Power supply energy analysis method and system for medium-voltage power distribution network

Abstract

The invention provides a power supply energy analysis method and system for a medium-voltage power distribution network. The method comprises the following steps: collecting multi-modal data of the medium-voltage power distribution network; constructing a health state evaluation model, and predicting a health score and a fault risk of the transformer of the medium-voltage power distribution network according to the fusion features after dimension reduction; constructing an optimization target of the power electronic equipment of the medium-voltage distribution network according to the health score and the fault probability of the transformer; constructing a multi-modal load prediction model to predict the load change of the power electronic equipment in the future; constructing a multi-objective optimization function, and generating an optimal scheduling strategy; and executing the optimal scheduling strategy to actually control the medium-voltage power distribution network equipment. According to the method, the defects of a traditional method in the aspects of equipment monitoring, load prediction and scheduling optimization are overcome, and the overall management level and coping capacity of a power grid are improved by introducing a leading-edge artificial intelligence algorithm and a data fusion technology.

Description

Technical Field

[0001] The present invention belongs to the technical field of medium voltage distribution networks, and in particular relates to a power supply energy analysis method and system for medium voltage distribution networks. Background Art

[0002] The medium-voltage distribution network is a key link in the power transmission and distribution system. It is responsible for converting high-voltage electricity into medium-voltage electricity suitable for end users and distributing it. The energy analysis and dispatching management of traditional medium-voltage distribution networks mainly rely on static load forecasting and manual experience judgment, which faces many challenges in actual operation. First, with the development of social economy and the diversification of power demand, the load characteristics of medium-voltage distribution networks have become more complex. Traditional load forecasting models cannot accurately cope with frequent fluctuations and short-term changes in loads, resulting in reduced power supply reliability and energy waste. Secondly, a large number of power equipment (such as transformers, power electronic equipment, etc.) in medium-voltage distribution networks may deteriorate rapidly due to long-term high-load operation or harsh environment. Traditional regular maintenance methods are difficult to detect potential equipment failures and hidden dangers in time, which can easily cause sudden power outages. Third, the variability of environmental conditions (such as meteorological conditions, ambient temperature, etc.) has a significant impact on power loads and equipment status, and traditional dispatching strategies usually fail to fully consider these external factors, resulting in delayed response when dealing with extreme weather or emergencies.

[0003] In the current research and technological development, although some systems have begun to introduce certain intelligent means, such as load forecasting and equipment monitoring based on big data, these methods still have limitations. Most systems lack the ability to comprehensively analyze multi-dimensional data, such as load data, equipment health data, meteorological and environmental data, etc., which have not been effectively integrated, resulting in the accuracy and efficiency of power supply optimization being still unsatisfactory. At the same time, the existing intelligent systems are mostly focused on data analysis, lack effective real-time response mechanisms and adaptive scheduling strategies, and it is difficult to provide stable and reliable power supply when the power grid status changes rapidly. In addition, the algorithm models of many systems are relatively simple, such as relying only on traditional time series analysis or simple machine learning models, and it is difficult to fully utilize the advantages of the latest algorithms such as deep learning and reinforcement learning to achieve more refined and intelligent management of the power grid. Therefore, the existing technology still has significant deficiencies in improving the energy utilization efficiency, equipment safety and power supply stability of medium-voltage distribution networks. Summary of the invention

[0004] The purpose of the present invention is to design a power supply energy analysis method and system for medium-voltage distribution networks, and to achieve comprehensive intelligent optimization management of medium-voltage distribution networks by introducing a variety of advanced artificial intelligence algorithms and data fusion technologies. By integrating transformer health status assessment, dynamic optimization control of power electronic equipment, and multimodal fusion analysis of meteorological and environmental data, the present invention can not only improve the accuracy of power grid load forecasting, but also realize real-time monitoring and intelligent scheduling of equipment status, significantly improving the overall operation efficiency and safety of the power grid.

[0005] In order to achieve the above object, a first aspect of the present invention provides a method for analyzing power supply energy for a medium voltage distribution network, the method comprising the following steps: Step 1: Collect multimodal data of the medium voltage distribution network and perform preprocessing, including: S101. Design a standardization formula with absolute median deviation to standardize the original data of different modes respectively; S102, dynamically adjusting denoising parameters using adaptive wavelet denoising combined with nonlinear smoothing on the standardized data, retaining important feature information, and obtaining filtered wavelet decomposition coefficients; S103, introducing a multimodal dynamic time warping algorithm to perform time alignment processing on the filtered wavelet decomposition coefficients; S104, after the time alignment process, using a feature extraction method based on Laplace eigenmap to perform feature extraction, fusion and dimensionality reduction process to obtain fusion features after dimensionality reduction; Step 2: Build a health status assessment model to predict the health score and fault risk of the transformer in the medium-voltage distribution network based on the fusion features after dimensionality reduction, and design equipment optimization measures to be executed when the threshold is triggered, including: S201, using the fusion features after dimension reduction as input vectors of a health status assessment model, wherein each input vector contains multimodal data of the transformer, and constructing an input vector set according to the input vectors, and constructing a dynamic weighted health status assessment model according to the input vector set, wherein the dynamic weighted health status assessment model evaluates the health score of the transformer according to the dynamic changes of the current feature value and historical data; S202. construct a fault probability prediction model based on a Bayesian dynamic network according to the health score of the transformer, and use the fault probability prediction model to calculate the failure probability of the transformer under given conditions; S203. Determine whether to trigger equipment optimization measures based on the health score and failure probability of the transformer: If the health score is lower than the preset threshold, or the failure probability is higher than the set risk threshold, the equipment optimization strategy is executed; Step 3: Construct the optimization target of the power electronic equipment of the medium voltage distribution network according to the health score and failure probability of the transformer, and optimize the operating parameters of the power electronic equipment, including: S301, constructing an optimization objective function for power electronic equipment based on the current health status score and failure probability of the transformer, minimizing reactive power loss and harmonic distortion, and ensuring that the equipment operates within a health status threshold; S302, constructing an optimization model based on adaptive model predictive control, performing optimal control on power electronic equipment in a dynamic environment, calculating a control input vector, and using a dynamic optimization algorithm to solve the optimal control input vector so that the objective function is minimized while satisfying the safety constraints of health status and failure probability; S303, adjusting the operating parameters of the power electronic equipment according to the optimal control input vector; Step 4: Construct a multi-modal load prediction model to predict the future load changes of power electronic equipment based on the adjusted operating parameters of the power electronic equipment and the current state of the power electronic equipment, including: S401, constructing an extended multimodal feature input vector, constructing a multimodal deep spatiotemporal convolutional network to perform load forecasting according to the multimodal feature input vector, and at the same time introducing a regularization term based on a hybrid kernel function before the output of the multimodal deep spatiotemporal convolutional network, and then outputting the predicted load forecast; S402, constructing a dispatch strategy optimization objective function according to load forecasting to obtain a power grid dispatch strategy; Step 5: Construct a multi-objective optimization function to optimize the power grid dispatching strategy in a dynamic environment with multiple constraints and generate the optimal dispatching strategy, including: Combine load forecasting, equipment cost, energy loss and power quality constraints to design a multi-objective scheduling optimization function. Combine particle swarm optimization algorithm and reinforcement learning to build a dynamic scheduling optimization model. Optimize the optimal scheduling strategy of the power grid under dynamic load and equipment status changes and solve the optimal scheduling strategy. Step 6: Execute the optimal dispatching strategy to actually control the medium-voltage distribution network equipment, and introduce an adaptive optimization feedback mechanism to dynamically adjust the dispatching strategy by real-time monitoring of equipment status and load response to ensure stable operation of the power grid in a changing environment.

[0006] Furthermore, the method of adaptive wavelet denoising combined with nonlinear smoothing specifically includes: Perform multi-scale wavelet decomposition on the standardized data, extract signal features in different frequency bands, and calculate the wavelet decomposition coefficients respectively; For each wavelet decomposition coefficient, an adaptive filtering strategy based on frequency band energy ratio is introduced to calculate and obtain the filtered wavelet decomposition coefficient; The filtered wavelet decomposition coefficients are post-processed by a nonlinear smoothing function to remove noise again; The multimodal dynamic time warping algorithm is implemented by minimizing the Euclidean distance between different data; The feature extraction method based on Laplace eigenmapping uses a Laplace mapping matrix to perform feature dimension reduction according to the global structure and local similarity of data.

[0007] Furthermore, the health score of the dynamic weighted health status assessment model is expressed as follows: , in, Indicates the health status score at the current moment, indicating the health level of the transformer; represents the bias term, which indicates the initial value of the health score; represents the regression coefficient, indicating the contribution of each characteristic to the health score; represents the first feature vector after dimensionality reduction Features The weight coefficient representing the historical characteristics; A function that represents the dynamic relationship between current features and historical features; The failure probability prediction formula of the failure probability prediction model is expressed as follows: , in, Indicates at time The probability of failure; Indicates the current health status score; Critical thresholds that indicate health status, indicating high-risk areas for failure; Represents the historical feature vector, which represents the health data at the past moment.

[0008] Furthermore, the S302 further includes: The optimization model based on adaptive model predictive control dynamically adjusts the optimization parameters according to the real-time equipment health status score and failure probability to minimize the optimization objective function. , while satisfying the safety operation constraints, the state update formula of the dynamic optimization control model is expressed as follows: , The control input update is expressed as follows: , in, Indicates the current time The device state vector of Represents the system state transfer matrix and control input matrix, describing the relationship between device state and control input; represents the control input vector, which indicates the equipment operating parameters that need to be adjusted; It represents the state adjustment item based on the health status score and the failure probability, reflecting the dynamic impact of the health status and the failure probability on the equipment status; represents the penalty coefficient, which is used to weigh the safety margin of health status and failure probability in the optimization objective; represents the optimization objective function; represents the health status score; Safety thresholds indicating health status; represents the probability of failure; A safety threshold representing the probability of failure; The optimal control input vector is solved by using a dynamic optimization algorithm, which is expressed as follows: , in, Represents the learning rate, which is used to control the step size of gradient descent; represents the objective function with respect to the control input The gradient indicates how to adjust To minimize the objective function; Through multiple iterations of the dynamic optimization algorithm, the optimal control input vector is found to ensure that the equipment operates safely in an optimized state.

[0009] Furthermore, the objective optimization function , which is expressed as follows: , in, Indicates reactive power loss, which indicates the energy loss of the equipment during operation; Indicates the harmonic distortion, indicating the impact of harmonics on power quality; Represents the weight coefficient used to balance the importance of loss and harmonic distortion in the optimization.

[0010] Further, in said S401, said constructing an extended multi-modal feature input vector specifically includes: adding historical load data, time characteristics and external influencing factors to the operating parameters and current state information of the power electronic equipment; The regularization term based on the hybrid kernel function is expressed as follows: , in, represents the regularization term, which is used to prevent overfitting and improve the robustness of the model; Represents the weight parameter in the MDSTCN model; Represents L2 norm regularization, which penalizes the complexity of model parameters; represents the hybrid kernel function, which measures the similarity between different feature vectors; n represents the total number of features, and m represents the total number of devices.

[0011] Furthermore, the scheduling strategy optimization objective function is expressed as follows: , in: represents the dispatch strategy optimization objective function, which measures the deviation between the predicted load and the actual load, the model complexity and the equipment operation cost; Indicates the length of the prediction time window; represents the predicted load value; Indicates the actual load; Represents the regularization weight coefficient, which controls the impact of model complexity on the objective function; Represents the cost weight coefficient, balancing the impact of load deviation and equipment cost; Indicates The operating cost of each device; Indicates The energy loss of a device.

[0012] Furthermore, the multi-objective scheduling optimization function , which is expressed as follows: , in: The weight coefficient representing the objective function is used to balance the load forecast error, equipment operating cost and power quality; Indicates the length of the prediction time window; Indicates the load forecast value; Indicates the actual load; Indicates The operating cost of each device; Indicates Energy loss of each device; Indicates Power quality deviation; The state update formula of the dynamic scheduling optimization model is as follows: , The control strategy optimization formula is as follows: , in, Indicates the current time The grid state vector of Indicates the current time The rate of change of state; Indicates the time step, which indicates the time interval between state updates; It represents the state adjustment item based on load forecast and power quality deviation, reflecting the dynamic impact of load and power quality on the power grid state; Represents the optimized control strategy vector, which represents the next moment The dispatch control parameters; Represents the penalty coefficient, which is used to weigh different constraints in the optimization objective; Indicates the control strategy Constraints of Indicates the status of the power grid constraints.

[0013] Furthermore, the iterative formula of the optimal scheduling strategy is expressed as follows: , in, Indicates The scheduling strategy for the next iteration; Indicates The scheduling strategy for the next iteration; Represents the learning rate, which is used to control the convergence speed of the optimization algorithm; Represents the objective function with respect to the control strategy The gradient indicates how to adjust To minimize the objective function.

[0014] In a second aspect of the present invention, a power supply energy analysis system for a medium voltage distribution network is provided, the system comprising: The data acquisition module is used to collect multi-modal data of the medium voltage distribution network and perform preprocessing. The execution steps specifically include: S101. Design a standardization formula with absolute median deviation to standardize the original data of different modes respectively; S102, dynamically adjusting denoising parameters using adaptive wavelet denoising combined with nonlinear smoothing on the standardized data, retaining important feature information, and obtaining filtered wavelet decomposition coefficients; S103, introducing a multimodal dynamic time warping algorithm to perform time alignment processing on the filtered wavelet decomposition coefficients; S104, after the time alignment process, using a feature extraction method based on Laplace eigenmap to perform feature extraction, fusion and dimensionality reduction process to obtain fusion features after dimensionality reduction; The health assessment module is used to build a health status assessment model, predict the health score and fault risk of the transformer in the medium-voltage distribution network based on the fusion features after dimension reduction, and design equipment optimization measures to be executed when the threshold is triggered. The specific execution steps include: S201, using the fusion features after dimension reduction as input vectors of a health status assessment model, wherein each input vector contains multimodal data of the transformer, and constructing an input vector set according to the input vectors, and constructing a dynamic weighted health status assessment model according to the input vector set, wherein the dynamic weighted health status assessment model evaluates the health score of the transformer according to the dynamic changes of the current feature value and historical data; S202. construct a fault probability prediction model based on a Bayesian dynamic network according to the health score of the transformer, and use the fault probability prediction model to calculate the failure probability of the transformer under given conditions; S203. Determine whether to trigger equipment optimization measures based on the health score and failure probability of the transformer: If the health score is lower than the preset threshold, or the failure probability is higher than the set risk threshold, the equipment optimization strategy is executed; The equipment optimization module is used to construct the optimization target of the power electronic equipment of the medium voltage distribution network according to the health score and fault probability of the transformer, and optimize the operating parameters of the power electronic equipment. The specific execution steps include: S301, constructing an optimization objective function for power electronic equipment based on the current health status score and failure probability of the transformer, minimizing reactive power loss and harmonic distortion, and ensuring that the equipment operates within a health status threshold; S302, constructing an optimization model based on adaptive model predictive control, performing optimal control on power electronic equipment in a dynamic environment, calculating a control input vector, and using a dynamic optimization algorithm to solve the optimal control input vector so that the objective function is minimized while satisfying the safety constraints of health status and failure probability; S303, adjusting the operating parameters of the power electronic equipment according to the optimal control input vector; The load prediction module is used to construct a multi-modal load prediction model to predict the future load changes of the power electronic equipment according to the adjusted operating parameters of the power electronic equipment and the current state of the power electronic equipment. The execution steps specifically include: S401, constructing an extended multimodal feature input vector, constructing a multimodal deep spatiotemporal convolutional network to perform load forecasting according to the multimodal feature input vector, and at the same time introducing a regularization term based on a hybrid kernel function before the output of the multimodal deep spatiotemporal convolutional network, and then outputting the predicted load forecast; S402, constructing a dispatch strategy optimization objective function according to load forecasting to obtain a power grid dispatch strategy; The multi-objective optimization module is used to construct a multi-objective optimization function, optimize the power grid dispatching strategy in a dynamic environment with multiple constraints, and generate the optimal dispatching strategy. The specific execution steps include: Combine load forecasting, equipment cost, energy loss and power quality constraints to design a multi-objective scheduling optimization function. Combine particle swarm optimization algorithm and reinforcement learning to build a dynamic scheduling optimization model. Optimize the optimal scheduling strategy of the power grid under dynamic load and equipment status changes and solve the optimal scheduling strategy. The system optimization module is used to execute the optimal dispatching strategy to actually control the medium-voltage distribution network equipment, and introduce an adaptive optimization feedback mechanism to dynamically adjust the dispatching strategy by real-time monitoring of equipment status and load response to ensure the stable operation of the power grid in a changing environment.

[0015] The beneficial technical effects of the present invention are at least as follows: The present invention introduces advanced deep learning algorithms (such as long short-term memory networks (LSTM)) to monitor and evaluate the health status of transformers in real time, including multiple key parameters such as temperature, load current, and oil temperature. By establishing a high-precision prediction model, the present invention can identify potential transformer failure risks in advance, provide early warnings, and avoid power outages caused by sudden equipment failures. This innovation directly solves the problems of untimely equipment status monitoring and insufficient fault prediction capabilities in the prior art.

[0016] The present invention uses convolutional neural network (CNN) and reinforcement learning algorithm to perform harmonic analysis and reactive power compensation optimization on power electronic equipment in medium voltage distribution network. The system can dynamically adjust the operation strategy of power electronic equipment according to real-time load changes and grid status to optimize power factor and minimize energy loss. This innovation effectively solves the problem of single control strategy and lack of dynamic optimization capability of power electronic equipment in the prior art, and significantly improves the energy efficiency and power quality of the grid.

[0017] The present invention uses a multimodal deep learning model to integrate meteorological data (such as temperature, humidity, wind speed) and environmental data (such as sunshine intensity and air quality) to achieve high-precision prediction of grid load changes. Based on the prediction results, the system conducts power dispatch and load distribution in advance to ensure power supply stability under extreme weather or special environmental conditions. This innovation makes up for the lack of consideration of environmental factors and insufficient prediction accuracy in the existing technology, and enhances the adaptability and response speed of the power grid under complex environmental conditions.

[0018] The present invention integrates multiple optimization strategies, and comprehensively analyzes transformer status, power electronic equipment operating parameters, load forecast results, and environmental data through a comprehensive optimization module to achieve global optimal dispatch of the power grid. The system uses edge computing technology to ensure low-latency data processing and decision-making capabilities, and can quickly respond and adjust strategies when the power grid status changes. This function solves the problem of delayed response and untimely adjustment of existing systems when responding to emergencies, and further improves the reliability and stability of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0020] Figure 1 The present invention is a flowchart of a method for analyzing power supply energy used in a medium voltage distribution network according to an embodiment of the present invention.

[0021] Figure 2 The present invention is a framework diagram of a power supply energy analysis system for a medium voltage distribution network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0023] In one or more embodiments, Figure 1 As shown, the present invention discloses a power supply energy analysis method for a medium voltage distribution network, the method comprising steps one to six, including: Step 1: Collect multimodal data of the medium voltage distribution network and perform preprocessing, including: S101. Design a standardization formula with absolute median deviation to standardize the original data of different modes respectively; Specifically, since multimodal data (e.g., transformer temperature, voltage, load current, ambient temperature, etc.) have different units and distributions, in order to ensure the comparability of different data sources, the present invention needs to standardize all data. Conventional standardization methods cannot effectively deal with various abnormal conditions in medium-voltage distribution networks (e.g., short-term voltage spikes or sensor drift). Therefore, the present invention introduces an adjustment term based on the median absolute deviation (MAD, Median Absolute Deviation) to enhance the ability to resist outliers. The standardization formula is: , in, For raw data (such as transformer temperature , current wait). is the data mean of the current window. is the standard deviation of the data in the current window. is the absolute median deviation, which measures the dispersion of the data. is the adjustment factor, which is used to balance the impact of standard deviation and MAD. For patent scenarios, it is set to To enhance robustness.

[0024] Calculation example: For the transformer temperature data with a sampling frequency of 1Hz in the past minute , the present invention calculates: Mean: ; Standard Deviation: ; MAD: ; Normalize data: ; This method can better maintain the stability of data when dealing with data outliers and ensure the accuracy of subsequent analysis.

[0025] S102, dynamically adjusting denoising parameters using adaptive wavelet denoising combined with nonlinear smoothing on the standardized data, retaining important feature information, and obtaining filtered wavelet decomposition coefficients; Specifically, the data of the medium voltage distribution network is often interfered by transient noise, such as switching noise of power equipment or external electromagnetic interference. Traditional denoising methods (such as mean filtering) will lose important dynamic information. In order to solve this problem, the present invention proposes an adaptive wavelet denoising combined with nonlinear smoothing method, which can dynamically adjust the denoising parameters and retain important feature information.

[0026] First, the standardized data Perform multi-scale wavelet decomposition to extract signal features in different frequency bands: , in, It is The coefficients of the layer wavelet decomposition represent the characteristics of the signal at this scale. It is The wavelet basis function of the layer.

[0027] For each wavelet decomposition coefficient , an adaptive filtering strategy based on frequency band energy ratio is introduced: , in, It is layer of energy. is the total energy.

[0028] The filtered coefficient The high-frequency noise is filtered out through the exponential decay strategy to retain the signal characteristics.

[0029] Next, the wavelet reconstructed signal is post-processed through a nonlinear smoothing function (such as a bilateral filter) to further remove noise: , Calculation example: Harmonic data for power electronics equipment , perform 3-layer wavelet decomposition to obtain energy , total energy .

[0030] Calculate the band energy ratio filter: ; Reconstruct the denoised signal: ; This denoising process can dynamically adjust the sensitivity to different frequency bands and improve the signal stability and information retention ability.

[0031] S103, introducing a multimodal dynamic time warping algorithm to perform time alignment processing on the filtered wavelet decomposition coefficients; Specifically, due to the inconsistent data acquisition frequency in the medium voltage distribution network (for example, the sampling frequency of transformer status data may be different from the sampling frequency of power electronic equipment), time alignment and feature extraction must be performed in order to perform unified analysis. Here, the present invention introduces multi-modal dynamic time warping (MDTW) to optimize the alignment process of multi-modal data.

[0032] Among them, time alignment is achieved by minimizing the Euclidean distance between different data: , in, and These are two groups of data samples that have been denoised and standardized respectively. is the time alignment path, finding the alignment with the minimum cost. is the time offset penalty coefficient, balancing spatial distance and time alignment.

[0033] Calculation example: Voltage for power electronics and transformer temperature , set the time penalty coefficient ; Compute time-aligned paths , so that the alignment cost is minimized;

[0034] Get the aligned data and , ensuring that multimodal data are analyzed on a uniform time scale;

[0035] This innovative time alignment method takes into account the time and amplitude deviations between different data and helps to improve the accuracy of multimodal data fusion.

[0036] S104, after the time alignment process, using a feature extraction method based on Laplace eigenmap to perform feature extraction, fusion and dimensionality reduction process to obtain fusion features after dimensionality reduction; Specifically, after completing the time alignment, multiple data sources need to be fused into a high-dimensional feature vector for unified analysis. To this end, the present invention proposes a feature extraction method based on Laplacian Feature Mapping (LFM) to retain the global structural information and local features of the data.

[0037] Feature fusion is expressed as: , Feature dimensionality reduction through Laplace eigenmap: , in It is the Laplace mapping matrix, which takes into account the global structure and local similarity of the data, ensuring that the features after dimensionality reduction can still accurately describe the power grid status.

[0038] Calculation example: For the fused data : Laplace Map Matrix ; Dimensionality reduction features ; This feature extraction and dimensionality reduction method can reduce the amount of calculation while retaining the main feature information, providing efficient and accurate input data for subsequent steps.

[0039] Step 2: Build a health status assessment model to predict the health score and fault risk of the transformer in the medium-voltage distribution network based on the fusion features after dimensionality reduction, and design equipment optimization measures to be executed when the threshold is triggered, including: S201, using the fusion features after dimension reduction as input vectors of a health status assessment model, wherein each input vector contains multimodal data of the transformer, and constructing an input vector set according to the input vectors, and constructing a dynamic weighted health status assessment model according to the input vector set, wherein the dynamic weighted health status assessment model evaluates the health score of the transformer according to the dynamic changes of the current feature value and historical data; Specifically, according to the feature vector extracted after dimensionality reduction , these feature vectors include key transformer status information (such as aligned transformer temperature, load current, ambient temperature, etc.), and the present invention defines these features as the input vectors of the health status assessment model .

[0040] , in, is the feature after dimensionality reduction, and each feature is the result of feature dimensionality reduction processing in step 1, such as Dimensionality reduction results.

[0041] Furthermore, using the input feature vector The present invention constructs a dynamic weighted health status assessment model, which evaluates the health score (HealthScore, ). This score is used to quantify the current operating health of the transformer, with lower values ​​indicating worse equipment condition.

[0042] The health score calculation formula is: , in, Indicates the health status score at the current moment, indicating the health level of the transformer. Represents the bias term, which indicates the initial value of the health score. represents the regression coefficient, indicating the contribution of each feature to the health score. represents the first feature vector after dimensionality reduction Features. Represents the weight coefficient of historical characteristics. A function that represents the dynamic relationship between current features and historical features (such as the rate of change, difference, etc. of features).

[0043] Calculation example: For the feature vector , assuming , and the dynamic characteristic function : , The health score reflects the overall health status of the current transformer.

[0044] S202. construct a fault probability prediction model based on a Bayesian dynamic network according to the health score of the transformer, and use the fault probability prediction model to calculate the failure probability of the transformer under given conditions; Specifically, in order to further evaluate the failure risk of the transformer, a failure probability prediction model based on a Bayesian dynamic network is used. This model takes into account the changing trend of the current health score and time series characteristics and calculates the failure probability of the transformer under given conditions. The failure probability prediction formula is: , in, Indicates at time The probability of failure. Indicates the current health status score. Indicates critical thresholds of health status, indicating areas of high risk of failure. Represents the historical feature vector, which represents the health data at the past moment.

[0045] Calculation example: Assuming current health score , the critical threshold is set to , the probability density function estimated based on historical data for , then the failure probability Calculated as: , This value means that under the current state, the probability of transformer failure is 75%.

[0046] S203. Determine whether to trigger equipment optimization measures based on the health score and failure probability of the transformer: If the health score is lower than the preset threshold, or the failure probability is higher than the set risk threshold, the equipment optimization strategy is executed; Specifically, according to the health score and failure probability , the system determines whether to trigger equipment optimization measures (for example, adjust load, increase cooling power, etc.). Below the preset threshold , or the probability of failure Above the set risk threshold , the device optimization strategy is executed. The optimization trigger conditions are: , in, Indicates the health score threshold for device optimization. : Risk threshold of failure probability. : Failure probability at the current moment.

[0047] Calculation example: If the current health score And set the device optimization threshold ,but: , As a result, the system triggers equipment optimization measures, such as adjusting electrical loads or increasing cooling.

[0048] Step 3: Construct the optimization target of the power electronic equipment of the medium voltage distribution network according to the health score and failure probability of the transformer, and optimize the operating parameters of the power electronic equipment, including: S301, constructing an optimization objective function for power electronic equipment based on the current health status score and failure probability of the transformer, minimizing reactive power loss and harmonic distortion, and ensuring that the equipment operates within a health status threshold; Specifically, the optimization target of the power electronic equipment is determined based on the output data. The optimization of power electronic equipment (such as reactive power compensation devices, harmonic filters, etc.) needs to consider the health status score of the current transformer. and failure probability The goal is to minimize reactive power loss and harmonic distortion while ensuring that the equipment operates within the health threshold. Suppose the optimization objective function for: , in, Indicates reactive power loss, which indicates the energy loss of the equipment during operation. Indicates harmonic distortion, which indicates the impact of harmonics on power quality. Represents the weight coefficient used to balance the importance of loss and harmonic distortion in the optimization. The optimization constraint is the health score and failure probability The set safety threshold cannot be exceeded: , in, The safety threshold of the health status indicates the minimum safety score for device operation. The safety threshold of the failure probability indicates the maximum allowed failure probability.

[0049] S302, constructing an optimization model based on adaptive model predictive control, performing optimal control on power electronic equipment in a dynamic environment, calculating a control input vector, and using a dynamic optimization algorithm to solve the optimal control input vector so that the objective function is minimized while satisfying the safety constraints of health status and failure probability; Specifically, in order to achieve optimal control of power electronic equipment under dynamic conditions, an optimization model based on adaptive model predictive control (AMPC) is constructed. AMPC scores the real-time equipment health status. and failure probability , dynamically adjust the optimization parameters to minimize the objective function , while satisfying the safety operation constraints. The state update formula of the dynamic optimization control model is: , The control input update formula is: , in, Indicates the current time The device state vector includes operating parameters such as voltage, current, and temperature. Represents the system state transfer matrix and control input matrix, describing the relationship between device state and control input. Represents the control input vector, which indicates the equipment operating parameters that need to be adjusted (such as compensation voltage, filter current, etc.). It represents the status adjustment item based on the health status score and failure probability, reflecting the dynamic impact of the health status and failure probability on the equipment status. Represents the penalty coefficient, which is used to weigh the safety margin of the health state and failure probability in the optimization objective.

[0050] Used to adjust the state equation according to the current equipment health status and failure probability to ensure that the equipment can still be effectively optimized under different health conditions. and Ensure that the optimization results do not violate the safety constraints of health status and failure probability.

[0051] Furthermore, a dynamic optimization algorithm (such as gradient descent or Newton's method) is used to solve the control input , so that the objective function Minimum, while satisfying the safety constraints of health state and failure probability. Calculate the optimal control input The iteration formula is: , in, Represents the learning rate, which is used to control the step size of gradient descent. represents the objective function with respect to the control input The gradient indicates how to adjust To minimize the objective function.

[0052] Calculation example: Assume that the initial control input is , weight coefficient , health status score , failure probability , safety threshold , penalty coefficient , learning rate .

[0053] Optimized control input for the first iteration Calculated as: Calculate the objective function (Calculated based on actual data and ); Computing Gradients ; Update control input ; Through multiple iterations, the optimal control input is found , ensuring that the equipment operates safely in an optimized state.

[0054] S303, adjusting the operating parameters of the power electronic equipment according to the optimal control input vector; Specifically, according to the optimal control input obtained , adjust the operating parameters of power electronic equipment to minimize reactive power loss and harmonic distortion. Specific adjustments include changing the compensation amount of the reactive power compensation device, adjusting the filtering characteristics of the filter, etc., to dynamically respond to changes in the health status of the transformer and ensure that the equipment operates in the best condition.

[0055] The implementation steps are as follows: Read the control input obtained by optimization calculation ;

[0056] Update the operating parameters of the equipment (such as voltage adjustment, compensation current setting, etc.);

[0057] Monitor the operating status of the adjusted equipment in real time to ensure compliance with safety requirements.

[0058] Step 4: Construct a multi-modal load prediction model to predict the future load changes of power electronic equipment based on the adjusted operating parameters of the power electronic equipment and the current state of the power electronic equipment, including: S401, constructing an extended multimodal feature input vector, constructing a multimodal deep spatiotemporal convolutional network to perform load forecasting according to the multimodal feature input vector, and at the same time introducing a regularization term based on a hybrid kernel function before the output of the multimodal deep spatiotemporal convolutional network, and then outputting the predicted load forecast; Specifically, the optimized device parameters and current status information are extracted from the output to form a multimodal feature input vector In addition, in order to improve the accuracy of the prediction, the present invention further expands the feature vector by adding historical load data, time characteristics (such as season, time period) and external influencing factors (such as weather conditions). The expanded feature vector Contains optimized compensation voltage ( )、Filter current( )、Current Voltage( )、Current( )、Power Factor( )、Historical Load( ), time characteristics( ), and external factors ( ) etc. The eigenvector is defined as: , in, Represents the optimized compensation voltage. Represents the optimized filter current. Indicates the current voltage status of the device. Indicates the current status of the device. Indicates the current power factor status of the device. Represents historical load data, capturing load variation patterns over time. Indicates time characteristics, such as time, week, season, etc., and indicates the periodicity of the load. Represents external influencing factors, such as weather conditions (temperature, humidity, wind speed, etc.).

[0059] By expanding the feature input vector, the present invention more comprehensively captures various influencing factors of power grid load changes, which helps to improve the accuracy and reliability of prediction.

[0060] Furthermore, using the expanded feature vector , this paper proposes an innovative multi-modal deep spatiotemporal convolutional network (MDSTCN) for load forecasting. MDSTCN can effectively capture the complex spatiotemporal relationship between multimodal data, and is particularly suitable for fusion prediction of high-dimensional, multi-source heterogeneous data. The load forecasting formula is: , in, Indicates the predicted future moment of grid load. Represents the Multimodal Deep Spatiotemporal Convolutional Network (MDSTCN) function, mapping extended input features and historical spatiotemporal relations to future loads. Indicates the current time The extended multimodal feature input vector. Represents the spatiotemporal state of MDSTCN, capturing the spatiotemporal dependencies and patterns of the input data.

[0061] Calculation example: Assume that the current extended feature input vector is , space-time state , predicted by the MDSTCN model: , Forecasted Load Indicates the grid load at the next moment.

[0062] Furthermore, in order to improve the robustness of the load forecasting model, especially in the case of data noise or abnormalities, a regularization term based on a hybrid kernel function (HKF) is introduced into the model. This regularization term can simultaneously consider the similarity between linear and nonlinear features, ensuring that the model has stronger robustness when dealing with different feature scales and correlations. The regularization term is defined as: , in, represents a regularization term, which is used to prevent overfitting and improve the robustness of the model. Represents the weight parameter in the MDSTCN model. Represents L2 norm regularization, which penalizes the complexity of model parameters. represents the mixed kernel function, which measures the similarity between different feature vectors and is defined as: , in, Represents the regularization coefficient, which controls the weights of the two regularization terms. Represents the bandwidth parameter of the kernel function, which determines the smoothness of the Gaussian kernel.

[0063] Regularization term The L2 norm part in ( ) is used to limit the model parameters to be too large and prevent overfitting; Hybrid kernel function part It is used to capture the nonlinear relationship between data and improve the robustness of the model at different feature scales and correlations.

[0064] S402, constructing a dispatch strategy optimization objective function according to load forecasting to obtain a power grid dispatch strategy; Specific, based on the predicted load The invention further generates a dispatch strategy for the power grid by introducing a regularization term. The strategy needs to consider the current load level, predicted load changes, equipment operating status, and cost constraints to ensure that the load demand of the power grid at a future time is met and the equipment operates in a safe and efficient state. for: , in, It represents the dispatch strategy optimization objective function, which measures the deviation between the predicted load and the actual load, the model complexity and the equipment operation cost. Indicates the length of the prediction time window. Represents the predicted load value. Indicates the actual load. Represents the regularization weight coefficient, which controls the impact of model complexity on the objective function. Represents the cost weight coefficient, balancing the impact of load deviation and equipment cost. Indicates The operating cost of a device. Indicates The energy loss of a device.

[0065] Calculation example: Assuming the forecast time window The predicted load value is The actual load is , regularization weight coefficient , equipment operating costs , energy loss , cost weight coefficient , the regularization term , then the scheduling strategy optimization objective function is calculated as: , By optimizing the objective function , obtain the optimal dispatching strategy to ensure the stable and efficient operation of the power grid.

[0066] Step 5: Construct a multi-objective optimization function to optimize the power grid dispatching strategy in a dynamic environment with multiple constraints and generate the optimal dispatching strategy, including: Combine load forecasting, equipment cost, energy loss and power quality constraints to design a multi-objective scheduling optimization function. Combine particle swarm optimization algorithm and reinforcement learning to build a dynamic scheduling optimization model. Optimize the optimal scheduling strategy of the power grid under dynamic load and equipment status changes and solve the optimal scheduling strategy. Specifically, based on the predicted load data, the present invention first defines the objective function of dispatch optimization. The goal of dispatch optimization is to minimize the total equipment operating cost and energy loss while ensuring that the power quality of the power grid (such as voltage stability and frequency stability) remains within a reasonable range. The present invention proposes an innovative multi-objective optimization function , which takes into account load forecasting error, equipment cost, energy loss and power quality constraints. for: , in, Represents the weight coefficient of the objective function, which is used to balance load forecast error, equipment operating cost and power quality. Indicates the length of the prediction time window. Indicates the actual load. Indicates The operating costs of a device (such as electricity usage, equipment maintenance, etc.). Indicates The energy loss of a device. Indicates The power quality deviation (such as voltage deviation, frequency deviation, etc.)

[0067] The objective function takes into account load forecast error, operating cost and power quality to ensure that the dispatch strategy is both economical and reliable.

[0068] Furthermore, in order to optimize scheduling in a dynamic environment, the present invention proposes a multi-objective dynamic optimization algorithm (MODOA). MODOA combines the advantages of particle swarm optimization (PSO) and reinforcement learning (RL), and can generate the optimal scheduling strategy under dynamic load and equipment state changes. The state update formula of the dynamic scheduling optimization model is: , The control strategy optimization formula is: , in, Indicates the current time The grid state vector includes voltage, current, power, etc. Indicates the current time The rate of state change. Represents the time step, which indicates the time interval between state updates. It represents the state adjustment item based on load forecast and power quality deviation, reflecting the dynamic impact of load and power quality on the power grid state. Represents the optimized control strategy vector, which represents the next moment The dispatching control parameters (such as generator output, voltage regulation, etc.) Represents the penalty coefficient, which is used to weigh different constraints in the optimization objective. Indicates the control strategy constraints, such as equipment output limitations, safe operation limitations, etc. Indicates the status of the power grid constraints, such as voltage stability, frequency stability, etc.

[0069] Status update formula Ensure that the grid status can dynamically change according to predicted load and power quality.

[0070] Control strategy optimization formula Ensure that the generated scheduling strategy is optimal while satisfying the constraints.

[0071] Furthermore, hybrid optimization algorithms (such as particle swarm optimization combined with reinforcement learning) are used to solve the optimal scheduling strategy. , so that the scheduling optimization objective function Minimum, while satisfying all constraints. The iterative formula for solving the optimal scheduling strategy is: , in, Indicates The scheduling strategy for the iteration. Indicates The scheduling strategy for the iteration. Represents the learning rate, which is used to control the convergence speed of the optimization algorithm. Represents the objective function with respect to the control strategy The gradient indicates how to adjust To minimize the objective function.

[0072] Calculation example: Assuming the initial scheduling strategy , weight coefficient , load forecasting , actual load , equipment operating costs , energy loss , power quality deviation , penalty coefficient , learning rate .

[0073] Optimized scheduling strategy for the first iteration Calculated as: Calculate the objective function ; Computing Gradients ; Update Scheduling Strategy ; Through multiple iterations, the optimal scheduling strategy is found , ensuring that the power grid meets all constraints and achieves optimal performance in dispatch optimization.

[0074] Furthermore, according to the optimal scheduling strategy obtained , adjust the operating parameters of the power grid (such as generator output, transformer tap position, capacitor bank switching, etc.) to ensure that load demand is met, while minimizing operating costs and energy losses and keeping power quality within the set range. The implementation steps are as follows: Read the optimal scheduling strategy ; Update the operating parameters of power grid equipment; Monitor the grid operation status in real time to ensure compliance with safety and quality requirements.

[0075] Step 6: Execute the optimal dispatching strategy to actually control the medium-voltage distribution network equipment, and introduce an adaptive optimization feedback mechanism to dynamically adjust the dispatching strategy by real-time monitoring of equipment status and load response to ensure stable operation of the power grid in a changing environment.

[0076] Specifically, according to the calculated optimal scheduling strategy , execute the control operation of the power grid equipment. This includes adjusting the generator output power, transformer tap position, reactive power compensation, capacitor bank switching, etc. During the execution process, it is necessary to collect the state data of the power grid (such as voltage, current, frequency, etc.) in real time to ensure that the execution effect meets expectations. The dispatch execution formula is: , in, Represents the grid state vector after the strategy is executed. Represents the grid state vector before execution. According to the optimal scheduling strategy Control operations for making adjustments.

[0077] Furthermore, during the strategy execution process, the sensor network monitors the grid status in real time and records the current status. . Calculate the error between the actual execution state and the expected state for subsequent adaptive optimization feedback adjustment. The error calculation formula is: , in, Represents at time The state error is used to identify the deviation of the current policy execution.

[0078] Furthermore, according to the calculated error vector , an Adaptive Optimization Feedback Algorithm (AOFA) is introduced. AOFA compensates for the errors in the execution of the current strategy by adjusting the scheduling strategy in real time.

[0079] The adjustment strategy formula is: , in, Represents the optimal scheduling strategy at the next moment after adjustment. Represents the adaptive feedback gain coefficient, which is used to control the speed of error adjustment.

[0080] Furthermore, according to the adjusted scheduling strategy , continue to perform control operations and monitor the status, forming a closed-loop feedback control system. Through repeated iterations, the dispatching strategy is gradually optimized to enable the power grid system to achieve a stable and efficient operation state.

[0081] In one or more embodiments, Figure 2 As shown, the present invention discloses a power supply energy analysis system for a medium voltage distribution network, the system comprising: The data acquisition module 101 is used to collect multi-modal data of the medium voltage distribution network and perform pre-processing. The execution steps specifically include: S101. Design a standardization formula with absolute median deviation to standardize the original data of different modes respectively; S102, dynamically adjusting denoising parameters using adaptive wavelet denoising combined with nonlinear smoothing on the standardized data, retaining important feature information, and obtaining filtered wavelet decomposition coefficients; S103, introducing a multimodal dynamic time warping algorithm to perform time alignment processing on the filtered wavelet decomposition coefficients; S104, after the time alignment process, using a feature extraction method based on Laplace eigenmap to perform feature extraction, fusion and dimensionality reduction process to obtain fusion features after dimensionality reduction; The health assessment module 102 is used to build a health status assessment model, predict the health score and fault risk of the transformer of the medium voltage distribution network according to the fusion features after dimension reduction, and design equipment optimization measures to be executed when the threshold is triggered. The execution steps specifically include: S201, using the fusion features after dimension reduction as input vectors of a health status assessment model, wherein each input vector contains multimodal data of the transformer, and constructing an input vector set according to the input vectors, and constructing a dynamic weighted health status assessment model according to the input vector set, wherein the dynamic weighted health status assessment model evaluates the health score of the transformer according to the dynamic changes of the current feature value and historical data; S202. construct a fault probability prediction model based on a Bayesian dynamic network according to the health score of the transformer, and use the fault probability prediction model to calculate the failure probability of the transformer under given conditions; S203. Determine whether to trigger equipment optimization measures based on the health score and failure probability of the transformer: If the health score is lower than the preset threshold, or the failure probability is higher than the set risk threshold, the equipment optimization strategy is executed; The equipment optimization module 103 is used to construct the optimization target of the power electronic equipment of the medium voltage distribution network according to the health score and the fault probability of the transformer, and optimize the operating parameters of the power electronic equipment. The execution steps specifically include: S301, constructing an optimization objective function for power electronic equipment based on the current health status score and failure probability of the transformer, minimizing reactive power loss and harmonic distortion, and ensuring that the equipment operates within a health status threshold; S302, constructing an optimization model based on adaptive model predictive control, performing optimal control on power electronic equipment in a dynamic environment, calculating a control input vector, and using a dynamic optimization algorithm to solve the optimal control input vector so that the objective function is minimized while satisfying the safety constraints of health status and failure probability; S303, adjusting the operating parameters of the power electronic equipment according to the optimal control input vector; The load prediction module 104 is used to construct a multi-modal load prediction model to predict the future load change of the power electronic equipment according to the adjusted operating parameters of the power electronic equipment and the current state of the power electronic equipment. The execution steps specifically include: S401, constructing an extended multimodal feature input vector, constructing a multimodal deep spatiotemporal convolutional network to perform load forecasting according to the multimodal feature input vector, and at the same time introducing a regularization term based on a hybrid kernel function before the output of the multimodal deep spatiotemporal convolutional network, and then outputting the predicted load forecast; S402, constructing a dispatch strategy optimization objective function according to load forecasting to obtain a power grid dispatch strategy; The multi-objective optimization module 105 is used to construct a multi-objective optimization function, optimize the power grid dispatching strategy under a dynamic environment and with multiple constraints, and generate an optimal dispatching strategy. The execution steps specifically include: Combine load forecasting, equipment cost, energy loss and power quality constraints to design a multi-objective scheduling optimization function. Combine particle swarm optimization algorithm and reinforcement learning to build a dynamic scheduling optimization model. Optimize the optimal scheduling strategy of the power grid under dynamic load and equipment status changes and solve the optimal scheduling strategy. The system optimization module 106 is used to execute the optimal dispatching strategy to actually control the medium-voltage distribution network equipment, and introduce an adaptive optimization feedback mechanism to dynamically adjust the dispatching strategy by real-time monitoring of equipment status and load response to ensure the stable operation of the power grid in a changing environment.

[0082] These are just some preferred embodiments of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of implementing the above embodiments and making equivalent changes according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A power supply energy analysis method for a medium voltage distribution network, characterized in that: The method comprises the following steps: Step 1: Collect multimodal data of the medium voltage distribution network and perform preprocessing, including: S101. Design a standardization formula with absolute median deviation to standardize the original data of different modes respectively; S102, dynamically adjusting denoising parameters using adaptive wavelet denoising combined with nonlinear smoothing on the standardized data, retaining important feature information, and obtaining filtered wavelet decomposition coefficients; S103, introducing a multimodal dynamic time warping algorithm to perform time alignment processing on the filtered wavelet decomposition coefficients; S104, after the time alignment process, using a feature extraction method based on Laplace eigenmap to perform feature extraction, fusion and dimensionality reduction process to obtain fusion features after dimensionality reduction; Step 2: Build a health status assessment model to predict the health score and fault risk of the transformer in the medium-voltage distribution network based on the fusion features after dimensionality reduction, and design equipment optimization measures to be executed when the threshold is triggered, including: S201, using the fusion features after dimension reduction as input vectors of a health status assessment model, wherein each input vector contains multimodal data of the transformer, and constructing an input vector set according to the input vectors, and constructing a dynamic weighted health status assessment model according to the input vector set, wherein the dynamic weighted health status assessment model evaluates the health score of the transformer according to the dynamic changes of the current feature value and historical data; S202. construct a fault probability prediction model based on a Bayesian dynamic network according to the health score of the transformer, and use the fault probability prediction model to calculate the failure probability of the transformer under given conditions; S203. Determine whether to trigger equipment optimization measures based on the health score and failure probability of the transformer: If the health score is lower than the preset threshold, or the failure probability is higher than the set risk threshold, the equipment optimization strategy is executed; Step 3: Construct the optimization target of the power electronic equipment of the medium voltage distribution network according to the health score and failure probability of the transformer, and optimize the operating parameters of the power electronic equipment, including: S301, constructing an optimization objective function for power electronic equipment based on the current health status score and failure probability of the transformer, minimizing reactive power loss and harmonic distortion, and ensuring that the equipment operates within a health status threshold; S302, constructing an optimization model based on adaptive model predictive control, performing optimal control on power electronic equipment in a dynamic environment, calculating a control input vector, and using a dynamic optimization algorithm to solve the optimal control input vector so that the objective function is minimized while satisfying the safety constraints of health status and failure probability; S303, adjusting the operating parameters of the power electronic equipment according to the optimal control input vector; Step 4: Construct a multi-modal load prediction model to predict the future load changes of power electronic equipment based on the adjusted operating parameters of the power electronic equipment and the current state of the power electronic equipment, including: S401, constructing an extended multimodal feature input vector, constructing a multimodal deep spatiotemporal convolutional network to perform load forecasting according to the multimodal feature input vector, and at the same time introducing a regularization term based on a hybrid kernel function before the output of the multimodal deep spatiotemporal convolutional network, and then outputting the predicted load forecast; S402, constructing a dispatch strategy optimization objective function according to load forecasting to obtain a power grid dispatch strategy; Step 5: Construct a multi-objective optimization function to optimize the power grid dispatching strategy in a dynamic environment with multiple constraints and generate the optimal dispatching strategy, including: Combine load forecasting, equipment cost, energy loss and power quality constraints to design a multi-objective scheduling optimization function. Combine particle swarm optimization algorithm and reinforcement learning to build a dynamic scheduling optimization model. Optimize the optimal scheduling strategy of the power grid under dynamic load and equipment status changes and solve the optimal scheduling strategy. Step 6: Execute the optimal dispatching strategy to actually control the medium-voltage distribution network equipment, and introduce an adaptive optimization feedback mechanism to dynamically adjust the dispatching strategy by real-time monitoring of equipment status and load response to ensure stable operation of the power grid in a changing environment.

2. A method for analyzing power supply energy for a medium voltage distribution network according to claim 1, characterized in that: The method of adaptive wavelet denoising combined with nonlinear smoothing specifically includes: Perform multi-scale wavelet decomposition on the standardized data, extract signal features in different frequency bands, and calculate the wavelet decomposition coefficients respectively; For each wavelet decomposition coefficient, an adaptive filtering strategy based on frequency band energy ratio is introduced to calculate and obtain the filtered wavelet decomposition coefficient; The filtered wavelet decomposition coefficients are post-processed by a nonlinear smoothing function to remove noise again; The multimodal dynamic time warping algorithm is implemented by minimizing the Euclidean distance between different data; The feature extraction method based on Laplace eigenmapping uses a Laplace mapping matrix to perform feature dimension reduction according to the global structure and local similarity of data.

3. A method for analyzing power supply energy for a medium voltage distribution network according to claim 1, characterized in that: The health score of the dynamic weighted health status assessment model is expressed as follows: ， in, Indicates the health status score at the current moment, indicating the health level of the transformer; represents the bias term, which indicates the initial value of the health score; represents the regression coefficient, indicating the contribution of each characteristic to the health score; represents the first feature vector after dimensionality reduction Features The weight coefficient representing the historical characteristics; A function that represents the dynamic relationship between current features and historical features; The failure probability prediction formula of the failure probability prediction model is expressed as follows: ， in, Indicates at time Probability of failure; Indicates the current health status score; Critical thresholds that indicate health status, indicating high-risk areas for failure; Represents the historical feature vector, which represents the health data at the past moment.

4. A method for analyzing power supply energy for a medium voltage distribution network according to claim 1, characterized in that: The S302 further includes: The optimization model based on adaptive model predictive control dynamically adjusts the optimization parameters according to the real-time equipment health status score and failure probability to minimize the optimization objective function. , while satisfying the safety operation constraints, the state update formula of the dynamic optimization control model is expressed as follows: ， The control input update is expressed as follows: ， in, Indicates the current time The device state vector of Represents the system state transfer matrix and control input matrix, describing the relationship between device state and control input; represents the control input vector, which indicates the equipment operating parameters that need to be adjusted; It represents the state adjustment item based on the health status score and the failure probability, reflecting the dynamic impact of the health status and the failure probability on the equipment status; represents the penalty coefficient, which is used to weigh the safety margin of health status and failure probability in the optimization objective; represents the optimization objective function; represents the health status score; Safety threshold indicating health status; represents the probability of failure; A safety threshold representing the probability of failure; The optimal control input vector is solved by using a dynamic optimization algorithm, which is expressed as follows: ， in, Represents the learning rate, which is used to control the step size of gradient descent; represents the objective function with respect to the control input The gradient indicates how to adjust To minimize the objective function; Through multiple iterations of the dynamic optimization algorithm, the optimal control input vector is found to ensure that the equipment operates safely in an optimized state.

5. A method for analyzing power supply energy for a medium voltage distribution network according to claim 1, characterized in that: The objective optimization function , which is expressed as follows: ， in, Indicates reactive power loss, which indicates the energy loss of the equipment during operation; Indicates the harmonic distortion, indicating the impact of harmonics on power quality; Represents the weight coefficient used to balance the importance of loss and harmonic distortion in the optimization.

6. A method for analyzing power supply energy for a medium voltage distribution network according to claim 1, characterized in that: In S401, the construction of the extended multi-modal feature input vector specifically includes: adding historical load data, time characteristics and external influencing factors to the operating parameters and current state information of the power electronic equipment; The regularization term based on the hybrid kernel function is expressed as follows: ， in, represents the regularization term, which is used to prevent overfitting and improve the robustness of the model; Represents the weight parameter in the MDSTCN model; Represents L2 norm regularization, which penalizes the complexity of model parameters; represents the hybrid kernel function, which measures the similarity between different feature vectors; n represents the total number of features, and m represents the total number of devices.

7. A method for analyzing power supply energy for a medium voltage distribution network according to claim 6, characterized in that: The scheduling strategy optimization objective function is expressed as follows: ， in: represents the dispatch strategy optimization objective function, which measures the deviation between the predicted load and the actual load, the model complexity and the equipment operation cost; Indicates the length of the prediction time window; represents the predicted load value; Indicates the actual load; Represents the regularization weight coefficient, which controls the impact of model complexity on the objective function; Represents the cost weight coefficient, balancing the impact of load deviation and equipment cost; Indicates The operating cost of each device; Indicates The energy loss of a device.

8. A method for analyzing power supply energy for a medium voltage distribution network according to claim 1, characterized in that: The multi-objective scheduling optimization function , which is expressed as follows: ， in, The weight coefficient representing the objective function is used to balance the load forecast error, equipment operating cost and power quality; Indicates the length of the prediction time window; Indicates the load forecast value; Indicates the actual load; Indicates The operating cost of each device; Indicates Energy loss of each device; Indicates Power quality deviation; The state update formula of the dynamic scheduling optimization model is as follows: ， The control strategy optimization formula is as follows: ， in, Indicates the current time The grid state vector of Indicates the current time The rate of change of state; Indicates the time step, which indicates the time interval between state updates; It represents the state adjustment item based on load forecast and power quality deviation, reflecting the dynamic impact of load and power quality on the power grid state; Represents the optimized control strategy vector, which represents the next moment The dispatch control parameters; Represents the penalty coefficient, which is used to weigh different constraints in the optimization objective; Indicates the control strategy Constraints of Indicates the status of the power grid constraints.

9. A method for analyzing power supply energy for a medium voltage distribution network according to claim 1, characterized in that: The iterative formula of the optimal scheduling strategy is expressed as follows: ， in, Indicates The scheduling strategy for the next iteration; Indicates The scheduling strategy for the next iteration; Represents the learning rate, which is used to control the convergence speed of the optimization algorithm; Represents the objective function with respect to the control strategy The gradient indicates how to adjust To minimize the objective function; Represents the penalty coefficient, which is used to weigh different constraints in the optimization objective.

10. A power supply energy analysis system for a medium voltage distribution network, characterized in that: The system comprises: The data acquisition module is used to collect multi-modal data of the medium voltage distribution network and perform preprocessing. The execution steps specifically include: S101. Design a standardization formula with absolute median deviation to standardize the original data of different modes respectively; S102, dynamically adjusting denoising parameters using adaptive wavelet denoising combined with nonlinear smoothing on the standardized data, retaining important feature information, and obtaining filtered wavelet decomposition coefficients; S103, introducing a multimodal dynamic time warping algorithm to perform time alignment processing on the filtered wavelet decomposition coefficients; S104, after the time alignment process, using a feature extraction method based on Laplace eigenmap to perform feature extraction, fusion and dimensionality reduction process to obtain fusion features after dimensionality reduction; The health assessment module is used to build a health status assessment model, predict the health score and fault risk of the transformer in the medium-voltage distribution network based on the fusion features after dimension reduction, and design equipment optimization measures to be executed when the threshold is triggered. The specific execution steps include: S201, using the fusion features after dimension reduction as input vectors of a health status assessment model, wherein each input vector contains multimodal data of the transformer, and constructing an input vector set according to the input vectors, and constructing a dynamic weighted health status assessment model according to the input vector set, wherein the dynamic weighted health status assessment model evaluates the health score of the transformer according to the dynamic changes of the current feature value and historical data; S202. construct a fault probability prediction model based on a Bayesian dynamic network according to the health score of the transformer, and use the fault probability prediction model to calculate the failure probability of the transformer under given conditions; S203. Determine whether to trigger equipment optimization measures based on the health score and failure probability of the transformer: If the health score is lower than the preset threshold, or the failure probability is higher than the set risk threshold, the equipment optimization strategy is executed; The equipment optimization module is used to construct the optimization target of the power electronic equipment of the medium voltage distribution network according to the health score and fault probability of the transformer, and optimize the operating parameters of the power electronic equipment. The specific execution steps include: S301, constructing an optimization objective function for power electronic equipment based on the current health status score and failure probability of the transformer, minimizing reactive power loss and harmonic distortion, and ensuring that the equipment operates within a health status threshold; S302, constructing an optimization model based on adaptive model predictive control, performing optimal control on power electronic equipment in a dynamic environment, calculating a control input vector, and using a dynamic optimization algorithm to solve the optimal control input vector so that the objective function is minimized while satisfying the safety constraints of health status and failure probability; S303, adjusting the operating parameters of the power electronic equipment according to the optimal control input vector; The load prediction module is used to construct a multi-modal load prediction model to predict the future load changes of the power electronic equipment according to the adjusted operating parameters of the power electronic equipment and the current state of the power electronic equipment. The execution steps specifically include: S401, constructing an extended multimodal feature input vector, constructing a multimodal deep spatiotemporal convolutional network to perform load forecasting according to the multimodal feature input vector, and at the same time introducing a regularization term based on a hybrid kernel function before the output of the multimodal deep spatiotemporal convolutional network, and then outputting the predicted load forecast; S402, constructing a dispatch strategy optimization objective function according to load forecasting to obtain a power grid dispatch strategy; The multi-objective optimization module is used to construct a multi-objective optimization function, optimize the power grid dispatching strategy in a dynamic environment with multiple constraints, and generate the optimal dispatching strategy. The specific execution steps include: Combine load forecasting, equipment cost, energy loss and power quality constraints to design a multi-objective scheduling optimization function. Combine particle swarm optimization algorithm and reinforcement learning to build a dynamic scheduling optimization model. Optimize the optimal scheduling strategy of the power grid under dynamic load and equipment status changes and solve the optimal scheduling strategy. The system optimization module is used to execute the optimal dispatching strategy to actually control the medium-voltage distribution network equipment, and introduce an adaptive optimization feedback mechanism to dynamically adjust the dispatching strategy by real-time monitoring of equipment status and load response to ensure the stable operation of the power grid in a changing environment.

Images