Distribution network power supply multi-dimensional auxiliary decision analysis method and system based on machine learning

Abstract

The invention provides a machine learning-based multi-dimensional auxiliary decision analysis method and system for power supply of a distribution network. The method comprises the following steps of: S1, acquiring historical data and historical electrical load data of the distribution network; s2, obtaining a power supply demand prediction model through the long and short term memory network and the historical data of the distribution network; s3, obtaining a load change model through a time sequence prediction algorithm and historical electrical load data; s4, identifying the real-time operation data of the distribution network based on the output of the steps S2 and S3 to obtain a power supply capacity change trend and a load margin; s5, establishing a comprehensive objective function based on the output in the step S4, performing optimization solution on the objective function through a genetic algorithm, and outputting an optimal power supply decision scheme; and S6, executing the output of the step S5, monitoring the power supply operation parameters of the distribution network in real time, and dynamically adjusting the optimal power supply decision scheme based on the power supply operation parameters of the distribution network. According to the invention, the reliability of distribution network power supply is improved, and the dynamic adjustment and optimization of the distribution network power supply decision scheme are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network power supply decision analysis, and in particular to a distribution network power supply multi-dimensional auxiliary decision analysis method and system based on machine learning. Background Art

[0002] With the rapid development of the power system and the continuous growth of electricity demand, the power supply guarantee of the distribution network faces increasingly severe challenges. The power supply guarantee of the distribution network mainly relies on conventional measurement equipment, including the collection of basic operating parameters such as voltage and current, and the analysis of power supply demand based on historical statistical data and empirical formulas. Load management adopts a fixed solution to balance the distribution by adjusting transformers and switches, and power supply decisions are often based on a single indicator and lack the ability to comprehensively optimize. Operation monitoring relies on regular inspections and simple equipment, and abnormal handling mostly relies on manual intervention, which lacks flexibility and dynamism. When dealing with complex and changing power supply needs, it is difficult to achieve accurate prediction and rapid response.

[0003] The existing distribution network power supply decision-making methods mainly rely on simple statistical models, which have obvious deficiencies in power supply demand forecasting accuracy, load change analysis, and decision-making plan flexibility. The prediction models used cannot effectively capture the complex changing characteristics of power supply demand, load analysis lacks multi-dimensional considerations, decision-making plans are difficult to adjust dynamically, and the optimization goals are too single, which seriously restricts the reliability of distribution network power supply. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a distribution network power supply multi-dimensional auxiliary decision-making analysis method and system based on machine learning, which solves the problems that the prior art lacks multi-dimensional considerations for load analysis, the decision-making scheme is difficult to dynamically adjust, and the optimization target is too single.

[0005] To achieve the above-mentioned object of the invention, the first aspect of the present invention provides a multi-dimensional auxiliary decision-making analysis method for power distribution network based on machine learning, the method comprising:

[0006] S1. Obtain historical distribution network data and historical power load data;

[0007] S2. construct an initial power supply demand prediction model through a long short-term memory network, train and verify the initial power supply demand prediction model based on the distribution network historical data, and obtain a power supply demand prediction model;

[0008] S3. constructing an initial load change model through a time series prediction algorithm, training and verifying the initial load change model based on the historical power load data, and obtaining a load change model;

[0009] S4. Based on the power supply demand prediction model and the load change model, identify the real-time operation data of the distribution network to obtain the power supply capacity change trend and load margin;

[0010] S5. Construct a comprehensive objective function based on the power supply capacity change trend and load margin, optimize and solve the objective function through a genetic algorithm, and output an optimal power supply decision plan;

[0011] S6. Execute the optimal power supply decision plan, monitor the power supply operation parameters of the distribution network in real time, and dynamically adjust the optimal power supply decision plan based on the power supply operation parameters of the distribution network.

[0012] Furthermore, in step S1, historical distribution network data and historical power load data are obtained, specifically including:

[0013] S11, collecting initial distribution network historical data and initial historical power load data respectively;

[0014] S12, performing time serialization processing and feature extraction on the initial distribution network historical data to obtain distribution network historical data;

[0015] S13. Perform periodic analysis and outlier processing on the initial historical power load data to obtain historical power load data.

[0016] Furthermore, in step S2, a power supply demand prediction model is constructed, which specifically includes:

[0017] The initial power supply demand forecasting model is constructed by using the long short-term memory network, and the distribution network historical data is input into the initial power supply demand forecasting model to obtain the power supply demand forecasting model. The calculation formula of the power supply demand forecasting model is:

[0018]

[0019] Where f(t) is the predicted power demand at time t, W c is the model center weight, H(t) is the hidden state at time t, α 0 is the adjustment coefficient, D i (t) is the data processed by data source i at time t, n is the number of data sources, W i is the weight coefficient of data source i, α and L i are the first error correction coefficient and the second error correction coefficient, β is the trend adjustment coefficient, and φ(t) is the trend function;

[0020] Further, step S3 includes:

[0021] S31. Use long short-term memory network to establish a deep learning model to extract the long-term change characteristics of historical power load data;

[0022] S32. Using an autoregressive integrated moving average model to analyze the periodic fluctuation characteristics and short-term fluctuation characteristics of historical power load data;

[0023] S33, combining the prediction results of the deep learning model and the autoregressive integrated moving average model through an adaptive weight fusion algorithm to obtain a combined prediction result;

[0024] S34, inputting the combined prediction result into the initial load change model for training and verification to obtain a load change model, wherein the calculation formula of the load change model is:

[0025]

[0026] Loss = λ 1 MSE+λ 2 MAE+λ 3 ·RMSPE

[0027]

[0028] in, is time t fh The predicted power load at the time, L(t fh -1) is time t fh The actual power load at the time, X j (t fh ) is time t fh The jth feature that affects the power load at time , m is the number of features that affect the power load, For X j (t fh ) feature weight, δ(t fh ) is time t fh , θ is the adjustable parameter of the actual power load, γ is the adjustable parameter affecting the power load, η is the adjustable parameter of the external interference factor, Loss is the loss function value of the initial load change model, MSE is the mean square error, MAE is the mean absolute error, RMSPE is the root mean square percentage error, λ 1 is the mean square error, λ 2 Mean absolute error, λ 3 is the weight coefficient of the root mean square percentage error, Accuracy is the accuracy, σ p is the mean value of the predicted power load, and μ is the mean value of the actual power load.

[0029] Further, step S4 includes:

[0030] S41, collecting real-time operation data of the distribution network in operation through sensors and monitoring equipment;

[0031] S42, identifying the real-time operation data of the distribution network based on the power supply demand prediction model and the load change model to obtain a real-time predicted power supply demand and a real-time predicted power load;

[0032] S43, classifying and regressing the real-time predicted power demand and the real-time predicted power load through a support vector machine algorithm to obtain a power supply capacity change trend in the future period, wherein the power supply capacity change trend is calculated as follows:

[0033] ΔP(t gd )=τ 1 ·V(t gd )+τ 2 ·I(t gd )+τ 3 ·F(t gd )

[0034] Among them, ΔP(t gd ) is time t gd The power supply capacity change trend at t gd ) is time t gd The voltage at the time, I(t gd ) is time t gd The current at the time, F(t gd ) is time t gd The frequency at time, τ 1 is the voltage weight coefficient, τ 2 is the current weight coefficient, τ 3 is the frequency weight coefficient;

[0035] S44, calculating the real-time predicted power load based on a fuzzy logic algorithm to obtain a load margin, wherein the calculation formula of the load margin is:

[0036]

[0037] MU(t yd ) is time t yd Load margin at time, P pred (t yd ) is time t yd The real-time predicted power load at the time of acnual (t yd ) is time t yd The real-time actual power load at that time.

[0038] Further, step S5 includes:

[0039] S51. Determine a power supply capacity index and a load margin index based on a power supply capacity change trend and a load margin;

[0040] S52, using a weight allocation method to assign different importance weights to the power supply capacity index and the load margin index, and constructing a comprehensive objective function, the calculation formula of the comprehensive objective function is:

[0041]

[0042] Among them, F is the comprehensive objective function value, C k is the actual value of the kth power supply capability indicator, L l is the actual value of the lth load margin index, C max,k The maximum value of the power supply capability index is standardized, L max,l is the maximum value of the load margin index, w k is the weight coefficient of the power supply capability index, v l is the weight coefficient of the load margin index, N is the number of power supply capacity indicators, and M is the number of load margin indicators;

[0043] S53, optimizing and solving the comprehensive objective function by a genetic algorithm, initializing the population, and generating an initial population;

[0044] S54, evaluating the fitness of each individual in the initial population based on the comprehensive objective function, and selecting the individual with the highest fitness based on the elite retention strategy;

[0045] S55. Generate a new generation of individuals through partial random crossover and mutation, and generate new individuals through repeated iterations. When the iteration reaches a preset number of iterations or the fitness change meets the requirements, stop the iteration and output the optimal power supply decision plan.

[0046] Further, step S6 includes:

[0047] S61. According to the optimal power supply decision plan, adjust the operating parameters of the equipment in the distribution network and redistribute the load demand of the equipment;

[0048] S62, through the distribution network control center, the adjusted operating parameters and the reallocated load demand are sent to the corresponding devices in the form of control commands;

[0049] S63, deploying IoT sensors on key equipment in the distribution network, and collecting power supply operation parameters in real time based on the IoT sensors;

[0050] S64, cleaning, filtering and normalizing the power supply operation parameters to obtain distribution network power supply operation data, and transmitting the distribution network power supply operation data to a centralized monitoring platform for real-time storage and backup;

[0051] S65. Analyze the current power supply status based on the distribution network power supply operation data, identify potential anomalies and operation bottlenecks, and obtain status assessment results;

[0052] S66. According to the status assessment results and in combination with historical data, a corresponding adjustment strategy is formulated and executed to dynamically adjust the optimal power supply decision plan.

[0053] A second aspect of the present invention provides a multi-dimensional auxiliary decision-making analysis system for distribution network power supply based on machine learning, the system comprising:

[0054] Data acquisition module, used to obtain historical distribution network data and historical power load data;

[0055] A power supply demand module is used to construct an initial power supply demand prediction model through a long short-term memory network, and train and verify the initial power supply demand prediction model based on the distribution network historical data to obtain a power supply demand prediction model;

[0056] A load change module, used to construct an initial load change model through a time series prediction algorithm, train and verify the initial load change model based on the historical power load data, and obtain a load change model;

[0057] A trend margin module, based on the power demand prediction model and the load change model, identifies the real-time operation data of the distribution network to obtain the power supply capacity change trend and load margin;

[0058] The target optimization module is used to construct a comprehensive target function through the power supply capacity change trend and the load margin, optimize and solve the comprehensive target function through the genetic algorithm, and output the optimal power supply decision plan;

[0059] The scheme adjustment module is used to execute the optimal power supply decision scheme, monitor the distribution network power supply operation parameters in real time, and dynamically adjust the optimal power supply decision scheme based on the distribution network power supply operation parameters.

[0060] A third aspect of the present invention provides an electronic device, comprising: at least one processor, at least one memory, a communication interface and a bus;

[0061] The processor, memory, and communication interface communicate with each other via the bus, the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

[0062] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. By acquiring historical data of distribution network and power load data, and using long short-term memory network to build a power supply demand forecasting model, using time series forecasting algorithm to establish load change model, combining multi-dimensional data to analyze load change characteristics, based on power supply demand and load change model, identify real-time operation data of distribution network, obtain power supply capacity change trend and load margin, and build and solve comprehensive objective function according to genetic algorithm, output optimal power supply decision plan, realize dynamic adjustment and optimization of decision plan, continuously optimize power supply decision through real-time monitoring and feedback mechanism, improve reliability and economy of distribution network power supply, overcome the limitations of traditional methods, and realize efficient power supply guarantee.

[0065] (2) A power supply demand forecasting model is constructed through the long short-term memory network. Its deep learning ability is combined to effectively capture the long-term variation characteristics of power supply demand. Time series modeling is performed through the recursive neural network structure, and the adaptive learning rate optimization algorithm is used to dynamically adjust the network parameters, thereby improving the accuracy of power supply demand forecasting.

[0066] (3) By combining the long short-term memory network and the autoregressive integrated moving average model, a comprehensive analysis of the long-term change characteristics and periodic fluctuation characteristics of historical electricity load data is achieved. Through the adaptive weight fusion algorithm, the prediction results of the deep learning model and the traditional statistical model are combined to construct an accurate load change model, which improves the accuracy of load change prediction and enhances the adaptability to complex load changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0068] Figure 1 A flow chart of a multi-dimensional auxiliary decision-making analysis method for power distribution network based on machine learning provided in an embodiment of the present invention;

[0069] Figure 2 A structural diagram of a multi-dimensional auxiliary decision-making analysis system for distribution network power supply based on machine learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0071] Reference Figure 1 The first aspect of this embodiment provides a multi-dimensional auxiliary decision-making analysis method for distribution network power supply based on machine learning. Specifically, this embodiment obtains historical data of the distribution network and power load data, and uses a long short-term memory network to build a power supply demand prediction model, adopts a time series prediction algorithm to establish a load change model, and combines multi-dimensional data to analyze the load change characteristics. Based on the power supply demand and load change model, the real-time operation data of the distribution network is identified, the power supply capacity change trend and load margin are obtained, a comprehensive objective function is constructed, and the comprehensive objective function is solved by a genetic algorithm to output the optimal power supply decision plan, so as to realize dynamic adjustment and optimization of the decision plan. Through real-time monitoring and feedback mechanism, the power supply decision is continuously optimized, the reliability and economy of the distribution network power supply are improved, the limitations of traditional methods are overcome, and the efficiency of power supply guarantee is realized.

[0072] The method comprises:

[0073] S1. Obtain historical distribution network data and historical power load data.

[0074] In step S1, historical distribution network data and historical power load data are obtained, specifically including:

[0075] S11, respectively collecting initial distribution network historical data and initial historical power load data. The original distribution network historical data includes distribution network equipment operation status data, distribution network fault record data and distribution network maintenance record data.

[0076] S12: Perform time serialization processing and feature extraction on the initial distribution network historical data to obtain distribution network historical data.

[0077] S13, periodically analyzing and processing the abnormal value of the initial historical power load data to obtain the historical power load data. The original historical power load data includes user power consumption data, load power data and load change rate data.

[0078] Specifically, this embodiment ensures the integrity of basic data by comprehensively collecting multi-dimensional data such as the operating status, fault records and maintenance records of distribution network equipment, as well as load data such as user power consumption, load power and change rate, performs time serialization and feature extraction on the distribution network historical data to make the data time-series, performs periodic analysis and outlier processing on historical power load data, effectively removes data noise and improves data reliability.

[0079] The distribution network historical data includes distribution network equipment operation status data, distribution network fault record data and distribution network maintenance record data. The distribution network equipment operation status data is used to reflect the working status of the equipment; the distribution network fault record data is used to record the fault conditions that occur during system operation; the distribution network maintenance record data is used to record the maintenance information of the equipment. After the above raw data is processed by time series and feature extraction, the distribution network historical data that can be used for model training is formed.

[0080] The historical power load data includes user power consumption data, load power data and load change rate data; the user power consumption data is used to record the user's actual power consumption; the load power data is used to reflect the system's load level; the load change rate data is used to characterize the dynamic characteristics of load changes. After periodic analysis and outlier processing, the above raw data forms historical power load data that can be used for model training.

[0081] S2. Construct an initial power supply demand prediction model through a long short-term memory network, train and verify the initial power supply demand prediction model based on the distribution network historical data, and obtain a power supply demand prediction model.

[0082] In step S2, a power supply demand prediction model is constructed, which specifically includes:

[0083] The initial power supply demand forecasting model is constructed by using the long short-term memory network, and the distribution network historical data is input into the initial power supply demand forecasting model to obtain the power supply demand forecasting model. The calculation formula of the power supply demand forecasting model is:

[0084]

[0085] Where f(t) is the predicted power demand at time t, W c is the model center weight, H(t) is the hidden state at time t, α 0 is the adjustment coefficient, D i (t) is the data processed by data source i at time t, n is the number of data sources, W i is the weight coefficient of data source i, α and L i are the first error correction coefficient and the second error correction coefficient respectively, β is the trend adjustment coefficient, and φ(t) is the trend function.

[0086] Specifically, the long short-term memory network is used to train and verify the power supply demand forecasting model, the recursive neural network structure is used for time series modeling, and the network parameters are dynamically adjusted through the adaptive learning rate optimization algorithm. The specific steps for constructing the power supply demand forecasting model through the long short-term memory network are as follows:

[0087] 1) Data preprocessing:

[0088] Collect and clean distribution network historical data, including:

[0089] Equipment operating status data (voltage, current, frequency, etc.), fault record data (fault type, occurrence time, duration), maintenance record data (maintenance type, maintenance cycle, maintenance effect), ensure the integrity and consistency of the data, and standardize the data, such as: time serializing the data, using 15 minutes as the basic sampling interval, and mapping the data to the [-1,1] interval through standardization to meet the input requirements of the long short-term memory network model.

[0090] 2) Model construction:

[0091] Select the long short-term memory network structure, determine the number of layers in the network and the number of neurons in each layer, define the structure of the input layer, hidden layer, and output layer, and ensure that the model can process time series data.

[0092] A three-layer LSTM network structure was constructed: the input layer consisted of 128 neurons; the hidden layer consisted of 256 neurons; the output layer consisted of 1 neuron; the time step was set to 96 (corresponding to 24 hours); and the batch size was set to 64.

[0093] 3) Model initialization:

[0094] Initialize various parameters of the model, including center weight, data source weight coefficient, adjustment coefficient, error correction coefficient and trend adjustment coefficient, and determine the form of the trend function in order to capture the changing trend of power supply demand in the model.

[0095] The model center weight is randomly initialized in the interval [-0.1, 0.1]; the initial values ​​of the data source weight coefficients are all set to 1 / n; the initial value of the adjustment coefficient is set to 0.8; the error correction coefficients and are initialized to 0.3 and 0.7 respectively; and the initial value of the trend adjustment coefficient is set to 0.5.

[0096] 4) Model training:

[0097] The standardized distribution network historical data is input into the long short-term memory network model, and the recursive neural network structure is used for time series modeling. The adaptive learning rate optimization algorithm (such as Adam or RMSprop) is used to dynamically adjust the network parameters to improve the training efficiency and prediction accuracy of the model. Specifically, the Adam optimizer is used, the initial learning rate is set to 0.001, the recursive neural network structure is used, the time window size is 24, the training round (epoch) is set to 100, and the learning rate is updated every 50 batches.

[0098] 5) Model Validation:

[0099] Use the validation data set to evaluate the model to ensure the generalization ability of the model, adjust the model parameters according to the validation results, and optimize the model performance.

[0100] 20% of the data was used as the validation set, and the evaluation indicators included: mean square error (MSE) target value <0.05, mean absolute error (MAE) target value <0.08, and accuracy target value >90%.

[0101] 6) Model output:

[0102] The trained LSTM network model outputs the predicted power demand at time t, and the final prediction result is calculated using the model center weight, hidden state, adjustment coefficient, etc. The predicted power demand at time t is output, including: voltage prediction value (accurate to 0.1V), current prediction value (accurate to 0.01A), and frequency prediction value (accurate to 0.01Hz).

[0103] This embodiment introduces multiple adjustment coefficients and error correction coefficients into the calculation formula to achieve fine adjustment of the prediction results, taking into account the weight influence of multiple data sources. i , improve the comprehensiveness of the prediction, use recursive neural network structure for time series modeling, which can effectively capture the timing characteristics of power supply demand, dynamically adjust network parameters through adaptive learning rate optimization algorithm, so that the model can adaptively adapt to data changes, and introduce trend function φ(t) to model the changing trend of power supply demand. The hidden state H(t) is used to retain and transmit historical information, which improves the long-term memory ability of the model and improves the accuracy of power supply demand prediction.

[0104] S3. constructing an initial load change model through a time series prediction algorithm, training and verifying the initial load change model based on the historical power load data, and obtaining a load change model;

[0105] Step S3 includes:

[0106] S31. Use long short-term memory network to establish a deep learning model to extract the long-term change characteristics of historical power load data.

[0107] S32. Use the autoregressive integrated moving average model to analyze the periodic fluctuation characteristics and short-term fluctuation characteristics of historical electricity load data.

[0108] S33. Combining the prediction results of the deep learning model and the autoregressive integrated moving average model through an adaptive weight fusion algorithm to obtain a combined prediction result.

[0109] S34, inputting the combined prediction result into the initial load change model for training and verification to obtain a load change model, wherein the calculation formula of the load change model is:

[0110]

[0111] Loss = λ 1 MSE+λ 2 MAE+λ 3 ·RMSPE

[0112]

[0113] in, is time t fh The predicted power load at the time, L(t fh -1) is time t fh The actual power load at j (t fh ) is time t fh The jth feature that affects the power load at time , m is the number of features that affect the power load, For X j (t fh ) feature weight, δ(t fh ) is time t fh , θ is the adjustable parameter of the actual power load, γ is the adjustable parameter affecting the power load, η is the adjustable parameter of the external interference factor, Loss is the loss function value of the initial load change model, MSE is the mean square error, MAE is the mean absolute error, RMSPE is the root mean square percentage error, λ 1 is the mean square error, λ 2 Mean absolute error, λ 3 is the weight coefficient of the root mean square percentage error, Accuracy is the accuracy, σ p is the mean value of the predicted power load, and μ is the mean value of the actual power load.

[0114] Specifically, the historical power load data is divided into a training set and a test set, and the initial load change model is trained and verified by the training set and the test set. The initial load change model is evaluated by cross-validation, and the mean square error and mean absolute error are used as model performance evaluation indicators. The model parameters are adjusted according to the verification results. This embodiment extracts the long-term change characteristics of the historical power load data through a long short-term memory network, uses an autoregressive integrated moving average model to analyze the periodic fluctuation and short-term fluctuation characteristics, and uses an adaptive weight fusion algorithm to optimize the combination of the prediction results of the two models. By introducing three evaluation indicators: mean square error MSE, mean absolute error MAE, and root mean square percentage error RMSPE, as well as the weight coefficient λ 1 , 2 , 3A weighted combination of different evaluation indicators is performed, and the accuracy indicator Accuracy is used for comprehensive evaluation to improve the reliability of model verification. The historical electricity load data is divided into training sets and test sets for model training and verification. The cross-validation method is used to evaluate the model performance. The model parameters are dynamically adjusted according to the verification results to improve the accuracy of the model.

[0115] S4. Based on the power supply demand prediction model and the load change model, the real-time operation data of the distribution network is identified to obtain the power supply capacity change trend and the load margin.

[0116] Step S4 includes:

[0117] S41. Collect real-time distribution network operation data in real time through sensors and monitoring equipment.

[0118] S42: Based on the power supply demand prediction model and the load change model, the real-time operation data of the distribution network is identified to obtain a real-time predicted power supply demand and a real-time predicted power load. The real-time predicted power supply demand includes voltage, current, and frequency.

[0119] S43, classifying and regressing the real-time predicted power demand and the real-time predicted power load through a support vector machine algorithm to obtain a power supply capacity change trend in the future period, wherein the power supply capacity change trend is calculated as follows:

[0120] ΔP(t gd )=τ 1 ·V(t gd )+τ 2 ·I(t gd )+τ 3 ·F(t gd )

[0121] Among them, ΔP(t gd ) is time t gd The power supply capacity change trend at t gd ) is time t gd The voltage at the time, I(t gd ) is time t gd The current at the time, F(t gd ) is time t gd The frequency at time, τ 1 is the voltage weight coefficient, τ 2 is the current weight coefficient, τ 3 is the frequency weight coefficient.

[0122] S44, calculating the real-time predicted power load based on a fuzzy logic algorithm to obtain a load margin, wherein the calculation formula of the load margin is:

[0123]

[0124] MU(t yd ) is time t yd Load margin at time, P pred (t yd ) is time t yd The real-time predicted power load at the time of acnual (t yd ) is time t yd The real-time actual power load at that time.

[0125] Specifically, this embodiment collects distribution network operation data in real time through sensors and monitoring equipment, obtains key power supply parameters such as voltage, current, and frequency in real time, ensures the timeliness and accuracy of the data, identifies real-time data based on the power demand prediction model and the load change model, obtains real-time predicted power demand and real-time predicted power load, and realizes real-time evaluation of the power supply status. By using the support vector machine algorithm for classification and regression analysis, the power supply capacity change trend is accurately predicted, and the fuzzy logic algorithm is used to calculate the load margin to improve the rationality of the calculation results.

[0126] S5. Construct a comprehensive objective function based on the power supply capacity change trend and load margin, optimize and solve the objective function through a genetic algorithm, and output an optimal power supply decision plan.

[0127] Step S5 includes:

[0128] S51. Determine a power supply capacity index and a load margin index based on a power supply capacity change trend and a load margin.

[0129] S52, using a weight allocation method to assign different importance weights to the power supply capacity index and the load margin index, and constructing a comprehensive objective function, the calculation formula of the comprehensive objective function is:

[0130]

[0131] Among them, F is the comprehensive objective function value, C k is the actual value of the kth power supply capability indicator, L l is the actual value of the lth load margin index, C max,k The maximum value of the power supply capability index is standardized, L max,l is the maximum value of the load margin index, w k is the weight coefficient of the power supply capability index, v l is the weight coefficient of the load margin index, N is the number of power supply capacity indicators, and M is the number of load margin indicators.

[0132] S53, optimizing and solving the comprehensive objective function through a genetic algorithm, initializing the population, and generating an initial population.

[0133] S54. Evaluate the fitness of each individual in the initial population based on the comprehensive objective function, and select the individual with the highest fitness based on the elite retention strategy.

[0134] S55. Generate a new generation of individuals through partial random crossover and mutation, and generate new individuals through repeated iterations. When the iteration reaches a preset number of iterations or the fitness change meets the requirements, stop the iteration and output the optimal power supply decision plan.

[0135] Specifically, this embodiment simultaneously considers the power supply capacity index and the load margin index, assigns reasonable importance weights to different indicators through the weight distribution method, and realizes the unified optimization of multiple objectives by constructing a comprehensive objective function. This embodiment uses a genetic algorithm to optimize and solve the comprehensive objective function, retains high-quality individuals with high fitness through an elite retention strategy, and uses partial random crossover and mutation mechanisms to avoid local optimality. Based on iterative optimization, the quality of the solution is continuously improved, and reasonable iteration termination conditions (such as preset iteration times or fitness change requirements) are set, so as to output a fully optimized optimal power supply decision plan.

[0136] S6. Execute the optimal power supply decision plan, monitor the power supply operation parameters of the distribution network in real time, and dynamically adjust the optimal power supply decision plan based on the power supply operation parameters of the distribution network.

[0137] Step S6 includes:

[0138] S61. According to the optimal power supply decision plan, the operating parameters of the equipment in the distribution network are adjusted to redistribute the load demand of the equipment. The operating parameters include switch status, transformer tap position, and distributed energy output power.

[0139] S62. The adjusted operating parameters and reallocated load demand are sent to corresponding devices in the form of control commands through the distribution network control center.

[0140] S63: deploy IoT sensors on key equipment in the distribution network, and collect power supply operation parameters in real time based on the IoT sensors. The power supply operation parameters include voltage, current, frequency, and power factor.

[0141] S64. Clean, filter and normalize the power supply operation parameters to obtain distribution network power supply operation data, and transmit the distribution network power supply operation data to a centralized monitoring platform for real-time storage and backup.

[0142] S65. Based on the distribution network power supply operation data, analyze the current power supply status, identify potential anomalies and operation bottlenecks, and obtain status assessment results.

[0143] S66. According to the status evaluation results and in combination with historical data, formulate and execute corresponding adjustment strategies to dynamically adjust the optimal power supply decision plan. Continuously monitor the adjustment effect, and verify the performance improvement of the adjusted optimal power supply decision plan through a feedback mechanism.

[0144] In a specific embodiment, according to the optimal power supply decision plan, the operating parameters of each device in the distribution network are adjusted, including the switch state, the transformer tap position, and the output power of distributed energy. The load demand of each node is redistributed, and the adjusted operating parameters and load distribution plan are sent to each relevant device in the form of a control command through the distribution network control center. IoT sensors are deployed on key equipment in the distribution network to collect power supply operation parameters such as voltage, current, frequency, and power factor in real time. The collected power supply operation parameters are cleaned, filtered, and normalized to obtain the distribution network power supply operation data, and the data is transmitted to the centralized monitoring platform for real-time storage and backup. Based on the distribution network power supply operation data, the current power supply status is analyzed, potential anomalies and operation bottlenecks are identified, and the status evaluation results are obtained. According to the status evaluation results and combined with historical data, a corresponding adjustment strategy is formulated, the adjustment strategy is executed, and the optimal power supply decision plan is adjusted. The adjustment effect is continuously monitored, and the performance improvement of the adjusted optimal power supply decision plan is verified through a feedback mechanism.

[0145] Specifically, this embodiment ensures the stability of the power supply system by accurately adjusting the operating parameters and load distribution of the distribution network equipment, and ensures the timeliness of the power supply operation data by deploying IoT sensors and collecting real-time data. This embodiment also ensures the continuous optimization of the power supply decision plan by analyzing the power supply status in real time, identifying anomalies and bottlenecks, and making dynamic adjustments based on historical data.

[0146] As another optional embodiment, refer to Figure 2 The present invention also provides a multi-dimensional auxiliary decision-making analysis system for power distribution network based on machine learning, the system comprising:

[0147] The data acquisition module is used to obtain historical distribution network data and historical power load data.

[0148] The power supply demand module is used to construct an initial power supply demand prediction model through a long short-term memory network, train and verify the initial power supply demand prediction model based on the distribution network historical data, and obtain a power supply demand prediction model.

[0149] The load change module is used to construct an initial load change model through a time series prediction algorithm, and train and verify the initial load change model based on the historical power load data to obtain a load change model.

[0150] The trend margin module identifies the real-time operation data of the distribution network based on the power supply demand prediction model and the load change model to obtain the power supply capacity change trend and load margin.

[0151] The target optimization module is used to construct a comprehensive objective function through the power supply capacity change trend and load margin, optimize and solve the comprehensive objective function through a genetic algorithm, and output the optimal power supply decision plan.

[0152] The scheme adjustment module is used to execute the optimal power supply decision scheme, monitor the distribution network power supply operation parameters in real time, and dynamically adjust the optimal power supply decision scheme based on the distribution network power supply operation parameters.

[0153] Specifically, the data acquisition module of this embodiment can obtain comprehensive distribution network historical data and historical power load data, and ensure the accuracy of the data by time serialization processing and feature extraction of the data. The power supply demand module and the load change module respectively use long short-term memory networks and time series prediction algorithms to construct and train power supply demand prediction models and load change models, which can identify and predict power supply demand and load change trends. The trend margin module identifies distribution network operation data in real time and provides power supply capacity change trends and load margin information. The target optimization module optimizes and solves the comprehensive objective function constructed based on power supply capacity and load margin through genetic algorithms, and outputs the optimal power supply decision plan. The scheme adjustment module can dynamically adjust the optimal power supply decision plan according to real-time monitoring data to ensure the stability of the power supply system.

[0154] As another optional embodiment, the present invention further provides an electronic device, comprising: at least one processor, at least one memory, a communication interface and a bus.

[0155] Among them, the processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement a multi-dimensional auxiliary decision-making analysis method for distribution network power supply based on machine learning.

[0156] As another optional embodiment, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions enable the computer to implement all or part of the steps of a multi-dimensional auxiliary decision-making analysis method for power distribution network based on machine learning described in the embodiment of the present invention. The storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0157] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-dimensional auxiliary decision analysis method for distribution network power supply based on machine learning, characterized in that: The following steps are involved: S1. Obtain historical distribution network data and historical power load data; S2. construct an initial power supply demand prediction model through a long short-term memory network, train and verify the initial power supply demand prediction model based on the distribution network historical data, and obtain a power supply demand prediction model; S3. constructing an initial load change model through a time series prediction algorithm, training and verifying the initial load change model based on the historical power load data, and obtaining a load change model; S4. Based on the power supply demand prediction model and the load change model, identify the real-time operation data of the distribution network to obtain the power supply capacity change trend and load margin; S5. Construct a comprehensive objective function based on the power supply capacity change trend and load margin, optimize and solve the objective function through a genetic algorithm, and output an optimal power supply decision plan; S6. Execute the optimal power supply decision plan, monitor the power supply operation parameters of the distribution network in real time, and dynamically adjust the optimal power supply decision plan based on the power supply operation parameters of the distribution network.

2. A method for multi-dimensional auxiliary decision-making analysis of power distribution network based on machine learning as claimed in claim 1, characterized in that: In step S1, historical distribution network data and historical power load data are obtained, specifically including: S11, collecting initial distribution network historical data and initial historical power load data respectively; S12, performing time serialization processing and feature extraction on the initial distribution network historical data to obtain distribution network historical data; S13. Perform periodic analysis and outlier processing on the initial historical power load data to obtain historical power load data.

3. The method for multi-dimensional auxiliary decision-making analysis of power distribution network based on machine learning as claimed in claim 1, characterized in that: In step S2, a power supply demand prediction model is constructed, which specifically includes: The initial power supply demand forecasting model is constructed by using the long short-term memory network, and the distribution network historical data is input into the initial power supply demand forecasting model to obtain the power supply demand forecasting model. The calculation formula of the power supply demand forecasting model is: Where f(t) is the predicted power demand at time t, W c is the model center weight, H(t is the hidden state at time t, α0 is the adjustment coefficient, D i (t) is the data processed by data source i at time t, n is the number of data sources, W i is the weight coefficient of data source i, α and L i are the first error correction coefficient and the second error correction coefficient respectively, β is the trend adjustment coefficient, and φ(t) is the trend function.

4. A method for multi-dimensional auxiliary decision-making analysis of power distribution network based on machine learning as claimed in claim 3, characterized in that: Step S3 includes: S31. Use long short-term memory network to establish a deep learning model to extract the long-term change characteristics of historical power load data; S32. Using an autoregressive integrated moving average model to analyze the periodic fluctuation characteristics and short-term fluctuation characteristics of historical power load data; S33, combining the prediction results of the deep learning model and the autoregressive integrated moving average model through an adaptive weight fusion algorithm to obtain a combined prediction result; S34, inputting the combined prediction result into the initial load change model for training and verification to obtain a load change model, wherein the calculation formula of the load change model is: Loss=λ1·MSE+λ2·MAE+λ3·RMSPE in, is time t fh The predicted power load at the time, L(t fh -1) is time t fh The actual power load at j (t fh ) is time t fh The jth feature that affects the power load at time , m is the number of features that affect the power load, For X j (t fh ) feature weight, δ(t fh ) is time t fh , θ is the adjustable parameter of the actual power load, γ is the adjustable parameter affecting the power load, η is the adjustable parameter of the external interference factor, Loss is the loss function value of the initial load change model, MSE is the mean square error, MAE is the mean absolute error, RMSPE is the root mean square percentage error, λ1 is the mean square error, λ2 is the mean absolute error, λ3 is the weight coefficient of the root mean square percentage error, Accuracy is the accuracy, σ p is the mean value of the predicted power load, and μ is the mean value of the actual power load.

5. A method for multi-dimensional auxiliary decision-making analysis of power distribution network based on machine learning as claimed in claim 4, characterized in that: Step S4 includes: S41, collecting real-time operation data of the distribution network in operation through sensors and monitoring equipment; S42, identifying the real-time operation data of the distribution network based on the power supply demand prediction model and the load change model to obtain a real-time predicted power supply demand and a real-time predicted power load; S43, classifying and regressing the real-time predicted power demand and the real-time predicted power load through a support vector machine algorithm to obtain a power supply capacity change trend in the future period, wherein the power supply capacity change trend is calculated as follows: ΔP(t gd )=τ1·V(t gd )+τ2·I(t gd )+τ3·F(t gd ) Among them, ΔP(t gd ) is time t gd The power supply capacity change trend at t gd ) is time t gd The voltage at the time, I(t gd ) is time t gd The current at the time, F(t gd ) is time t gd The frequency at which the voltage is measured, τ1 is the weight coefficient of the voltage, τ2 is the weight coefficient of the current, and τ3 is the weight coefficient of the frequency; S44, calculating the real-time predicted power load based on a fuzzy logic algorithm to obtain a load margin, wherein the calculation formula of the load margin is: MU(t yd ) is time t yd Load margin at time, P pred (t yd ) is time t yd The real-time predicted power load at the time of acnual (t yd ) is time t yd The real-time actual power load at that time.

6. The method for multi-dimensional auxiliary decision-making analysis of power distribution network based on machine learning as claimed in claim 1, characterized in that: Step S5 includes: S51. Determine a power supply capacity index and a load margin index based on a power supply capacity change trend and a load margin; S52, using a weight allocation method to assign different importance weights to the power supply capacity index and the load margin index, and constructing a comprehensive objective function, the calculation formula of the comprehensive objective function is: Among them, F is the comprehensive objective function value, C k is the actual value of the kth power supply capability indicator, L l is the actual value of the lth load margin index, C max,k The maximum value of the power supply capability index is standardized, L max,l is the maximum value of the load margin index, w k is the weight coefficient of the power supply capability index, v l is the weight coefficient of the load margin index, N is the number of power supply capacity indicators, and M is the number of load margin indicators; S53, optimizing and solving the comprehensive objective function by a genetic algorithm, initializing the population, and generating an initial population; S54, evaluating the fitness of each individual in the initial population based on the comprehensive objective function, and selecting the individual with the highest fitness based on the elite retention strategy; S55. Generate a new generation of individuals through partial random crossover and mutation, and generate new individuals through repeated iterations. When the iteration reaches a preset number of iterations or the fitness change meets the requirements, stop the iteration and output the optimal power supply decision plan.

7. The method for multi-dimensional auxiliary decision-making analysis of power distribution network based on machine learning as claimed in claim 1, characterized in that: Step S6 includes: S61. According to the optimal power supply decision plan, adjust the operating parameters of the equipment in the distribution network and redistribute the load demand of the equipment; S62, through the distribution network control center, the adjusted operating parameters and the reallocated load demand are sent to the corresponding devices in the form of control commands; S63, deploying IoT sensors on key equipment in the distribution network, and collecting power supply operation parameters in real time based on the IoT sensors; S64, cleaning, filtering and normalizing the power supply operation parameters to obtain distribution network power supply operation data, and transmitting the distribution network power supply operation data to a centralized monitoring platform for real-time storage and backup; S65. Analyze the current power supply status based on the distribution network power supply operation data, identify potential anomalies and operation bottlenecks, and obtain status assessment results; S66. According to the status assessment results and in combination with historical data, a corresponding adjustment strategy is formulated and executed to dynamically adjust the optimal power supply decision plan.

8. A multi-dimensional auxiliary decision analysis system for distribution network power supply based on machine learning, characterized in that: The system comprises: Data acquisition module, used to obtain historical distribution network data and historical power load data; A power supply demand module is used to construct an initial power supply demand prediction model through a long short-term memory network, and train and verify the initial power supply demand prediction model based on the distribution network historical data to obtain a power supply demand prediction model; A load change module, used to construct an initial load change model through a time series prediction algorithm, train and verify the initial load change model based on the historical power load data, and obtain a load change model; A trend margin module, based on the power demand prediction model and the load change model, identifies the real-time operation data of the distribution network to obtain the power supply capacity change trend and load margin; The target optimization module is used to construct a comprehensive target function through the power supply capacity change trend and the load margin, optimize and solve the comprehensive target function through the genetic algorithm, and output the optimal power supply decision plan; The scheme adjustment module is used to execute the optimal power supply decision scheme, monitor the distribution network power supply operation parameters in real time, and dynamically adjust the optimal power supply decision scheme based on the distribution network power supply operation parameters.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other via the bus, the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images