A standardized intelligent control system for a substation

Abstract

The present invention relates to the field of intelligent control technology for substations, particularly to a standardized intelligent control system for substations. By obtaining the current, voltage, power factor, temperature and humidity, operating status, insulation status and ambient temperature and humidity of the substation unit in real time, real-time response to the operating status of the equipment is achieved. The gradient boosting decision tree is combined with the variational autoencoder and the self-attention mechanism to extract the state vector features, accurately identify complex abnormal states and early detect potential faults, greatly improving the monitoring efficiency and early warning accuracy. Through the long short-term memory network model combined with historical and real-time load data, the future power demand is accurately predicted, overcoming the limitations of traditional static load management. Based on the predicted power demand, the reinforcement learning model is used to dynamically adjust the number of operating substation units, achieving the optimal allocation of resources while meeting the demand, and significantly improving the overall operating efficiency of the substation.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of substations, and particularly to a standardized intelligent control system for substations. Background Art

[0002] Substations are important nodes in the power transmission and distribution system. The management mode usually relies on manual inspections and regular maintenance, making it difficult to respond promptly to changes in equipment operating status and fluctuations in load demand. This management method not only has problems such as low monitoring efficiency and poor equipment fault warning capabilities, but also increases operating costs.

[0003] Existing substation monitoring systems usually include basic parameter collection and fault alarm functions, but lack intelligent analysis means for complex operating states within the substation. Although some systems have introduced data collection and analysis modules, they mostly use alarm methods based on fixed thresholds, unable to dynamically adapt to changes in equipment status and difficult to handle complex abnormal situations. Due to the inability to effectively identify and predict potential equipment failures, the reliability and warning accuracy of current systems are limited. In addition, existing systems mostly use static load distribution methods in load management, unable to accurately predict real-time load changes and future power demands, resulting in low overall operating efficiency of the substation, difficulty in adapting to fluctuations in power demand, and further causing waste of resources due to the simultaneous operation of too many transformation units. Summary of the Invention

[0004] To solve the above problems, the present invention provides a standardized intelligent control system for substations, which solves the problems of low monitoring efficiency, poor warning capabilities, static load management, and resource waste in traditional substations through intelligent monitoring, dynamic fault detection, load prediction, and reinforcement learning control strategies.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A standardized intelligent control system for substations, comprising: a distributed monitoring module, a fault detection module, a load monitoring module, and an intelligent control module. The distributed monitoring module and the fault detection module are connected in sequence, and the load monitoring module and the intelligent control module are connected;

[0007] The distributed monitoring module is used to obtain the active parameters and passive parameters of the transformation units in the substation. The active parameters include current, voltage, power factor, temperature and humidity, operating status, and insulation status. The passive parameters include the ambient temperature and humidity inside and outside the substation;

[0008] The fault detection module is used to receive the active parameters and passive parameters of a single substation unit, standardize them to generate a substation unit state vector, perform preliminary anomaly classification based on the gradient boosting decision tree, and determine the abnormal state; according to the abnormal state, use the variational autoencoder and self-attention mechanism to extract features from the state vector to detect potential faults.

[0009] The load monitoring module is used to obtain and store the substation load data.

[0010] The intelligent control module is used to predict future power demand data based on the long short-term memory network through historical substation load data and real-time substation load data, and control the number of operating substation units without anomalies and potential faults based on the future power demand data through a reinforcement learning model. The number of operations is negatively correlated with the reward value in the reinforcement learning model.

[0011] Further, the substation units in the substation include: main transformer, high-voltage switchgear, low-voltage switchgear, circuit breaker, bus system, cable joint, and lightning arrester.

[0012] Further, the distributed monitoring module includes current sensors, voltage sensors, temperature and humidity sensors, insulation detectors, and an embedded operation monitoring system.

[0013] Further, the steps of receiving the active parameters of a single substation unit, standardizing them to generate a substation unit state vector, and performing preliminary anomaly classification based on the gradient boosting decision tree include the following steps:

[0014] Denoise and fuse the active parameters and passive parameters to generate a high-dimensional feature matrix.

[0015] Based on the high-dimensional feature matrix, use an adaptive standardization model to generate a real-time dynamic state vector.

[0016] Through the gradient boosting decision tree, generate a threshold matrix according to the historical high-dimensional feature matrix set, and embed the adaptive threshold of each parameter into the state vector.

[0017] Compare the state vector with the threshold matrix, and classify the state vector hierarchically based on the hierarchical path of the gradient boosting decision tree to determine the abnormal state. The abnormal states include no anomaly, fluctuation, and fault.

[0018] Further, the gradient boosting decision tree is constructed through the following steps:

[0019] Clean and normalize the historical high-dimensional feature matrix set.

[0020] Based on the parameters in the feature matrix, use the SHAP value to screen out the key parameters that have a significant impact on anomaly classification.

[0021] Based on the filtered key parameters, an initial decision tree model is constructed. With the goal of minimizing the classification error, the initial branching paths are set according to the distribution characteristics of the parameters in different abnormal states.

[0022] In each training iteration, the sample weights are adjusted according to the classification error of the previous round, and a new decision tree is constructed through the residual minimization strategy. The new decision tree is combined with the previous decision tree to generate a gradient boosting decision tree model.

[0023] Furthermore, the formula of the gradient boosting decision tree is as follows:

[0024] ;

[0025] where is the prediction model of the m-th iteration; is the prediction model of the previous round; is the learning rate; is the number of samples in the training dataset; is the i-th sample; is the i-th sample the optimal weight corresponding to the residual of; is the weak decision tree generated in this round.

[0026] Furthermore, the feature extraction of the state vector using the variational autoencoder and self-attention mechanism according to the abnormal state includes the following steps:

[0027] Input the state vector into the encoding layer of the variational autoencoder, and perform high-dimensional feature encoding on the state vector through several layers of neural networks to generate a latent variable representation;

[0028] Apply the self-attention mechanism to the latent variable representation output by the variational autoencoder encoding to calculate the attention weights of each dimension feature;

[0029] Input the feature representation processed by the self-attention mechanism into the decoding layer of the variational autoencoder to generate a reconstructed state vector;

[0030] Calculate the difference between the reconstructed state vector and the original state vector to determine the abnormal feature vector, and match the corresponding potential fault according to the abnormal feature vector.

[0031] Furthermore, the prediction of future power demand data based on the long short-term memory network through historical substation load data and real-time substation load data includes the following steps:

[0032] Denoise, smooth, and normalize the historical load data and real-time load data, convert them into a time series data matrix, and segment and construct time windows to generate a standardized load data input matrix;

[0033] The standardized load data input matrix is partitioned along the time dimension, and the load quantity and time stamps are mapped into a multi-dimensional feature vector sequence;

[0034] Time labels and load change rates are added to the feature sequences within each time window to form a time-dependent matrix as the input to the long short-term memory network model;

[0035] The time-dependent matrix is input into the long short-term memory network model, and forward and backward propagation training is performed on each input time window. The model parameters are optimized through gradient descent;

[0036] The real-time load data sequence is input into the trained long short-term memory network model, and the memory cells are used to update the state of the sequence features;

[0037] The future power demand data is output through the output layer.

[0038] Furthermore, the construction steps of the reinforcement learning model include:

[0039] The predicted future power demand data is used as the environmental input state of the reinforcement learning model, and the operating states and demand loads of each substation unit in the substation are initialized to form the initial state matrix of the system;

[0040] For the operating states of the substation units, a set of control actions is constructed, including increasing, decreasing, or maintaining the number of operating substation units;

[0041] A reward function is constructed such that the number of operations is negatively correlated with the reward value in the reinforcement learning model;

[0042] Through policy decision-making based on the current state matrix and the reward function at each time step, the action with the highest reward value is selected to adjust the number of operating substation units;

[0043] The load response, reward value, and state change data after each executed action are recorded and fed back to the model to update the state matrix and optimize the reinforcement learning model.

[0044] The beneficial effects of the present invention are as follows: By obtaining the active parameters (including current, voltage, power factor, temperature and humidity, operating status, and insulation status) and passive parameters (including ambient temperature and humidity) of the substation unit in real time, the present invention can respond in a timely manner to changes in the operating status of the equipment. The gradient boosting decision tree and variational autoencoder (VAE) are combined with the self-attention mechanism to extract features from the state vector, realizing accurate identification of complex abnormal states and early detection of potential faults, effectively improving the monitoring efficiency of the system and the accuracy of fault warning. Different from the traditional alarm method based on fixed thresholds, the present invention uses machine learning algorithms to dynamically generate thresholds through an intelligent decision-making module and perform anomaly classification in real time, enabling the system to flexibly adapt to changes in the equipment state. Through intelligent analysis of the collected multi-dimensional data, the system can accurately identify various operating states, support the handling of complex abnormal situations, and realize intelligent analysis and dynamic response to complex states in the substation. By introducing the long short-term memory (LSTM) model, the historical load data and real-time load data are combined to achieve accurate prediction of future power demand. The LSTM model can capture the time series characteristics of the load data, ensuring accurate prediction results when the power demand changes and overcoming the limitations of static allocation of load management in traditional systems. Based on the future power demand data, this solution uses a reinforcement learning model to control the number of operating substation units. By dynamically adjusting the number of operating substation units, resource waste is minimized while meeting the power demand. The reinforcement learning model continuously optimizes the control strategy by setting a reward function negatively correlated with the number of operations, achieving efficient resource allocation and optimal operating status, and improving the overall operating efficiency of the substation. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a schematic structural diagram of a standardized intelligent control system for a substation in the present invention.

[0046] Figure 2 is a flowchart of the construction steps of the gradient boosting decision tree in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] Please refer to Figure 1 - Figure 2 As shown, the present invention relates to a standardized intelligent control system for a substation, including: a distributed monitoring module, a fault detection module, a load monitoring module, and an intelligent control module. The distributed monitoring module and the fault detection module are connected in sequence, and the load monitoring module and the intelligent control module are connected;

[0048] The distributed monitoring module is used to obtain the active parameters and passive parameters of the substation unit. The active parameters include current, voltage, power factor, temperature and humidity, operating status, and insulation status. The passive parameters include the ambient temperature and humidity inside and outside the substation;

[0049] The fault detection module is used to receive the active parameters and passive parameters of a single substation unit, standardize them to generate a substation unit state vector, perform preliminary anomaly classification based on the gradient boosting decision tree, and determine the abnormal state; according to the abnormal state, use the variational autoencoder and self-attention mechanism to extract features from the state vector to detect potential faults;

[0050] The load monitoring module is used to obtain and store the substation load data;

[0051] The intelligent control module is used to predict the future power demand data based on the historical substation load data and the real-time substation load data through the long short-term memory network, and control the operation quantity of the substation units without anomalies and potential faults based on the future power demand data through the reinforcement learning model. The operation quantity is negatively correlated with the reward value in the reinforcement learning model.

[0052] In some embodiments, the standardized intelligent control system of the substation mainly consists of a distributed monitoring module, a fault detection module, a load monitoring module, and an intelligent control module. The distributed monitoring module is composed of high-precision sensors and embedded monitoring devices, and its specific function is to obtain and monitor the active parameters and passive parameters of the substation unit in real time. To ensure the accuracy of parameter acquisition, current sensors, voltage sensors, power factor sensors, and temperature and humidity sensors are connected to the processing unit of the monitoring module through high-precision data interfaces. Inside, the analog signals are converted into digital signals through an ADC converter, and further data fusion and denoising processing are carried out. By transmitting the active parameters and passive parameters to the fault detection module together, all-round monitoring of the equipment state is ensured. The fault detection module is designed based on a multi-layer hardware architecture to perform efficient preliminary anomaly analysis after receiving the monitoring data. First, the hardware platform of this module is built with an FPGA and a GPU parallel processor to support the real-time operation of the gradient boosting decision tree algorithm. The FPGA is used to generate and update the threshold matrix in real time, and the GPU performs data classification calculations. The parameter data received by the detection module is standardized on the FPGA to generate the state vector of the substation unit, and preliminary anomaly classification is carried out through the gradient boosting decision tree under the collaborative calculation of the GPU. If an abnormal state is detected, the module will use a variational autoencoder (VAE) combined with a self-attention mechanism for feature extraction to screen out the feature information that may be related to potential faults. The encoding layer of the variational autoencoder is deployed on the GPU to further accelerate the compression and mapping of high-dimensional features, and the self-attention mechanism is used to extract key features to identify complex fault patterns, thereby improving the fault detection accuracy of the system. The load monitoring module is responsible for the continuous collection and storage of load data to ensure the traceability of historical power load data. This module uses non-volatile storage devices (such as EEPROM and NAND flash). The data is synchronously written to local storage during collection and periodically transmitted to a remote database. The stored load data will be used as the input of the intelligent control module for training the load prediction model. The intelligent control module mainly realizes the load prediction and dynamic control of the substation through a long short-term memory network (LSTM) and a reinforcement learning model. In terms of hardware design, the prediction calculation of the LSTM model is completed through the cooperation of the CPU and the GPU. The GPU accelerates matrix operations to improve the prediction response speed. The reinforcement learning model is used to dynamically adjust the operating state of the substation unit according to the predicted load demand. Specifically, this module combines the load prediction data and adjusts the operating quantity through the reinforcement learning model. The reward function in the model is set to be negatively correlated to punish excessive or insufficient operating quantities to ensure the adaptive ability of the system when the load demand changes. The final control signal is generated by the processing unit of the intelligent control module and regulates the actual operating quantity of each substation unit through a high-precision drive circuit, thereby achieving the optimal allocation and efficient management of resources.

[0053] Furthermore, the power conversion unit in the substation includes: main transformer, high-voltage switchgear, low-voltage switchgear, circuit breaker, bus system, cable joint, and lightning arrester.

[0054] Specifically, as the core power conversion equipment, the main transformer is connected between the incoming line and the outgoing line of the substation. The input and output voltage, current values, and environmental parameters such as temperature and humidity are monitored through the current and voltage sensors in the distributed monitoring module. The monitoring module standardizes the operating parameters of the main transformer, generates a real-time state vector, and performs anomaly recognition through the gradient boosting decision tree of the fault detection module to ensure the stability and reliability of the operation of the main transformer.

[0055] High-voltage switchgear is used to connect or disconnect high-voltage lines, and parameters such as its operating state, temperature, and insulation state are collected by sensors embedded inside the equipment. To improve the system response speed and reliability, the high-voltage switchgear transmits the collected data to the fault detection module in real time through FPGA parallel processing hardware, and extracts key abnormal features by combining the self-attention mechanism to enable a quick response when a fault or abnormal operating state occurs. The low-voltage switchgear is mainly responsible for distributing electric energy to other equipment inside the substation and external loads. The monitoring system arranges multi-channel temperature and humidity sensors and voltage and current sensors inside the low-voltage switchgear, and the load monitoring module collects and stores the load state of the low-voltage switchgear in real time. The intelligent control module will dynamically adjust the operating state of the low-voltage switchgear based on the load prediction results, combined with the reinforcement learning model, to ensure the efficiency of load distribution. The function of the circuit breaker is to protect the circuit and prevent equipment damage caused by overload or short circuit. The circuit breaker status monitoring uses high-precision current sensors and thermistors to detect abnormal states such as overcurrent and temperature rise. The distributed monitoring module transmits the collected status data to the fault detection module, and classifies it through the gradient boosting decision tree. If an abnormality is found, the system immediately triggers the fault detection process to conduct an in-depth analysis of potential faults. The busbar system connects various equipment and is used for electric energy distribution. The operating parameters such as the temperature and current-carrying capacity of the busbar system are obtained in real time by the distributed monitoring module and transmitted to the load monitoring module for reference during load prediction. If the load prediction results show that the busbar load is close to the upper limit, the intelligent control module adjusts the number of operating substation units through the reinforcement learning model to ensure the safe load distribution of the busbar. As a key component connecting various equipment, the cable joint is vulnerable to the influence of environmental temperature and load fluctuations, and abnormal temperature rise may occur. The system monitors the temperature change of the cable joint through a temperature sensor, and uses the variational autoencoder model combined with the self-attention mechanism in the fault detection module to identify potential faults, providing early warning for the abnormal temperature rise of the cable joint. Lightning arresters are used to protect equipment from lightning strikes. The current sensor and the grounding resistance monitoring module inside the lightning arrester detect the number of discharges and the grounding resistance value in real time, and the distributed monitoring module transmits this data to the fault detection module for standardization processing to ensure the normal operation of the lightning arrester during lightning strikes, and automatically analyzes and alarms when abnormalities occur to improve the lightning protection ability and safety of the system.

[0056] Further, the distributed monitoring module includes a current sensor, a voltage sensor, a temperature and humidity sensor, an insulation detector, and an embedded operation monitoring system.

[0057] Specifically, the current sensor is used to monitor the current flowing through the substation in real time to ensure that the equipment is within a safe load range. The high-precision current sensor converts the collected current signal into a digital signal through a low-noise signal amplifier and a filter circuit, and transmits it to the embedded operation monitoring system for processing. In the case of high load or abnormal current fluctuations, the current sensor will trigger a real-time alarm, prompting the fault detection module to further analyze the cause of the abnormal current. The voltage sensor is responsible for monitoring the voltage state of the substation, converting the analog voltage signal into a digital signal through a voltage divider circuit and an ADC module and then transmitting it to the monitoring system. In order to improve the accuracy of voltage monitoring, the voltage sensor uses multi-layer electrical isolation technology to avoid interference and overvoltage damage. When the voltage fluctuation exceeds the set threshold, the system will automatically trigger the fault detection module and further analyze whether there is a potential fault in combination with other parameters. The temperature and humidity sensor is used to collect temperature and humidity data inside the substation to prevent equipment aging or insulation performance degradation due to abnormal temperature and humidity. The sensor has a built-in microprocessor that can convert analog signals into standard digital signals through a protocol conversion interface and transmit them to the embedded operation monitoring system. When the temperature or humidity exceeds the standard, the system will automatically record environmental data and link the insulation detector to conduct insulation status inspection. The embedded operation monitoring system is the core part of the distributed monitoring module, which is responsible for receiving, processing and transmitting the data information and operation status data collected by each sensor. The embedded system integrates FPGA and MCU to collect various data through high-speed interfaces.

[0058] Furthermore, the receiving of active parameters of a single substation unit and standardizing the generation of a substation unit state vector, and performing preliminary abnormality classification based on a gradient boosting decision tree include the following steps:

[0059] De-noise the active and passive parameters and fuse them to generate a high-dimensional feature matrix;

[0060] Based on the high-dimensional feature matrix, an adaptive normalization model is used to generate real-time dynamic state vectors;

[0061] Generate a threshold matrix based on the historical high-dimensional feature matrix set through the gradient boosting decision tree, and embed the adaptive threshold of each parameter into the state vector;

[0062] The state vector is compared with the threshold matrix, and the state vector is classified based on the hierarchical path of the gradient boosting decision tree to determine the abnormal state, which includes no abnormality, fluctuation and failure.

[0063] In some embodiments, first, a multi-level denoising process is applied to the active parameters and passive parameters. The system uses the wavelet transform algorithm for multi-scale denoising to separate the high-frequency noise components and the actual useful signals in the signals. In the initial denoising step, a low-pass filter is used to weaken the instantaneous high-frequency noise during the acquisition process to ensure the stability of the core feature data. Next, the signal is decomposed into different frequency bands through adaptive wavelet denoising, the low-frequency components reflecting features such as current and voltage are retained, and the interference signals are removed. The denoised data is fused with the historical data to generate a high-dimensional feature matrix to provide an accurate feature representation. Based on the high-dimensional feature matrix, the system uses an adaptive normalization model to unify the scales of each feature. First, a statistical analysis of the historical data distribution of each parameter is performed, including the dynamic update of the mean, variance, minimum value, and maximum value, to generate a standardized feature description. The adaptive normalization model maps each feature to the same numerical interval by normalizing the data in the feature matrix, so that the data has the same scale in different feature dimensions. For example, the normalization of the current parameter increases the weight due to its large dynamic range, while the temperature parameter is given a lower weight due to its small fluctuation range. At the same time, the model dynamically adjusts the normalization weights of the parameters to cope with the influence of load fluctuations or environmental changes. For example, when the load increases, the adaptive normalization model automatically relaxes the parameter weights of the current and power factor to capture more significant change trends in the state vector. Next, the system performs a preliminary anomaly classification on the state vector through the gradient boosting decision tree (GBDT) obtained by training. First, the GBDT model is trained based on the high-dimensional feature matrix set. In each round of iteration, weak classifiers are gradually established, and the model is iteratively optimized through the residual minimization strategy to ensure that the model can generate an accurate threshold matrix. The threshold matrix contains the upper and lower limits of each parameter in different states, representing the safety ranges of the system in the "no anomaly", "fluctuation", and "fault" states. For example, the thresholds of parameters such as current and insulation resistance are dynamically adjusted according to historical data and environmental characteristics. After the threshold matrix is generated, these upper and lower limits are embedded in the state vector to form a set of dynamic standardized judgment benchmarks. The generated real-time state vector is compared with the threshold matrix, and each parameter in the state vector is classified step by step through the hierarchical path of the GBDT. Specifically, the GBDT model first screens out the features that may have anomalies through the hierarchical path and compares whether the fluctuation amplitude of the parameter meets the classification conditions of "fluctuation" or "fault". During the classification process, the GBDT model gives priority to sensitive parameters such as current and voltage to ensure that the first-step screening can quickly and accurately identify key anomalies. When a certain parameter exceeds the set safety threshold range, the system will further trigger the determination of the "fluctuation" or "fault" state. For example, if the current value deviates from the normal range but does not reach a serious level, the system determines the state as "fluctuation"; while when the current deviation exceeds the tolerance and persists, it is determined as "fault" and immediately marked for in-depth processing by the subsequent analysis module.

[0064] Further, the gradient boosting decision tree is constructed through the following steps:

[0065] Clean and normalize the historical high-dimensional feature matrix set;

[0066] Based on the parameters in the feature matrix, screen out the key parameters that have a significant impact on abnormal classification through SHAP values;

[0067] Based on the screened key parameters, construct an initial decision tree model, aiming to minimize the classification error, and set the initial branching path according to the distribution characteristics of the parameters in different abnormal states;

[0068] In each training iteration, adjust the sample weights according to the classification error of the previous round, and construct a new decision tree through the residual minimization strategy. Combine the new decision tree with the previous decision tree to generate a gradient boosting decision tree model.

[0069] In some embodiments, first, the historical high-dimensional feature matrix set is cleaned and normalized. The feature matrix set contains multi-dimensional parameters such as current, voltage, temperature, humidity, etc. Each parameter may be affected by noise and generate outliers at different time periods. During the cleaning process, the system uses outlier detection methods (such as the three-standard-deviation method) to identify and remove the values that do not conform to the normal range, thereby improving the stability of the data. After the cleaning is completed, the feature parameters in the matrix are normalized so that the data is mapped between 0 and 1, ensuring that the scales of different parameters in the model are consistent, and providing basic support for the accuracy of the subsequent model. After the data preprocessing, the system screens the parameters in the high-dimensional feature matrix based on the feature importance algorithm SHAP (SHapley Additive exPlanations) to determine the key parameters that have a significant impact on the abnormal classification. During the SHAP value calculation process, the system evaluates the marginal contribution of each feature to the classification result one by one, and assigns higher weights to those features that have a greater impact on the model prediction result. For example, in the abnormal classification of the substation unit, the current and insulation resistance may have higher impact weights, while parameters such as temperature and humidity have less impact. Through the analysis of the SHAP values, the system screens out the key parameters with high importance, thereby effectively reducing the feature space, reducing the computational complexity of the model, and at the same time improving the sensitivity of the model to abnormal states. Based on the screened key parameters, the system constructs an initial decision tree model and sets the initial branch path with the goal of minimizing the classification error. When constructing the initial decision tree, the system divides the training data set into multiple layers of nodes and constructs branches according to the distribution characteristics of the key parameters in different abnormal states. For example, when the current exceeds a specific threshold but the temperature and humidity are still within the normal range, the system classifies it as the "fluctuation" state; if both the current and insulation resistance exceed the threshold, it is classified as the "fault" state. The division of each branch node is based on indicators such as information gain or Gini index to ensure that the data is optimally split at each layer of nodes, thereby improving the classification accuracy of the initial model. After the initial decision tree is constructed, the system enters the iterative training stage, with the residual minimization strategy as the core, and gradually optimizes the model. In each iteration, the system adjusts the weights of the samples based on the classification error of the previous round, assigns higher weights to the misclassified samples to increase the attention of the model in the subsequent decision trees. According to the adjusted sample weights, the system constructs a new decision tree to supplement the samples that the previous decision tree failed to correctly classify. In this process, through the residual minimization strategy, the new decision tree is combined with the previous decision tree layer by layer to ensure that the model can classify the abnormal features in the data set more accurately. The overall output result of the model is obtained by the integrated weighted average of multiple decision trees, and finally a complete gradient boosting decision tree model is formed.

[0070] Furthermore, the formula of the gradient boosting decision tree is as follows:

[0071] ;

[0072] Among them, is the prediction model for the m-th round of iteration; is the prediction model of the previous round; is the learning rate; is the number of samples in the training dataset; is the i-th sample; is the i-th sample the optimal weight corresponding to the residual; is the weak decision tree generated in this round.

[0073] Specifically, starting from an initial model, the prediction error is gradually reduced. By adding a new weak decision tree in each round of iteration, the fitting effect of the model on the samples is improved. In each round of iteration, the new decision tree focuses on fitting the error part of the previous round of model, thereby gradually optimizing the accuracy of the overall model. First, the system starts from a simple initial model, which usually uses the average value of the target variable in the training set as the predicted value. Since this initial model cannot accurately predict in most cases, certain errors will occur. The goal of GBDT is to reduce this error through iterative rounds, making the model gradually approach the true value. In each round of iteration, the system calculates the error of the current model for all samples, that is, the difference between the true value and the predicted value of each sample, and this difference is called the residual. In each round, the residual reflects the features that the current model fails to capture accurately. Therefore, the system constructs a new decision tree based on these residuals. The new decision tree is called a "weak decision tree", which is specifically used to fit these residuals to make up for the deficiencies of the current model. To control the learning process of the model and prevent overfitting, the system introduces a parameter called the "learning rate". The role of the learning rate is to control the influence of the new decision tree on the model in each round. By introducing the learning rate into the model, the system can ensure appropriate adjustment of the model in each round of iteration without causing large fluctuations in the prediction results. Each newly generated decision tree in each round is assigned a weight when added to the model, and this weight is determined by the optimization strategy of minimizing the residual. This means that for samples with larger current errors, the system will give higher weights to ensure that the next round of decision tree pays more attention to these difficult-to-fit samples. Therefore, the GBDT model gradually forms a more accurate prediction model by continuously accumulating the results of each round of weak decision trees. The final GBDT model is the weighted accumulation result of all the weak decision trees generated during the iterative process. In testing or actual applications, when new data is input, the system calculates through all the decision trees and accumulates the results of each tree to generate the final prediction result.

[0074] Further, the feature extraction of the state vector using the variational autoencoder and the self-attention mechanism according to the abnormal state includes the following steps:

[0075] Input the state vector into the encoding layer of the variational autoencoder, and perform high-dimensional feature encoding on the state vector through several layers of neural networks to generate a latent variable representation;

[0076] Apply the self-attention mechanism to the latent variable representation output by the variational autoencoder encoding to calculate the attention weights of each dimension feature;

[0077] Input the feature representation processed by the self-attention mechanism into the decoding layer of the variational autoencoder to generate a reconstructed state vector;

[0078] Calculate the difference between the reconstructed state vector and the original state vector, determine the abnormal feature vector, and match the corresponding potential fault according to the abnormal feature vector.

[0079] Specifically, the intelligent control system of the substation uses the variational autoencoder (VAE) and the self-attention mechanism to extract features from the state vector of the substation unit, so as to identify abnormal states and extract possible fault features. This process realizes high-dimensional feature encoding, feature weight allocation, and reconstruction difference calculation through a multi-layer neural network to determine abnormal states. First, the state vector is input into the encoding layer of the VAE. The encoder of the VAE consists of several layers of neural networks, which are used to compress the input high-dimensional state vector and extract its latent features. During the encoding process, the system extracts the high-dimensional features in the state vector layer by layer according to the weights and activation functions of each layer, and compresses them into a latent variable representation. This latent variable representation contains the key features of the state vector and retains the statistical information of the original state vector, providing a concise and effective representation for subsequent feature analysis. Taking multiple monitoring parameters such as current and voltage as examples, the encoding layer captures the joint characteristics of these parameters in a specific state through high-dimensional compression, enabling the latent variable to more effectively express the correlation between these features. Next, the self-attention mechanism is applied to the latent variable representation output by the VAE encoding to calculate the importance weights of each dimension of features. The role of the self-attention mechanism is to highlight the features that have a key impact on fault detection while reducing the impact of other unimportant features. The self-attention mechanism calculates the attention weights between different feature dimensions by constructing query, key, and value matrices. Specifically, the system first performs similarity matching of the latent variable representation with itself, and calculates the attention weight matrix according to the similarity between each dimension. For key parameters such as current and voltage in the substation unit, the self-attention mechanism will automatically assign higher weights to these features to ensure that the influence of these parameters is highlighted in subsequent analysis. For example, in an abnormal state with a significant change in current, the self-attention mechanism will assign a higher weight to the current feature, enabling the system to focus on these key parameters. Subsequently, the feature representation processed by the self-attention mechanism is input into the decoding layer of the VAE to generate a reconstructed state vector. The decoding layer restores the feature vector step by step to a space consistent with the input dimension to realize the reconstruction of the original state. The decoding layer has a similar structure to the encoding layer, and restores the encoded and compressed information to an approximate representation of the original state through layer-by-layer reverse restoration. Through this process, the system can generate a reconstructed state vector, whose value is close to the original state vector, but the feature structure is adjusted by the self-attention mechanism, highlighting the key features. Finally, the system calculates the difference between the reconstructed state vector and the original state vector to determine the abnormal feature vector. The difference calculation usually uses the mean square error (MSE) or a similarity index to represent the deviation degree of the reconstructed state from the original state. The magnitude of the difference value directly reflects the system's judgment on the normality of the current state, and the larger the difference, the higher the likelihood of abnormality.For example, under normal circumstances, the difference between the reconstructed vector and the original vector should be small; if a significant difference occurs, the system will mark this difference as an abnormal feature vector. By comparing this abnormal feature vector with a preset fault template or historical data, the potential fault type that matches it is determined. In this way, the system can not only effectively identify the current abnormal state but also initially locate the possible fault causes, thus providing an important basis for subsequent fault handling.

[0080] Further, the predicting of future power demand data based on the long short-term memory network from historical substation load data and real-time substation load data includes the following steps:

[0081] Denoise, smooth, and normalize the historical load data and real-time load data, convert them into a time series data matrix, and segment and construct time windows to generate a standardized load data input matrix;

[0082] Chunk the standardized load data input matrix along the time dimension, and map the load quantity and time stamps into a multi-dimensional feature vector sequence;

[0083] Add time labels and load change rates to the feature sequences within each time window to form a time-dependent matrix as the input to the long short-term memory network model;

[0084] Input the time-dependent matrix into the long short-term memory network model, perform forward and backward propagation training on each input time window, and optimize the model parameters through gradient descent;

[0085] Input the real-time load data sequence into the trained long short-term memory network model, and use the memory units to update the state of the sequence features;

[0086] Output the future power demand data through the output layer.

[0087] In some embodiments, first, the historical load data and real-time load data are denoised, smoothed, and normalized. The denoising operation removes high-frequency noise in the data through a filtering algorithm to eliminate the impact of abnormal fluctuations on load forecasting; the smoothing process uses a moving average or weighted average method to make the data curve more stable, facilitating the model to capture long-term trends. Subsequently, the normalization operation converts the data to a unified scale range (usually between 0 and 1) to ensure that data of different magnitudes have consistent numerical weights in the model. The processed data is organized into a time series data matrix, and time windows are constructed in segments at fixed time intervals to generate a standardized load data input matrix, providing a standardized data structure for subsequent model training. Based on the generated standardized load data input matrix, the data is chunked along the time dimension. This step maps the load quantity and timestamp into a multi-dimensional feature vector sequence, enabling the LSTM to capture the time-dependent relationship of load changes. Through chunking, the load data in the time series is converted into a structured input that can be understood by the LSTM. The chunked data not only retains the continuity of the load data but also is provided to the model in a segmented form, thus enhancing the flexibility and accuracy of prediction. Based on the chunked feature vector sequence, additional information such as time tags and load change rates is added to the feature sequence of each time window to generate a time-dependent matrix. The time tags indicate the relative positions of the time windows in the sequence, enabling the model to more effectively understand the time series relationship between short-term and long-term loads; the load change rate reflects the change speed and trend of the current load compared to the previous load, providing auxiliary information for the model to identify load peaks and valleys. The time-dependent matrix serves as the input matrix for the LSTM model, ensuring that the model can receive multi-dimensional time series features during training. The time-dependent matrix is input into the LSTM model, and the model starts the forward and backward propagation training for the input of each time window. During this process, the "memory unit" of the LSTM saves state information according to the data of the previous time window to capture the long-term trend of load changes; at the same time, the "forget gate" mechanism is used to discard useless information at appropriate times to ensure that the model can dynamically adjust the transmission and memory states of the data. Through forward propagation, the model updates the state based on the current input, and in backward propagation, the model optimizes the parameters based on the gradient descent method, gradually reducing the prediction error, thereby enhancing the accuracy and generalization ability of the model. After completing the model training, the real-time load data sequence is input into the trained LSTM model to update the state of the memory unit. The LSTM model can rely on its "long short-term memory" mechanism at each time step to adjust the state of the load data in real time and gradually accumulate the time series features of the data, enabling the prediction to more accurately reflect the changing trend of the real-time load. Finally, through the output layer of the LSTM, the model outputs the future power demand data. The output results include the predicted loads at multiple future time points. The model combines short-term memory and long-term memory through multiple rounds of iteration to generate accurate predicted values for future loads.This predicted data can be used as the input of the intelligent control module to support the substation in dynamically allocating loads and scheduling resources, thereby improving the response speed and efficiency of the system during actual operation. Through this LSTM model prediction process, the system can adapt to the fluctuations in power demand in real time and accurately, realizing the optimal management of power resources.

[0088] Furthermore, the steps for constructing the reinforcement learning model include:

[0089] Taking the predicted future power demand data as the environmental input state of the reinforcement learning model, initializing the operating states and demand loads of each substation unit in the substation to form the initial state matrix of the system;

[0090] For the operating states of the substation units, construct a set of control actions, including increasing, decreasing, or maintaining the number of operating substation units;

[0091] Construct a reward function such that the number of operations is negatively correlated with the reward value in the reinforcement learning model;

[0092] Through making policy decisions based on the current state matrix and the reward function at each time step, select the action with the highest reward value to adjust the number of operating substation units;

[0093] Record the load response, reward value, and state change data after each executed action, and feedback them to the model for updating the state matrix and optimizing the reinforcement learning model.

[0094] In some embodiments, first, the predicted future power demand data is used as the environmental input state of the reinforcement learning model, and at the same time, the current operating states and demand loads of each substation unit in the substation are initialized. The system organizes the initial parameters of each substation unit (such as the current load, power, current, etc.) and the predicted load demand into the initial state matrix of the system. This state matrix represents the overall operating state of the current system, including the working states and load demands of each substation unit, which provides a complete environmental description for the reinforcement learning model, enabling the model to understand the operating conditions of each substation unit under the given load demand. For the operating states of the substation units, the system constructs a set of control actions to cope with the fluctuations in load demand. The set of control actions includes increasing, decreasing, or maintaining the number of operating substation units, which are specifically defined as: increasing the number of operating units means starting additional substation units to cope with the rising future load; decreasing the number of operating units means shutting down some substation units to save energy; and maintaining the number of operating units is used to maintain the current working state when the load demand is stable. This set of actions provides operation options for the reinforcement learning model, enabling the system to make flexible adjustments under different load conditions. To ensure that the model can balance efficient operation and resource conservation when operating the substation units, the system constructs a reward function such that the number of operating units is negatively correlated with the reward value. The reward function considers the impact of the number of operating units on resource consumption. Increasing the number of operating units will result in a decrease in the reward value, while decreasing the number of operating units will increase the reward value. In this way, the model will tend to select fewer substation units on the premise of meeting the load demand to optimize resource utilization. For example, when the load demand is at a peak, the model will balance the load demand and resource cost through the reward function and minimize the number of operating units as much as possible while meeting the demand. In actual operation, the system makes policy decisions based on the current state matrix and the reward function at each time step, selects the action with the highest reward value, and adjusts the number of operating substation units. At each time step, the reinforcement learning model calculates the reward value corresponding to each action based on the current system state and the future load prediction, and selects the action that can maximize the reward to ensure that the system can achieve optimal control at future time points. For example, if the prediction shows that the load is about to rise, the model will moderately increase the number of operating substation units; while when the load drops, the model will choose to decrease the number of operating units to save resources. After each execution of an action, the system records the load response, reward value, and state change data after the execution of the action, and feeds these data back to the reinforcement learning model to update the state matrix and optimize the model's policy. In this feedback process, the model updates the policy parameters to make future decisions more accurate and efficient, thereby gradually enhancing the intelligence and adaptability of the model. This process continues, enabling the reinforcement learning model to continuously adapt to the actual load demand of the substation while also accumulating and optimizing its control strategy to achieve the optimal allocation and efficient operation of resources.

[0095] The above embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A standardized intelligent control system for a substation, characterized in that: include: A distributed monitoring module, a fault detection module, a load monitoring module and an intelligent control module, wherein the distributed monitoring module and the fault detection module are connected in sequence, and the load monitoring module and the intelligent control module are connected; The distributed monitoring module is used to obtain active parameters and passive parameters of the substation units in the substation; The fault detection module is used to receive active parameters and passive parameters of a single substation and standardize them to generate a substation state vector, perform preliminary abnormal classification based on a gradient boosting decision tree, and determine the abnormal state; according to the abnormal state, use a variational autoencoder and a self-attention mechanism to extract features from the state vector and detect potential faults; The load monitoring module is used to obtain and store substation load data; The intelligent control module is used to predict future power demand data through historical substation load data and real-time substation load data based on the long short-term memory network, and control the operating quantity of substation units without abnormalities and potential faults through a reinforcement learning model based on the future power demand data, and the operating quantity is negatively correlated with the reward value in the reinforcement learning model; The receiving active parameters and passive parameters of a single substation unit and standardizing and generating a substation unit state vector, and performing preliminary abnormality classification based on a gradient boosting decision tree comprises: Active parameters and passive parameters are fused to generate a high-dimensional feature matrix, and a real-time dynamic state vector is generated based on the matrix; According to the historical high-dimensional feature matrix set, a threshold matrix is ​​generated, and the adaptive threshold of each parameter is embedded in the state vector; Compare the state vector with the threshold matrix and classify the state vector based on the hierarchical path of the gradient boosting decision tree; The active parameters include current, voltage, power factor, temperature and humidity, operating status and insulation status, and the passive parameters include ambient temperature and humidity inside and outside the substation.

2. A standardized intelligent control system for a substation according to claim 1, characterized in that: The substation unit includes a main transformer, a high-voltage switch cabinet, a low-voltage switch cabinet, a circuit breaker, a busbar system, a cable connector and a lightning arrester.

3. A standardized intelligent control system for a substation according to claim 1, characterized in that: The distributed monitoring module includes a current sensor, a voltage sensor, a temperature and humidity sensor, an insulation detector and an embedded operation monitoring system.

4. A standardized intelligent control system for a substation according to claim 1, characterized in that: The receiving active parameters of a single substation unit and standardizing and generating a substation unit state vector, and performing preliminary abnormal classification based on a gradient boosting decision tree specifically include the following steps: De-noise the active and passive parameters and fuse them to generate a high-dimensional feature matrix; Based on the high-dimensional feature matrix, an adaptive normalization model is used to generate real-time dynamic state vectors; Generate a threshold matrix based on the historical high-dimensional feature matrix set through the gradient boosting decision tree, and embed the adaptive threshold of each parameter into the state vector; The state vector is compared with the threshold matrix, and the state vector is classified based on the hierarchical path of the gradient boosting decision tree to determine the abnormal state, which includes no abnormality, fluctuation and failure.

5. A standardized intelligent control system for a substation according to claim 4, characterized in that: The gradient boosting decision tree is constructed by the following steps: Clean and normalize the historical high-dimensional feature matrix set; Based on the parameters in the feature matrix, the key parameters that have a significant impact on anomaly classification are screened out through the SHAP value; Based on the selected key parameters, an initial decision tree model is constructed with the goal of minimizing the classification error, and the initial branch path is set according to the distribution characteristics of the parameters under different abnormal conditions; In each training iteration, the sample weights are adjusted according to the classification error of the previous round, and a new decision tree is constructed through the residual minimization strategy. The new decision tree is combined with the previous decision tree to generate a gradient boosting decision tree model.

6. A standardized intelligent control system for a substation according to claim 5, characterized in that: The formula for the gradient boosting decision tree is as follows: ； in, is the prediction model of the mth iteration; is the prediction model of the previous round; is the learning rate; is the number of samples in the training data set; is the i-th sample; is the i-th sample The optimal weight corresponding to the residual of ; Weak decision tree generated for this round.

7. A standardized intelligent control system for a substation according to claim 5, characterized in that: The method of extracting features from the state vector using a variational autoencoder and a self-attention mechanism according to the abnormal state includes the following steps: The state vector is input into the encoding layer of the variational autoencoder, and the state vector is encoded with high-dimensional features through several layers of neural networks to generate latent variable representation; Apply the self-attention mechanism to the latent variable representation of the variational autoencoder output and calculate the attention weight of each dimension feature; The feature representation processed by the self-attention mechanism is input into the decoding layer of the variational autoencoder to generate a reconstructed state vector; The difference between the reconstructed state vector and the original state vector is calculated to determine the abnormal feature vector, and the corresponding potential fault is matched according to the abnormal feature vector.

8. A standardized intelligent control system for a substation according to claim 1, characterized in that: The method of predicting future power demand data based on the long short-term memory network by using historical substation load data and real-time substation load data includes the following steps: De-noising, smoothing and normalizing the historical load data and real-time load data, converting them into a time series data matrix, and constructing time windows in segments to generate a standardized load data input matrix; The standardized load data input matrix is ​​divided into blocks according to the time dimension, and the load amount and timestamp are mapped into a multi-dimensional feature vector sequence; Add time labels and load change rates to the feature sequences in each time window to form a time dependency matrix as the input of the long short-term memory network model; The time dependency matrix is ​​input into the long short-term memory network model, and forward and backward propagation training is performed on each input time window, and the model parameters are optimized by gradient descent; Input the real-time load data sequence into the trained long short-term memory network model, and use the memory unit to update the state of the sequence features; The future electricity demand data is output through the output layer.

9. A standardized intelligent control system for a substation according to claim 8, characterized in that: The steps of constructing the reinforcement learning model include: The predicted future power demand data is used as the environmental input state of the reinforcement learning model to initialize the operating state and demand load of each substation unit in the substation to form the initial state matrix of the system; According to the operation status of the substation unit, a set of control actions is constructed, including increasing, decreasing or maintaining the number of substation units in operation; Construct the reward function so that the number of runs is negatively correlated with the reward value in the reinforcement learning model; By making strategic decisions based on the current state matrix and reward function in each time step, the action with the highest reward value is selected to adjust the number of substation operations; The load response, reward value, and state change data after each action are recorded and fed back to the model to update the state matrix and optimize the reinforcement learning model.

Images