Water, wind and light combined power prediction and control method, system and medium

Abstract

The invention provides a water, wind and light combined power prediction and control method and system and a medium, and belongs to the technical field of water, wind and light field station control, and the method comprises the steps: scheduling a CPU module to collect real-time operation data, and issuing a prediction trigger instruction; the prediction processing CPU module receives the prediction trigger instruction, reads and preprocesses the real-time operation data, and executes reasoning calculation to obtain a prediction result; generating a scheduling instruction according to the prediction result and the real-time operation data; the control execution CPU module sends the generated dispatching instruction to the hydroelectric generating set through a bus, and feeds back real-time output data after the hydroelectric generating set executes dispatching; and taking the data as new real-time operation data, cyclically operating the steps, and entering the next period. According to the invention, a bus-type multi-CPU architecture is utilized, complex prediction and optimization calculation loads are shared to different modules for parallel processing, and unified message communication and task scheduling are combined, so that the intelligent decision-making capability is realized, and the harsh requirement of real-time control of an electric power field is met.

Description

Technical Field

[0001] This invention belongs to the field of water, wind and solar power station control technology, and particularly relates to a method, system and medium for predicting and controlling the combined power of water, wind and solar. Background Technology

[0002] With the significant increase in installed capacity of wind, solar, and hydropower, renewable energy is gradually becoming an important part of the power grid. However, wind and solar power generation are characterized by significant intermittency, volatility, and instability, posing a significant challenge to the frequency and voltage control of the power grid. To mitigate the power fluctuations of renewable energy, traditional power grids mainly rely on thermal power, hydropower, and energy storage equipment for peak shaving and frequency regulation. However, with the continuous increase in the proportion of new energy sources, relying solely on hydropower or energy storage for compensation is no longer sufficient to meet the demands for rapid response and refined dispatch.

[0003] Currently, the control systems of most new energy power plants still adopt a centralized or loosely distributed architecture: 1. Centralized PLC / DCS control—Data acquisition, prediction and scheduling calculations, and execution control are all completed on the same controller or a small number of processors. This approach has limited computing power, making it difficult to run complex power prediction algorithms and optimized scheduling models in parallel, resulting in slow control response and difficulty in tracking rapid fluctuations in wind and solar power.

[0004] 2. Distributed Multi-Module Architecture—This architecture deploys data acquisition, AI prediction, scheduling decisions, and execution control across multiple hardware units, interconnected locally via CAN bus, Modbus RTU, or Industrial Ethernet. While this architecture improves hardware redundancy and scalability, it often only breaks down modules at the functional level, lacking a unified communication protocol and real-time closed-loop mechanism, making it difficult to achieve millisecond-level predictions. decision making Execute linkage.

[0005] 3. Edge-cloud collaborative scheduling—Lightweight predictive control is performed on edge servers, while deep optimization scheduling is performed in the cloud. This relies on network conditions, and communication latency and uncertainty can affect local rapid response. Furthermore, the separation of edge and cloud functions makes it impossible to form a completely closed-loop adaptive adjustment.

[0006] In summary, existing technologies generally suffer from the following shortcomings: insufficient computing power and real-time performance, making it impossible to support parallel computation of complex prediction and optimization algorithms; separation between prediction and control, with a lack of unified timing and standardized communication leading to data asynchrony; high expansion and maintenance costs, requiring large-scale hardware and software modifications to integrate new algorithms or energy storage devices; and weak fault tolerance, with single-point failures often causing control interruptions.

[0007] Therefore, there is an urgent need for a technical solution that can realize the integrated closed loop of wind and solar power prediction and hydropower dispatch control in the same bus system through modular multi-CPU collaboration, so as to meet the requirements of high real-time performance, high reliability and scalability for large-scale new energy grid connection. Summary of the Invention

[0008] In view of the shortcomings of the existing technology, the present invention provides a method, system and medium for predicting and controlling the combined power of water, wind and solar power.

[0009] In a first aspect, the system includes a predictive processing CPU module, a control execution CPU module, and a scheduling CPU module; wherein each CPU module is interconnected and communicates on a bus through a bus communication mechanism. The prediction processing CPU module includes a wind power prediction processing CPU and a photovoltaic prediction processing CPU. It uses independent CPU cores to undertake the inference calculation tasks of wind power and photovoltaic power prediction algorithms, and is used to run artificial intelligence prediction models for wind power generation and photovoltaic power generation. The control execution CPU module includes a control CPU, which is used to execute the regulation and control strategies of the hydropower unit; wherein, the control CPU is used to adjust the output of the hydropower unit according to the scheduling instructions; The scheduling CPU module is the main control unit for bus communication, including a scheduling CPU, which is used for overall task scheduling and data acquisition. The scheduling CPU is used to trigger predictive calculations, collect data from each module, issue scheduling instructions, and coordinate real-time communication and task allocation between CPU modules.

[0010] Furthermore, the control CPU is specifically used for the start-up and shutdown control of the hydropower unit and the adjustment of its output. The bus adopts a backplane bus or an industrial Ethernet high-bandwidth, low-latency communication link, supporting plug-and-play access for each CPU board.

[0011] Furthermore, the CPU modules communicate and interconnect on the bus through a bus communication mechanism, specifically by using a standardized communication protocol and message format for communication interconnection; wherein the message format includes message type, source / target identifier, timestamp, and data payload.

[0012] Secondly, the methods based on the above system include, S1. At the beginning of each control cycle, the scheduling CPU module collects real-time operating data of wind farms, photovoltaic arrays and hydropower stations, and sends prediction trigger commands to the prediction processing CPU module. S2. After receiving the prediction trigger command, the prediction processing CPU module reads and preprocesses the real-time running data, and inputs the preprocessed real-time running data into the artificial intelligence prediction model to perform inference calculations and obtain the prediction results. S3. Based on the prediction results and real-time operation data, generate scheduling instructions through rule set control strategies and optimized scheduling models; S4. The control execution CPU module sends the generated scheduling instructions to the hydropower unit via the bus. After the hydropower unit executes the scheduling, it feeds back the real-time output data. S5. Use the feedback real-time output data as new real-time operating data, run S1-S4, and enter the next cycle.

[0013] Furthermore, the real-time operating data of the wind farm, photovoltaic array and hydropower station includes recent historical wind power output and photovoltaic power output data, relevant meteorological observations and forecasts, and grid load conditions.

[0014] Furthermore, the preprocessing of real-time running data includes filling in missing values, normalizing, and aligning timestamps on the real-time running data; The artificial intelligence prediction model includes a wind LSTM model and a photovoltaic LSTM model; wherein, both the wind LSTM model and the photovoltaic LSTM model adopt a three-layer stacked LSTM network structure, including three sequentially stacked LSTM hidden layers and a final output layer.

[0015] Furthermore, the rule set control strategy compares the prediction results with the system's target load or planned output demand to determine the required direction and basic magnitude of hydropower regulation for the current cycle. The optimized scheduling model uses the comparison results of the rule set control strategy as an initial reference. Under the condition of satisfying the constraints, it solves the optimal power output command of hydropower by constructing a linear programming model. The constraints include hydropower output limit constraints, hydraulic operating condition constraints, power change rate constraints, and system power balance constraints.

[0016] Furthermore, the method also includes the scheduling CPU module arranging the time flow of the method.

[0017] Furthermore, the method also includes the introduction of optimization measures, specifically including data buffering and pipelined processing, task prioritization and real-time scheduling, algorithm elastic degradation, parallel computing and load balancing, redundancy backup design and anomaly monitoring and fault handling.

[0018] Thirdly, a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described methods for predicting and controlling combined water, wind, and solar power.

[0019] Compared with the prior art, the present invention has the following advantages: This invention utilizes a multi-CPU module-based integrated method for hydropower-wind-solar power prediction and control. Through hardware topology and communication design, it achieves deep synergy between wind and solar power prediction and hydropower regulation and control. By employing a bus-based multi-CPU architecture, the complex prediction and optimization computational load is distributed to different modules for parallel processing. Combined with unified message communication and rigorous task scheduling, the system possesses both highly intelligent decision-making capabilities and meets the stringent requirements of real-time power field control.

[0020] Other features and advantages of the invention 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 invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims, and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 The diagram shows a system module schematic of a combined hydropower, wind power, and solar power prediction and control method according to the present invention.

[0023] Figure 2 A schematic flowchart of a method for predicting and controlling the combined power of water, wind, and solar energy according to the present invention is shown. Detailed Implementation

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

[0025] like Figure 1 As shown, the present invention discloses a water, wind, and solar power prediction and control system, which adopts a bus-type multi-CPU module collaborative architecture to construct a water, wind, and solar power prediction and control system. It includes: a prediction processing CPU module, a control execution CPU module, and a scheduling CPU module. Each CPU module is interconnected through a high-speed bus and has a clear division of labor and works collaboratively.

[0026] In this embodiment, the prediction processing CPU module includes a wind power prediction processing CPU and a photovoltaic prediction processing CPU. Independent CPU cores are used to undertake the inference calculation tasks of wind power and photovoltaic power prediction algorithms. This is used to run artificial intelligence prediction models for wind power generation and photovoltaic power generation, ensuring the real-time performance of prediction calculations without affecting other control tasks.

[0027] The control execution CPU module includes a control CPU, which is used to execute the regulation and control strategies of the hydropower unit. The control CPU is responsible for adjusting the output of the hydropower unit according to the scheduling instructions, including the start-up and shutdown control of the hydropower unit and the adjustment of output increase and decrease, so as to ensure that the hydropower unit can respond quickly to the fluctuations in wind and solar power.

[0028] The scheduling CPU module, as the main control unit for bus communication, includes a scheduling CPU for overall task scheduling and data acquisition. The scheduling CPU is responsible for triggering predictive calculations, collecting data from each module, issuing scheduling instructions, and coordinating real-time communication and task allocation between CPU modules.

[0029] Optionally, the CPU modules are interconnected via a unified high-speed communication bus to form a master control station. This high-speed bus can utilize a backplane bus or industrial Ethernet, providing high bandwidth and low latency communication links, and supports plug-and-play access for each CPU board. Through this modular topology, each CPU module can independently complete its own tasks while also enabling data exchange and instruction synchronization via the bus. In terms of layout, all CPU modules are installed in the same rack, sharing a power supply and clock synchronization mechanism to ensure consistent system timing, thereby improving system computing power and reliability. Alternatively, when additional functionality is required, it can be achieved by adding a CPU module without modifying the original architecture, offering good scalability and flexibility.

[0030] In this embodiment, the CPU modules communicate and interconnect on the bus through a bus communication mechanism; wherein, the bus communication mechanism is a standardized communication protocol and message format.

[0031] The message format uses a frame structure for encapsulation, including: (1) Message type: The message function category represented by a pre-agreed code; such as prediction results, scheduling instructions, control commands, status heartbeats, and abnormal alarms.

[0032] (2) Source / Target Identifier: This indicates the CPU module ID that sends the message and the CPU module ID that receives the message (or broadcast flag); for example, the source ID of the prediction result message sent by the wind prediction processing CPU or the photovoltaic prediction processing CPU is the prediction processing CPU module, and the target ID can be the scheduling CPU module and the control execution CPU module (broadcast to the relevant modules).

[0033] (3) Timestamp: It is accompanied by a high-precision time mark to indicate the moment when the message is generated or the data is collected; the timestamp can realize the data alignment of multiple modules, ensuring that each module is executed based on the same time base and avoiding timing chaos.

[0034] (4) Data payload: The specific content carried according to the message type; for example, the prediction result message contains the predicted values ​​of wind power and photovoltaic power for several future periods, the dispatch instruction message contains the target output value of hydropower units or start-up and shutdown commands for the next dispatch cycle, and the status message contains equipment operating status parameters, etc. The length of the data payload part is variable, but it is identified by parsing the type field in the message header.

[0035] In this embodiment, the bus communication mechanism adopts a master-slave control combined with an arbitration strategy: typically, the scheduling CPU acts as the bus master, uniformly scheduling the communication timing, and each CPU module sends messages after obtaining bus access permissions.

[0036] Optionally, for critical scheduling instructions, an acknowledgment mechanism is also adopted, that is, after the control execution CPU module receives the scheduling instruction, it feeds back the execution result or acknowledgment signal through the bus to ensure that the instructions are delivered and executed consistently.

[0037] In this embodiment, the bus supports high-speed parallel transmission. When multiple CPUs send data almost simultaneously, hardware arbitration ensures that messages are ordered and do not conflict. Furthermore, with the help of high-speed bus communication, the data synchronization cycle between each CPU module can reach the millisecond level, which improves the data consistency and command synchronization reliability between different CPU modules and avoids the problems of low communication efficiency, data asynchrony, or timing disorder in traditional multi-CPU module distributed control.

[0038] In this embodiment, the mainstream industrial Ethernet communication protocol Modbus TCP is used to realize data exchange between CPU modules. Each CPU module communicates through a request / response frame format, achieving reliable transmission of prediction data and control commands within the multi-CPU architecture. The control execution CPU module acts as the Modbus master (client), periodically polling the wind power prediction processing CPU, photovoltaic prediction processing CPU, etc., which act as slaves (servers), to obtain prediction data and real-time operating conditions, and then issuing control commands. The Modbus TCP communication frame consists of a 7-byte protocol header and a function data field, including: transaction ID (2 bytes), protocol ID (2 bytes, Modbus default is 0x0000), length (2 bytes, indicating the number of subsequent data bytes), unit ID (1 byte, slave address), function code (1 byte), and data field (n bytes). For example, when the control execution CPU module requests to read a holding register at address 0x0001 in the wind power prediction processing CPU, the following Modbus TCP message frame (hexadecimal) can be constructed: 1234 00 00 00 06 01 03 00 01 00 01. Where 0x1234 is the transaction ID, 0x0000 is the Modbus protocol identifier, 0x0006 is the data length, 0x01 is the unit ID (target slave address), 0x03 is the function code indicating reading the holding register, and the subsequent 0x0001 and 0x0001 are the starting register address and the number of registers to be read, respectively. After receiving the request frame, the wind power forecasting CPU will return a response frame containing function code 0x03, data length, and register values ​​according to the Modbus protocol.

[0039] Based on the above system, this invention provides a method for predicting and controlling the combined power output of hydropower, wind power, and solar power, meeting the requirements of high real-time performance, high reliability, and scalability for large-scale grid connection of new energy sources. Figure 2 As shown, the steps include: S1. At the beginning of each control cycle, the scheduling CPU module collects real-time operating data of wind farms, photovoltaic arrays and hydropower stations, and sends prediction trigger commands to the prediction processing CPU module.

[0040] Optionally, real-time operational data of wind farms, photovoltaic arrays, and hydropower stations, including recent historical wind power output and photovoltaic power output data, relevant meteorological observations and forecasts (such as wind speed, wind direction, solar radiation intensity, temperature, etc.), and grid load information (obtained as needed).

[0041] S2. After receiving the prediction trigger command, the prediction processing CPU module reads and preprocesses the real-time running data, and inputs the preprocessed real-time running data into the artificial intelligence prediction model to perform prediction calculations and obtain the prediction results.

[0042] S2.1 Predictive processing CPU module receives and preprocesses real-time running data.

[0043] In this embodiment, the real-time running data is preprocessed, including missing value imputation, normalization, and timestamp alignment.

[0044] S2.2 The preprocessed real-time operating data is fed into the artificial intelligence prediction model for inference to obtain the prediction results (wind power output prediction value and photovoltaic power output prediction value) for the future prediction period (e.g., within the next 1 to 15 minutes, every minute or more finely granular).

[0045] In this embodiment, for wind power prediction, a trained wind LSTM model can be used to predict wind turbine power output in the short term by inputting past power and meteorological sequences. The wind LSTM model utilizes historical wind farm output data from the past 60 minutes and corresponding meteorological monitoring data. For example, meteorological monitoring data may include real-time operating data such as average wind speed, wind direction, temperature, and wind turbine output power per minute during that period. The aforementioned 60 minutes of real-time operating data are input into the wind LSTM model in chronological order.

[0046] Optionally, to enhance the learning effect on time-dependent patterns, this invention employs a three-layer stacked LSTM network structure, comprising three sequentially stacked LSTM hidden layers and a final output layer. Each LSTM hidden layer can contain, for example, 64, 32, or 16 memory units, extracting short-term variation features of wind power layer by layer. The wind LSTM model is finally connected to a fully connected output layer, which uses a linear activation function to output the predicted value. Since the output of wind power is a continuous real number, the number of nodes in the output layer is 1 (i.e., predicting the power value for the next scheduling cycle).

[0047] Optionally, the training of the wind LSTM model can be supervised by historical operating data, and the LSTM network weights can be optimized by minimizing the mean square error between the predicted and actual values. During the prediction run, the latest 60-minute real-time operating data is received as input data at certain intervals, forward calculation is performed to obtain the future short-term power prediction results, and the prediction results are stored in the prediction processing CPU module so as to control the CPU module to poll and read them.

[0048] For photovoltaic (PV) power, a PV LSTM model similar to that used for wind power is employed to perform ultra-short-term forecasts of PV power plant output. Input data consists of the historical power output of the PV system over the past 60 minutes and the corresponding meteorological data series.

[0049] In this embodiment, for photovoltaic power generation, the main meteorological factors affecting output power include solar irradiance (light intensity), ambient temperature, photovoltaic panel surface temperature, solar altitude angle, etc. The present invention selects representative features, such as the horizontal solar irradiance (GHI) and ambient temperature recorded every minute, to form a multivariate time series input together with historical power.

[0050] In this embodiment, the photovoltaic LSTM model structure also adopts a three-layer stacked architecture: it includes three successively stacked LSTM hidden layers and a regression output layer. The size of the hidden layers can be set according to the complexity of the photovoltaic power data, for example, 50, 20, and 10 units per layer (adjustable according to data characteristics). The photovoltaic LSTM model predicts short-term power changes by learning the changing patterns of sunlight and temperature over the past hour and the dynamic relationship of photovoltaic output. The output layer provides the predicted photovoltaic output power P pv,pred(t+5min) for the next scheduling cycle, which is a single value (continuous value, unit kW or MW).

[0051] In this embodiment, the training method of the photovoltaic LSTM model is similar to that of the wind power LSTM model, which is also based on historical measured data for training and optimization. Since photovoltaic power output has daily periodicity and weather correlation, the accuracy can be improved by increasing the dimensions of training data (such as the latest weather forecast information) in the model design.

[0052] In this embodiment, because the system of the present invention adopts a multi-CPU module architecture, it allows the LSTM models of wind power and photovoltaics to run in parallel on different CPU cores, or in parallel via multiple threads within the same module, thereby accelerating the overall prediction calculation speed. In addition, the artificial intelligence prediction model can also output prediction confidence or error estimation as additional fields. For example, the LSTM model can provide prediction interval confidence based on the training error, and this information can be used as a reference for scheduling decisions.

[0053] In another embodiment of the present invention, the prediction results may be post-processed (e.g., additional confidence assessment, smoothing of abnormal peaks, etc.) and encapsulated into a prediction result message for reporting to the bus; wherein the payload of the prediction result message includes the prediction time range, time resolution and the corresponding prediction power sequence.

[0054] Optionally, the prediction result messages can be temporarily stored on the scheduling CPU module for subsequent scheduling optimization. At the same time, the prediction processing CPU module can also save the prediction result messages for data accumulation for adaptive correction of artificial intelligence prediction models or retraining of offline models. The entire prediction process takes very little time (typically completing model inference in the seconds), and because it uses a dedicated CPU module for operation, it will not block the execution of other critical tasks such as water and electricity control.

[0055] S3. Based on the prediction results and real-time operation data, determine the direction and initial value of hydropower output adjustment through rule set control strategy, and construct a linear programming model to solve for the optimal hydropower output value and generate scheduling instructions.

[0056] S3.1 Control the execution CPU module to read the prediction results and real-time running data, and select to call the rule set control or optimize the scheduling strategy according to the set control strategy type.

[0057] In this embodiment, the rule set control strategy compares the forecast results with the system's target load or planned output demand to determine the required direction and basic magnitude of hydropower regulation for the current cycle; for example, setting the system's total output reference value or load demand as... .

[0058] When the sum of wind power forecast and solar power forecast Significantly lower than When the "insufficient" scenario is entered, the rule set control strategy is to increase hydropower output to make up for the shortfall. when Significantly higher When there is a surplus of wind and solar power, the system enters the "surplus" scenario. The rule set control strategy is to reduce hydropower output or reserve peak-shaving capacity to avoid wind and solar power curtailment. When the two are close, it is considered a "balanced" scenario, and the rule set control strategy is to keep the hydropower output approximately the same as the previous cycle.

[0059] Based on the aforementioned rule-based decision-making, the control execution CPU module initially determines the direction (increase or decrease) and approximate value of the hydropower unit output adjustment for this cycle. The rule set control strategy reflects the empirical strategies and safety constraints of power plant operation, such as ensuring that hydropower output does not exceed upper or lower limits and avoiding power shortages through rapid response. However, relying solely on fixed rules is insufficient to guarantee global optimization; therefore, this method further introduces a linear programming model to refine the optimization of hydropower output. The linear programming scheduling part uses the rule-based decision results as an initial reference. Under the constraint conditions, it constructs a linear programming model to obtain the optimal power output command for hydropower. The decision variable in the linear programming model is the power generation P of the hydropower unit in the next scheduling cycle. hyd (Or, when there are multiple generating units, it is the output combination of each unit); the objective function of the linear programming model can be flexibly set according to the scheduling requirements. For example, a typical linear programming model is to make the total output of "hydro + wind + solar" as close as possible to the load demand, reducing the surplus or deficit caused by wind and solar fluctuations; equivalently, the objective function can also be set to maximize wind and solar utilization and minimize wind and solar curtailment, and to absorb the fluctuations of new energy sources by increasing hydropower peak shaving; among them, the typical objective function of the linear programming model is expressed as follows,

[0060] Where T is the set of scheduling periods; L t The system load demand for time period t; , , Actual output of wind power, solar power, and hydropower during time period t; The predicted output of wind and solar power for time period t; These are the weighting coefficients for each sub-objective.

[0061] Optionally, the objective function of the linear programming model may also include, This refers to the fluctuation of the remaining load. This is due to the deviation in wind power output. For photovoltaic power output deviation; This refers to the amount of air that is forcibly abandoned. This refers to the amount of light discarded. In this embodiment, based on the above objective function, the constraints of the linear programming model include: hydropower output limit constraints, hydraulic operating condition constraints, power change rate constraints, and system power balance constraints; wherein, the hydropower output limit constraints are: ≤ ≤ Ensure that the power output of the hydropower unit is within the allowable range (from the unit's minimum technical output to its maximum output). Hydraulic operating condition constraints can be considered if necessary, taking into account the current reservoir water level and flow constraints, to ensure that optimized scheduling does not violate reservoir scheduling rules (e.g., water level does not rise to the flood limit level, and reservoir release for flood control when necessary; in the short term, this constraint is usually fixed to the output range permitted by the current operating conditions). The power change rate constraint is | |≤Δmax, limiting the hydropower unit's ramp rate to no more than the technically permissible value, to avoid excessive turbine adjustment; The system power balance constraint is, in the case of an independent microgrid, the constraint that the total output of hydropower, wind power, and solar power can be increased to balance the local load, i.e. + + = Or, in the case of grid connection, ensure that the power surplus or deficit is within the adjustment range, etc.

[0062] Optionally, rule set control or optimization scheduling strategies can be used in combination; when the problem is relatively simple or requires a fast response, the rule set control strategy is used, while during stable periods, the optimization scheduling strategy is run periodically to obtain a better solution.

[0063] S4. The control execution CPU module sends the generated scheduling instructions to the hydropower unit via the bus. After the hydropower unit executes the scheduling, it feeds back the real-time output data to each CPU module.

[0064] In this embodiment, once a new scheduling instruction is generated, the control execution CPU module packages the scheduling instruction into a control command message and sends it to the hydropower actuator via the bus. The command content may be the desired power setpoint of the hydropower unit, start / stop instructions, or output change rate, such as: "Increase the hydropower output to 50MW within the next minute" or "Start Unit 2 and increase the output to the rated value". For pumped storage units, it may also include instructions such as "Switch to pumping mode and run for N minutes at a certain time".

[0065] Optionally, the command message includes a timestamp and priority, which the receiver (hydropower unit control unit) uses to perform the corresponding operation immediately or at regular intervals.

[0066] In this embodiment, after the hydropower unit performs regulation, the scheduling CPU module monitors the actual changes in wind power, photovoltaic and hydropower output in real time, and feeds back to each CPU module through the bus.

[0067] Optionally, the control execution CPU module also compares the actual scheduling effect with the prediction results. If it finds that the prediction error (the actual wind and solar power output deviates from the prediction) is large, the strategy is dynamically adjusted in the control decision of the next cycle. For example, if the predicted wind speed is higher than expected, resulting in actual wind power being lower than the predicted value, the control execution CPU module increases the hydropower compensation in the subsequent cycle; conversely, if the actual photovoltaic output is higher than expected, causing excess, the hydropower output reduction is increased in advance or the pumping time is extended. This adaptive control driven by prediction error and confidence level closely links the depth of hydropower regulation with the fluctuations of renewable energy, improving the stability of combined output and the ability to track plans.

[0068] S5. Use the feedback real-time output data as new real-time operating data, run the above steps S1-S4, and enter the next cycle.

[0069] In this embodiment, the control execution CPU module can compare the deviation between the prediction and the actual output in subsequent cycles based on the real-time output data fed back, and correct the scheduling strategy when necessary (e.g., adjusting the rule threshold or notifying the upper-level scheduling when the continuous deviation increases). At the same time, if an unexpected situation occurs during the execution of the previous cycle (such as the hydropower unit failing to reach the commanded output, or the wind turbine being cut off, resulting in the actual wind power being lower than the prediction), the control execution CPU module will make corrections and compensations in the decision-making of the next cycle.

[0070] In another embodiment of the present invention, in order to ensure the efficient and orderly execution of the above-mentioned prediction and control tasks, the scheduling CPU module also performs a unified time flow arrangement for the entire system tasks, which includes, Data Acquisition and Broadcast (Time T0): At the beginning of each control cycle, the scheduling CPU module acquires real-time operating data (power output, water level, weather, etc.) from the wind farm, photovoltaic array, and hydropower station, encapsulates this data into status messages, and broadcasts them to other modules via the bus. This step ensures that the prediction and control modules operate based on the latest data.

[0071] Triggering Prediction Calculation (Time T1): The scheduling CPU module sends a prediction trigger command to the prediction processing CPU module (or triggers automatically at a pre-agreed period). The prediction processing CPU module begins reading the newly acquired input data, loads the corresponding AI prediction model, and executes wind / solar power prediction calculations. Since the prediction processing CPU module operates independently, the calculation process can be performed in parallel with other operations.

[0072] Execution of control adjustments (times T1-T2): While the predictive processing CPU is calculating, the control execution CPU module can utilize the prediction results from the previous cycle to continuously perform fine-tuning control of the hydropower output to smooth the power output at the current moment. That is, while the prediction for the current cycle is still being calculated, the control commands from the previous cycle may still be being executed or finely adjusting the current output. This task overlap leverages the parallel advantages of multiple CPU modules, allowing control to proceed without waiting for a new prediction to complete, ensuring the continuity of control actions.

[0073] Prediction Result Release (Time T2): The prediction processing CPU module completes the wind and solar power prediction calculation for this cycle within a very short time (e.g., within a few seconds), and then immediately packages the results into a prediction result message, which is sent to the scheduling CPU module and the control execution CPU module via the bus. The bus communication latency is very small (milliseconds), so it can be assumed that each module has received the new prediction data at time T2. The scheduling CPU records the prediction results and, if necessary, performs simple smoothing or reliability checks, and then notifies the control execution CPU module to adopt the new data.

[0074] Scheduling decision calculation (times T2-T3): After receiving the latest prediction results, the control execution CPU module immediately runs the scheduling decision algorithm (rule set control strategy or optimized scheduling model) to calculate the new hydropower output plan. Thanks to the multi-CPU architecture, the control execution CPU module has sufficient computing resources at this time, as the prediction task has been completed by other CPUs. Even with an optimized scheduling model, modern industrial control execution CPU modules can solve the problem in milliseconds to hundreds of milliseconds (the algorithm's solution accuracy can be adjusted according to model complexity and constraint scale to meet real-time requirements). If the scheduling CPU module undertakes some decision calculation functions, it will also assist in completing calculations or allocating tasks during this period.

[0075] Issuance of Control Commands (Time T3): Once the new control scheme is calculated, the control execution CPU module immediately sends the hydropower control commands to the execution layer via the bus. Typically, the scheduling CPU module, acting as the bus master, coordinates the allocation of bus transmission time slots to the control execution CPU module at time T3, ensuring its commands are sent with the highest priority. Upon receiving the commands, the hydropower unit control unit executes the corresponding power adjustments. This process includes operations such as driving the turbine governor, adjusting the guide vane opening, or starting and stopping the pumps. The transmission delay from generation to execution of critical control commands is strictly controlled to the millisecond level.

[0076] Feedback and preparation for the next cycle (time T4): The hydropower unit adjusts its output according to the instructions. The new output value and the status of the entire wind-solar-hydro system are collected by sensors and fed back (returning to data acquisition in step 1). The scheduling CPU module continuously monitors the operating status of each module. If any step is detected as not being completed within the predetermined time window (e.g., prediction calculation timeout, delay in control command issuance), abnormal handling or emergency strategies (such as directly using the previous result or a simplified decision-making strategy) will be triggered to ensure that the control loop is not interrupted. Subsequently, the system enters the next control cycle.

[0077] The intervals between key time points in the above process have been rigorously designed to meet real-time control requirements. Through multi-CPU parallel processing and high-speed bus communication, the total latency of the prediction-decision-execution closed loop is minimized. For example, in a typical implementation, the complete prediction and scheduling calculations can be completed within tens to hundreds of milliseconds every 5-minute scheduling cycle, leaving most of the time for actual execution and smooth adjustment. This ensures that hydropower can promptly adjust within short cycles of 1 to 5 minutes when wind power or photovoltaic output changes drastically. For higher real-time requirements (such as grid frequency regulation), the system can also operate on a second-level cyclical basis: the prediction module performs rolling second-level power predictions, and the control module adjusts hydropower output every second. Due to the high efficiency of bus communication and decision algorithms, the response latency of key control loops can be reduced to below 20 milliseconds when needed, meeting the requirements for rapid control of the power system.

[0078] In another embodiment of the present invention, taking a scheduling cycle of 5 minutes as an example, the entire process of the above-mentioned multi-CPU collaborative prediction and control scheme is described, and the following steps are executed sequentially in each cycle: a. Data Acquisition Phase (Cycle Start): At the start of each 5-minute cycle, the control execution CPU module acquires the latest operational data via the communication network. Acting as a Modbus master, the control execution CPU module sends data read requests to the wind power forecasting CPU and the photovoltaic forecasting CPU to obtain their respective real-time status data. For example, the control execution CPU module reads the current output of the wind farm, the wind turbine power over the past 60 minutes, and key meteorological data, as well as the current output of the photovoltaic power station, the photovoltaic power over the past 60 minutes, and irradiance. In addition, the control execution CPU module also acquires operational parameters such as the current output of the hydropower station and the reservoir water level (provided via local I / O or by the hydropower control module). All acquired data is updated in the memory of the control execution CPU module for subsequent forecasting and decision-making calculations. The entire data acquisition process is completed within hundreds of milliseconds via high-speed Ethernet, ensuring sufficient time for subsequent processing.

[0079] b. Power Prediction Calculation Stage: Upon receiving data requests from the control execution CPU module or based on internal timing, the wind power prediction processing CPU and the photovoltaic prediction processing CPU each execute the latest round of power prediction calculations at the beginning of the cycle. Specifically, the wind power prediction processing CPU inputs the updated past 60-minute wind power sequence and corresponding meteorological sequence into its three-layer LSTM model, and obtains the wind power prediction value for the next moment (i.e., within the next 5 minutes) through forward calculation. Similarly, the photovoltaic prediction processing CPU inputs data such as photovoltaic power and irradiance from the past 60 minutes into its LSTM model to calculate the predicted photovoltaic power for the next 5 minutes. The two prediction processing CPUs store the calculated prediction results into their respective registers or preset memory areas. Subsequently, the control execution CPU module reads the prediction results stored in the wind power prediction processing CPU's memory address via Modbus messages. Values, and the photovoltaic prediction processing CPU memory The control execution CPU module has now obtained the wind and solar power prediction results for the current cycle. It should be noted that, due to the relatively small computational load of the prediction model (only a few layers of LSTM network inference), the inference time on modern CPUs is in the millisecond range and negligible, and will not affect the real-time scheduling of the 5-minute cycle. At the end of the prediction phase, each module holds key information for the end of the cycle: wind and solar power prediction values ​​and real-time operating data.

[0080] c. Scheduling Decision Phase: The control execution CPU module, based on the obtained... and Based on the system's operational goals and constraints, the adjustment commands for the hydropower units in the next cycle are calculated and determined. First, preset rules are applied to analyze the difference between wind and solar power output and the target value: the predicted total output is calculated. = + and compared with reference load or planned power Comparison. If < If the power supply is insufficient for the next 5 minutes, it will enter the hydropower generation mode; if > If the renewable power is excessive, the system enters a power reduction (hydropower storage) mode; if the two are roughly equal, the power is considered balanced, and no significant adjustment is needed. Based on the judgment result, the control execution CPU module generates the initial hydropower output command value; for example, In the "insufficient" mode, the initial output of hydropower can be reduced. = +( This means increasing output to make up for the difference; Under the "surplus" mode, it can make = ( This means reducing power output to absorb excess power generation; In balanced mode, ≈ The above calculations must remain unchanged (the calculations must be limited to the upper and lower limits of hydropower output).

[0081] Next, the CPU module controls the execution of the process and uses this initial instruction to construct a linear programming optimization model. Taking into account the constraints, it solves the problem to obtain the adjusted optimized output. ; For example, in "surplus" mode, if the initial calculation requires hydropower to be reduced to near minimum output but considers the minimum output limit of the generating units, then LP optimization will output... = And some wind and solar power may still need to be discarded; For example, in "insufficient" mode, if the difference is large enough to exceed the maximum output of the hydropower, the LP will output at saturation. = And leave gaps to await other backup supplies.

[0082] Linear programming can simultaneously optimize secondary objectives, such as smoothing out changes in hydropower output and considering pre-balancing scheduling for the next few cycles, but within a short 5-minute cycle, the primary objective is to achieve the current balance. Rule-based decision-making provides direction and LP (Literal Programming) provides quantitative optimization, ultimately determining the... It satisfies the real-time balance requirements while adhering to the constraints of hydropower operation, thereby optimizing dispatching decisions.

[0083] d. Command Issuance and Execution Phase: The control execution CPU module sends the optimized hydropower output command to the hydropower station actuator via the communication network for adjustment. Specifically, the control execution CPU module, acting as a Modbus master, sends a write register command to the hydropower unit controller (e.g., the PLC module of the turbine governor, configured as a Modbus slave). Write to its control register address. The message frame format is similar to the aforementioned read, but function codes 0x06 or 0x10 indicate writing to one or more registers, and the data area contains the destination register address and... The hydroelectric controller receives the command and immediately analyzes the target power value, adjusting the guide vane opening and unit output to the set level through its internal control loop. If there are multiple turbines, the total output command can be sent to each unit according to a predetermined strategy (such as load sharing or allocation based on efficiency curves). This process typically completes unit response within a few seconds.

[0084] e. Feedback and Closed-Loop Phase: After the hydropower units perform regulation, the system enters the operation monitoring state. Field sensors and each CPU module continuously collect real-time output data from wind, solar, and hydropower, providing this data to the control execution CPU module at the start of the next cycle. Through this real-time feedback, the control execution CPU module can compare the deviation between predictions and actual outputs in subsequent cycles and correct the scheduling strategy if necessary (e.g., adjusting rule thresholds or notifying upper-level scheduling when continuous deviations increase). Simultaneously, if unforeseen circumstances occur during the previous cycle (e.g., hydropower units failing to reach commanded output, wind turbines being cut off resulting in actual wind power lower than predicted), the control execution CPU module will correct and compensate in the next cycle's decision-making. This cyclical scheduling mechanism forms a complete closed-loop control: data acquisition → prediction calculation → optimization decision → command execution → performance feedback → next cycle, ensuring the entire hydro-wind-solar combined system can continuously and stably operate according to the optimization target. This cycle repeats every 5 minutes, with the control strategy continuously adjusted based on the latest data to achieve dynamic response and smoothing of wind and solar fluctuations. This embodiment uses 5 minutes as an example. In actual systems, the scheduling cycle can be set to any reasonable interval such as 1 minute or 15 minutes as needed, and the principle is similar.

[0085] In another embodiment of the present invention, to further improve system performance and reliability, the present invention also introduces optional optimization measures, including data buffering, redundant backup, and anomaly handling strategies, etc. 1. Data Buffering and Pipeline Processing: A data buffering queue mechanism is employed in bus communication and within modules. All sensor data, prediction results, and control commands are first written to the buffer during transmission / reception, and then asynchronously read and processed by the target task. This decoupled design avoids real-time tasks directly blocking and waiting for data. For example, the prediction processing CPU reads sensor data from the buffer queue for calculation, and writes the result to the output buffer queue after the calculation is completed; the control execution CPU module reads the prediction result from its input buffer. This not only improves parallelism but also enables task pipelined processing, meaning that when the prediction of a certain cycle has not yet ended, the control execution CPU module can process the task of the previous cycle or prepare for the next step. Data buffering also allows for batch data transfer and DMA direct storage on the high-speed bus, reducing communication overhead. By optimizing buffer management, the system improves bus throughput efficiency while ensuring real-time performance and mitigates the impact of sudden data traffic on the system.

[0086] 2. Task Priority and Real-time Scheduling: To ensure the timely completion of critical control tasks, the system assigns priorities to various tasks. Core tasks such as predictive calculations, control decisions, and command issuance have the highest priority, while non-critical tasks such as data logging and model updates have lower priority. The scheduling CPU runs a real-time scheduling algorithm, allocating CPU time slices based on task priority and deadline. For example, if control calculations are not yet complete and the deadline is approaching, the scheduling CPU will interrupt secondary tasks to ensure that the control task has exclusive CPU access, guaranteeing timely issuance of control commands. Furthermore, by utilizing the redundancy of multiple CPUs, tasks of different priorities can be distributed across different CPUs for parallel execution, completely avoiding latency caused by high- and low-priority tasks competing for the same processor.

[0087] 3. Flexible Algorithm Degradation: To address potential computational peaks in complex algorithms, the system pre-defines an algorithm degradation strategy. This allows for the use of simplified algorithms when necessary to ensure timeliness. For example, if the optimization scheduling model fails to converge to the optimal solution within a specified time, the control execution CPU module will directly output the current optimal feasible solution at the deadline or switch to a rule-based control strategy, sacrificing some optimization performance to ensure timely execution of control measures. Similarly, if the AI ​​prediction model fails to generate new results in time due to abnormal input, the system will automatically use the prediction from the previous cycle or temporarily replace it with a simple extrapolation algorithm. This dual-scheme mechanism enables the system to have flexible scheduling capabilities, pursuing optimization while quickly degrading to a safer approach under pressure, avoiding control failure due to algorithm timeouts.

[0088] 4. Parallel Computing and Load Balancing: Leveraging the parallel capabilities of multiple CPU modules, the system achieves dynamic load balancing. The scheduling CPU module continuously monitors metrics such as CPU utilization and operating temperature of each module. When a module is found to be overloaded or experiencing a performance bottleneck, some tasks can be migrated to other idle CPUs (supporting task sandboxing and hot-migrating mechanisms). For example, when wind power and photovoltaic forecasting calculations simultaneously consume significant CPU resources, one forecasting task can be offloaded to another idle forecasting CPU module, or, in a configuration with a CPU / FPGA coprocessor, some computation can be offloaded to a dedicated acceleration unit. This load balancing prevents slowdowns caused by overloading a single CPU. Furthermore, the multiple CPUs can act as backups for each other; if one CPU fails, the scheduling CPU immediately switches its tasks to a backup module, ensuring continuous and stable system operation.

[0089] 5. Redundancy Backup Design: Redundant hardware and software backup schemes are designed for critical functions. On the hardware side, dual predictive processing CPUs and dual control execution CPU modules can be configured, operating as a primary and backup module or in parallel for verification. When the primary module fails or outputs abnormally, the backup module seamlessly takes over the task. For example, the two predictive processing CPU modules can simultaneously perform predictive calculations on the same set of data and cross-validate the results. If the difference in results exceeds a threshold, an alarm is issued indicating a model anomaly. Similarly, the backup control execution CPU module normally receives all predictive messages synchronously and runs the scheduling algorithm in the background, but does not issue commands. When the primary control execution CPU module fails, the backup control execution CPU module immediately takes over command sending responsibilities, ensuring uninterrupted power and water regulation. Data redundancy is also implemented on the software side. For example, copies of critical states and predictive results are cached at the sending end, and a mirror image exists on the backup control execution CPU module to prevent data loss due to single-point failures. This redundancy design improves system reliability and fault tolerance, making it suitable for power dispatching scenarios with extremely high continuity requirements.

[0090] 6. Anomaly Monitoring and Fault Handling: A complete anomaly handling strategy is defined. The scheduling CPU monitors the heartbeat and status messages of each module. Upon detecting anomalies such as communication timeouts, data inconsistencies, or module unresponsiveness, it takes immediate action. For example, if a prediction module times out without a result, the scheduling CPU will notify the control module to operate in the previous prediction or safety mode, and record the error for future maintenance. If a control module fails, the system switches to the backup control execution CPU module and notifies the hydropower station to switch to local automatic regulation mode to temporarily maintain stability. For bus communication failures, the system can switch to a degraded communication link or activate emergency stop control to ensure safety. All anomalies and their handling processes are recorded for future analysis and improvement.

[0091] The foregoing description and accompanying drawings fully illustrate embodiments of the invention to enable those skilled in the art to practice them. Other embodiments may include structural and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Some portions and features of some embodiments may be included or substituted for portions and features of other embodiments. Embodiments of the invention are not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from their scope. The scope of the invention is limited only by the appended claims.

Claims

1. A water, wind, light combined power prediction and control system, characterized in that, The system includes a predictive processing CPU module, a control execution CPU module, and a scheduling CPU module; wherein, each CPU module communicates and interconnects on a bus through a bus communication mechanism. The prediction processing CPU module includes a wind power prediction processing CPU and a photovoltaic prediction processing CPU. It uses independent CPU cores to undertake the inference calculation tasks of wind power and photovoltaic power prediction algorithms, and is used to run artificial intelligence prediction models for wind power generation and photovoltaic power generation. The control execution CPU module includes a control CPU, which is used to execute the regulation and control strategies of the hydropower unit; wherein, the control CPU is used to adjust the output of the hydropower unit according to the scheduling instructions; The scheduling CPU module is the main control unit for bus communication, including a scheduling CPU, which is used for overall task scheduling and data acquisition. The scheduling CPU is used to trigger predictive calculations, collect data from each module, issue scheduling instructions, and coordinate real-time communication and task allocation between CPU modules.

2. The water, wind, and solar power prediction and control system of claim 1, wherein, The control CPU is specifically used for the start-up and shutdown control of the hydropower unit and the adjustment of its output. The bus adopts a backplane bus or an industrial Ethernet high-bandwidth, low-latency communication link, supporting plug-and-play access for each CPU board.

3. The water, wind, and solar power prediction and control system of claim 1, wherein, The CPU modules communicate and interconnect on the bus through a bus communication mechanism, specifically by using a standardized communication protocol and message format for communication interconnection; wherein the message format includes message type, source / target identifier, timestamp, and data payload.

4. A water, wind, light combined power prediction and control method based on the system of any one of claims 1-3, characterized in that, The method includes, S1. At the beginning of each control cycle, the scheduling CPU module collects real-time operating data of wind farms, photovoltaic arrays and hydropower stations, and sends prediction trigger commands to the prediction processing CPU module. S2. After receiving the prediction trigger command, the prediction processing CPU module reads and preprocesses the real-time running data, and inputs the preprocessed real-time running data into the artificial intelligence prediction model to perform inference calculations and obtain the prediction results. S3. Based on the prediction results and real-time operation data, generate scheduling instructions through rule set control strategies and optimized scheduling models; S4. The control execution CPU module sends the generated scheduling instructions to the hydropower unit via the bus. After the hydropower unit executes the scheduling, it feeds back the real-time output data. S5. Use the feedback real-time output data as new real-time operating data, run S1-S4, and enter the next cycle.

5. The method of claim 4, wherein the water, wind, and light power prediction and control method further comprises: The real-time operating data of the wind farm, photovoltaic array and hydropower station includes recent historical wind power output and photovoltaic power output data, relevant meteorological observations and forecasts, and grid load conditions.

6. The method of claim 4, wherein the water, wind, and light power prediction and control method further comprises: The preprocessing of real-time running data includes filling missing values, normalizing, and aligning timestamps in the real-time running data; The artificial intelligence prediction model includes a wind LSTM model and a photovoltaic LSTM model; wherein, both the wind LSTM model and the photovoltaic LSTM model adopt a three-layer stacked LSTM network structure, including three sequentially stacked LSTM hidden layers and a final output layer.

7. The method of claim 4, wherein the water, wind, and light power prediction and control method further comprises: The rule set control strategy compares the prediction results with the system's target load or planned output demand to determine the direction and basic magnitude of hydropower regulation required in the current cycle. The optimized scheduling model uses the comparison results of the rule set control strategy as an initial reference. Under the condition of satisfying the constraints, it solves the optimal power output command of hydropower by constructing a linear programming model. The constraints include hydropower output limit constraints, hydraulic operating condition constraints, power change rate constraints, and system power balance constraints.

8. The method of claim 4, wherein the water, wind, and light power prediction and control method further comprises: The method further includes the scheduling CPU module scheduling the time flow of the method.

9. The method of claim 4, wherein the method further comprises: The method also includes optimization measures, specifically including data buffering and pipelined processing, task prioritization and real-time scheduling, algorithm elastic degradation, parallel computing and load balancing, redundancy backup design and anomaly monitoring and fault handling.

10. 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 water, wind, and solar power prediction and control method according to any one of claims 4-9.

Images