An AI-based reactive power compensation optimization method and system

Abstract

The application relates to the technical field of power grid operation and maintenance, and discloses a reactive power compensation optimization method and system based on AI, which comprises the following steps: integrating four-dimensional features of traffic flow, power grid load, user preference and equipment health degree, constructing a space-time correlation graph model, realizing adaptive evolution of a topological structure through dynamic graph convolution and an attention mechanism, fusing space-time features of a graph neural network and BiGRU time sequence modeling, injecting an equipment health degree attenuation factor, generating a charging demand prediction result with uncertainty quantization, expanding an agent state space and improving a reward function, and embedding an equipment health degree constraint in a policy network and a proximal optimization. The application realizes real-time sensing of an equipment deterioration state, intelligently adjusts a reactive power compensation strategy, prolongs the service life of old equipment while ensuring voltage stability, realizes feedforward-feedback compound control and dynamic rescheduling, effectively suppresses a deviation accumulation effect, and ensures accurate matching between scheduling instructions and actual output.

Description

Technical Field

[0001] The present invention relates to the field of power grid operation and maintenance, and more specifically, to an AI-based reactive power compensation optimization method and system. Background Art

[0002] Urban distribution networks are responsible for providing a stable and reliable power supply to both residential and commercial users. However, as the number of urban power grid users continues to increase and the load structure becomes increasingly complex, optimizing reactive power compensation has become a key issue in distribution network operation and maintenance.

[0003] Traditional reactive power compensation methods usually rely on empirical rules and manual adjustments, which are difficult to adapt to the ever-changing load characteristics and easily lead to low voltage and power loss problems. In recent years, the rapid development of artificial intelligence (AI) technology has provided new ideas for solving reactive power compensation problems in distribution networks. Summary of the Invention

[0004] The present invention provides an AI-based reactive power compensation optimization method and system to solve the technical problems in related technologies, such as the traditional static scheduling method relying on empirical rules and manual adjustment, being difficult to adapt to the ever-changing load characteristics, and easily leading to low voltage and power loss, especially the impact of power output deviation of old equipment on scheduling.

[0005] The present invention provides an AI-based reactive power compensation optimization method, comprising the following steps:

[0006] S100 integrates four-dimensional features of traffic flow, grid load, user preferences, and device health to build a spatiotemporal correlation graph model. It achieves adaptive evolution of topology through dynamic graph convolution and attention mechanism.

[0007] S200, which integrates the spatiotemporal features of graph neural networks with BiGRU time series modeling, injects device health attenuation factors, and generates charging demand prediction results with quantified uncertainty;

[0008] S300, which expands the agent state space and improves the reward function, embeds device health constraints in the policy network and proximal optimization, and enables distributed decision-making that considers device reliability;

[0009] S400 performs global resource allocation in the cloud and multi-objective optimization at the edge, achieving cloud-edge collaborative solution through the ADMM algorithm and generating scheduling instructions with reliability margins;

[0010] S500, which integrates historical deviations, scheduling instructions, and equipment status data, uses a TCN-GAT hybrid model to predict future execution deviations and assess uncertainty;

[0011] S600 combines feedforward scheduling instructions with PID feedback compensation to dynamically adjust power output and implement device priority management, ensuring instruction security through encryption verification;

[0012] S700 monitors accumulated load deviations and triggers a rescheduling mechanism, building a multi-objective model that considers equipment reliability to achieve smooth switching between new and old instructions;

[0013] S800 aggregates real-time execution data, uses sliding window statistics and anomaly filtering mechanisms, recursively updates device health indicators and implements normalization processing.

[0014] Furthermore, in S100, the following steps are specifically included:

[0015] S110, node feature initialization: construct a multi-dimensional feature vector for each road node, covering traffic, power grid, user and device status information;

[0016] S120, edge weight calculation: construct an adjacency matrix based on traffic flow transfer probability to reflect the dynamic association of the road network;

[0017] S130, Spatiotemporal Attention Generation: Capturing spatiotemporal correlations through multi-head attention mechanism;

[0018] S140, dynamic graph convolution update: integrating spatiotemporal features to update graph structure weights;

[0019] S150, health feature normalization: normalize the device health indicators.

[0020] Furthermore, in S200, the following steps are specifically included:

[0021] S210, Spatiotemporal Graph Convolution Feature Extraction: Based on Dynamic Graph Perform multi-level spatial feature aggregation;

[0022] S220, Time Series Modeling: Temporal dependency modeling of historical load series;

[0023] S230, Reliability Attenuation Factor Injection: Adjust the prediction weight based on the equipment health;

[0024] S240, multimodal feature fusion: fusing spatiotemporal features with attenuation factors to generate the final prediction;

[0025] The final prediction is calculated as follows:

[0026] ;

[0027] in, represents the feature projection matrix, It represents the historical mean value for the same period. Indicates time The charging demand forecast value, Representation device At the moment The hidden state, Indicates the total number of devices, Representation device Reliability attenuation factor;

[0028] S250, Uncertainty Quantification: Evaluating confidence intervals for predictions.

[0029] Furthermore, in S300, the following steps are specifically included:

[0030] S310, state space expansion: incorporating device health indicators into the agent's observed state;

[0031] S320, Reliability Weighted Reward Design: Introducing a device health penalty term into the basic reward function;

[0032] S330, the policy network improvement means injecting reliability attention into the policy network output layer;

[0033] S340, Proximal Policy Optimization: Policy Gradient Update with Reliability Constraints;

[0034] S350, distributed policy execution: each agent generates actions based on local observations;

[0035] The calculation formula of action is as follows:

[0036] ;

[0037] in, represents the execution action of device i at time t, represents the action space, represents the optimized policy network, Indicates the current state of device i, represents the health index of device i, represents the temperature coefficient, Represents the optimized strategy parameters.

[0038] Furthermore, in S400, the following steps are specifically included:

[0039] S410, Cloud-based Global Resource Allocation: Cross-regional power capacity allocation and transmission loss optimization;

[0040] S420, Edge Multi-Objective Optimization: Balancing User Costs and Grid Stability under Equipment Reliability Constraints;

[0041] S430, Constraint Consistency Coordination: Decoupling cloud-edge optimization through the ADMM algorithm;

[0042] S440, robust margin calculation: dynamically adjusts the power margin based on device health;

[0043] S450, generating a scheduling instruction: generating a scheduling instruction set with a reliability identifier;

[0044] The calculation formula for the scheduling instruction set is as follows:

[0045] ;

[0046] in:

[0047] ;

[0048] in, Represents the scheduling instruction set, Indicates the total number of devices, represents the final dispatch power of device k, Represents the health indicator of device k, represents the safety margin of device k, represents the locally calculated power value, Indicates the power value of cloud computing, Indicates the local power weight coefficient.

[0049] Furthermore, in S500, the following steps are specifically included:

[0050] S510, multi-source data fusion: integrating equipment historical data, dispatch instructions, and health indicators;

[0051] S520, temporal feature extraction means using deep TCN to capture long-range dependency patterns;

[0052] S530, spatial correlation modeling means capturing the spatial correlation between devices through a graph attention network;

[0053] S540, Cross-modal Fusion Prediction: Fusion of spatiotemporal features to generate biased predictions;

[0054] The calculation formula for the deviation prediction is as follows:

[0055] ;

[0056] in, represents the fused projection matrix, represents the feature concatenation operation, represents the bias term, represents the forecast deviation at time t+1, Representation layer normalization operation, Represents the time series characteristics, Represents spatial features;

[0057] S550, Uncertainty Quantification: Evaluating the properties of probability distributions of predicted values.

[0058] Furthermore, in S600, the following steps are specifically included:

[0059] S610, feedforward control item generation: generating a reference control variable based on the robust scheduling instruction;

[0060] S620, feedback compensation term calculation: dynamically adjust the control amount according to the real-time execution deviation;

[0061] S630, dynamic limiting processing: combining equipment reliability index to constrain control output;

[0062] S640, device priority reallocation means dynamically adjusting the control order according to health;

[0063] S650, instruction security verification: ensures the integrity of control instructions based on digital signatures;

[0064] The calculation formula for the final encryption instruction is as follows:

[0065] ;

[0066] in, Indicates the timestamp, Represents ECC elliptic curve signature, represents the SHA-256 hash function, Represents the exclusive OR operation, Indicates the adjusted control instruction, Indicates the device health indicator. Indicates the device priority. Indicates the final encryption instruction.

[0067] Furthermore, in S700, the following steps are specifically included:

[0068] S710, deviation accumulation monitoring: real-time detection of the integral effect of grid load deviation;

[0069] S720, rescheduling trigger judgment: triggering a rescheduling mechanism based on multiple thresholds;

[0070] S730, Robust Rescheduling Modeling: Constructing a multi-objective optimization model considering equipment reliability;

[0071] S740, distributed collaborative solution: uses the improved ADMM algorithm to achieve cloud-edge collaborative solution;

[0072] S750, dynamic instruction switching: achieving seamless transition between old and new scheduling instructions;

[0073] The calculation formula for the final power command is as follows:

[0074] ;

[0075] ;

[0076] in, represents the dynamic switching coefficient, Indicates the decay coefficient of old instructions, represents the switching rate parameter, Indicates the switching start time, Indicates the current moment, Represents the final power instruction, Indicates the new power command, Indicates the original power command.

[0077] Furthermore, in S800, the following steps are specifically included:

[0078] S810, multi-dimensional data aggregation: integrating real-time execution data and historical health records;

[0079] S820, sliding window statistical analysis: calculating the deviation distribution characteristics of the device within the window period;

[0080] S830, abnormal data filtering: eliminating abnormal measurements based on the improved 3σ criterion;

[0081] S840, recursive health update: exponentially weighted update with adaptive forgetting factor;

[0082] S850, health normalization processing: using improved Min-Max normalization processing.

[0083] The present invention further proposes an AI-based reactive power compensation optimization system for executing the steps of the aforementioned AI-based reactive power compensation optimization method, including:

[0084] Dynamic graph processing module: Integrates traffic flow, grid load, user behavior, and device health data in real time to construct a dynamic graph with spatiotemporal correlation characteristics, and continuously optimizes network topology expression through adaptive graph convolution;

[0085] Spatiotemporal prediction module: This module integrates graph neural networks and bidirectional time series models, combined with equipment reliability attenuation factors, to generate charging demand forecasts for future periods and quantify the uncertainty of the forecast results.

[0086] Policy Optimization Module: This module embeds device health constraints into the multi-agent reinforcement learning framework, dynamically adjusts the reward function and policy network structure, and implements reliability-aware distributed decision optimization.

[0087] Cloud-edge scheduling module: This module uses a layered optimization architecture to achieve coordination between global resource allocation and local multi-objective optimization, and employs a distributed algorithm to solve scheduling solutions with reliability margins.

[0088] Deviation prediction module: This module predicts device execution deviations based on a time-series-spatial hybrid model and evaluates the credibility of prediction results by combining historical data with real-time status.

[0089] Execution control module: Integrates feedforward scheduling instructions with feedback compensation mechanisms, dynamically adjusts power output, implements device priority management, and ensures control instruction security through encryption verification;

[0090] Dynamic scheduling module: monitors system operation deviations and triggers the rescheduling mechanism to achieve dynamic optimization and smooth transition of resource allocation while ensuring equipment reliability;

[0091] Health management module: Continuously collects equipment operation data, updates health indicators through sliding window statistics and exception filtering, and provides dynamic equipment status assessment for the entire system.

[0092] The beneficial effects of the present invention are:

[0093] By integrating dynamic spatiotemporal perception with multi-level intelligent decision-making, the present invention effectively improves the responsiveness of urban distribution networks to complex reactive power compensation demands and achieves precise and adaptive control of equipment health.

[0094] The system can autonomously coordinate grid resources, significantly enhance power supply quality and stability, reduce harmonic interference and power loss, and extend the service life of key equipment;

[0095] Through a closed-loop mechanism of prediction-compensation-optimization, it effectively prevents the risk of failure caused by voltage fluctuations and provides an intelligent and highly reliable reactive power compensation solution for high-density load scenarios.

[0096] This invention senses the degradation status of equipment in real time and intelligently adjusts the reactive power compensation strategy, thereby extending the service life of old equipment while ensuring voltage stability. Through feedforward-feedback composite control and dynamic rescheduling, it effectively suppresses the cumulative effect of deviations and ensures that the dispatch instructions are accurately matched with the actual output. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 This is a flow chart of an AI-based reactive compensation optimization method proposed by the present invention;

[0098] Figure 2This is a structural block diagram of an AI-based reactive compensation optimization system proposed in the present invention.

[0099] In the figure: 101, dynamic graph processing module; 102, spatiotemporal prediction module; 103, strategy optimization module; 104, cloud-edge scheduling module; 105, deviation prediction module; 106, execution control module; 107, dynamic scheduling module; 108, health management module. DETAILED DESCRIPTION

[0100] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0101] like Figure 1-Figure 2 As shown, an AI-based reactive compensation optimization method includes the following steps:

[0102] S100, Dynamic Graph Neural Network (DGNN): Integrates four-dimensional features of traffic flow, grid load, user preferences, and device health to build a spatiotemporal correlation graph model. It achieves adaptive topological evolution through dynamic graph convolution and attention mechanisms.

[0103] In one embodiment of the present invention, the following steps are specifically included:

[0104] S110, node feature initialization: for each road segment node Construct a multi-dimensional feature vector covering traffic, power grid, user and device status information;

[0105] formula:

[0106] ;

[0107] in, Indicates road section exist Traffic flow at any given moment, Indicates road section Associated grid node loads, Indicates road section User's charging preference index, , Indicates road section The standard deviation of the charging pile's historical power deviation, i.e., the health indicator, Indicates road section At the moment The eigenvector of , Represents the set of all road segment nodes in the road network;

[0108] S120, edge weight calculation: construct an adjacency matrix based on traffic flow transfer probability to reflect the dynamic association of the road network;

[0109] formula:

[0110] ;

[0111] in, represents the road segment similarity coefficient, Indicates that the road To the road section of traffic volume, Indicates road section The set of adjacent road segments, Indicates road section arrive The edge weights of represents the traffic flow from section i to section k at time t, represents the total outflow traffic volume of road section i at time t;

[0112] S130, Spatiotemporal Attention Generation: Capturing spatiotemporal correlations through multi-head attention mechanism;

[0113] formula:

[0114] ;

[0115] ;

[0116] in, denote the query / key transformation matrix, Indicates the The output projection of the attention head, represents the dimension of a single attention head, represents the number of attention heads, Indicates the The spatial attention matrix of the attention head, represents the fused spatial attention matrix, represents the attention head splicing operation, represents the normalization function, represents the matrix transpose, road section At the moment The eigenvector of

[0117] S140, dynamic graph convolution update: integrating spatiotemporal features to update graph structure weights;

[0118] formula:

[0119] ;

[0120] Temporal difference operator:

[0121] ;

[0122] ;

[0123] in, represents the dynamic graph convolution kernel parameters, represents the tensor concatenation operation, Indicates time The graph weight matrix of represents the graph weight matrix at the next moment, Indicates time The traffic flow matrix, represents the traffic flow matrix at the previous moment, represents the time difference of traffic flow, represents the time difference of health, represents the Hadamard product (element-wise product), represents the dynamic graph convolutional network operation, Indicates the current road segment Standard deviation of charging pile historical power deviation;

[0124] S150, health feature normalization: normalize the device health indicators;

[0125] formula:

[0126] ;

[0127] in, represents the mean value of the health index, represents the standard deviation of the health index, Indicates calculating the expected value of the standard deviation of the power deviation of all devices, Indicates the standard deviation of the power deviation of all devices. Indicates the normalized current road segment The standard deviation of the charging pile's historical power deviation, i.e. the standardized health indicator;

[0128] Output product: standardized dynamic graph structure;

[0129] ;

[0130] in:

[0131] ;

[0132] in, Represents the normalized dynamic graph structure, represents the normalized node features, represents the normalized edge weight, represents the normalized graph weight matrix;

[0133] S200, demand forecasting: Integrates graph neural network spatiotemporal features with BiGRU time series modeling, injects device health attenuation factors, and generates charging demand forecasts with quantified uncertainty.

[0134] In one embodiment of the present invention, the following steps are specifically included:

[0135] S210, Spatiotemporal Graph Convolution Feature Extraction: Based on Dynamic Graph Perform multi-level spatial feature aggregation;

[0136] formula:

[0137] ;

[0138] ;

[0139] in, represents the initial node feature matrix, represents the adjacency matrix with self-connection added, represents the trainable parameters, Indicates the The node feature matrix of the layer, express The identity matrix, Indicates the total number of nodes, represents the rectified linear unit activation function, represents the power of the degree matrix;

[0140] S220, Time Series Modeling: Temporal dependency modeling of historical load series;

[0141] formula:

[0142] ;

[0143] in, Representing history Charging demand during the period, represents the final graph convolution output, Indicates time The hidden state output of represents the time step, Indicates the size of the historical time window, Indicates the length of historical data, represents a bidirectional gated recurrent unit network;

[0144] S230, Reliability Attenuation Factor Injection: Adjust the prediction weight based on the equipment health;

[0145] formula:

[0146] ;

[0147] in, represents the health attenuation coefficient, represents the standardized health index, Representation device The reliability attenuation factor, represents the S-type activation function;

[0148] S240, multimodal feature fusion: fusing spatiotemporal features with attenuation factors to generate the final prediction;

[0149] formula:

[0150] ;

[0151] in, represents the feature projection matrix, It represents the historical mean value for the same period. Indicates time The charging demand forecast value, Representation device At the moment The hidden state, Indicates the total number of devices;

[0152] S250, Uncertainty Quantification: Evaluating confidence intervals for predictions;

[0153] formula:

[0154] ;

[0155] in, represents the standard deviation of the prediction results, represents the gradient of the predicted value to the hidden state, represents the uncertainty of the hidden state, represents the basic uncertainty constant, Indicates the total number of devices;

[0156] S300, MAPPO with enhanced reliability: This expands the agent state space and improves the reward function, embeds device health constraints in the policy network and proximal optimization, and enables distributed decision-making that takes device reliability into account.

[0157] In one embodiment of the present invention, the following steps are specifically included:

[0158] S310, state space expansion: incorporating device health indicators into the agent's observed state;

[0159] formula:

[0160] ;

[0161] in, represents the state vector of device i at time t, represents the normalized node feature vector, represents the predicted value of charging demand, represents the forecast uncertainty, Represents the equipment reliability attenuation factor;

[0162] S320, Reliability Weighted Reward Design: Introducing a device health penalty term into the basic reward function;

[0163] formula:

[0164] ;

[0165] in, represents the new reward function value, Represent the weight coefficients of the basic reward items of operating income, user cost and grid stability index respectively, represents operating income, represents the user cost, represents the grid stability index, represents the health penalty coefficient, Represents the health indicator of device k, represents the historical execution deviation, Indicates the total number of devices;

[0166] S330, the policy network improvement means injecting reliability attention into the policy network output layer;

[0167] formula:

[0168] ;

[0169] in, represents the action probability output by the policy network, represents the policy network weight matrix, represents the policy network bias term, Indicates status The hidden layer representation of represents the reliability scaling factor, represents the health index of device i, represents the action space, represents a candidate action, The corresponding hidden layer representation;

[0170] S340, Proximal Policy Optimization: Policy Gradient Update with Reliability Constraints;

[0171] formula:

[0172] ;

[0173] Constraints:

[0174] ;

[0175] in, represents the updated policy parameters, Indicates the current policy parameters, represents the action probability under the new strategy, represents the action probability under the old strategy, represents the advantage function, represents the KL divergence penalty coefficient, represents the health penalty coefficient, represents the KL divergence, represents the KL divergence constraint threshold, represents the health constraint threshold, represents the KL divergence between the new and old strategies, Indicates the expected value of all device health indicators;

[0176] S350, distributed policy execution: each agent generates actions based on local observations;

[0177] formula:

[0178] ;

[0179] in, represents the execution action of device i at time t, represents the action space, represents the optimized policy network, Indicates the current state of device i, represents the health index of device i, represents the temperature coefficient, Represents the optimized strategy parameters;

[0180] S400, hierarchical scheduling decision: Global resource allocation is performed in the cloud, multi-objective optimization is performed at the edge, cloud-edge collaborative solution is achieved through the ADMM algorithm, and scheduling instructions with reliability margin are generated;

[0181] In one embodiment of the present invention, the following steps are specifically included:

[0182] S410, Cloud-based Global Resource Allocation: Cross-regional power capacity allocation and transmission loss optimization;

[0183] formula:

[0184] ;

[0185] ;

[0186] ;

[0187] ;

[0188] in, Indicates the total number of regions, represents the operating income of region n, represents the transmission loss weight coefficient, represents the transmission loss in region n, represents the grid power of region n, Indicates the maximum capacity of the power grid, Represents the system-level health indicator, Indicates the upper and lower limits of power in area n;

[0189] S420, Edge Multi-Objective Optimization: Balancing User Costs and Grid Stability under Equipment Reliability Constraints;

[0190] formula:

[0191] ;

[0192] ;

[0193] ;

[0194] in, Indicates the total number of users, represents the cost of user u, represents the preference coefficient of user u, represents the attenuation factor of user u, represents the power deviation penalty coefficient, represents the power prediction deviation, represents the maximum power of device k, Represents the health indicator of device k, represents the optimized strategy parameters, Represents local policy parameters, represents the policy deviation threshold;

[0195] S430, Constraint Consistency Coordination: Decoupling cloud-edge optimization through the ADMM algorithm;

[0196] formula:

[0197] ;

[0198] in, represents the step length, represents the penalty factor, represents the cloud constraint space, represents the auxiliary variable for the kth iteration, represents the objective function, represents the augmented Lagrangian function, represents the Lagrange multiplier, represents the grid power, Represents the local power prediction value, represents the projection operator;

[0199] S440, robust margin calculation: dynamically adjusts the power margin based on device health;

[0200] formula:

[0201] ;

[0202] in, represents the safety margin of device k, represents the maximum power of device k, represents the predicted power of device k, Represents the health indicator of device k, represents the inverse function of the standard normal distribution, represents the rectified linear unit function;

[0203] S450, generating a scheduling instruction: generating a scheduling instruction set with a reliability identifier;

[0204] formula:

[0205] ;

[0206] in:

[0207] ;

[0208] in, Represents the scheduling instruction set, Indicates the total number of devices, represents the final dispatch power of device k, Represents the health indicator of device k, represents the safety margin of device k, represents the locally calculated power value, Indicates the power value of cloud computing, represents the local power weight coefficient;

[0209] S500, Execution Deviation Prediction: Integrates historical deviations, scheduling instructions, and equipment status data, using a TCN-GAT hybrid model to predict future execution deviations and assess uncertainty;

[0210] In one embodiment of the present invention, the following steps are specifically included:

[0211] S510, multi-source data fusion: integrating equipment historical data, dispatch instructions, and health indicators;

[0212] formula:

[0213] ;

[0214] in, represents the historical execution deviation, Represents the final scheduling instruction, Indicates the health of the device. represents the attenuation factor, Indicates the length of the historical data window, Indicates the current moment, Indicates the device index;

[0215] S520, temporal feature extraction means using deep TCN to capture long-range dependency patterns;

[0216] formula:

[0217] ;

[0218] in, represents the dilated convolution kernel, Represents the time series feature output, represents the input data sequence, Represents a weight matrix The dilated convolution operation, " represents a standard one-dimensional causal convolution operation, ensuring that the current output only depends on the historical input;

[0219] S530, spatial correlation modeling means capturing the spatial correlation between devices through a graph attention network;

[0220] formula:

[0221] ;

[0222] ;

[0223] in, represents the topological neighbors of device k, represents node feature embedding, represents the query / key-value transformation matrix, represents the value transformation matrix, represents the attention weight, represents the spatial feature output, represents the hidden layer dimension, Represents a multi-head attention splicing operation;

[0224] S540, Cross-modal Fusion Prediction: Fusion of spatiotemporal features to generate biased predictions;

[0225] formula:

[0226] ;

[0227] in, represents the fused projection matrix, represents the feature concatenation operation, represents the bias term, represents the forecast deviation at time t+1, Representation layer normalization operation, Represents the time series characteristics, Represents spatial features;

[0228] S550, Uncertainty Quantification: Evaluating the probability distribution properties of predicted values;

[0229] formula:

[0230] ;

[0231] in, represents the numerical stability term, represents the forecast uncertainty, represents the time series feature variance, represents the spatial feature variance;

[0232] S600, real-time control with feedback correction: combines feedforward scheduling instructions with PID feedback compensation to dynamically adjust power output and implement device priority management, ensuring instruction security through encryption verification;

[0233] In one embodiment of the present invention, the following steps are specifically included:

[0234] S610, feedforward control item generation: generating a reference control variable based on the robust scheduling instruction;

[0235] formula:

[0236] ;

[0237] in, Represents the final scheduling instruction value, represents the prediction deviation value, Indicates the device health indicator. Indicates the rated power of the device. represents the feedforward control output, represents the S-type activation function;

[0238] S620, feedback compensation term calculation: dynamically adjust the control amount according to the real-time execution deviation;

[0239] formula:

[0240] ;

[0241] ;

[0242] in, represents the proportionality coefficient, represents the integral coefficient, represents the differential coefficient, represents the integration window length, Indicates historical bias, Indicates the actual execution power, represents the target power, Indicates feedback compensation amount Indicates the current moment;

[0243] S630, dynamic limiting processing: combining equipment reliability index to constrain control output;

[0244] formula:

[0245] ;

[0246] in, Indicates the upper and lower limits of the device's operating power. Indicates the device health indicator. It means that x is restricted to the interval [a,b]. represents the feedforward control quantity, Indicates the feedback compensation amount, represents the adjusted control output;

[0247] S640, device priority reallocation means dynamically adjusting the control order according to health;

[0248] formula:

[0249] ;

[0250] in, represents the sensitivity of reward to power, represents the device attenuation factor, represents the time decay coefficient, Indicates the equipment running time, Indicates the device health indicator. Indicates the priority value of device k;

[0251] S650, instruction security verification: ensures the integrity of control instructions based on digital signatures;

[0252] formula:

[0253] ;

[0254] in, Indicates the timestamp, Represents ECC elliptic curve signature, represents the SHA-256 hash function, Represents the exclusive OR operation, Indicates the adjusted control instruction, Indicates the device health indicator. Indicates the device priority. Indicates the final encryption instruction;

[0255] S700, dynamic resource rescheduling: monitors accumulated load deviations and triggers a rescheduling mechanism, building a multi-objective model that considers equipment reliability to achieve smooth switching between new and old instructions;

[0256] In one embodiment of the present invention, the following steps are specifically included:

[0257] S710, deviation accumulation monitoring: real-time detection of the integral effect of grid load deviation;

[0258] formula:

[0259] ;

[0260] ;

[0261] in, Indicates real-time load deviation, Indicates the actual load value, represents the predicted load value, represents the time decay coefficient, Indicates the monitoring start time, Indicates the current moment, Indicates the total number of load measurement points, represents the integral variable;

[0262] S720, rescheduling trigger judgment: triggering a rescheduling mechanism based on multiple thresholds;

[0263] formula:

[0264] ;

[0265] ;

[0266] ;

[0267] in, represents the integration threshold, represents the instantaneous threshold, represents the minimum reliability threshold, Indicates the base load value, Indicates the length of the evaluation time, represents the reliability factor of device k, Indicates the load deviation value;

[0268] S730, Robust Rescheduling Modeling: Constructing a multi-objective optimization model considering equipment reliability;

[0269] formula:

[0270] ;

[0271] ;

[0272] ;

[0273] ;

[0274] ;

[0275] in, Represent the weight coefficients, Indicates the power compensation amount, represents the equipment reliability factor, Indicates the maximum allowable adjustment amount, Indicates load deviation, Represents the health indicator of device k, Indicates the power adjustment amount, Indicates the new power instruction, Indicates the original power command, Respectively represent the upper and lower limits of power, represents the new policy parameters, Represents the original policy parameters, Indicates the policy adjustment threshold. Indicates the total number of devices;

[0276] S740, distributed collaborative solution: uses the improved ADMM algorithm to achieve cloud-edge collaborative solution;

[0277] Cloud:

[0278] ;

[0279] edge:

[0280] ;

[0281] renew:

[0282] ;

[0283] in, represents the inertia coefficient, represents the reliability weight, represents the penalty factor, represents the augmented Lagrangian function, represents the auxiliary variable for the kth iteration, represents the power variable at the kth iteration, represents the Lagrange multiplier of the kth iteration, represents the equipment reliability factor, represents the updated auxiliary variable, represents the updated power variable, represents the updated Lagrange multiplier;

[0284] S750, dynamic instruction switching: achieving seamless transition between old and new scheduling instructions;

[0285] formula:

[0286] ;

[0287] ;

[0288] in, represents the dynamic switching coefficient, Indicates the decay coefficient of old instructions, represents the switching rate parameter, Indicates the switching start time, Indicates the current moment, Represents the final power instruction, Indicates the new power command, Indicates the original power command;

[0289] S800, online health update: aggregates real-time execution data, uses sliding window statistics and anomaly filtering mechanisms, recursively updates device health indicators and implements normalization;

[0290] In one embodiment of the present invention, the following steps are specifically included:

[0291] S810, multi-dimensional data aggregation: integrating real-time execution data and historical health records;

[0292] formula:

[0293] ;

[0294] ;

[0295] in, Indicates real-time execution deviation, Indicates the actual execution power, represents the target power instruction, represents the prediction deviation value, Represents historical health indicators, Indicates the device priority. represents the reliability attenuation factor, Indicates the current moment;

[0296] S820, sliding window statistical analysis: calculating the deviation distribution characteristics of the device within the window period;

[0297] formula:

[0298] ;

[0299] ;

[0300] ;

[0301] in, Indicates the size of the statistics window. represents the time decay weight, represents the attenuation coefficient, represents the mean deviation, represents the original standard deviation, represents the power deviation at time τ, Indicates the current moment, To indicate a historical moment;

[0302] S830, abnormal data filtering: eliminating abnormal measurements based on the improved 3σ criterion;

[0303] formula:

[0304] ;

[0305] in, represents the reliability factor adjustment threshold, Indicates the adjusted power deviation, represents the original power deviation, represents the mean deviation, represents the original standard deviation;

[0306] S840, recursive health update: exponentially weighted update with adaptive forgetting factor;

[0307] formula:

[0308] ;

[0309] ;

[0310] in, represents the adaptive forgetting factor, Indicates the rated power of the device. Indicates the health index at the current moment, Indicates the health index at the previous moment, Indicates the adjusted power deviation;

[0311] S850, health normalization processing: using improved Min-Max normalization processing;

[0312] formula:

[0313] ;

[0314] in, represents the mean value of system health, represents the numerical stability term of health, Indicates the maximum health value based on all system devices. represents the normalized health index, Represents the raw health indicator.

[0315] In one embodiment of the present invention, an AI-based reactive power compensation optimization system is further provided, which is used to execute the steps in the above-mentioned AI-based reactive power compensation optimization method, and includes the following modules:

[0316] Dynamic Graph Processing Module 101: Integrates traffic flow, grid load, user behavior, and device health data in real time to construct a dynamic graph with spatiotemporal correlation characteristics, and continuously optimizes network topology representation through adaptive graph convolution.

[0317] Spatiotemporal prediction module 102: This module integrates the graph neural network and the bidirectional time series model, combines the device reliability attenuation factor, generates a charging demand forecast for the future period, and quantifies the uncertainty of the forecast results;

[0318] Policy Optimization Module 103: Embeds device health constraints in a multi-agent reinforcement learning framework, dynamically adjusts the reward function and policy network structure, and implements reliability-aware distributed decision optimization.

[0319] Cloud-edge scheduling module 104: Through the hierarchical optimization architecture, the global resource allocation and the local multi-objective optimization are coordinated, and the distributed algorithm is used to solve the scheduling scheme with reliability margin;

[0320] Deviation prediction module 105: Based on the time-space hybrid model, the equipment execution deviation is predicted, and the credibility of the prediction result is evaluated in combination with historical data and real-time state;

[0321] Execution control module 106: The feedforward scheduling instruction and the feedback compensation mechanism are integrated, the power output is dynamically adjusted, and the equipment priority management is implemented, and the control instruction safety is ensured through encryption verification;

[0322] Dynamic scheduling module 107: The system running deviation is monitored and the rescheduling mechanism is triggered, and the dynamic optimization and smooth transition of resource allocation are realized under the premise of ensuring equipment reliability;

[0323] Health management module 108: The equipment running data is continuously collected, the health index is updated through the sliding window statistics and abnormal filtering, and the dynamic equipment state evaluation is provided for the whole system.

[0324] Scenario background: A large commercial complex faces the challenge of charging management during the holiday peak period: when 300 fast charging piles in the underground parking lot are running at the same time, the reactive power fluctuation caused by electric vehicle charging harmonics leads to voltage drop in the power distribution room, and the start-stop of the air conditioning system in the commercial area aggravates the load peak-valley difference.

[0325] The flow of the reactive power compensation optimization method is as follows:

[0326] Dynamic perception:

[0327] The roadside camera and the charging pile upload the traffic data and the equipment state in real time, and the system constructs the dynamic atlas containing the parking lot topology, the charging demand hot area and the transformer load rate.

[0328] Demand prediction:

[0329] The AI model predicts the charging peak of the cinema dispersal period 2 hours in advance, and identifies that the charging pile in C area will have 3 times the concentrated demand of the daily demand.

[0330] Intelligent scheduling:

[0331] The cloud coordinates the commercial area photovoltaic energy storage system to increase the output, the edge node adjusts the B1 layer charging pile group to the "power flexible mode", and applies for dynamic capacity increase permission to the power grid.

[0332] Deviation compensation:

[0333] When the actual output of old charging piles fluctuates, the system automatically reduces the reactive compensation of adjacent healthy equipment to offset the fluctuation and maintain bus voltage stability.

[0334] Emergency Response:

[0335] When the elevator group control system suddenly started up and caused voltage flicker, the system switched to the backup compensation plan within 200ms and converted the charging piles in area D to voltage support mode.

[0336] Application value: During actual testing during the commercial district's summer sales peak period, the system successfully avoided power distribution room tripping incidents caused by charging load shocks. At the same time, dynamic scheduling reduced the reactive power loss of the central air-conditioning system, significantly improving the mall's overall energy efficiency level.

[0337] Operation and maintenance personnel can use the three-dimensional visualization interface to grasp the power quality status of each area in real time, realizing the transformation from passive repair to active prevention.

[0338] The above describes the embodiments of the present invention, but the present invention is not limited to the above specific implementation methods. The above specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms under the guidance of the present invention, all of which are protected by the present invention.

Claims

1. A reactive power compensation optimization method based on AI, characterized in that: The following steps are involved: S100 integrates four-dimensional features of traffic flow, grid load, user preferences, and device health to build a spatiotemporal correlation graph model, and achieves adaptive evolution of topology structure through dynamic graph convolution and attention mechanism; Here are the steps: S110, node feature initialization: construct a multi-dimensional feature vector for each road node, covering traffic, power grid, user and device status information; S120, edge weight calculation: construct an adjacency matrix based on traffic flow transfer probability to reflect the dynamic association of the road network; S130, Spatiotemporal Attention Generation: Capturing spatiotemporal correlations through multi-head attention mechanism; S140, dynamic graph convolution update: integrating spatiotemporal features to update graph structure weights; S150, health feature normalization: normalize the device health indicators; S200, which integrates the spatiotemporal features of graph neural networks with BiGRU time series modeling, injects device health attenuation factors, and generates charging demand prediction results with quantified uncertainty; Here are the steps: S210, Spatiotemporal Graph Convolutional Feature Extraction: Multi-level spatial feature aggregation based on dynamic graphs; S220, Time Series Modeling: Temporal dependency modeling of historical load series; S230, Reliability Attenuation Factor Injection: Adjust the prediction weight based on the equipment health; S240, multimodal feature fusion: fusing spatiotemporal features with attenuation factors to generate the final prediction; S250, Uncertainty Quantification: Evaluating confidence intervals for predictions; S300, which expands the agent state space and improves the reward function, embeds device health constraints in the policy network and proximal optimization, and enables distributed decision-making that considers device reliability; Here are the steps: S310, state space expansion: incorporating device health indicators into the agent's observed state; S320, Reliability Weighted Reward Design: Introducing a device health penalty term into the basic reward function; S330, the policy network improvement means injecting reliability attention into the policy network output layer; S340, Proximal Policy Optimization: Policy Gradient Update with Reliability Constraints; S350, distributed policy execution: each agent generates actions based on local observations; S400 performs global resource allocation in the cloud and multi-objective optimization at the edge, achieving cloud-edge collaborative solution through the ADMM algorithm and generating scheduling instructions with reliability margins; Here are the steps: S410, Cloud-based Global Resource Allocation: Cross-regional power capacity allocation and transmission loss optimization; S420, Edge Multi-Objective Optimization: Balancing User Costs and Grid Stability under Equipment Reliability Constraints; S430, Constraint Consistency Coordination: Decoupling cloud-edge optimization through the ADMM algorithm; S440, robust margin calculation: dynamically adjusts the power margin based on device health; S450, generating a scheduling instruction: generating a scheduling instruction set with a reliability identifier; S500, which integrates historical deviations, scheduling instructions, and equipment status data, uses a TCN-GAT hybrid model to predict future execution deviations and assess uncertainty; Here are the steps: S510, multi-source data fusion: integrating equipment historical data, dispatch instructions, and health indicators; S520, temporal feature extraction means using deep TCN to capture long-range dependency patterns; S530, spatial correlation modeling means capturing the spatial correlation between devices through a graph attention network; S540, Cross-modal Fusion Prediction: Fusion of spatiotemporal features to generate biased predictions; S600 combines feedforward scheduling instructions with PID feedback compensation to dynamically adjust power output and implement device priority management, ensuring instruction security through encryption verification; Here are the steps: S610, feedforward control item generation: generating a reference control variable based on the robust scheduling instruction; S620, feedback compensation term calculation: dynamically adjust the control amount according to the real-time execution deviation; S630, dynamic limiting processing: combining equipment reliability index to constrain control output; S640, device priority reallocation means dynamically adjusting the control order according to health; S650, instruction security verification: ensures the integrity of control instructions based on digital signatures; S700 monitors accumulated load deviations and triggers a rescheduling mechanism, building a multi-objective model that considers equipment reliability to achieve smooth switching between new and old instructions; Here are the steps: S710, deviation accumulation monitoring: real-time detection of the integral effect of grid load deviation; S720, rescheduling trigger judgment: triggering a rescheduling mechanism based on multiple thresholds; S730, Robust Rescheduling Modeling: Constructing a multi-objective optimization model considering equipment reliability; S740, distributed collaborative solution: uses the improved ADMM algorithm to achieve cloud-edge collaborative solution; S750, dynamic instruction switching: achieving seamless transition between old and new scheduling instructions; S800 aggregates real-time execution data, uses sliding window statistics and anomaly filtering mechanisms, recursively updates device health indicators and implements normalization processing.

2. The AI-based reactive compensation optimization method according to claim 1, characterized in that: In S240, the calculation formula for the final prediction is as follows: ； in, represents the feature projection matrix, It represents the historical mean value for the same period. Indicates time The charging demand forecast value, Representation device At the moment The hidden state, Indicates the total number of devices, Representation device Reliability degradation factor.

3. The AI-based reactive compensation optimization method according to claim 2, characterized in that: In S350, the calculation formula of the action is as follows: ； in, represents the execution action of device i at time t, represents the action space, represents the optimized policy network, Indicates the current state of device i, represents the health index of device i, represents the temperature coefficient, Represents the optimized strategy parameters.

4. The AI-based reactive compensation optimization method according to claim 3, characterized in that: In S450, the calculation formula of the scheduling instruction set is as follows: ； in: ； in, Represents the scheduling instruction set, Indicates the total number of devices, represents the final dispatch power of device k, Represents the health indicator of device k, represents the safety margin of device k, represents the locally calculated power value, Indicates the power value of cloud computing, Indicates the local power weight coefficient.

5. The AI-based reactive compensation optimization method according to claim 4, characterized in that: In S540, the calculation formula for the deviation prediction is as follows: ； in, represents the fused projection matrix, represents the feature concatenation operation, represents the bias term, represents the forecast deviation at time t+1, Representation layer normalization operation, Represents the time series characteristics, Represents spatial features; S550, Uncertainty Quantification: Evaluating the properties of probability distributions of predicted values.

6. The AI-based reactive compensation optimization method according to claim 5, characterized in that: In S650, the calculation formula of the final encryption instruction is as follows: ； in, Indicates the timestamp, Represents ECC elliptic curve signature, represents the SHA-256 hash function, Represents the exclusive OR operation, Indicates the adjusted control instruction, Indicates the device health indicator. Indicates the device priority. Indicates the final encryption instruction.

7. The AI-based reactive compensation optimization method according to claim 6, characterized in that: In S750, the calculation formula of the final power command is as follows: ； ； in, represents the dynamic switching coefficient, Indicates the decay coefficient of old instructions, represents the switching rate parameter, Indicates the switching start time, Indicates the current moment, Represents the final power command, Indicates the new power command, Indicates the original power command.

8. The AI-based reactive compensation optimization method according to claim 7, characterized in that: In S800, the following steps are specifically included: S810, multi-dimensional data aggregation: integrating real-time execution data and historical health records; S820, sliding window statistical analysis: calculating the deviation distribution characteristics of the device within the window period; S830, abnormal data filtering: eliminating abnormal measurements based on the improved 3σ criterion; S840, recursive health update: exponentially weighted update with adaptive forgetting factor; S850, health normalization processing: using improved Min-Max normalization processing.

9. An AI-based reactive compensation optimization system, characterized in that: The steps for executing the AI-based reactive compensation optimization method according to any one of claims 1 to 8 include: Dynamic graph processing module: Integrates traffic flow, grid load, user behavior, and device health data in real time to construct a dynamic graph with spatiotemporal correlation characteristics, and continuously optimizes network topology expression through adaptive graph convolution; Spatiotemporal prediction module: This module integrates graph neural networks and bidirectional time series models, combined with equipment reliability attenuation factors, to generate charging demand forecasts for future periods and quantify the uncertainty of the forecast results. Policy Optimization Module: This module embeds device health constraints into the multi-agent reinforcement learning framework, dynamically adjusts the reward function and policy network structure, and implements reliability-aware distributed decision optimization. Cloud-edge scheduling module: This module uses a layered optimization architecture to achieve coordination between global resource allocation and local multi-objective optimization, and employs a distributed algorithm to solve scheduling solutions with reliability margins. Deviation prediction module: This module predicts device execution deviations based on a time-series-spatial hybrid model and evaluates the credibility of prediction results by combining historical data with real-time status. Execution control module: Integrates feedforward scheduling instructions with feedback compensation mechanisms, dynamically adjusts power output, implements device priority management, and ensures control instruction security through encryption verification; Dynamic scheduling module: monitors system operation deviations and triggers the rescheduling mechanism to achieve dynamic optimization and smooth transition of resource allocation while ensuring equipment reliability; Health management module: Continuously collects equipment operation data, updates health indicators through sliding window statistics and exception filtering, and provides dynamic equipment status assessment for the entire system.

Images