A method, apparatus, and readable storage medium for dispatching and control of a distributed generation system.

Abstract

This invention provides a scheduling and control method, apparatus, and readable storage medium for a distributed generation system, belonging to the field of distributed generation system scheduling and control technology. To address the technical problems of difficulty in unifying multi-objective scheduling, unverifiable node data, low trust levels, and delayed scheduling response in distributed generation systems, the technical solution adopted is as follows: An edge computing module is deployed at each node of the distributed generation system to collect the real-time operating status of each node. A lightweight prediction model is used for local modeling and training to predict the node's generation capacity and load demand within a set prediction period in real time. The trust value of each node is updated based on its prediction accuracy, scheduling response latency, communication stability, and instruction execution status in the previous scheduling period. This invention is applied to the scheduling and control of distributed generation systems.

Description

Technical Field

[0001] This invention provides a scheduling and control method, device, and readable storage medium for a distributed generation system, belonging to the field of scheduling and control technology for distributed generation systems. Background Technology

[0002] With the widespread grid integration of renewable energy and the development of the energy internet, the proportion of distributed generation systems (such as photovoltaic, wind power, gas turbines, and energy storage devices) in the energy structure is constantly increasing. These generation systems are characterized by multi-source heterogeneity, wide distribution, and strong dynamism, which makes the traditional dispatch and control methods based on central power sources inadequate when dealing with complex and ever-changing distributed energy systems in the following ways:

[0003] Traditional scheduling models rely on a central controller to obtain global information for unified scheduling, which suffers from long scheduling cycles and instruction execution delays, making it difficult to respond to dynamic changes in each node in real time. Existing technologies mostly employ fixed model prediction or rule-driven approaches, failing to fully leverage the edge computing and prediction capabilities of distributed nodes, and are unable to quickly perceive and adaptively adjust to sudden load fluctuations and localized abnormal operating states. Due to varying node prediction capabilities, the central scheduling system struggles to accurately adopt scheduling suggestions from each node, limiting global optimization capabilities and lacking a comprehensive evaluation mechanism for node execution capabilities and communication stability. These shortcomings all affect the feasibility and stability of scheduling instructions. Furthermore, in an open distributed environment, data transmission and scheduling instructions face the risk of tampering and forgery; existing systems lack effective encryption authentication and trust management mechanisms, making it difficult to ensure the secure and reliable operation of the system. Summary of the Invention

[0004] To address the technical problems of difficulty in unifying multi-objective scheduling, unverifiable node data, low trust levels, and delayed scheduling response in distributed generation systems, this invention provides a scheduling control method for distributed generation systems, comprising the following scheduling control steps:

[0005] S1: Deploy edge computing modules at the nodes of the distributed generation system to collect the real-time operating status of each node. Then, use a lightweight prediction model to perform local modeling and training based on the historical power generation data, current operating status data, user load curves, and weather forecast information of the collected nodes, and predict the power generation capacity and load demand of each node in real time within the set prediction period.

[0006] A dynamic feature selection mechanism is adopted to automatically screen key influencing factors and adjust model parameters based on the correlation between node type and historical power generation data;

[0007] S2: Update the trust value of each node based on its prediction accuracy, scheduling response latency, communication stability, and instruction execution status in the previous scheduling cycle.

[0008] S3: Each node generates locally optimized scheduling suggestions based on local forecast data, its own energy storage status, and scheduling objectives, and uploads the scheduling suggestions to the central scheduling platform through encrypted hash digest.

[0009] S4: The central dispatch platform constructs a multi-objective dispatch optimization model with weighted factors based on the dispatch suggestions of each node, the current status of the node, and the trust value. It generates the globally optimal dispatch instruction by solving the optimization model and dispatches the distributed generation system based on the instruction.

[0010] The specific method for predicting the power generation capacity and load demand of the node within the set prediction period in step S1 is as follows:

[0011] Several lightweight prediction models are deployed on each node to form a model set. After each model runs independently, multiple predicted values ​​{y1, y2, ..., y} are obtained. n Then, a weighted fusion method is used to integrate the prediction results of each model to obtain the weighted fusion prediction value. The calculation formula is:

[0012] ;

[0013] Among them, w i For model M i The weights;

[0014] Further, the confidence level of the final predicted value after fusing the prediction results of multiple different prediction models for the same node is calculated. Based on the dispersion of the predicted values ​​from multiple models, the standard deviation is estimated as the prediction confidence level σ. The calculation formula is as follows:

[0015] ;

[0016] The final prediction output is: ;

[0017] Where z is the confidence interval coefficient, and if σ exceeds the preset tolerance threshold δ, it indicates that the prediction result is unstable;

[0018] A fault tolerance threshold δ is set based on the prediction confidence level σ. If σ > δ, the current prediction is considered unreliable, and more node data needs to be requested, a backup scheduling strategy needs to be issued, or the current prediction output needs to be corrected based on the historical average.

[0019] The specific method for screening key influencing factors and adjusting model parameters in step S1 is as follows:

[0020] The edge computing module calculates the prediction confidence σ of the fused output based on the dispersion of each prediction result. If the prediction confidence σ exceeds a preset fault tolerance threshold δ, the edge computing module will mark the prediction as a low-confidence output and trigger one of the following fault tolerance control strategies:

[0021] Request neighboring nodes to share their prediction results for collaborative correction;

[0022] Use historical average load forecasts as rollback values;

[0023] Automatically switch to the locally predefined backup policy number;

[0024] Finally, each node, based on its current energy storage status, forecast results, and known scheduling objectives, automatically generates a scheduling suggestion that is locally optimal for the current scheduling cycle.

[0025] The specific method for updating the trust value of each node in step S2 is as follows:

[0026] The trust value of each node is updated using the following weighted cumulative method:

[0027] ;

[0028] Among them, T i (t) is the trust value from the previous period, β is the memory factor used to balance historical values ​​with current performance, and S i (t) is the overall score for the current period;

[0029] After normalizing the key performance indicators of each distributed scheduling node, a weighted summation model is used to obtain a comprehensive score, calculated as follows:

[0030] ;

[0031] in, It is the prediction accuracy. It is the normalized value of the response delay. It's the communication success rate. It represents the success rate of executing instructions, and w1, w2, w3, and w4 are the importance weights of each dimension.

[0032] The specific method for generating local optimization scheduling suggestions in step S3 is as follows:

[0033] Each node constructs a lightweight prediction model based on its own historical data sequence and current environmental state. Each node generates a local optimization problem based on its local prediction data and scheduling objective, and outputs a suggested power. and energy storage dispatch quantity ΔSOC i The expression for the optimization problem is: ;

[0034] in, This is a penalty function for scheduling deviation, lag, etc., where a, b, and c are weighting coefficients. Let represent the predicted power generation capacity of node i at a future time t+τ. Let τ be the predicted load demand of node i at a future time t+τ, where τ is the scheduling delay.

[0035] The specific method for constructing the lightweight prediction model is as follows:

[0036] Input historical power generation sequence, current meteorological parameters, and user load curves; adopt a multi-model integrated structure and perform weighted fusion prediction. Based on the prediction results, the nodes calculate the future net load and remaining available energy storage capacity, and input them into the optimization model.

[0037] Each node constructs the following local optimization problem as its objective function based on the predicted data:

[0038] The power output node is expected to output during the scheduling cycle, determined by the energy storage charge / discharge quantity ΔSOC. i To compensate for discrepancies between power generation forecasting and load forecasting, and to optimize the dispatch power at nodes. It is closer to the scheduling goals of the central platform, enabling nodes to passively execute scheduling when there is no energy storage, and to make flexible adjustments when energy storage is available;

[0039] The energy storage term in the objective function is used to constrain ΔSOC. i The amplitude, determined by ΔSOC i Adjusting the output power and controlling energy storage safety, the objective function is expressed as follows:

[0040] ;

[0041] Among them, the scheduling deviation penalty function Dev measures the degree of deviation between the suggested power and the predicted supply and demand gap; the lag penalty function Lag controls the rate of change of power output to avoid sudden changes; and the SOC usage penalty SOCPenalty limits the frequent deep charging and discharging of energy storage to ensure battery life. It is the predicted power generation of node i. It is the actual execution power of node i in the previous scheduling cycle.

[0042] The specific method for uploading the scheduling suggestion to the central scheduling platform using encrypted hash digest in step S3 is as follows:

[0043] The node performs a hash calculation on the encrypted data to generate a unique digest. The node then sends the uploaded content to the central scheduling platform. The expression for the uploaded content is:

[0044] ;

[0045] Among them, H i The summary is used to centrally verify that the data has not been tampered with, E i The encrypted data is used by the center to decrypt and restore the original scheduling suggestion, ID. i It indicates the data source, while the timestamp records the upload time and is used for scheduling time-series synchronization or subsequent source tracing;

[0046] The following hash signature encryption method is used to upload the data to the central dispatch platform:

[0047] The SHA-256 hash H(S) is used to compute a digest of the scheduling suggestion data S.

[0048] Use the node's private key to digitally sign H(S) to generate the signature Sig(S);

[0049] The content uploaded by the node is: [Node ID, timestamp, S digest H(S), signature Sig(S)];

[0050] The central scheduling platform verifies the signature using the node's public key and confirms the integrity of the proposed content by comparing the digest. If the hash verification fails, the scheduling proposal is rejected and marked as data tampering or communication corruption.

[0051] The central scheduling platform constructs a multi-objective optimization model with a trust factor based on the scheduling suggestions uploaded by each node, the current node operating status, and the trust value.

[0052] The specific method for generating the globally optimal scheduling instruction in step S4 is as follows:

[0053] The optimization model, by setting different objective functions and combining node trust values ​​with various system constraints, forms a multi-objective optimization problem. Each objective function has a corresponding weight factor, which is dynamically adjusted according to the actual operational needs of the system. The expression for the objective function is:

[0054] ;

[0055] Among them, f i (x) is the i-th objective function, w i It is the weight factor of the i-th objective function;

[0056] The multi-objective optimization model is solved by an optimization algorithm. In each iteration, the objective function value is evaluated based on the current solution, and the control variables are adjusted to achieve the global optimum. During the optimization process, the algorithm will gradually adjust the power generation and energy storage status of each node until a solution that minimizes the objective function is found.

[0057] Once the optimization model finds the optimal solution, the central dispatch platform will generate specific dispatch instructions based on the solution results. The dispatch instructions include the power generation requirements of each node, the energy storage status adjustment requirements of each node, and the control instructions within the dispatch cycle.

[0058] A dispatch control device for a distributed generation system includes the following control module:

[0059] The data acquisition and prediction module is used to deploy edge computing modules at the nodes of distributed generation, collect real-time operating status data of each node, and make local predictions of the node's power generation capacity and load demand based on a lightweight prediction model.

[0060] The dynamic trust assessment module is used to update the trust value of each node based on its prediction accuracy, scheduling response latency, communication stability, and instruction execution status in the previous scheduling cycle.

[0061] The scheduling suggestion generation module is used by each node to generate local optimization scheduling suggestions based on local forecast data, its own energy storage status and scheduling objectives, and upload the scheduling suggestions to the central scheduling platform through encrypted hash digest.

[0062] The optimization scheduling module is used by the central scheduling platform to construct a multi-objective scheduling optimization model with weighted factors based on the scheduling suggestions of each node, the current status of the node, and the trust value. The global optimal scheduling instruction is generated by solving the optimization model.

[0063] A readable storage medium storing a computer program, which, when executed by a processor, implements a scheduling and control method for the distributed generation system.

[0064] The advantages of this invention compared to existing technologies are as follows: This invention provides a dispatch control scheme for distributed generation systems with multi-node collaborative prediction, trust assessment mechanism, dispatch suggestion fusion, and global optimal solution capabilities, which can improve system operating efficiency, prediction accuracy, and dispatch security; This invention, by deploying lightweight local prediction models at each distributed node, enables nodes to make short-term predictions of their own generation capacity and load demand, significantly improving the real-time performance and accuracy of predictions, reducing dependence on the central dispatch platform, and facilitating rapid response to local changes; Each node generates local dispatch suggestions based on local information, and the central dispatch platform comprehensively considers the dispatch suggestions, operating status, and historical trust values ​​of each node to construct a multi-objective optimization model with weighted factors, realizing the transformation from centralized command issuance to fusion-based collaborative optimization, improving the synergy of global dispatch, and thus meeting the core requirements of next-generation smart grid dispatch. Attached Figure Description

[0065] The present invention will be further described below with reference to the accompanying drawings:

[0066] Figure 1 This is a flowchart of the steps in the distributed generation system scheduling and control method of the present invention;

[0067] Figure 2 This is a control flowchart of the distributed generation system dispatch control device of the present invention;

[0068] Figure 3 This is a schematic diagram of the structure of the computer-readable storage medium of the present invention;

[0069] Figure 4 A schematic diagram of the structure of the computing device for dispatch control of distributed generation systems according to the present invention;

[0070] In the diagram, the numbers represent: 30 for optical disc, 40 for computing device, 401 for processing unit, 402 for system memory, 403 for bus, 404 for external device, 405 for I / O interface, 406 for network adapter, 4021 for random access memory, 4022 for cache memory, 4023 for storage medium, 4024 for program module, and 4025 for utility tool. Detailed Implementation

[0071] like Figure 1 As shown, the present invention provides a scheduling and control method for a distributed generation system, specifically including the following scheduling and control steps:

[0072] S1: Deploy edge computing modules at the nodes of distributed generation to collect the real-time operating status of each node and use a lightweight prediction model to make local predictions on the node's power generation capacity and load demand.

[0073] Preferably, the lightweight prediction model is based on the historical power generation data, current operating status data, user load curves and weather forecast information of the collection node, and performs local modeling and training to predict the power generation capacity and load demand of the node in real time within a set prediction period. Then, a dynamic feature selection mechanism is adopted to automatically filter key influencing factors and adjust model parameters according to the correlation between node type and historical power generation data.

[0074] Preferably, several lightweight prediction models are deployed on each node to form a model set. After each model runs independently, multiple predicted values ​​are obtained, represented as {y1, y2, ..., y...}. n Then, a weighted fusion method is used to integrate the prediction results of each model, and the weighted fusion prediction value is... The calculation formula is:

[0075] ;

[0076] Among them, w i For model M i The weights;

[0077] Furthermore, the confidence level of the final predicted value after fusing the prediction results of multiple different prediction models for the same node is calculated. Based on the dispersion of the predicted values ​​from multiple models, the standard deviation σ is estimated as the confidence level. The calculation formula is as follows:

[0078] ;

[0079] The final prediction output is: ;

[0080] Where z is the confidence interval coefficient, and if σ exceeds the preset tolerance threshold δ, it indicates that the prediction result is unstable;

[0081] The fault tolerance threshold δ is set based on the confidence level σ. If σ > δ, the current prediction is considered unreliable, and more node data can be requested, a backup scheduling strategy can be issued, or the current prediction output can be corrected based on the historical average.

[0082] The edge computing module calculates the prediction confidence σ of the fused output based on the dispersion of each prediction result. If the prediction confidence σ exceeds the preset fault tolerance threshold δ, the edge computing module will mark the prediction as a low-confidence output and trigger one of the following fault tolerance control strategies:

[0083] Request neighboring nodes to share their prediction results for collaborative correction;

[0084] Use historical average load forecasts as rollback values;

[0085] Automatically switch to the locally predefined backup policy number.

[0086] Each node, based on its current energy storage status, forecast results, and known scheduling objectives, automatically generates a scheduling suggestion that is locally optimal for the current scheduling cycle.

[0087] S2: Update the trust value of each node based on its prediction accuracy, scheduling response latency, communication stability, and instruction execution status in the previous scheduling cycle.

[0088] Preferably, the trust value updates of each node are calculated using the following weighted cumulative method:

[0089] ;

[0090] Among them, T i (t) is the trust value from the previous period, β is the memory factor used to balance historical values ​​with current performance, and S i (t) is the overall score for the current period;

[0091] After normalizing the key operational performance indicators of each distributed scheduling node (such as distributed power nodes, energy storage nodes, and load nodes), a weighted summation model is used to obtain a comprehensive score, calculated as follows:

[0092] ;

[0093] in, It is the prediction accuracy. It is the normalized value of the response delay. It's the communication success rate. It represents the success rate of executing instructions, and w1, w2, w3, and w4 are the importance weights of each dimension.

[0094] In one embodiment of the present invention, it is assumed that the current node N7 has the following historical data: T7(t)=0.78, prediction accuracy A7=0.87, response delay normalization R7=0.2, communication success rate C7=0.95, instruction execution success rate E7=0.92, and weights w1=0.3, w2=0.2, w3=0.25, w4=0.25;

[0095] Therefore, the corresponding comprehensive score can be obtained by calculating according to the above formula:

[0096] S7(t) = 0.8885;

[0097] Setting dynamic β:

[0098] β7 = 0.665;

[0099] The final updated trust value is:

[0100] T7(t+1)=0.8163;

[0101] Furthermore, in the next cycle of the scheduling optimization model, the weight of node N7 will participate in global scheduling based on the trust value T7(t+1)=0.8163; if the trust value continuously drops below the tolerance threshold (such as 0.6), it will be marked as a "low reliability node" and will not be used as a core scheduling input; the platform can trigger alarms or automatically redeploy the edge model based on the trust trend to repair performance.

[0102] S3: Each node generates locally optimized scheduling suggestions based on local forecast data, its own energy storage status, and scheduling objectives, and uploads the scheduling suggestions to the central scheduling platform through encrypted hash digest.

[0103] Preferably, based on historical power generation data sequences, nodes construct lightweight prediction models, and each node generates a local optimization problem based on local prediction data and scheduling objectives, outputting suggested power. and energy storage dispatch quantity ΔSOC i The expression for the optimization problem is:

[0104] ;

[0105] in, This is a penalty function for scheduling deviation, lag, etc., where a, b, and c are weighting coefficients. Let represent the predicted power generation capacity of node i at a future time t+τ. Let τ be the predicted load demand of node i at a future time t+τ, where τ is the scheduling delay.

[0106] Preferably, the node processes the encrypted data E. i A hash calculation is performed to generate a unique digest. The node then sends the uploaded content to the central scheduling platform. The expression for the uploaded content is:

[0107] ;

[0108] Among them, H i This is a summary used to centrally verify that the data has not been tampered with. i It is encrypted data, used by the center to decrypt and restore the original scheduling suggestion, ID. i The timestamp is used to indicate the source of the data and to record the upload time for scheduling time-series synchronization or subsequent tracing.

[0109] After receiving the uploaded content, the central scheduling platform performs integrity verification. If the hash verification fails, the scheduling suggestion is rejected and marked as data tampering or communication corruption.

[0110] Preferably, the following hash signature encryption method is used to upload the data to the central dispatch platform:

[0111] The SHA-256 hash H(S) is used to compute a digest of the scheduling suggestion data S.

[0112] Use the node's private key to digitally sign H(S) to generate the signature Sig(S);

[0113] The content uploaded by the node is: [Node ID, timestamp, S digest H(S), signature Sig(S)];

[0114] The central scheduling platform verifies signatures using node public keys to ensure that data has not been tampered with;

[0115] The central dispatch platform confirms the completeness of the suggested content by comparing the summary.

[0116] The central dispatch platform constructs a multi-objective optimization model with a trust factor based on the dispatch suggestions uploaded by each node, the current operating status of the node, and the trust value. It comprehensively considers: minimizing the system-level supply and demand balance error; the matching degree between the predicted values ​​of each node and the dispatch objectives; optimizing the utilization efficiency of energy storage devices; maximizing the overall dispatch success rate; and minimizing the collaborative operation cost between nodes.

[0117] Preferably, the specific method for each node to construct a lightweight prediction model based on its own historical data sequence and current environmental state is as follows:

[0118] Input historical power generation sequence, current meteorological parameters, and user load curves; adopt a multi-model integrated structure and perform weighted fusion prediction. Based on the prediction results, the nodes calculate the future net load and remaining available energy storage capacity, and input them into the optimization model.

[0119] Each node constructs the following local optimization problem based on the predicted data: Suggest the output power the node wants to output during the scheduling cycle; Energy storage charge / discharge quantity ΔSOC. i To compensate for discrepancies between power generation forecasting and load forecasting, and to optimize the dispatch power at nodes. It more closely aligns with the central platform's scheduling objectives; without energy storage, nodes passively execute scheduling; with energy storage, nodes can flexibly adjust, reducing the risks associated with prediction bias; the energy storage term in the objective function is used to constrain ΔSOC. i The magnitude, ultimately, ΔSOC i On the one hand, it is a regulating variable (adjusting output power), and on the other hand, it is a constraining variable (controlling energy storage safety), playing a dual role in balancing supply and demand and protecting the battery in the objective function;

[0120] The expression for constructing the objective function is:

[0121] ;

[0122] Among them, the scheduling deviation penalty function Dev measures the degree of deviation between the suggested power and the predicted supply and demand gap; the lag penalty function Lag controls the rate of change of power output to avoid sudden changes; and the SOC usage penalty SOCPenalty limits the frequent deep charging and discharging of energy storage to ensure battery life. It is the predicted power generation of node i. It is the actual execution power of node i in the previous scheduling cycle.

[0123] In another embodiment of the present invention, assuming that node N5 is a household energy storage + photovoltaic unit, the prediction result is as follows:

[0124] Forecasted photovoltaic power generation for the next 15 minutes: 2.8kW; Forecasted load: 3.5kW; Current battery SOC: 60%, allowable charge / discharge range: 40%~90%; Previous cycle dispatch output: 2.0kW; Node current optimization results are calculated as follows:

[0125] Input prediction error: Energy required = 0.7kW, Optimized output suggestion: Suggested output power = 3.0kW, Energy storage discharge error compensation = -2%;

[0126] Penalty function calculation illustration: Deviation penalty = 2.3, Lag penalty = 1.0, Energy storage penalty = 0.0004;

[0127] With α=0.5, β=0.3, and γ=0.2, the objective function score is:

[0128] The target value is 1.4501. The scheduling suggestion corresponding to this minimized result is submitted to the central scheduling platform.

[0129] Furthermore, assume that node N7 uploads the following data to the platform at t=10:00: scheduling suggestion = 2.4 kW, ΔSOC = -3%; raw data package D7 = {P, SOC, confidence = 0.92}; ciphertext E7 is obtained by RSA encryption; hash digest H7 = SHA-256(E7) is generated.

[0130] The uploaded data packet is:

[0131] {

[0132] "ID": "N_7",

[0133] "timestamp": "2025-04-23T10:00:00Z",

[0134] "hash": "81a213ef...97bc0",

[0135] "cipher": "<base64 encoded E_7> "

[0136] }

[0137] This data packet is generated and uploaded by node N7 at a specific time. It contains node ID, timestamp, encrypted data, and corresponding hash digest. The scheduling suggestions and status information are encrypted using RSA to form a cipher, and the hash field is the SHA-256 digest of the cipher, which is used by the platform to verify whether the data has been tampered with. Through this mechanism, the uploaded data is ensured to have secure and trustworthy characteristics such as traceability of source, immutability of content, and information encryption protection, supporting the central platform to effectively receive and reliably decide on the scheduling suggestions of each node.

[0138] Central platform verification results: If Hash(E7′) = H7, verification is successful → proceed to decryption and scheduling; if the hash does not match → data is rejected, and the platform issues a notification: the data uploaded by node N7 failed the integrity verification, and has been tampered with or damaged, and the scheduling suggestion has been rejected.

[0139] S4: The central dispatch platform constructs a multi-objective dispatch optimization model with weighted factors based on the dispatch suggestions of each node, the current status of the node, and the trust value. It generates the globally optimal dispatch instruction by solving the optimization model and dispatches the distributed generation system based on the instruction.

[0140] Preferably, the model, by setting different objective functions and combining node trust values ​​with various system constraints, forms a multi-objective optimization problem. Each objective function has a corresponding weight factor, which is dynamically adjusted according to the actual operating requirements of the system. The objective functions are:

[0141] ;

[0142] Among them, f i (x) is the i-th objective function, w i It is the weight factor of the i-th objective function;

[0143] The multi-objective optimization model is solved by an optimization algorithm. In each iteration, the objective function value is evaluated based on the current solution, and the control variables are adjusted to achieve the global optimum. During the optimization process, the algorithm will gradually adjust the power generation and energy storage status of each node until a solution that minimizes the objective function is found. Once the optimization model finds the optimal solution, the central scheduling platform will generate specific scheduling instructions based on the solution results. The scheduling instructions include the power generation requirements of each node, the energy storage status adjustment requirements of each node, and other control instructions within the scheduling cycle.

[0144] In another embodiment of the present invention, the system initializes the scheduling scheme of each node using a heuristic algorithm. Assuming there are 10 nodes, the objective function for optimization is calculated as follows: Assuming the system demand is 50kW and the predicted power generation of the nodes is 48kW, the deviation is f1(x) = (50-48). 2 =4 kW 2 If the node response delay is 1 second, the lag f2(x) = 1; the scheduling impact is penalized according to the node trust value. Assuming that the trust value of node 3 is T3 = 0.7, its impact on scheduling will be weakened.

[0145] Assuming a node with a low trust value, T3 = 0.7, its objective function weights will be adjusted according to the formula. Assuming w3 is initially 0.25, after adjustment, when the trust value is low, w3 will be reduced to 0.2, minimizing the node's impact on the global optimization. Using a heuristic algorithm, each iteration of the optimization process evaluates the objective function value based on the current solution and adjusts the control variables. After each iteration, the optimal solution is gradually optimized while minimizing the objective function. This process includes: updating the power generation and energy storage state of each node based on the current solution; evaluating the objective function value of the new solution and comparing whether there is a better solution; if the node's trust value is low, the optimization algorithm will give it less adjustment weight to avoid unreliable nodes having an excessive impact on scheduling.

[0146] Once the optimal solution is found, the central scheduling platform will generate specific scheduling instructions based on the solution results. The generated instruction format is as follows:

[0147] {

[0148] "node_id": "Node_1",

[0149] "schedule": {

[0150] "P_cmd": 3.5,

[0151] "SOC_delta": -1.2,

[0152] "valid_time": "2025-04-23T14:00:00Z - 14:15:00Z",

[0153] "confidence": 0.95

[0154] }

[0155] }

[0156] This scheduling instruction is generated by the central scheduling platform after optimization. It explicitly specifies that the power output instruction P_cmd to be executed by node "Node_1" within a specific time period is 3.5 kW, the energy storage adjustment amount SOC_delta is -1.2%, and it comes with a prediction confidence of 0.95. This is used to guide the node to operate according to the plan and to evaluate the reliability of the instruction, so as to ensure the accuracy and stability of scheduling execution.

[0157] As shown in Table 1 below, through multi-node scheduling tests in a simulated environment, it can be seen that the system using this multi-objective optimization method exhibits significant advantages compared with the traditional scheduling method.

[0158] Table 1. Comparison of performance data obtained by different scheduling methods.

[0159]

[0160] This invention effectively improves the reliability and robustness of system scheduling by dynamically adjusting the influence of each node in the scheduling process; and effectively improves the efficiency of scheduling by quickly obtaining the global optimal solution through a heuristic optimization algorithm.

[0161] like Figure 2 As shown, to implement the above-mentioned scheduling and control method, the present invention also provides a scheduling and control device for a distributed generation system, which includes:

[0162] The data acquisition and prediction module is used to deploy edge computing modules at distributed generation nodes to collect the real-time operating status of each node and make local predictions of the node's short-term generation capacity and load demand based on a lightweight prediction model.

[0163] The dynamic trust assessment module is used to update the trust value of each node based on its prediction accuracy, scheduling response latency, communication stability, and instruction execution status in the previous scheduling cycle.

[0164] The scheduling suggestion generation module is used by each node to generate local optimization scheduling suggestions based on local forecast data, its own energy storage status and scheduling objectives, and upload the scheduling suggestions to the central scheduling platform through encrypted hash digest.

[0165] The optimization scheduling module is used by the central scheduling platform to construct a multi-objective scheduling optimization model with weighted factors based on the scheduling suggestions of each node, the current status of the node, and the trust value. The global optimal scheduling instruction is generated by solving the optimization model.

[0166] Furthermore, the present invention also provides a readable storage medium for dispatch control of distributed generation systems, such as... Figure 3 As shown, the computer-readable storage medium is an optical disc 30, on which a computer program (i.e., a program product) is stored. When the computer program is run by the processor, it implements the steps described in the above method implementation. For example, it deploys an edge computing module at the distributed generation node, collects the real-time operating status of each node, and performs local prediction of the node's short-term generation capacity and load demand based on a lightweight prediction model; it updates the trust value of each node based on the prediction accuracy, scheduling response latency, communication stability, and instruction execution status of each node in the previous scheduling cycle; each node generates locally optimized scheduling suggestions based on local prediction data, its own energy storage status, and scheduling objectives, and uploads the scheduling suggestions to the central scheduling platform through encrypted hash digest; the central scheduling platform constructs a multi-objective scheduling optimization model with weighted factors based on the scheduling suggestions of each node, the current status of the node, and the trust value, and generates a globally optimal scheduling instruction by solving the optimization model; the specific implementation of each step will not be repeated here.

[0167] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media.

[0168] Furthermore, to realize the scheduling and control of distributed generation systems, this invention also provides corresponding computing devices, such as... Figure 4 The block diagram shown is of an exemplary computing device 40 for implementing an embodiment of the present invention. The computing device 40 may be a computer system or a server.

[0169] The components of the computing device 40 may include, but are not limited to: one or more processors or processing units 401, system memory 402, and bus 403 connecting different system components (including system memory 402 and processing unit 401).

[0170] The computing device 40 typically includes a variety of computer system readable media, which can be any available media that can be accessed by the computing device 40, including volatile and non-volatile media, removable and non-removable media.

[0171] System memory 402 may include computer system readable media in the form of volatile memory, such as random access memory 4021 (RAM) and / or cache memory 4022. Computing device 40 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage medium 4023 (ROM) may be used to read and write non-removable, non-volatile magnetic media (…). Figure 4 (Not shown in the image, usually referred to as "hard drive"). Although not shown in... Figure 4The diagram illustrates that disk drives for reading and writing to removable non-volatile disks (e.g., "floppy disks") and optical disc drives for reading and writing to removable non-volatile optical discs (e.g., CD-ROMs, DVD-ROMs, or other optical media) can be provided. In these cases, each drive can be connected to bus 403 via one or more data media interfaces. System memory 402 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention. A program / utility 4025 having a set (at least one) of program modules 4024 may be stored, for example, in system memory 402, and such program modules 4024 include, but are not limited to, operating systems, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment. Program modules 4024 generally perform the functions and / or methods of the embodiments described in the present invention.

[0172] The computing device 40 can also communicate with one or more external devices 404 (such as a keyboard, pointing device, display, etc.). This communication can be performed through the input / output I / O interface 405. Furthermore, the computing device 40 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 406. The network adapter 406 communicates with other modules of the computing device 40 (such as the processing unit 401, etc.) via a bus 403. Although... Figure 4 As not shown, it can be used in conjunction with computing device 40 with other hardware and / or software modules.

[0173] The processing unit 401 executes various functional applications and data processing by running programs stored in the system memory 402. For example, it deploys edge computing modules at distributed generation nodes to collect the real-time operating status of each node and performs local predictions of the node's short-term generation capacity and load demand based on a lightweight prediction model. It updates the trust value of each node based on its prediction accuracy, scheduling response latency, communication stability, and instruction execution status in the previous scheduling cycle. Each node generates locally optimized scheduling suggestions based on local prediction data, its own energy storage status, and scheduling objectives, and uploads the scheduling suggestions to the central scheduling platform through encrypted hash digest. The central scheduling platform constructs a multi-objective scheduling optimization model with weighted factors based on the scheduling suggestions of each node, the current status of the node, and the trust value, and generates the globally optimal scheduling instruction by solving the optimization model.

[0174] This invention proposes a distributed scheduling architecture in which each node has local intelligent modeling and autonomous scheduling capabilities. The overall system achieves high-efficiency optimization through "node autonomy + central collaboration". This structure significantly improves the system's flexibility and scalability, and is suitable for distributed energy and microgrid scenarios.

[0175] This invention deploys a lightweight prediction model at each node, which performs local modeling and real-time prediction based on historical power generation data, current operating status, user load curves and weather information. This approach processes data and generates suggestions in advance at the edge, which greatly reduces the computational pressure on the central platform and improves the real-time performance and accuracy of scheduling decisions.

[0176] This invention introduces a node trust value mechanism. Each node dynamically calculates its trust value based on dimensions such as prediction accuracy, response latency, communication success rate, and instruction execution rate. The central platform constructs a multi-objective optimization scheduling model with weighted factors based on the node trust value, current status, and their respective scheduling suggestions, thereby achieving refined and differentiated scheduling and making scheduling control more intelligent.

[0177] Regarding data security, this invention provides a scheduling data reporting method that combines encryption and hash digest. When each node uploads a suggestion, it first encrypts the data and generates a unique digest value for integrity verification. After receiving the data, the central platform first performs hash verification on the digest. If any abnormal data is found, the data is rejected and the source of the abnormality is marked. This mechanism significantly enhances the reliability of scheduling data and the system's resistance to attacks, and has high practical value.

[0178] Regarding the stability assurance of scheduling prediction, this invention improves the robustness of prediction through a multi-model fusion mechanism, while calculating the confidence level of the prediction results. If the prediction is unreliable, it automatically falls back to the backup strategy or requests data support from other nodes to ensure the steady-state operation of the system and has fault tolerance and prediction quality discrimination mechanisms.

[0179] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dispatch control of a distributed power generation system, characterized by: The scheduling control step includes the following steps: S1: deploying an edge computing module at each node of the distributed power generation system to collect real-time running states of the nodes, and then using a lightweight prediction model to locally model and train based on historical power generation data, current running state data, user load curve, and weather prediction information of the nodes to predict power generation capacity and load demand of each node within a set prediction period; A dynamic feature selection mechanism is used to automatically select key influencing factors and adjust model parameters according to the correlation between node types and historical power generation data; S2: updating the trust value of each node according to the prediction accuracy, scheduling response time delay, communication stability, and instruction execution of each node in the last scheduling period; S3: each node generates a locally optimized scheduling suggestion based on local prediction data, its own energy storage status, and scheduling targets, and uploads the scheduling suggestion to the central scheduling platform through an encrypted hash digest method; S4: the central scheduling platform constructs a multi-objective scheduling optimization model with a weight factor based on the scheduling suggestions of each node, the current state and trust value of the node, solves the optimization model to generate a globally optimal scheduling instruction, and schedules the distributed power generation system based on the instruction.

2. The method of claim 1, wherein: The specific method for predicting the power generation capacity and load demand of each node within a set prediction period in step S1 is as follows: A plurality of lightweight prediction models are deployed on each node to form a model set. After each model is independently run, a plurality of prediction values {y1, y2,..., y n} are obtained. Then, the prediction results of each model are integrated using a weighted fusion method to obtain a weighted fusion prediction value . The calculation formula is as follows: ； where w i are the weights of the model M i . Further calculate the confidence of the final prediction value after fusing the prediction results of multiple different prediction models for the same node, and estimate the standard deviation as the prediction confidence σ according to the dispersion degree of the prediction values of multiple models, the calculation formula is: ； The final prediction output is: ; Where z is the confidence interval coefficient, if σ exceeds the preset tolerance threshold δ, it means that the prediction result is unstable; Set a fault tolerance threshold δ according to the prediction confidence σ, if σ> δ, it is considered that the current prediction is not reliable, and more node data needs to be requested, a backup scheduling strategy needs to be issued, or the current prediction output needs to be corrected based on the historical average value.

3. The method of claim 2, wherein: The specific method for selecting key influencing factors and adjusting model parameters in step S1 is as follows: The edge computing module calculates the prediction confidence σ of the fused prediction results according to the dispersion degree of each prediction result, if the prediction confidence σ exceeds the preset fault tolerance threshold δ, the edge computing module will mark the prediction result as a low-confidence output, and trigger one of the following fault control strategies: Request surrounding nodes to share their prediction results for collaborative correction; Enable historical average load prediction as a rollback value; Automatically switch to a locally predefined backup strategy number; Finally, each node automatically generates a locally optimal scheduling suggestion for the current scheduling period based on its current energy storage status, prediction result, and known scheduling target.

4. The method of claim 3, wherein: The specific method for updating the trust value of each node in step S2 is as follows: The trust value of each node is updated in the following weighted cumulative form: ； where T i (t) is the trust value of the last period, β is the memory factor, used to balance the historical value with the current performance, S i (t) is the comprehensive score of the current period; After normalizing the key performance indicators of each distributed scheduling node, a weighted addition model is used to obtain a comprehensive score, the calculation formula is: ； wherein, is a prediction accuracy rate, is a normalized value of a response time delay, is a communication success rate, is a success rate of executing an instruction, w1, w2, w3, w4 are importance weights of respective dimensions.

5. The method of claim 4, wherein: The specific method for generating a locally optimized scheduling suggestion in step S3 is as follows: Each node constructs a lightweight prediction model based on its own historical data sequence and current environment state, and each node generates a local optimization problem based on local prediction data and scheduling target, and outputs a recommended power and the energy storage scheduling amount ΔSOC i The expression of the optimization problem is: ; wherein, is a penalty function for the dispatch deviation, hysteresis, a, b, c are weight coefficients, is the predicted generation capacity of node i at future time t+τ, is the predicted load demand of node i at future time t+τ, τ is the dispatch delay.

6. The method of claim 5, wherein: The specific method for constructing a lightweight prediction model is as follows: The input history power generation sequence, current meteorological parameters, and user load curve are input into the optimization model, and a multi-model integration structure is adopted to perform weighted fusion prediction. Each node constructs the following local optimization problem as a target function according to the prediction data: The suggested output power node is the power that the node wants to output in the scheduling period, and the energy storage charging and discharging amount ΔSOC is the amount of energy storage charging and discharging i The deviation between power generation prediction and load prediction is compensated, and the scheduling power of the node is The scheduling target of the central platform is closer, and the node is passively executed in the scheduling without energy storage, and is flexibly adjusted with energy storage. The energy storage term in the objective function is used to constrain ΔSOC i The amplitude of ΔSOC i Adjust the output power and control the energy storage safety, and the expression of the objective function is: ； where Dev is the dispatch deviation penalty function, which measures the deviation degree of the recommended power from the predicted supply-demand gap; Lag is the hysteresis penalty function, which controls the rate of power output change to avoid sudden changes; and SOCPenalty is the SOC usage penalty, which limits the frequent deep charging and discharging of the energy storage to ensure the battery life. is the predicted power generation of node i, is the actual execution power of node i in the last dispatch cycle.

7. The method of claim 6, wherein: The specific method for uploading the scheduling suggestion to the central scheduling platform in step S3 is as follows: The node performs hash calculation on the encrypted data to generate a unique digest, and the node sends the uploaded content to the central scheduling platform, and the expression of the uploaded content is as follows: ； Wherein, H i is the digest for the center to verify that the data has not been tampered with, E i is the ciphertext data for the center to decrypt to restore the original scheduling suggestion, ID i is the data source, and timestamp is the time when the record is uploaded, used for scheduling timing synchronization or subsequent tracing. The following hash signature encryption method is used to upload to the central scheduling platform: An SHA-256 is used to calculate the digest H(S) of the scheduling suggestion data S; The node private key is used to digitally sign H(S) to generate a signature Sig(S); The node uploads the content as follows: [node ID, timestamp, S digest H(S), signature Sig(S)]; The central scheduling platform verifies the signature by using the node public key, and confirms the integrity of the suggestion content according to the digest comparison. If the hash verification fails, the scheduling suggestion is rejected, and is marked as data tampering or communication damage; The central scheduling platform constructs a multi-objective optimization model with a trust factor according to the scheduling suggestions uploaded by each node, the current node running state and the trust value.

8. The method of claim 7, wherein: The specific method for generating the global optimal scheduling instruction in step S4 is as follows: The optimization model forms a multi-objective optimization problem by setting different target functions, combining the node trust value and the constraints of the system, each target function has a corresponding weight factor, the weight factor is dynamically adjusted according to the actual running demand of the system, and the expression of the target function is as follows: ； wherein f i (x) is the ith objective function, w i is the weight factor of the ith objective function; The multi-objective optimization model is solved by an optimization algorithm, the target function value is evaluated according to the current solution at each iteration, and the control variable is adjusted to achieve the global optimal target. During the optimization process, the algorithm gradually adjusts the power generation of each node and the energy storage state until a solution that minimizes the target function is found. Once the optimal solution of the optimization model is solved, the central scheduling platform generates specific scheduling instructions according to the solution, and the scheduling instructions include the power generation requirement of each node, the energy storage state adjustment requirement of each node and the control instruction in the scheduling period.

9. A control device for implementing a dispatch control method of a distributed power generation system as claimed in claim 1, characterized by: The control module includes the following: The prediction module is used to deploy an edge computing module at the node of distributed power generation, collect real-time running state data of each node, and perform local prediction on the power generation capacity and load demand of the node based on a lightweight prediction model; The dynamic trust evaluation module is used to update the trust value of each node according to the prediction accuracy, scheduling response time delay, communication stability and instruction execution of the node in the last scheduling period; The scheduling suggestion generation module is used to generate a local optimization scheduling suggestion based on the local prediction data, the energy storage state of the node and the scheduling target, and upload the scheduling suggestion to the central scheduling platform by using an encryption hash digest method; The optimization scheduling module is used to construct a multi-objective scheduling optimization model with a weight factor based on the scheduling suggestion of each node, the current state and the trust value of the node by the central scheduling platform, and generate a global optimal scheduling instruction by solving the optimization model.

10. A readable storage medium characterized by: The readable storage medium has stored thereon a computer program, and the computer program is executed by the processor to implement the method according to any one of claims 1 to 8.

Images