Intelligent control method and system for power supply system

Abstract

The present application relates to the technical field of artificial intelligence control of power supply systems, and in particular to an intelligent control method and system for power supply systems. The present application collects data through sensors calibrated periodically, processes the data through a dynamic weighted fusion algorithm to obtain space-time characteristics, and predicts the working conditions based on the characteristics output through a reinforcement learning prediction model. A decision model generates a control strategy accordingly, and the instructions generated by the decomposed control strategy are sent to the execution nodes after the power verification of the computing power, and the clock is kept synchronized. The system calculates the execution state representation value according to the execution state feedback, dynamically determines the control state of the power supply system, and then adjusts the power verification period and the clock alignment period of the execution nodes based on the control state, realizes the closed-loop optimization of system resource configuration and synchronization accuracy, and improves the control robustness and efficiency. The present application improves the precision of adaptive control of power supply systems.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence control of power supply systems, and in particular to an intelligent control method and system for power supply systems. BACKGROUND

[0002] With the popularization of new energy power generation technology and the increasing demand for power supply reliability of industrial loads and intelligent terminals, the operating conditions of power supply systems are becoming increasingly complex. The existing control methods of power supply systems mainly have the following defects:

[0003] 1. Control strategy is fixed: mostly using preset threshold or traditional PID control algorithm, relying on manual experience to set parameters, unable to adapt to power input fluctuations and load dynamic changes, resulting in poor power supply stability and low energy utilization efficiency.

[0004] 2. Data utilization is insufficient: the multi-dimensional data generated during the operation of the power supply system is not effectively fused and analyzed, which cannot accurately predict the trend of operating conditions and cannot realize forward-looking control.

[0005] 3. Fault diagnosis lags behind: traditional fault detection relies on hardware redundancy design, which can only alarm and protect after the fault occurs, and cannot identify potential fault hazards in advance, resulting in high system downtime probability and high maintenance cost.

[0006] 4. Poor cross-device collaboration: each unit in the distributed power supply system is controlled independently, lacking a collaborative scheduling mechanism based on global optimization goals, resulting in low overall operating efficiency and inability to adapt to the collaborative operation requirements of microgrids and smart grids.

[0007] In the prior art, although some attempts have been made to apply intelligent technology to power control, most of them are limited to simple adaptation of a single algorithm, and have not formed a full-link intelligent system of "data collection-fusion analysis-intelligent decision-making-collaborative execution-fault self-healing", and have not fully combined computing resource optimization and real-time data processing technology, making it difficult to meet the high-precision and high-reliability control requirements of complex power supply systems.

[0008] Chinese Patent Publication No. CN119906277A discloses a power supply intelligent management control system, relating to the field of power management, which includes a power supply processing module for accessing and transmitting alternating current energy and performing step-down, rectification filtering, inversion and voltage conversion processing; a current sampling module for current sampling control, signal conversion and amplification filtering; a voltage comparison module for setting an overvoltage threshold and voltage comparison and performing shunt degree detection; a power supply processing module for converting and amplifying filtering processing of the sampled signals after shunting, rectification and storage; an intelligent control module for signal reception and module control; an electric energy adjustment module for voltage adjustment; an output processing module for output filtering and electric energy superposition processing and power supply for the output module.

[0009] Therefore, the prior art has the following problems: the power system control method has poor adaptability, low data utilization rate and poor collaboration, resulting in the problem that the power system cannot accurately perform adaptive control. SUMMARY

[0010] To this end, the present application provides an intelligent control method and system for a power system to overcome the defects of the prior art power system control method, such as poor adaptability, low data utilization rate and poor collaboration, resulting in the problem that the power system cannot accurately perform adaptive control.

[0011] To achieve the above-mentioned purpose, the present application provides an intelligent control method for a power system, comprising:

[0012] Real-time collection of multi-dimensional operation data by sensors deployed at key nodes of the power system, and preprocessing of the collected multi-dimensional operation data, wherein the sensors are periodically calibrated;

[0013] Temporal and spatial fusion of the preprocessed multi-dimensional operation data based on an improved fusion algorithm to obtain fusion feature data, wherein the improved fusion algorithm is obtained by dynamically adjusting the weight coefficients of each data source;

[0014] Inputting the fusion feature data into a periodically updated reinforcement learning prediction model to obtain a working condition prediction result of the power system within a predetermined time period;

[0015] Based on the working condition prediction result and the predetermined control target, an adaptive control strategy is generated using a reinforcement learning decision model;

[0016] The adaptive control strategy is decomposed into execution instructions, and based on the data amount used by the execution instructions, it is issued to the execution nodes with periodic verification of residual computing power through a dual-mode communication method, and the execution state data of each execution node is fed back in real time, wherein the clock of the execution node is periodically aligned;

[0017] The control state of the power system is determined based on the execution state representation value calculated based on the execution state data of each execution node, and the verification period of the residual computing power of the execution node is adjusted based on the control state, and the alignment period of the clock of the execution node is adjusted based on the control state after adjusting the verification period.

[0018] Further, the process of determining the control state of the power supply system based on the execution state data of each execution node includes: calculating the instruction response accuracy and the response time compliance rate of each execution node respectively; calculating the average value of the instruction response accuracy of the plurality of execution nodes to obtain a first average value; calculating the average value of the response time compliance rate of the plurality of execution nodes to obtain a second average value; performing weighted summation on the first average value and the second average value to obtain the execution state characteristic value; comparing the execution state characteristic value with a preset characteristic value; if the execution state characteristic value is less than the preset characteristic value, determining that the control state of the power supply system is unqualified; and if the execution state characteristic value is greater than or equal to the preset characteristic value, determining that the control state of the power supply system is qualified.

[0019] Further, the method further includes: in the case where the control state of the power supply system is determined to be unqualified, monitoring the load rates corresponding to the plurality of execution nodes; counting the number of load rates greater than a preset load rate, and comparing the number with a preset number; and if the number is greater than the preset number, adjusting the verification period of the residual computing power of the execution node based on the difference between the number and the preset number.

[0020] Further, the process of adjusting the verification period of the residual computing power of the execution node based on the difference between the number and the preset number includes: reducing the verification period of the residual computing power of the execution node based on the difference between the number and the preset number, and the reduction amplitude of the verification period is proportional to the difference.

[0021] Further, the method further includes: in the case where the control state of the power supply system is unqualified after the verification period of the residual computing power of the execution node is adjusted, calculating the ratio of the second average value to a preset average value; and if the ratio is less than a preset ratio, adjusting the alignment period of the clock of the execution node based on the difference between the preset ratio and the ratio.

[0022] Further, the process of adjusting the alignment period of the clock of the execution node based on the difference between the preset ratio and the ratio includes: reducing the alignment period of the clock of the execution node based on the difference between the preset ratio and the ratio, and the reduction amplitude of the alignment period is proportional to the difference.

[0023] Further, the method further includes: in the case where the control state of the power supply system is unqualified after the alignment period of the clock of the execution node is adjusted, repeating the adjustment of the alignment period of the clock of the execution node at least once until the adjustment times is less than a preset number of times and the control state of the power supply system is qualified or the adjustment times is equal to the preset number of times; if the control state of the power supply system is unqualified after the adjustment is stopped, calculating the confidence of the reinforcement learning prediction model corresponding to a plurality of historical time points, and calculating the average value of the plurality of confidences; and if the average value is less than a preset average value, adjusting the training iteration number of the reinforcement learning prediction model based on the ratio of the average value to the preset average value.

[0024] Further, the process of adjusting the training iteration number of the reinforcement learning prediction model based on the ratio of the average value to the preset average value comprises: increasing the training iteration number of the reinforcement learning prediction model based on the ratio of the average value to the preset average value, and the increase amplitude of the training iteration number is inversely proportional to the ratio.

[0025] Further, the method further comprises: in the case that the control state of the power supply system after adjusting the training iteration number of the reinforcement learning prediction model is unqualified, calculating the difference between the execution state characteristic value corresponding to the plurality of historical moments and the preset characteristic value; calculating the variance of the difference, and if the variance is less than a preset variance, increasing the number of key nodes of data collection based on the ratio of the variance to the preset variance, and the increase amplitude of the number of key nodes is inversely proportional to the ratio.

[0026] To achieve the above-mentioned purposes, the present application provides an intelligent control system for a power supply system, which is used to implement the intelligent control method for a power supply system described in any one of the above-mentioned embodiments, comprising:

[0027] The acquisition unit is used to acquire multi-dimensional operation data in real time through sensors deployed at each key node of the power supply system, and pre-process the acquired multi-dimensional operation data, wherein the sensors are periodically calibrated;

[0028] The fusion unit is connected with the acquisition unit, and is used to perform space-time fusion on the pre-processed multi-dimensional operation data based on an improved fusion algorithm to obtain fusion feature data, wherein the improved fusion algorithm is obtained by dynamically adjusting the weight coefficients of each data source;

[0029] The prediction unit is connected with the fusion unit, and is used to input the fusion feature data into a periodically updated reinforcement learning prediction model to obtain a working condition prediction result of the power supply system within a preset time period;

[0030] The control strategy generation unit is connected with the prediction unit, and is used to generate an adaptive control strategy by using a reinforcement learning decision model based on the working condition prediction result and a preset control target;

[0031] The control unit is connected with the control strategy generation unit, and is used to decompose the adaptive control strategy into execution instructions, and under the dual-mode communication mode, the execution instructions are issued to the execution nodes with periodic verification of residual computing power, and the execution state data of each execution node is fed back in real time, wherein the clocks of the execution nodes are periodically aligned;

[0032] An analysis unit connected to the control unit is used to determine the control state of the power supply system based on the execution state representation value calculated based on the execution state data of each execution node, adjust the verification period of the remaining computing power of the execution node based on the control state of the power supply system, and adjust the alignment period of the clock of the execution node based on the control state of the power supply system after adjusting the verification period.

[0033] Compared with the prior art, the beneficial effects of the present application are that the present application collects data through the periodically calibrated sensor, obtains the space-time features through the dynamic weighted fusion algorithm, and outputs the working condition prediction based on the features through the reinforcement learning prediction model, and the decision model generates a control strategy accordingly, and the instructions generated by the decomposed control strategy are issued to the execution node after the computing power verification through the dual-mode communication, and the clock is kept synchronized, the system calculates the execution state representation value based on the execution state feedback, dynamically determines the control state of the power supply system, and then adjusts the computing power verification period and the clock alignment period of the execution node based on the control state, realizes the closed-loop optimization of system resource configuration and synchronization accuracy, and improves the control robustness and efficiency. The present application improves the precision of adaptive control of the power supply system.

[0034] Further, the present application determines the control state of the power supply system based on the comparison result of the execution state representation value and the preset representation value, which can more accurately determine the control state of the power supply system, so that subsequent adjustment can be more effectively performed according to the cause, thereby further improving the precision of adaptive control of the power supply system.

[0035] Further, the present application determines the cause of the unqualified control state of the power supply system based on the number of execution nodes corresponding to the load rate greater than the preset load rate, which can perform subsequent adjustment of related parameters based on more accurate causes, thereby further improving the precision of adaptive control of the power supply system.

[0036] Further, the present application reduces the verification period of the remaining computing power of the execution node based on the difference between the number and the preset number, which can avoid the situation that the response time deviation of the execution node increases due to the excessive processing tasks of part of the nodes, thereby further reducing the influence on the control state of the power supply system, and further improving the precision of adaptive control of the power supply system.

[0037] Further, the present application determines the cause of the unqualified control state of the power supply system based on the ratio of the second average value to the preset average value, which can more accurately determine the cause of the unqualified control state of the power supply system, so that subsequent adjustment can be more effectively performed based on the cause, thereby further improving the precision of adaptive control of the power supply system.

[0038] Further, the present application reduces the alignment period of the clock of the execution node based on the preset ratio and the difference of the ratio, can make the clocks of the execution nodes more synchronized, so that the control action is more coordinated, and the calculation of the response time is more accurate, thereby further improving the precision of the adaptive control of the power supply system.

[0039] Further, the present application determines the reason for the unqualified control state of the power supply system based on the average of the confidence of the reinforcement learning prediction model corresponding to the plurality of historical moments, can more accurately determine the reason, so that subsequent adjustment can be more effective, thereby further improving the precision of the adaptive control of the power supply system.

[0040] Further, the present application increases the training iteration number of the reinforcement learning prediction model based on the ratio of the average value and the preset average value, can make the reinforcement learning prediction model training more sufficient, so that the output result is more accurate, further improves the cooperativity, and thereby further improves the precision of the adaptive control of the power supply system.

[0041] Further, the present application adjusts the number of key nodes of data acquisition based on the variance of the difference between the execution state representation value corresponding to the plurality of historical moments and the preset representation value, can make the working condition covered by the data more comprehensive, make the data more diverse, so that the output result of the model is more accurate, and thereby further improves the precision of the adaptive control of the power supply system. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The structure schematic diagram of the intelligent control system for the power supply system of the embodiment of the present application;

[0043] Figure 2 The step flow chart of the intelligent control method for the power supply system of the embodiment of the present application;

[0044] Figure 3 The step flow chart of the determination based on the comparison result of the execution representation value and the preset representation value of the embodiment of the present application;

[0045] Figure 4 The step flow chart of the determination based on the control state after adjusting the alignment period of the clock of the execution node of the embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to make the purpose and advantages of the present application more clear and explicit, the present application is further described below in combination with the embodiments; it should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the present application.

[0047] The preferred embodiments of the present application will be described below with reference to the drawings. Those skilled in the art will appreciate that the embodiments are only used to explain the technical principles of the present application, and are not intended to limit the protection scope of the present application.

[0048] It should be noted that in the description of the present application, unless explicitly defined and limited, the terms "mounting", "connecting", "connecting" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be connected inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0049] Please refer to Figure 1 As shown in the figure, it is a structural schematic diagram of the intelligent control system of the power supply system according to the embodiment of the present application.

[0050] The system comprises a collection unit, a fusion unit, a prediction unit, a control strategy generation unit, a control unit and an analysis unit.

[0051] The collection unit is used to collect multi-dimensional operation data in real time through sensors deployed at key nodes of the power supply system, and to pre-process the collected multi-dimensional operation data, wherein the sensors are periodically calibrated;

[0052] The fusion unit is connected with the collection unit, and is used to perform space-time fusion on the pre-processed multi-dimensional operation data based on an improved fusion algorithm to obtain fusion feature data, wherein the improved fusion algorithm is obtained by dynamically adjusting the weight coefficients of each data source;

[0053] The prediction unit is connected with the fusion unit, and is used to input the fusion feature data into a periodically updated reinforcement learning prediction model to obtain a working condition prediction result of the power supply system within a preset time period;

[0054] The control strategy generation unit is connected with the prediction unit, and is used to generate an adaptive control strategy based on the working condition prediction result and a preset control target by using a reinforcement learning decision model;

[0055] The control unit is connected with the control strategy generation unit, and is used to decompose the adaptive control strategy into execution instructions, and to issue the execution instructions to execution nodes with periodic verification of residual computing power through a dual-mode communication mode based on the data amount used by the execution instructions, and to feed back execution state data of each execution node in real time, wherein the clock of the execution node is periodically aligned;

[0056] The analysis unit is connected with the control unit, which is used to determine the control state of the power system based on the execution state data calculated by each execution node, and adjust the verification period of the remaining computing power of the execution node based on the control state of the power system, and adjust the alignment period of the clock of the execution node based on the control state of the power system after adjusting the verification period.

[0057] Specifically, the key node refers to a physical location and a logical link in the power system which has a decisive influence on state awareness, safety and stability or control effectiveness. Its definition includes: being in a pivotal position in the system topology, carrying key state measurements, affecting the effectiveness of protection and control actions such as relay installation points and AGC / AVC execution endpoints, or being a weak link in the safe operation of the system such as heavy load equipment and harmonic source access points. Non-key nodes are auxiliary monitoring points with high data redundancy, less influence on global state inference and control objectives, or indirectly inferred through key node data. The definition needs to be determined comprehensively by combining power structure, operation history and simulation analysis.

[0058] Specifically, the multi-dimensional operation data includes power input parameters, power output parameters, energy storage unit state parameters, load characteristic parameters, environmental parameters and equipment state parameters.

[0059] Specifically, the preprocessing includes data cleaning, data normalization and data completion, wherein the data cleaning adopts 3σ criterion to remove abnormal data; the data normalization adopts min-max standardization method to map the data to the interval [0, 1]; the data completion adopts an LSTM-based missing data completion algorithm to fill in the missing time series data.

[0060] Specifically, the improved weighted Bayesian fusion algorithm first standardizes the multi-dimensional operation data after preprocessing to form a time-aligned spatial data matrix; the algorithm assigns weights to each data source through a dynamic weight coefficient model, and the weights are adjusted online according to the real-time confidence of the sensor and the spatio-temporal correlation. In the Bayesian framework, the multi-source data is regarded as an independent observation of the system state, the influence strength of each data source likelihood function is modulated by dynamic weights, the posterior probability is updated recursively, the asynchronous and heterogeneous observation data in space-time dimension are fused, and the spatio-temporal fused feature data, i.e. the state estimation vector, is output.

[0061] Specifically, the reinforcement learning prediction model adopts a DQN (Deep Q Network) improved model, and the improvement point of the model is reward design, that is, the cross-entropy loss is used for the classification task, the mean square error is used for the regression task, and the total reward is a weighted combination of the two. By dividing the power system working condition into four types of stable working condition, input fluctuation working condition, load mutation working condition and fault precursor working condition, the fusion feature data is taken as the input, and the working condition type and the predicted value of the key parameter are taken as the output. Among them, the problem modeling specifically includes: state space, that is, taking the fusion feature data (such as voltage, current, frequency, power and the like historical sequence) as the state input; action space, that is, defining four working condition types (stable, input fluctuation, load mutation, fault precursor) and the predicted value of the key parameter (such as voltage deviation, frequency change rate and the like), the key parameter prediction is usually a continuous value, so the action space needs to be expanded into a continuous space or discretized; reward function, that is, designing the reward according to the prediction accuracy. This process is prior art, which will not be repeated here.

[0062] Specifically, the reinforcement learning decision model adopts a PPO (Proximal Policy Optimization) algorithm, takes the fusion feature data containing the current real-time state and the predicted working condition as the state input, adopts the PPO algorithm framework, and the strategy network is responsible for outputting the control parameters in the continuous action space, including voltage adjustment parameters, current adjustment parameters, energy storage charging and discharging parameters and load distribution parameters; the execution effect of these actions is evaluated in real time by a weighted sum reward function. The PPO algorithm evaluates the state value and the preset control target through its value network, and uses mechanisms such as Clipped Surrogate Objective to constantly optimize the strategy network parameters under the premise of ensuring the stability of the strategy update step, so that the intelligent agent, that is, the decision model, can dynamically and adaptively adjust the control parameter sequence, and finally generate an adaptive control strategy.

[0063] Please refer to Figure 2 It is a step flow chart of the intelligent control method for the power system according to the embodiment of the application.

[0064] The steps of the intelligent control for the power system are specifically as follows:

[0065] S1, the acquisition unit acquires multi-dimensional operation data in real time through sensors deployed at key nodes of the power system, and pre-processes the acquired multi-dimensional operation data, wherein the sensors are periodically calibrated;

[0066] S2, the fusion unit connected with the acquisition unit performs space-time fusion on the pre-processed multi-dimensional operation data based on an improved fusion algorithm to obtain fusion feature data, wherein the improved fusion algorithm is obtained by dynamically adjusting the weight coefficients of each data source;

[0067] S3, inputting the fusion feature data to a periodically updated reinforcement learning prediction model through a prediction unit connected to the fusion unit to obtain a working condition prediction result of the power supply system in a preset time period;

[0068] S4, generating an adaptive control strategy by a control strategy generation unit connected to the prediction unit based on the working condition prediction result and a preset control target by using a reinforcement learning decision model;

[0069] S5, decomposing the adaptive control strategy into execution instructions by a control unit connected to the control strategy generation unit, and issuing the execution instructions to execution nodes with periodic verification of residual computing power through a dual-mode communication mode based on the data amount used by the execution instructions, and feeding back execution state data of each execution node in real time, wherein the clock of the execution node is periodically aligned;

[0070] S6, determining the control state of the power supply system based on the execution state representation value calculated based on the execution state data of each execution node by an analysis unit connected to the control unit, and adjusting the verification period of the residual computing power of the execution node based on the control state of the power supply system, and adjusting the alignment period of the clock of the execution node based on the control state of the power supply system after adjusting the verification period.

[0071] Please refer to Figure 3 The figure is a step flow chart of the embodiment of the application based on the comparison result of the execution representation value and the preset representation value.

[0072] Specifically, taking maintaining the stable operation of the regional micro-grid under new energy fluctuation as an example, and based on the hardware response limit of key devices such as photovoltaic inverters and energy storage converters, and the dynamic adjustment margin demand of safe operation of the power grid, and combined with the load fluctuation data, weather correlation curve and fault disturbance record obtained by subsequent corresponding preset or critical parameter value setting in historical operation analysis.

[0073] Specifically, the preset representation value L0=0.95, and the comparison process between the execution state representation value L and the preset representation value L0 is as follows:

[0074] If the execution state representation value L is less than the preset representation value L0, it is determined that the control state of the power supply system is unqualified;

[0075] If the execution state representation value L is greater than or equal to the preset representation value L0, it is determined that the control state of the power supply system is qualified.

[0076] Specifically, in the case of determining that the power system control state is unqualified, the load rates corresponding to the plurality of execution nodes are monitored; the plurality of load rates are compared with the preset load rate respectively; if the number of load rates greater than the preset load rate is greater than the preset number, it indicates that part of the node processing task is too heavy, which leads to the increase of the response time deviation of the execution node, and then affects the power system control state, and then the verification period of the remaining computing power of the execution node is adjusted based on the difference between the number and the preset number, wherein the preset difference P0 between the number and the preset number is 4, and the comparison process based on the difference P between the number and the preset number and the preset difference P0 is specifically as follows:

[0077] If the difference P between the number and the preset number is less than or equal to the preset difference P0, the verification period of the remaining computing power of the execution node is adjusted to 0.9 times of the original verification period, wherein the adjusted value is rounded up.

[0078] If the difference P between the number and the preset number is greater than the preset difference P0, the verification period of the remaining computing power of the execution node is adjusted to 0.73 times of the original verification period, wherein the adjusted value is rounded up.

[0079] Specifically, in the case of adjusting the verification period of the remaining computing power of the execution node, the ratio of the second average value to the preset value is calculated; if the ratio is less than the preset ratio, it indicates that the clocks of each execution node are not synchronized, which will lead to the deviation of the execution time point of the coordination control instruction on different nodes, so that the control action is not coordinated, and at the same time, it will also lead to the error of the calculation of the response time, so that the originally qualified response is judged as overtime or in advance, and then the alignment period of the clock of the execution node is adjusted based on the difference between the preset ratio and the ratio, wherein the preset difference Q0 between the preset ratio and the ratio is 0.13, and the comparison process based on the difference Q between the preset ratio and the ratio and the preset difference Q0 is specifically as follows:

[0080] If the difference Q between the preset ratio and the ratio is less than or equal to the preset difference Q0, the alignment period of the clock of the execution node is adjusted to 0.89 times of the original alignment period, wherein the adjusted value is rounded up.

[0081] If the difference Q between the preset ratio and the ratio is greater than the preset difference Q0, the alignment period of the clock of the execution node is adjusted to 0.71 times of the original alignment period, wherein the adjusted value is rounded up.

[0082] Please refer to Figure 4 It is a step flow chart for judging the control state after adjusting the alignment period of the clock of the execution node according to the embodiment of the application.

[0083] Specifically, in the case that the power system control state is unqualified after adjusting the alignment period of the clock of the execution node, the alignment period of the clock of the execution node is repeatedly adjusted at least once until the adjustment number is less than the preset number and the power system control state is qualified or the adjustment number is equal to the preset number; if the power system control state after stopping adjustment is unqualified, the confidence of the reinforcement learning prediction model corresponding to a plurality of historical moments is calculated, and the average value of a plurality of confidence values is calculated; if the average value is less than the preset average value, it means that the reinforcement learning prediction model is not fully trained, and the learned rule is more rough and less accurate, then the training iteration number of the reinforcement learning prediction model is adjusted based on the ratio of the average value to the preset average value, wherein the preset ratio R0 of the average value to the preset average value is 0.83, and the comparison process based on the ratio R of the average value to the preset average value and the preset ratio R0 is as follows:

[0084] If the ratio R of the average value to the preset average value is less than or equal to the preset ratio R0, the training iteration number of the reinforcement learning prediction model is adjusted to 2.4 times the original training iteration number, wherein the adjusted value is rounded up;

[0085] If the ratio R of the average value to the preset average value is greater than the preset ratio R0, the training iteration number of the reinforcement learning prediction model is adjusted to 1.7 times the original training iteration number, wherein the adjusted value is rounded up.

[0086] Specifically, in the case that the power system control state is unqualified after adjusting the training iteration number of the reinforcement learning prediction model, the difference between the execution state representation value corresponding to a plurality of historical moments and the preset representation value is calculated; the variance of the difference value is calculated, if the variance is less than the preset variance, it means that the working condition covered by the data is not comprehensive, so that the diversity of the data is insufficient, then the number of key nodes for data collection is adjusted based on the ratio of the variance to the preset variance, wherein the preset ratio T0 of the variance to the preset variance is 0.92, and the comparison process based on the ratio T of the variance to the preset variance and the preset ratio T0 is as follows:

[0087] If the ratio T of the variance to the preset variance is less than or equal to the preset ratio T0, the number of key nodes for data collection is adjusted to 2.5 times, wherein the adjusted value is rounded up;

[0088] If the ratio T of the variance to the preset variance is greater than the preset ratio T0, the number of key nodes for data collection is adjusted to 1.6 times, wherein the adjusted value is rounded up.

[0089] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after the changes or replacements will all fall within the protection scope of the present application.

[0090] The above only describes the preferred embodiments of the present application and is not intended to limit the present application; the present application can have various changes and variations for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An intelligent control method for a power supply system, characterized in that, include: Multi-dimensional operating data is collected in real time by sensors deployed at key nodes of the power system, and the collected multi-dimensional operating data is preprocessed, with the sensors being periodically calibrated. Based on the improved fusion algorithm, spatiotemporal fusion is performed on the preprocessed multi-dimensional running data to obtain fused feature data. The improved fusion algorithm is obtained by dynamically adjusting the weight coefficients of each data source. The fused feature data is input into a periodically updated reinforcement learning prediction model to obtain the power system's operating condition prediction results within a preset time period. Based on the predicted operating conditions and the preset control objectives, an adaptive control strategy is generated using a reinforcement learning decision model. The adaptive control strategy is decomposed into execution instructions, and based on the amount of data used by the execution instructions, they are sent to the execution nodes that periodically verify the remaining computing power through dual-mode communication, and the execution status data of each execution node is fed back in real time, wherein the clocks of the execution nodes are periodically aligned. The control state of the power system is determined by the execution state characterization value calculated based on the execution state data of each execution node, and the verification cycle of the remaining computing power of the execution node is adjusted based on the control state after the verification cycle is adjusted.

2. The intelligent control method for a power supply system according to claim 1, characterized in that, The process of determining the control state of the power system based on the execution state characterization values ​​calculated from the execution state data of each execution node includes: Calculate the instruction response accuracy and response time compliance rate for each execution node; Calculate the average of the instruction response accuracy of the multiple execution nodes to obtain a first average value; Calculate the average of the response time compliance rates of the multiple execution nodes to obtain a second average; The execution status characterization value is obtained by weighted summing of the first average value and the second average value. The execution status representation value is compared with the preset representation value; If the execution status characterization value is less than the preset characterization value, then the control status of the power system is determined to be unqualified. If the execution status characterization value is greater than or equal to the preset characterization value, then the control status of the power system is determined to be qualified.

3. The intelligent control method for a power supply system according to claim 2, characterized in that, The method further includes: If the control status of the power system is determined to be unqualified, monitor the load rate of multiple execution nodes. Count the number of items with a load rate greater than the preset load rate and compare the number with the preset number; If the quantity is greater than the preset quantity, the verification cycle of the remaining computing power of the execution node will be adjusted based on the difference between the quantity and the preset quantity.

4. The intelligent control method for a power supply system according to claim 3, characterized in that, The process of adjusting the verification cycle of the remaining computing power of the execution node based on the difference between the quantity and the preset quantity includes: The verification cycle of the remaining computing power of the execution node is reduced based on the difference between the quantity and the preset quantity, and the reduction in the verification cycle is proportional to the difference.

5. The intelligent control method for a power supply system according to claim 4, characterized in that, The method further includes: If the control status of the power system is unqualified after adjusting the verification cycle of the remaining computing power of the execution node, calculate the ratio of the second average value to the preset average value. If the ratio is less than the preset ratio, the clock alignment period of the execution node is adjusted based on the difference between the preset ratio and the ratio.

6. The intelligent control method for a power supply system according to claim 5, characterized in that, The process of adjusting the alignment period of the execution node's clock based on the difference between a preset ratio and the preset ratio includes: The clock alignment period of the execution node is reduced based on the difference between the preset ratio and the ratio, and the reduction in the alignment period is proportional to the difference.

7. The intelligent control method for a power supply system according to claim 6, characterized in that, The method further includes: If the control state of the power system is not qualified after adjusting the alignment period of the clock of the execution node, the alignment period of the clock of the execution node shall be adjusted at least once until the number of adjustments is less than the preset number and the control state of the power system is qualified, or the number of adjustments is equal to the preset number and the adjustment stops. If the control state of the power system after the adjustment is stopped is not qualified, calculate the confidence of the reinforcement learning prediction model corresponding to multiple historical time points, and calculate the average of multiple confidence scores; If the average value is less than the preset average value, the number of training iterations of the reinforcement learning prediction model will be adjusted based on the ratio of the average value to the preset average value.

8. The intelligent control method for a power supply system according to claim 7, characterized in that, The process of adjusting the number of training iterations of a reinforcement learning prediction model based on the ratio of the average value to a preset average value includes: The number of training iterations of the reinforcement learning prediction model is increased based on the ratio of the average value to the preset average value, and the increase in the number of training iterations is inversely proportional to the ratio.

9. The intelligent control method for a power supply system according to claim 8, characterized in that, The method further includes: If the control state of the power system is unqualified after adjusting the number of training iterations of the reinforcement learning prediction model, calculate the difference between the execution state representation value and the preset representation value corresponding to multiple historical moments. Calculate the variance of the difference. If the variance is less than the preset variance, increase the number of key data collection nodes based on the ratio of the variance to the preset variance. The increase in the number of key nodes is inversely proportional to the ratio.

10. An intelligent control system for a power supply system, used to implement the intelligent control method for a power supply system as described in any one of claims 1-9, characterized in that, include: The acquisition unit is used to collect multi-dimensional operating data in real time through sensors deployed at key nodes of the power system, and to preprocess the collected multi-dimensional operating data. The sensors are periodically calibrated. A fusion unit, connected to the acquisition unit, is used to perform spatiotemporal fusion on preprocessed multi-dimensional running data based on an improved fusion algorithm to obtain fused feature data. The improved fusion algorithm is obtained by dynamically adjusting the weight coefficients of each data source. A prediction unit, which is connected to the fusion unit, is used to input the fused feature data into a periodically updated reinforcement learning prediction model to obtain the power system's operating condition prediction results within a preset time period. A control strategy generation unit, which is connected to the prediction unit, is used to generate an adaptive control strategy based on the working condition prediction results and the preset control target using a reinforcement learning decision model. The control unit, which is connected to the control strategy generation unit, is used to decompose the adaptive control strategy into execution instructions, and send them to the execution nodes that periodically verify the remaining computing power based on the amount of data used by the execution instructions through dual-mode communication, and provide real-time feedback on the execution status data of each execution node, wherein the clocks of the execution nodes are periodically aligned. An analysis unit, connected to the control unit, is used to determine the control state of the power system based on the execution state characterization value calculated from the execution state data of each execution node, adjust the verification cycle of the remaining computing power of the execution node based on the control state of the power system, and adjust the clock alignment cycle of the execution node based on the control state of the power system after adjusting the verification cycle.

Images