predicted a surplus lasting 10 seconds and pumping was initiated, but a new forecast indicates the surplus will end soon, the rolling judgment will issue an early shutdown command or a reduction in pumping power to prevent over-pumping. Conversely, if the predicted deficit worsens and prolongs, rolling optimization will correspondingly extend the power generation operating time. By sliding windows and feedback corrections on each short cycle, start-up and shutdown timings can be dynamically adjusted, always making decisions based on the latest wind and solar output and load trends, achieving second-level adaptive scheduling.

[0108] Comparison of Rule-Based Logic and Optimization Decision-Making: The above start-up and shutdown strategies can be implemented in two ways: logical judgment based on preset rules, or decision variables embedded in an optimization model. The rule-based logic method uses manually set thresholds and durations, such as the aforementioned "power surplus triggers pumping after x seconds." If the condition is met, start-up and shutdown are executed; otherwise, no action is taken. This method is simple to implement, computationally fast, and easily applied in second-level control. However, its disadvantages are that the threshold and timing are difficult to accommodate all situations, potentially leading to untimely or excessively frequent start-ups and shutdowns (when the threshold is too low), and failure to achieve global optimum (e.g., not considering trends over longer time periods). The optimization method directly incorporates the unit's start-up and shutdown status as 0-1 decision variables into the scheduling model, determining whether to pump, generate electricity, or standby for each time period by solving for the overall optimum [as used in the above model]. [This indicates that] the optimization model comprehensively considers the trend of wind and solar power output within the prediction window and the constraints, automatically selecting the optimal start and stop times to optimize the objective function (such as minimizing power curtailment or minimizing costs). For example, when there is only a short-term surplus followed by a power shortage, the optimization may choose not to start pumping to avoid an invalid cycle of subsequent shutdown; conversely, when a long-term surplus is expected, the model will start pumping several cycles in advance. pump Set to 1 (start pumping). This method, based on mixed-integer linear programming (MILP), can obtain a better scheduling scheme and avoid the limitations of manual rules. However, its disadvantage is the high computational complexity, which places high demands on hardware and algorithm efficiency when solving problems on a second-level rolling basis. In practical applications, the two methods have their trade-offs: one can first use an optimization model to formulate start-up and shutdown plans offline or on a rolling basis, and then use rule logic to fine-tune and respond quickly at the real-time level; or in special emergency situations, predefined rules can be used to prioritize start-up and shutdown to ensure safety. Regardless of the method, the key is to do not omit the key mechanisms required for start-up and shutdown judgment based on detailed modeling: considering both the evolution trend of ΔP(t) and the unit response time, as well as following the physical constraints of the unit and frequency safety requirements, so that engineers can reproduce and implement the algorithm logic of second-level dynamic energy storage scheduling.

[0109] 4. Execution and Control: The pumped-storage control CPU reads the latest instructions and plans issued by the scheduling CPU and immediately executes the corresponding control operations based on the most recent instructions. For example, if the scheduling plan requires the pumped-storage unit to enter "pumping" mode and output a certain pumping power, the control CPU checks the unit's status (such as whether the pumps are running dry and whether the unit meets the start-up conditions), and then sends control signals to the unit's excitation system, guide vanes, or valve actuators to start the pumping operation and gradually adjust the power to the specified value. Simultaneously, if the plan requires the unit to start or stop at a future time, the control CPU will prepare in advance (such as closing the circuit a few seconds in advance). Throughout the process, the control CPU executes instructions according to priority: urgent instructions (such as frequency control commands) have the highest priority, followed by start / stop instructions in the scheduling plan, and finally power adjustment instructions. By implementing a priority queue within the control CPU, even when multiple instructions are received, the most critical instructions are ensured to be executed first.

[0110] 5. Real-time Monitoring and Feedback: While controlling the CPU to execute unit regulation, the system continuously monitors key operating indicators of the microgrid, including frequency, bus voltage, and power flow. When pumped-storage units operate, these indicators should tend to improve (e.g., frequency deviation decreases). The scheduling CPU and prediction CPU adjust their respective model states based on feedback—for example, the prediction CPU corrects prediction model deviations based on actual power changes, and the scheduling CPU judges whether the previous decisions have achieved the expected results based on the new SOC and frequency changes. If a significant discrepancy is found between actual and prediction, the system will quickly revise the scheduling scheme in the next cycle.

[0111] 6. Iterative Execution: The above process is executed every second (or less), continuously optimizing and executing scheduling strategies to achieve continuous closed-loop control. It is important to emphasize that due to the use of multi-CPU parallel processing, prediction calculation, optimization decision-making, and control execution are actually parallel and overlapping within a single cycle: the prediction CPU may still be calculating predictions for a more distant period, the scheduling CPU is already using existing predictions for current optimization, and the control CPU is executing instructions issued in the previous cycle. Information is exchanged via the bus within the same cycle, realizing pipelined operations, thus fully utilizing every second time window to complete complex calculations and control, enabling the system to have second-level or even sub-second-level response capabilities.

[0112] Through the aforementioned algorithms and architecture, this embodiment significantly improves the granularity of energy storage scheduling to the second level, thereby greatly enhancing the response speed and control effect to rapid power fluctuations. For example, when solar irradiance fluctuates drastically due to rapid changes in cloud cover, the system can react within seconds, coordinating pumped storage units to smooth out output changes. Compared to traditional systems that schedule by minutes or hours, frequency and voltage fluctuations are more effectively suppressed, and the stability of the microgrid is enhanced.

[0113] The following typical operating conditions illustrate the response process of the system in this application when wind speed changes, deepening the understanding of second-level collaborative scheduling.

[0114] Operating assumptions: The microgrid is currently operating off-grid, with a relatively stable load of approximately 50MW. The pumped-storage turbine is currently in power generation mode, outputting 10MW to support the load, and the turbine still has a significant amount of water remaining to continue generating electricity for some time. The wind farm is currently outputting approximately 40MW, and solar power output is close to zero (at night). At this point, the total power generation of the microgrid = 40MW wind power + 10MW pumped-storage = 50MW, perfectly matching the load with no frequency deviation. The pumped-storage turbine reservoir has a large remaining capacity (equivalent to approximately 20MW·h of pumped water).

[0115] Scenario 1: Sudden increase in wind speed leading to increased wind power output: A strong gust of wind suddenly passes through the wind farm at night, significantly increasing the wind speed within a short period. The total output power of the wind turbines rises from 40MW to a peak of 60MW within tens of seconds (exceeding the load demand by 10MW). The system modules respond according to the following steps: t=0s (Initial State): The prediction CPU continuously monitors the wind speed sensor and detects an upward trend in wind speed. Based on the latest data, the wind power prediction model is run, indicating that wind power will climb to approximately 60MW in about 30 seconds and may continue at high wind speeds for several minutes. The prediction CPU immediately publishes a "Wind Power Output Expected to Rise" event via the bus, along with a power prediction curve for the next 60 seconds. This shows that around t=20s, wind power output will exceed the 50MW load level, with a maximum surplus of approximately 10MW.

[0116] At t=1s: The scheduling CPU receives the predicted event and, after reading the curve, identifies a significant power generation surplus (up to approximately 10MW) between t=20s and t=60s. Simultaneously, the scheduling module obtains the current pumped storage status from the control CPU: the unit is in power generation mode at 10MW, the reservoir has sufficient capacity for pumping, and the unit needs to shut down and switch to pumping mode, estimated to take approximately 15 seconds. Based on this information, the scheduling algorithm decides to utilize pumped storage to absorb excess wind power to avoid wind curtailment. Since the predicted surplus will begin to appear after 20 seconds, and the unit needs 15 seconds to start pumping, the optimal strategy is to issue the pumping start command at t=5s: this allows the unit sufficient time to reduce its power generation mode to zero and switch, and it is expected to enter pumping mode around t≈20s to simultaneously absorb excess power.

[0117] At t=5s: The dispatch CPU issues a dispatch command to the pumped storage control CPU: "Gradually reduce the pumped storage power generation from 10MW to 0 within 5 seconds, and switch to pumping mode at t=15s, with a target pumping power of 10MW." The control CPU immediately executes: by adjusting the turbine guide vane opening, the power generation is linearly reduced, decreasing by approximately 2MW per second, until the unit's power generation drops to 0 at t=10s, at which point the unit enters no-load synchronous standby mode. Subsequently, according to the start-stop logic, the control CPU disconnects the generator from the grid at t=15s, switches the gate valve to activate the pumping pump, and starts the motor to drive the pump. Since it had already stabilized during no-load operation, the unit reaches pumping operation within a few seconds.

[0118] At t=20s: As predicted, the actual wind power output rose to approximately 55MW, exceeding the load by 5MW. The pumped-storage unit, as planned, entered pumping mode at t≈18s and ramped up to 5MW of pumping power, absorbing this excess 5MW from the microgrid. This rebalanced the system's power generation and load around 50MW, keeping the frequency stable without increasing. In the next few seconds, the wind power continued to rise to 60MW, and the excess 10MW was also absorbed by the pumped-storage unit's increased pumping power (the unit's pumping power was capped at 10MW). Therefore, around t=30s, although the wind power output reached its peak of 60MW, the microgrid did not waste energy or overclock: 50MW load + 10MW pumping power consumption = 60MW of power generation, exactly equal.

[0119] t=30s~60s: After a period of sustained wind, the wind begins to weaken. The prediction CPU, detecting the decrease in wind speed, issues a new event indicating "wind power output will decrease." Based on this, the dispatch CPU adjusts its plan: when wind power decreases to a level that matches the load (approximately t=60s), pumped storage should cease pumping to prevent insufficient supply. Based on this prediction, the dispatch CPU pre-issues the instruction: "Stop pumping at t=60s." The control CPU then shuts down the pumps at t=60s as instructed, and the units return to standby synchronization. Because wind power falls below the load, the microgrid experiences a power deficit of approximately 5MW (load 50MW, wind power only 45MW). After checking the pumped storage level, the dispatch CPU decides to restart pumped storage generation to fill the gap, and simultaneously instructs the units to switch to generation mode at t=60s. The control CPU successfully completes the mode switch within a few seconds, restoring a certain amount of power generation. Finally, at t≈75s, the unit was connected to the grid again in 5MW power generation mode, working together with the 45MW wind power to supply 50MW load, achieving a new power balance.

[0120] The above process demonstrates that this system achieves real-time coordination between pumped-storage units and wind power output throughout the entire process of wind speed increase, maintenance, and decrease: when wind is abundant, it quickly switches to pumping to absorb surplus power; when wind decreases, it promptly switches to power generation to supplement the deficit. From the perspective of the microgrid, the load receives stable power supply without significant fluctuations in frequency and voltage, and the wind farm is not forced to curtail wind power due to surplus power, resulting in a significant improvement in clean energy utilization. This effect is precisely due to the high-speed linkage and second-level optimization of the multi-CPU architecture—the prediction module anticipates in advance, the scheduling module plans ahead, and the control module executes on time, shortening the effective response time of the pumped-storage units by more than 10% compared to traditional methods (because traditional scheduling often lags behind the occurrence of fluctuations, while this solution acts in advance); the wind power curtailment rate is reduced by about 3%, essentially utilizing all short-term surplus power for pumped-storage instead of wasting it. Under this operating condition, the microgrid frequency remains consistently near its rated value, and the voltage deviation is also small, without drastic fluctuations. Practice has proven that the proposed solution greatly enhances the microgrid's adaptability to fluctuations in wind and solar power output.

[0121] It is worth mentioning that, in the above process, the coordination of each CPU is completed efficiently through the bus: the broadcast of wind power forecast events is delivered to the scheduling CPU within milliseconds; scheduling commands are quickly issued to the control CPU through the bus, and bus arbitration ensures that commands take effect immediately in emergency situations. The entire link operates without human intervention, in a fully automatic closed-loop manner, realizing true real-time intelligent energy scheduling.

[0122] To ensure the system operates stably in complex environments over a long period, this embodiment incorporates comprehensive security and fault-tolerance mechanisms at both the hardware and software levels: 1. Communication Reliability and Fault Monitoring: For bus communication, redundancy and monitoring strategies ensure reliable data transmission. For the CAN bus, dual-channel redundancy can be configured. When the primary bus fails, the backup bus automatically takes over, or the CAN's own error checking mechanism detects bus error frames and automatically retransmits them. The EtherCAT bus supports a ring redundancy topology; when communication is interrupted at one node, the signal can be transmitted in reverse to maintain network integrity. A heartbeat signal mechanism is implemented at the bus protocol layer: each CPU sends a heartbeat message periodically, and other modules monitor whether they receive the heartbeat on time to determine if the module is online and working normally. Once a CPU's heartbeat times out (e.g., predicting CPU failure or communication loss), the scheduling CPU and control CPU will enter a preset degraded mode. For example, if a module is predicted to lose connection, the scheduling CPU will activate an emergency scheduling mode: no longer relying on predicted data, but instead using the average value or trend inference over a recent period to make conservative scheduling decisions, while issuing an alarm and waiting for maintenance personnel to check. In this degraded mode, a new pumped-storage switching operation may not be initiated; instead, the current state is maintained or only simple adjustments are made based on local frequency / voltage feedback, prioritizing system stability. When communication is restored, the system smoothly switches back to normal mode.

[0123] 2. Module Fault Tolerance: Fault tolerance strategies are also designed for each CPU itself. If the prediction CPU malfunctions (e.g., program crashes or outputs obviously unreasonable data), its published data can be identified by the scheduling CPU's verification mechanism. The scheduling module sets a confidence range for the predicted values. If it receives data exceeding the physically possible range (e.g., wind power prediction is greater than the installed capacity of wind turbines), it will automatically ignore the data and mark the prediction module as faulty. At this time, the scheduling module also switches to emergency mode, performing "firefighting" control only based on real-time measurements (e.g., adjusting pumped storage power through frequency feedback). For scheduling CPU failures, the pumped storage control CPU has an independent local automatic control strategy: for example, a basic frequency / voltage control curve is pre-configured (similar to droop control). When upper-level instructions are lost, the pumped storage units can automatically increase or decrease output according to the microgrid frequency deviation to temporarily maintain steady state. This ensures that even if the central dispatch fails, the microgrid still has a certain degree of self-stabilization capability and will not go out of control. The control CPU itself generally uses a highly reliable microcontroller, and there is also an internal watchdog timer to monitor program operation. If the control CPU freezes, the hardware watchdog will reset the CPU within hundreds of milliseconds and put the unit into a safe mode (usually by disconnecting the unit's output, i.e., stopping the pumping unit or putting it into standby mode) to prevent accidents caused by erroneous control.

[0124] 3. Safety Protection and Priority Control: The system is designed with safety as the top priority, followed by performance optimization. For example, even if the scheduling algorithm requires a rapid switch in unit status, it must be reviewed by the control CPU's protection logic: the switch is only executed when all unit conditions are met and safety interlocks are closed; otherwise, it is delayed or refused, and an anomaly is reported. Furthermore, in emergency situations, the system sacrifices optimization plans for safety: when frequency / voltage exceeds limits, the scheduling module suspends all optimization calculations and immediately coordinates with the control module to adjust the pumped storage output to a safe range before gradually resuming normal scheduling. This design ensures that regardless of the algorithm's calculations, it will not violate the power system's safety constraints.

[0125] 4. Network and Information Security: Since multiple CPUs exchange data via the bus, this embodiment also considers information security measures to prevent harm from malicious commands or communication errors. Bus communication employs message verification and authentication; each command contains a CRC checksum, and execution only occurs after verification by the receiving end. For the publish / subscribe mechanism, topic authorization is used; only authenticated modules can publish critical control topics, preventing unknown devices from injecting commands. The system also deploys intrusion detection policies. If abnormal frequency or content of communication data occurs (e.g., a large number of abnormal commands in a short period), a security mode will be triggered to notify maintenance personnel for investigation, preventing network attacks.

[0126] 5. Redundancy and Recovery: Redundancy can be configured for critical hardware. For example, dual predictive CPUs can be set up for hot backup, with mutual verification to improve prediction reliability; important sensors (anemometers, power meters, etc.) also use dual redundant inputs, so that the other set of data can be used in case of an anomaly in either one. On the software side, data snapshots are taken periodically to save the current scheduling status and prediction trends. In the event of a system reset or fault restart, operation can be quickly restored from the snapshot, continuing the previous control plan without causing major disturbances due to the restart.

[0127] Through the aforementioned multi-layered security mechanisms, this system possesses self-monitoring, self-adaptive, and fault-tolerant operation capabilities even in complex real-world environments (communication noise, equipment failure, extreme operating conditions). In fact, distributed control structures have a natural reliability advantage over centralized control: when one module fails, other modules can still operate independently, and local problems will not immediately lead to global paralysis. For example, even if the predictive function is temporarily lost, control and scheduling can still maintain operation based on real-time feedback; if the scheduling module fails, the control module can still operate for a period of time according to simple rules. Because of this robust design, the system in this embodiment can maintain the energy balance of the microgrid stably over a long period and protect equipment and grid safety under various abnormal conditions.

[0128] Based on the above design, the bus-type multi-CPU dynamic energy storage collaborative scheduling system of this application will achieve better performance than traditional solutions in microgrids containing pumped hydro storage, wind power, and solar power. Faster response speed: Through parallel processing and second-level optimization, the start-up, shutdown, and power adjustment of pumped storage units can be planned in advance, shortening the actual response time. Simulation analysis shows that compared to the traditional mode that relies on manual or minute-level EMS decision-making, the average response delay from standby to output can be reduced by more than 10%. This means that frequency deviations or power gaps can be corrected earlier, ensuring power quality.

[0129] Higher renewable energy utilization: Due to fine-grained dispatching, short-term fluctuations in wind and solar power are mitigated locally as much as possible, eliminating the need for large margins or even curtailment due to forecast uncertainties, as was the case in the past. The curtailment rate of wind and solar power is expected to decrease; specific calculations show that the proportion of curtailed wind and solar power can be reduced by approximately 3 percentage points compared to existing dispatching strategies, improving the clean energy absorption capacity of microgrids. As the research points out, rolling corrections of planning deviations through second-level, small-scale dispatching help achieve dynamic economic dispatching and balance reliability.

[0130] More robust power quality control: Multi-CPU collaboration enables the system to suppress power fluctuations more promptly and effectively. Under typical operating conditions, the frequency fluctuation amplitude and voltage deviation of the microgrid are significantly reduced. Especially for off-grid microgrids, pumped storage can make reverse adjustments within the same second as wind and solar power changes, maintaining the frequency within a very small range above and below the rated value. This effect is equivalent to integrating the functions of traditional primary frequency regulation and secondary dispatch, continuously fine-tuning the power balance and improving the stability of microgrid operation.

[0131] Improved energy efficiency: Real-time optimization avoids unnecessary operation and redundant backup of energy storage devices. For example, by accurately predicting start-up and shutdown, pumped storage units avoid prolonged idling or frequent start-ups and shutdowns, significantly reducing unit losses and ineffective energy consumption. It is estimated that the overall energy efficiency ratio of the microgrid dispatch (the total power generation required to supply the same load) will improve, reducing unnecessary reserve power input. At the same time, the demand for reserve capacity is reduced, and the system can operate safely with a smaller margin, thereby saving operating costs.

[0132] Scalable modular design: The bus-based architecture makes the system easy to expand and upgrade. If new energy units (such as battery storage or diesel engines) are added in the future, only the corresponding control CPU needs to be added and connected to the bus to work together; if the prediction algorithm needs to be improved, the prediction CPU hardware or software can be upgraded without affecting other modules. This modular expansion capability is very convenient for microgrid evolution and the integration of new technologies. Those skilled in the art can select CPUs and buses with different performance levels according to the specific project needs to achieve the same collaborative control concept.

[0133] In summary, this embodiment details a second-level dynamic energy storage scheduling system that utilizes multi-CPU parallel collaboration and high-speed bus interaction to achieve coordinated pumped hydro storage and wind / solar power generation. This scheme organically combines control, prediction, and scheduling, maximizing renewable energy utilization and microgrid operating efficiency while ensuring stable and safe system operation. Those skilled in the art, by reading the above content and referring to the provided hardware / software configuration and algorithm description, can reproduce the embodiments of this application and apply them to the design of practical microgrid energy management systems. The scheme is not only applicable to off-grid or industrial park microgrids, but also has significant reference value in scenarios where a high proportion of renewable energy is integrated into the large power grid, and has positive implications for building new power systems.

[0134] Example 2 Based on the same inventive concept, this application also provides a bus-type multi-CPU dynamic collaborative scheduling device for hydro-wind-solar microgrids, such as... Figure 3 The device includes: The prediction unit is used to collect the operation data of the hydro-wind-solar microgrid in each cycle of the prediction CPU, and to obtain and publish the prediction results of wind and solar power and load based on the operation data using an ultra-short-term prediction algorithm. The scheduling unit is used by the scheduling CPU to obtain the start-up and shutdown time, optimal operating mode and power command of the pumped storage unit through rolling optimization and mathematical programming based on the prediction results and the status of the pumped storage unit, and then use these as scheduling commands. The control unit is used by the control CPU to execute the pumped storage unit mode conversion and power command according to the scheduling instructions and the status of the pumped storage unit, and to feed back the final status of the pumped storage unit after execution to the scheduling CPU to realize closed-loop control. The predictive CPU, scheduling CPU, and control CPU are connected via a bus, which adopts a publish / subscribe mode and employs priority arbitration and real-time interrupt mechanisms in the event of an emergency.

[0135] Optionally, the scheduling CPU uses rolling optimization and mathematical programming to obtain the start-up and shutdown times, optimal operating modes, and power commands of the pumped storage units based on the prediction results and the status of the pumped storage units, which are then used as scheduling commands. The scheduling CPU calculates the net power surplus based on the subscribed forecast results, and reads the current available regulation margin of the pumped storage and the time required for start-up and shutdown. Based on the net power surplus, it is determined whether there will be a sustained positive surplus or negative deficit in the future, and the pumped storage is planned to switch to charging or releasing mode. Based on the adjustment margin, start-up and shutdown time, and pumped storage unit status, the optimal operating mode and power command of the pumped storage unit are obtained through rolling optimization and mathematical programming. The start-up and shutdown times of pumped storage units are dynamically determined based on the forecast results.

[0136] Optionally, the step of determining whether a sustained positive surplus or negative deficit will occur in the future based on the net power surplus, and planning for pumped storage to enter charging or releasing mode, includes: If the net power surplus exceeds the pumped storage absorption threshold, the pumped storage is planned to enter the charging mode. The starting time of the surplus is selected as the target, and the pumping command is issued in advance of the pumping start time. If the unit is currently in the power generation mode, the shutdown switching time must be included in advance. If the net power surplus is negative for a long period of time, the pumped storage is planned to enter the energy release mode. The power generation start command will be issued in advance of the pumped storage start time from the moment the negative value appears. If the unit is pumping water at this time, the shutdown and reversal time needs to be included in advance. If the predicted fluctuation range is within the allowable range or the pumping capacity is limited and cannot fully cover the fluctuation, then the current state will be maintained.

[0137] Optionally, the step of obtaining the optimal operating mode and power command of the pumped storage unit through rolling optimization and mathematical programming based on the adjustment margin, start-up and shutdown time, and pumped storage unit status includes: The state of pumped storage units and their power output / absorption values ​​in each scheduling cycle are used as decision variables. The objective functions are to reduce the curtailment of new energy, reduce the number of unit start-ups and shutdowns, and reduce frequency fluctuations. The main constraints are power state constraints, power upper limit constraints, charge and discharge energy balance constraints, energy storage capacity upper and lower limit constraints, power change slope constraints, minimum start-up and shutdown duration constraints, system power balance constraints, and frequency support constraints. An objective function is established with the goals of minimizing wind and solar curtailment and minimizing frequency deviation. The minimum value of the objective function is solved under the main constraints to obtain the optimal operating mode and power command of the pumped storage unit for each time period.

[0138] Optionally, the step of dynamically deciding the start-up and shutdown time of the pumped storage unit based on the prediction results includes: When the wind and solar power is in net surplus and reaches the surplus power threshold for more than the first time, the pumping mode is activated. When the wind and solar power is in negative deficit and reaches the deficit power threshold for more than the second time, the power generation mode is activated. When the wind and solar power fluctuates between the surplus power threshold and the deficit power threshold, it remains unchanged. By scheduling the CPU to send scheduling instructions a preparation time in advance, and combining this with a sliding time window to dynamically adjust the start-up and shutdown time of the pumped storage unit using a second-level rolling mechanism.

[0139] Optionally, the control CPU executes pumped storage unit mode switching and power commands according to scheduling instructions and pumped storage unit status, and feeds back the final status of the pumped storage unit after execution to the scheduling CPU to achieve closed-loop control, including: The control CPU receives the scheduling instructions from the scheduling CPU and executes corresponding control according to the status of the pumped storage unit. If the scheduling instruction is a mode switch, the switch is completed according to the start-stop sequence. If the scheduling instruction is a fine-tuning, the set power output / absorption is achieved through the scheduler or converter control. During execution, the control CPU feeds back the final state, power and frequency of the pumped storage unit to the scheduling CPU via the bus to achieve closed-loop control.

[0140] Optionally, the bus adopts a publish / subscribe model and employs priority arbitration and real-time interruption mechanisms in the event of an emergency, including: The prediction CPU publishes new prediction results on the bus, and the scheduling CPU and control CPU subscribe to relevant topics to obtain updates asynchronously; If the frequency of the wind-solar microgrid deviates from the frequency deviation threshold, the highest priority emergency processing will be handled by the control CPU through priority arbitration / non-destructive arbitration. The bus will automatically allow high-priority instructions to preempt bus bandwidth, while low-priority messages will give way and wait. The emergency frequency adjustment strategy is executed in coordination between the scheduling CPU and the control CPU. Once the emergency event is handled, the normal process will be restored. If the CPU is predicted to acquire an urgent event between two cycles, the scheduling CPU will recalculate based on the new data. If a bus communication failure occurs, the system enters a safe mode, without scheduling or issuing mode switching commands.

[0141] Example 3 Based on the same inventive concept, the present invention also provides an electronic device. The electronic device of this embodiment includes at least one processor and at least one storage medium electrically connected to the processor, wherein the storage medium stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0142] Example 4 Based on the same inventive concept, the present invention also provides a storage medium storing instructions executable by at least one processor, the instructions being executed by at least one processor to enable at least one processor to perform the method described above.

[0143] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A bus-based multi-CPU dynamic collaborative scheduling method for hydro-wind-solar microgrids, characterized in that, include: The system uses a predictive CPU to collect operational data from the hydro-wind-solar microgrid in each cycle, and then uses an ultra-short-term prediction algorithm to obtain and publish the predicted results of wind and solar power and load based on the operational data. The scheduling CPU uses rolling optimization and mathematical programming to obtain the start-up and shutdown times, optimal operating modes, and power commands of the pumped storage units based on the prediction results and the status of the pumped storage units, which are then used as scheduling commands. The control CPU executes the pumped storage unit mode conversion and power command according to the scheduling instructions and the status of the pumped storage unit, and feeds back the final status of the pumped storage unit after execution to the scheduling CPU to realize closed-loop control. The predictive CPU, scheduling CPU, and control CPU are connected via a bus, which adopts a publish / subscribe mode and employs priority arbitration and real-time interrupt mechanisms in the event of an emergency.

2. The method as described in claim 1, characterized in that, The scheduling CPU uses rolling optimization and mathematical programming to obtain the start-up and shutdown times, optimal operating modes, and power commands of the pumped storage units based on prediction results and the status of the pumped storage units, which are then used as scheduling commands. The scheduling CPU calculates the net power surplus based on the subscribed forecast results, and reads the current available regulation margin of the pumped storage and the time required for start-up and shutdown. Based on the net power surplus, it is determined whether there will be a sustained positive surplus or negative deficit in the future, and the pumped storage is planned to switch to charging or releasing mode. Based on the adjustment margin, start-up and shutdown time, and pumped storage unit status, the optimal operating mode and power command of the pumped storage unit are obtained through rolling optimization and mathematical programming. The start-up and shutdown times of pumped storage units are dynamically determined based on the forecast results.

3. The method as described in claim 2, characterized in that, The step of determining whether a sustained positive surplus or negative deficit will occur in the future based on the net power surplus, and planning whether pumped storage will enter charging or releasing mode, includes: If the net power surplus exceeds the pumped storage absorption threshold, the pumped storage is planned to enter the charging mode. The starting time of the surplus is selected as the target, and the pumping command is issued in advance of the pumping start time. If the unit is currently in the power generation mode, the shutdown switching time must be included in advance. If the net power surplus is negative for a preset period, the pumped storage is scheduled to enter the energy release mode. The power generation start command is issued in advance of the pumped storage start time from the moment the negative value appears. If the unit is pumping water at this time, the shutdown and reversal time needs to be included in advance. If the predicted fluctuation range is within the allowable range or the pumping capacity is limited and cannot fully cover the fluctuation, then the current state will be maintained.

4. A method as described in claim 2, characterized in that, The optimal operating mode and power command of the pumped storage unit are obtained through rolling optimization and mathematical programming based on the adjustment margin, start-up and shutdown time, and pumped storage unit status, including: The state of pumped storage units and their power output / absorption values ​​in each scheduling cycle are used as decision variables. The objective functions are to reduce the curtailment of new energy, reduce the number of unit start-ups and shutdowns, and reduce frequency fluctuations. The main constraints are power state constraints, power upper limit constraints, charge and discharge energy balance constraints, energy storage capacity upper and lower limit constraints, power change slope constraints, minimum start-up and shutdown duration constraints, system power balance constraints, and frequency support constraints. An objective function is established with the goals of minimizing wind and solar curtailment and minimizing frequency deviation. The minimum value of the objective function is solved under the main constraints to obtain the optimal operating mode and power command of the pumped storage unit for each time period.

5. A method as described in claim 2, characterized in that, The dynamic decision-making process for starting and stopping pumped storage units based on prediction results includes: When the wind and solar power is in net surplus and reaches the surplus power threshold for more than the first time, the pumping mode is activated. When the wind and solar power is in negative deficit and reaches the deficit power threshold for more than the second time, the power generation mode is activated. When the wind and solar power fluctuates between the surplus power threshold and the deficit power threshold, it remains unchanged. By scheduling the CPU to send scheduling instructions a preparation time in advance, and combining this with a sliding time window to dynamically adjust the start-up and shutdown time of the pumped storage unit using a second-level rolling mechanism.

6. A method as described in claim 1, characterized in that, The control CPU executes pumped storage unit mode switching and power commands based on scheduling instructions and the pumped storage unit's status, and feeds back the final status of the pumped storage unit to the scheduling CPU after execution, thus achieving closed-loop control, including: The control CPU receives the scheduling instructions from the scheduling CPU and executes corresponding control according to the status of the pumped storage unit. If the scheduling instruction is a mode switch, the switch is completed according to the start-stop sequence. If the scheduling instruction is a fine-tuning, the set power output / absorption is achieved through the scheduler or converter control. During execution, the control CPU feeds back the final state, power and frequency of the pumped storage unit to the scheduling CPU via the bus to achieve closed-loop control.

7. A method as described in claim 1, characterized in that, The bus adopts a publish / subscribe model and employs priority arbitration and real-time interruption mechanisms in the event of an emergency, including: The prediction CPU publishes new prediction results on the bus, and the scheduling CPU and control CPU subscribe to relevant topics to obtain updates asynchronously; If the frequency of the wind-solar microgrid deviates from the frequency deviation threshold, the highest priority emergency processing will be handled by the control CPU through priority arbitration / non-destructive arbitration. The bus will automatically allow high-priority instructions to preempt bus bandwidth, while low-priority messages will give way and wait. The emergency frequency adjustment strategy is executed in coordination between the scheduling CPU and the control CPU. Once the emergency event is handled, the normal process will be restored. If the CPU is predicted to acquire an urgent event between two cycles, the scheduling CPU will recalculate based on the new data. If a bus communication failure occurs, the system enters a safe mode, without scheduling or issuing mode switching commands.

8. A bus-type multi-CPU dynamic collaborative scheduling device for hydropower, wind power, and solar power microgrids, characterized in that, The device includes: The prediction unit is used to collect the operation data of the hydro-wind-solar microgrid in each cycle of the prediction CPU, and to obtain and publish the prediction results of wind and solar power and load based on the operation data using an ultra-short-term prediction algorithm. The scheduling unit is used by the scheduling CPU to obtain the start-up and shutdown time, optimal operating mode and power command of the pumped storage unit through rolling optimization and mathematical programming based on the prediction results and the status of the pumped storage unit, and then use these as scheduling commands. The control unit is used by the control CPU to execute the pumped storage unit mode conversion and power command according to the scheduling instructions and the status of the pumped storage unit, and to feed back the final status of the pumped storage unit after execution to the scheduling CPU to realize closed-loop control. The predictive CPU, scheduling CPU, and control CPU are connected via a bus, which adopts a publish / subscribe mode and employs priority arbitration and real-time interrupt mechanisms in the event of an emergency.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the bus-based multi-CPU dynamic collaborative scheduling method for hydro, wind, and solar microgrids as described in any one of claims 1-7.

10. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the bus-based multi-CPU dynamic collaborative scheduling method for hydro-wind-solar microgrids as described in any one of claims 1-7.

Images