Distribution network planning problem automatic allocation solution measure algorithm library based on artificial intelligence

Abstract

The invention discloses a distribution network planning problem automatic allocation solution measure algorithm library based on artificial intelligence, which comprises the steps of constructing a distribution network planning typical problem definition standard library model, constructing an artificial intelligence algorithm library, automatically analyzing distribution network planning typical problem solution measures, and automatically distributing the distribution network planning typical problem solution measures. The method comprises the following steps: automatically analyzing power distribution network planning typical problem solving measure process visualization, machine learning and deep learning analysis engine, intelligent fault problem positioning, artificial intelligence-based dynamic load problem prediction, and transformer area low voltage problem cause intelligent analysis. According to the distribution network planning problem automatic allocation solution measure algorithm library based on artificial intelligence, a unified problem definition standard library is established, the definition standard of the distribution network planning problem is standardized, and the early-stage planning analysis efficiency and accuracy of a power grid can be improved; and power grid planning and construction can be guaranteed, and disaster prevention and reduction power supply planning and construction can be promoted actively.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis and processing, and in particular relates to an algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence. Background Art

[0002] The collection, integration, calculation and analysis of basic data for distribution network planning are complex and completely rely on manual work. With counties (districts) as units and grids as units, there is no unified problem definition standard. It is necessary to sort out the existing problems of the distribution network equipment one by one (substations, lines, and substations). There is no unified format numbering for power grid problems, which makes it inconvenient to track the resolution of the problems. In response to these two problems, the present invention establishes a unified problem definition standard library, standardizes the definition standards for distribution network planning problems, and sorts out the existing problems of the distribution network equipment one by one, including data such as specific equipment for the problem and specific manifestations of the problem. The lack of efficiency and accuracy in the early planning and analysis of the power grid indirectly leads to an increase in the number of emergency repairs and a longer time, resulting in customer complaints. Therefore, improving the accuracy of planning problems and the efficiency of analysis is conducive to the bottom-line planning and construction of power grids and the active promotion of disaster prevention and mitigation power supply planning and construction. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a standard for defining distribution network planning problems, improve the efficiency and accuracy of early-stage planning and analysis of power grids, and automatically allocate solution measures algorithm libraries for distribution network planning problems based on artificial intelligence, which is beneficial to the planning and construction of guaranteed power grids and actively promotes the planning and construction of disaster prevention and mitigation power sources.

[0004] The technical content of the present invention is: an algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence, including:

[0005] S1. Construct a standard library model for defining typical problems in distribution network planning. The typical problems in distribution network planning include four dimensions: problem causes, problem classification, investment strategies and grid indicators;

[0006] S2. Build an artificial intelligence algorithm library to automatically analyze solutions to typical problems in distribution network planning: Based on the standard library model for defining typical problems in distribution network planning, solidify the automatic allocation of solutions to typical problems in distribution network planning, integrate different types of problem solutions, fields and algorithms into an algorithm library, where the fields are specific data fields related to problems and solutions; build an adaptive algorithm selection mechanism to associate typical problem types and algorithms in distribution network planning: Suppose there is a set P representing all possible typical problems in distribution network planning, and a set R representing all algorithms. The mapping relationship is defined as the relationship PR:

[0007] PR=P×R,

[0008] Where P×R represents the Cartesian product of the problem and the algorithm, PR represents the set of problem-algorithm pairs, and for each problem p∈P, there is a unique corresponding algorithm r∈R, which is represented as f:P→R, where f(p)=R, indicating that the problem p is mapped to the algorithm r;

[0009] The algorithm for automatically allocating the solution measures includes: genetic algorithm and particle swarm optimization algorithm. The application steps of the genetic algorithm are: initializing the population, randomly generating 2-4 sets of candidate solutions, each solution represents a different device location and topology configuration; performing fitness evaluation, and using the fitness function to evaluate the fitness of the population. The expression of the fitness function is:

[0010] f(x)=α·Cost(x)+β·Reliability(x),

[0011] Among them, α and β are weight coefficients, Cost represents cost, and Reliability represents reliability;

[0012] Through crossover and mutation, individuals with higher fitness are selected for crossover and mutation to generate the next generation. The expression of crossover and mutation is:

[0013]

[0014] P mutated =P next +δ·N(0,σ),

[0015] Among them, P next represents the parameter value of the next generation of individuals, P parent1 represents the parameter value of one of the parent individuals, P parent2 Represents the parameter value of another parent individual, p c represents the probability of selecting the first parent individual, P mutated represents the parameter value after mutation, δ represents the scalar coefficient, and N(0,σ) represents a normal distribution with a mean of 0 and a standard deviation of σ;

[0016] The application steps of the particle swarm optimization algorithm are: initializing particles, each particle represents a topology configuration or fault recovery strategy, updating the speed of each particle, updating the position of each particle, updating the individual optimal position and the global optimal position, and achieving the global optimal position convergence to a smaller range of variation; the speed update formula of the particle swarm optimization algorithm is:

[0017]

[0018] Among them, v i (t+1) represents the velocity of the ith particle in the t+1th generation, v i(t) represents the velocity of the ith particle in the tth generation, ω represents the inertia weight, c1 and c2 represent acceleration constants, r1 and r2 represent random numbers uniformly distributed between [0,1], and pbest i represents the individual optimal position of the ith particle in the tth generation, gbest represents the global optimal position, and x i (t) represents the position of the i-th particle in the t-th generation;

[0019] The position update formula of the particle swarm optimization algorithm is:

[0020]

[0021] Among them, x i (t+1) represents the position of the ith particle in the t+1th generation, x i (t) represents the position of the ith particle in the tth generation, v i (t+1) represents the velocity of the i-th particle in the t+1 generation;

[0022] The expression for updating the individual optimal position of the particle swarm optimization algorithm is:

[0023]

[0024] Among them, p i (t+1) represents the individual optimal position of the i-th particle in the t+1th generation, p i (t) represents the individual optimal position of the i-th particle in the t-th generation, x i (t+1) represents the position of the i-th particle in the t+1 generation;

[0025] The expression for updating the global optimal position of the particle swarm optimization algorithm is:

[0026]

[0027] Among them, g(t+1) represents the global optimal position in the t+1th generation, p i (t+1) represents the individual optimal position of the i-th particle in the t+1th generation;

[0028] S3, automatic analysis and visualization of the solution process of typical problems in distribution network planning;

[0029] S4, machine learning and deep learning analysis engine: data collection, data preprocessing, model selection and training, model evaluation and optimization, anomaly detection and notification mechanism and continuous learning and feedback mechanism. The data collection includes the collection of historical load data and equipment status data;

[0030] S5. Intelligent fault location: Combined with deep learning models, the fault area can be quickly located by analyzing equipment data and historical fault records, and the specific location and impact range of the fault point can be displayed in a visual way;

[0031] S6. Prediction of dynamic load problems based on artificial intelligence: Based on time series analysis, artificial intelligence models are used to predict the future load change trend of the distribution network. Time series analysis is used to analyze and model historical load data. The autoregressive integral moving average model is used to model load changes. The formula is:

[0032] Y t =μ+φ1Y t-1 +φ2Y t-2 +...+φ p Y t-p +θ1∈ t-1 +θ2∈ t-2 +...+θ q ∈ t-q +∈ t ,

[0033] Among them, Y t represents the load at time t, μ represents the constant term, φ and θ both represent model parameters, and ∈t represents white noise;

[0034] In the load forecasting task, the load forecasting model uses the following algorithm to optimize the resource usage of edge computing. The formula is:

[0035]

[0036] Among them, T predict It represents the time required for load forecasting, T total represents the total processing time, C used Indicates the estimated amount of resources used, C total Indicates the total amount of resources;

[0037] The ant colony algorithm is used to optimize the parameters of the autoregressive integrated moving average model in time series analysis. The specific steps include: defining the problem, clarifying the target to be optimized, minimizing the prediction error of the ARIMA model, adjusting the number of autoregressive terms p, the difference order d and the number of moving average terms q; initializing the ant colony algorithm, defining a certain number of ants, simulating their search in the parameter space, initially allocating a certain amount of "pheromone" to each parameter value to affect the movement probability of the ants; the ants construct solutions, each ant selects a parameter combination (p, d, q) according to the current parameter pheromone concentration and heuristic information; updating pheromones, updating pheromones according to the performance of the ants, when the ants complete a path, they will add pheromones to the path, a good solution will increase pheromones, and a bad solution will reduce pheromones. The expression of the ant colony algorithm is:

[0038]

[0039] Where, ΔT ij represents the amount of pheromone added on edge i to j, Q represents the strength of pheromone, L k represents the length of the path completed by ant k;

[0040] Repeat the above steps until the optimal solution has no significant improvement in several consecutive iterations, or the change of pheromone is less than the threshold, and finally select the parameter combination with the highest pheromone concentration as the optimal parameter;

[0041] S7. Intelligent analysis of the causes of low voltage problems in the substation area: Through the principal component analysis of the deep learning analysis engine, the covariance matrix is ​​calculated, and the low voltage problems in the substation area caused by distribution transformers and low-voltage line problems are analyzed. Edge computing is used to process data at each node in the substation area, and data processing is moved closer to the data source. Edge devices are installed in the substation area to quickly collect and process real-time data on current, voltage, and load, check the monitoring data sent by the terminal, and monitor voltage and current fluctuations in real time. When voltage abnormalities are detected, the edge computing node can quickly analyze the data and issue an alarm in time. The formula for the edge computing is:

[0042]

[0043] Among them, R represents the comparative benefit, P AI is the prediction accuracy of the artificial intelligence model on edge computing, P cloud To calculate the prediction accuracy, P local The baseline accuracy of local processing is obtained; the artificial intelligence interpretability technology LIME is used in the analysis process. The workflow of the artificial intelligence interpretability technology LIME is as follows: input data points; generate perturbation data; predict the output of the perturbation data, use a complex model to predict the generated perturbation data, and record the prediction results of each perturbation instance; construct a weighted data set to assign weights to the perturbation data; train a local linear model, use the weighted perturbation data and the corresponding prediction results to train a simple interpretable model; visualize the explanation; the expression of the trained local linear model is:

[0044]

[0045] Where f represents a complex model and x′ i represents the generated neighboring samples, π(x′ i ) represents the weight of the generated neighboring samples, g represents a simple interpretable model, Ω(g) represents the complexity of the simple model, λ represents the regularization coefficient, and N represents the total number of neighboring samples;

[0046] S8. Intelligent optimization analysis of fault problems: Through principal component analysis of the deep learning analysis engine, the blockchain smart contract technology is applied, and sensors and smart devices are deployed at the same time to collect data from the power system in real time, and these data are uploaded to the chain. The data obtained from monitoring is recorded in the blockchain, and the corresponding thresholds are set in the system. The system automatically judges and automatically executes the corresponding response measures. When a fault is triggered, the smart contract will automatically record the time, location, situation and handling process of the fault, ensuring that all information is uploaded to the chain and the execution of all contracts can be traced on the blockchain.

[0047] Furthermore, the visualization of the process of automatic analysis of solutions to typical problems in distribution network planning is specifically as follows: the process of automatic allocation of solutions to typical problems in distribution network planning is solidified by using a JAVA workflow engine, each step of allocating solutions is recorded in detail in a database, and the automatic allocation process of solutions is checked to determine the operating status of power grid equipment.

[0048] Furthermore, the data collection includes: collection of sensor data and operation log data.

[0049] Furthermore, the data preprocessing includes: missing value processing, standardization and normalization, feature selection, principal component analysis and calculation of covariance matrix and eigenvalue and eigenvector solution. The missing value processing is specifically: processing using mean interpolation or linear interpolation. The expression of mean interpolation is: Among them, i represents the location of the missing data point, x i represents the value of the data point with missing data, n represents the number of samples, j represents the position of each non-missing data point in the data set, and x j represents the value of the jth data point in the data set; the standardized expression is: Wherein, z represents the standardized data value, x represents the original data value, μ represents the mean of the original data, and σ represents the standard deviation of the original data; the normalized expression is: Wherein, x′ represents the normalized data value, and x represents the original data value; the feature selection is: using K-fold cross-validation ratio to select correlation features; the calculation formula of the covariance matrix is: Where n represents the number of samples, X T represents the transpose of matrix X, X represents the data matrix; the solution expression of the eigenvalue and the eigenvector is: det(∑-λI)=0, wherein Σ represents the covariance matrix, λ represents the eigenvalue, and Ι represents the unit matrix.

[0050] Furthermore, the model selection and training includes: decision tree, support vector machine, regression analysis and central mean calculation of K-means clustering, and the expression of the decision tree is:

[0051] Where IG(D,A) represents the information gain obtained by partitioning feature A on data set D, IG represents information gain, D represents data set, A represents the feature used to partition data set, H represents entropy, H(D) represents the entropy of data set D, v represents the value of feature A, D v represents the subset of feature A in data set D whose value is v, H(D V ) represents the data subset D v The entropy of the support vector machine is expressed as: Where w represents the normal vector of the hyperplane, y i represents the label of the i-th sample, x i represents the feature vector of the i-th sample, w T x i Represents the weight vector w and the i-th sample feature vector x i The inner product of , b represents the intercept of the hyperplane; the expression of the regression analysis is: Among them, L represents the mean square error, N represents the number of samples, and y i represents the true value of the i-th sample, w represents the normal vector of the hyperplane, x i represents the eigenvector of the i-th sample, and b represents the intercept of the hyperplane; the expression for calculating the central mean of the K-means clustering is: Among them, μ j represents the centroid of the jth cluster, C j represents the set of all data points belonging to the jth cluster, x i Represents data cluster C j The i-th sample point of .

[0052] Furthermore, the model evaluation and optimization includes the following steps: cross-validation, hyperparameter adjustment, determination of evaluation indicators and mean square error optimization. The cross-validation is specifically: dividing the data set into k parts, taking out one part as the test set in turn, and the remaining k-1 parts as the training set; the hyperparameter adjustment is specifically: using the grid search method to find the best hyperparameter combination in the specified search space; the expression of the evaluation indicator is: Wherein, Accuracy represents the accuracy rate, TP represents the number of positive samples correctly predicted as positive by the model, TN represents the number of negative samples correctly predicted as negative by the model, FP represents the number of negative samples incorrectly predicted as positive by the model, and FN represents the number of positive samples incorrectly predicted as negative by the model; the expression of the mean square error is:

[0053] Among them, MSE represents mean square error, N represents the number of samples, and y i represents the true value of the i-th sample, Represents the predicted value of the i-th sample.

[0054] Furthermore, in the mean square error expression The expression is: Where w represents the normal vector of the hyperplane, x i represents the feature vector of the i-th sample, and b represents the intercept of the hyperplane; the mean square error optimization is specifically: optimizing the model parameters w and b to minimize the mean square error.

[0055] Furthermore, the anomaly detection and notification mechanism includes: performing problem diagnosis and analysis on the collected data, locating abnormal data and classifying problems; the expression of the anomaly detection algorithm is:

[0056]

[0057] Among them, Z i Represents data point x i The number of standard deviations from the population mean μ, x i represents the original data value, μ represents the mean of the original data, and σ represents the standard deviation of the original data;

[0058] If | Z i |>3, the data point is considered abnormal. The real-time warning is specifically to send the alarm information to the operation and maintenance personnel through API or internal notification system; the problem classification is specifically to classify the fault according to the severity and type of the problem.

[0059] Furthermore, the continuous learning and feedback mechanism includes: online learning and feedback tuning. The online learning is specifically: updating the model in real time and incorporating new data; the feedback tuning is specifically: using Q-learning or deep reinforcement learning methods to adjust the distribution network design strategy according to real-time feedback. The reinforcement learning model formula is:

[0060]

[0061] Among them, Q(s, a) represents the estimated value of performing action a in state s, r represents the immediate reward, γ represents the discount factor, α represents the learning rate, s represents the current state, a represents the action in state s, s′ represents the new state after taking action a immediately, and a′ represents the optimal action.

[0062] Beneficial effects of the present invention: an algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence of the present invention standardizes the definition standards of distribution network planning problems by establishing a unified problem definition standard library; by constructing an artificial intelligence algorithm library, typical solutions to distribution network planning problems are automatically analyzed, and automatic identification of typical solutions to distribution network planning problems is realized; by using machine learning and deep learning analysis engines to analyze real-time and historical data of the distribution network, abnormal patterns and potential problems are identified, and efficient fault prediction, load analysis and operation optimization are realized; by using artificial intelligence models to predict future load change trends in the distribution network, a scientific decision-making basis is provided, and the resource allocation and load management of the distribution network are optimized; in short, an algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence of the present invention establishes a unified problem definition standard library, standardizes the definition standards of distribution network planning problems, can improve the efficiency and accuracy of early planning analysis of the power grid, is conducive to the planning and construction of a guaranteed power grid and actively promotes the planning and construction of disaster prevention and mitigation power sources. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention.

[0064] Figure 1 It is a schematic diagram of a standard library model for defining typical distribution network planning problems based on artificial intelligence and an algorithm library for automatically allocating solutions to distribution network planning problems according to the present invention.

[0065] Figure 2 The invention discloses an automatic allocation flow chart of low voltage solutions for distribution areas based on an algorithm library of automatic allocation solutions for distribution network planning problems based on artificial intelligence.

[0066] Figure 3 The present invention is a schematic diagram of medium-voltage line segment optimization in intelligent fault problem positioning of an algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence.

[0067] Figure 4 This is a schematic diagram of load transfer optimization in the fault section of the intelligent fault problem location in the distribution network planning problem automatic allocation solution algorithm library based on artificial intelligence of the present invention.

[0068] Figure 5The present invention is a schematic diagram of automatic configuration of medium-voltage lines in intelligent fault problem location based on an artificial intelligence-based distribution network planning problem solution algorithm library.

[0069] Figure 6 It is a schematic diagram of line analysis based on artificial intelligence dynamic load problem prediction of an algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence of the present invention.

[0070] Figure 7 It is a schematic diagram of the analysis of lines with a load rate higher than 70% based on artificial intelligence dynamic load problem prediction in an algorithm library of automatic allocation solutions to distribution network planning problems based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0071] The following will be combined with the accompanying drawings of the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] like Figure 1-2 As shown, an algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence includes:

[0073] S1. Construct a standard library model for defining typical problems in distribution network planning. The typical problems in distribution network planning include four dimensions: problem causes, problem classification, investment strategies and grid indicators;

[0074] S2. Build an artificial intelligence algorithm library to automatically analyze solutions to typical problems in distribution network planning: Based on the distribution network planning and network planning typical problem definition standard library model, solidify the automatic allocation algorithm for solutions to typical problems in distribution network planning, integrate different types of problem solutions, fields and algorithms into an algorithm library, and the fields are specific data fields related to problems and solutions (taking the low voltage in the substation as an example, the automatic allocation process of solutions to the low voltage in the substation is as follows Figure 2 As shown: First, determine whether the voltage on the low-voltage side of the distribution transformer is qualified. If it is not qualified, determine whether the voltage on the high-voltage side of the distribution transformer is qualified. If it is not qualified, determine whether the voltage at the end of the medium-voltage line is qualified. If it is qualified, determine whether the bus voltage of the upper-level substation is qualified. If it is not qualified, assign solution 7 to the problem: solve it through the upper-level power grid), build an adaptive algorithm selection mechanism, and associate the typical problem types and algorithms of distribution network planning: set P, representing all typical problems of distribution network planning, set R represents all algorithms, and the mapping relationship is defined as relationship PR:

[0075] PR = P × R, where P × R represents the Cartesian product of the problem and the algorithm. PR is a set of problem-algorithm pairs. For each problem p∈P, there is a unique corresponding algorithm r∈R, expressed as f:P→R, where f(p) = R, indicating that problem p is mapped to algorithm r.

[0076] The algorithm for automatically allocating the solution measures includes: genetic algorithm and particle swarm optimization algorithm. The application steps of the genetic algorithm are: initializing the population, randomly generating 2-4 sets of candidate solutions, each solution represents a different device location and topology configuration; performing fitness evaluation, and using the fitness function to evaluate the fitness of the population. The expression of the fitness function is:

[0077] f(x)=α·Cost(x)+β·Reliability(x),

[0078] Among them, α and β are weight coefficients, Cost represents cost, and Reliability represents reliability;

[0079] Through crossover and mutation, individuals with higher fitness are selected for crossover and mutation to generate the next generation. The expression of crossover and mutation is:

[0080]

[0081] P mutated =P next +δ·N(0,σ),

[0082] Among them, P next represents the parameter value of the next generation of individuals, P parent1 represents the parameter value of one of the parent individuals, P parent2 Represents the parameter value of another parent individual, p c represents the probability of selecting the first parent individual, P mutated represents the parameter value after mutation, δ represents the scalar coefficient, and N(0,σ) represents a normal distribution with a mean of 0 and a standard deviation of σ;

[0083] The application steps of the particle swarm optimization algorithm are: initializing particles, each particle represents a topology configuration or fault recovery strategy, updating the speed of each particle, updating the position of each particle, updating the individual optimal position and the global optimal position, and achieving the global optimal position convergence to a smaller range of variation; the speed update formula of the particle swarm optimization algorithm is:

[0084]

[0085] Among them, v i (t+1) represents the velocity of the ith particle in the t+1th generation, v i(t) represents the velocity of the ith particle in the tth generation, ω represents the inertia weight, c1 and c2 represent acceleration constants, r1 and r2 represent random numbers uniformly distributed between [0,1], and pbest i represents the individual optimal position of the ith particle in the tth generation, gbest represents the global optimal position, and x i (t) represents the position of the i-th particle in the t-th generation;

[0086] The position update formula of the particle swarm optimization algorithm is:

[0087]

[0088] Among them, x i (t+1) represents the position of the ith particle in the t+1th generation, x i (t) represents the position of the ith particle in the tth generation, v i (t+1) represents the velocity of the i-th particle in the t+1 generation;

[0089] The expression for updating the individual optimal position of the particle swarm optimization algorithm is:

[0090]

[0091] Among them, p i (t+1) represents the individual optimal position of the i-th particle in the t+1th generation, p i (t) represents the individual optimal position of the i-th particle in the t-th generation, x i (t+1) represents the position of the i-th particle in the t+1 generation;

[0092] The expression for updating the global optimal position of the particle swarm optimization algorithm is:

[0093]

[0094] Among them, g(t+1) represents the global optimal position in the t+1th generation, p i (t+1) represents the individual optimal position of the i-th particle in the t+1th generation;

[0095] S3, automatic analysis and visualization of the solution process of typical problems in distribution network planning;

[0096] S4, machine learning and deep learning analysis engine: data collection, data preprocessing, model selection and training, model evaluation and optimization, anomaly detection and notification mechanism and continuous learning and feedback mechanism. The data collection includes the collection of historical load data and equipment status data;

[0097] S5. Intelligent fault location: Combined with deep learning models, the fault area can be quickly located by analyzing equipment data and historical fault records, and the specific location and impact range of the fault point can be displayed in a visual way. For example: Figure 3 As shown in the figure, the unreasonable segmentation of medium-voltage lines is analyzed: the fault point is on the main line, and the load distribution is considered to make reasonable segmentation; an intelligent circuit breaker is installed at the head end of the branch line where the fault point is located, and the branch line becomes the main line, and the distribution automation configuration is optimized. For branches, the original large branches are changed to main lines to form a ring network with other lines; it should be noted that: try to consider forming a ring network with other lines at the end or middle of the large branch line to avoid forming a new large branch line; when the branch becomes the main line, the original branch line can be set up with automatic segmentation according to the principle of the main line; the original branch line should be used as the main line, and the matching of the conductor cross-section should be considered; for large branches that do not meet the conditions for splitting, after setting an automatic boundary circuit breaker at the head end of the branch line, the branch line automatic segmentation load switch can be set up according to the principle of main line segmentation to segment the branch line; as shown in Figure 4 As shown in the figure, the fault section load transfer problem analysis: if there are new distribution points within the power supply range, the cause analysis will be based on the new distribution points first. If there are no new distribution points, consider whether the existing stations have the load shedding capability. It should be noted that when the contact point is adjusted to the middle and rear sections of the line, it is necessary to consider the simultaneous modification of the conductor cross-section. There should be no more than 3 contact points for overhead lines, ensuring that one of the contact points is the terminal contact. If the terminal of one overhead line is connected to the head section of another line, if the contact can satisfy the full line transfer of one line, it can be retained if the contact adjustment conditions are not met. Figure 5As shown in the figure, analysis of insufficient automation configuration of medium-voltage lines: Setting of automated section switches and automated interconnecting switches: both use load switches, and configure voltage and current type feeder automation and three-remote functions. Except for the branch line automated section switch configured with a single-side power supply PT, all others are configured with a double-side power supply PT; on the trunk line, there are 5 to 30 medium-voltage users and no more than 2,000 low-voltage users in the automated section. The line length in the section should not be less than 2 kilometers. For large branch lines that do not meet the conditions for splitting, branch line automated section load switches can be set; for some lines dedicated to public lines, or lines with a trunk length of no more than 2 kilometers, and with a small number of users and less than 2,000 low-voltage users, automated switches may not be set; automated circuit breaker setting: set at the head end of the branch line that meets the conditions or Special transformer user boundary: branch lines with 10 medium-voltage users or more than 1,000 low-voltage users; branch lines with high line failure rates or in difficult terrain for repair; branch lines with small hydropower access; when there are many users going out due to failures and management measures are ineffective, the public switch at the boundary with the special transformer users shall be changed to an automatic circuit breaker; when the length of the line trunk is greater than 15 kilometers, if the end exceeds the protection range of the substation outgoing line switch, it shall be treated as the head end of the exceeding part of the branch line. Except for the small hydropower branch line, the automatic circuit breaker shall be equipped with quick-break, zero-sequence overcurrent protection, reclosing and three-remote functions, and a single-side power supply PT shall be configured on the power supply side. If there is a small hydropower branch line, the automatic circuit breaker shall be equipped with additional protection functions such as overvoltage, high-frequency decoupling, and synchronous reclosing, and a double-side power supply PT shall be configured;

[0098] S6. Prediction of dynamic load problems based on artificial intelligence: Based on time series analysis, artificial intelligence models are used to predict the future load change trend of the distribution network. Time series analysis is used to analyze and model historical load data. The autoregressive integral moving average model is used to model load changes. The formula is:

[0099] Y t =μ+φ1Y t-1 +φ2Y t-2 +...+φ p Y t-p +θ1∈ t-1 +θ2∈ t-2 +...+θ q ∈ t-q +∈ t ,

[0100] Among them, Y t represents the load at time t, μ represents the constant term, φ and θ both represent model parameters, and ∈t represents white noise;

[0101] In the load forecasting task, the load forecasting model uses the following algorithm to optimize the resource usage of edge computing. The formula is:

[0102]

[0103] Among them, T predict It represents the time required for load forecasting, T total represents the total processing time, C used Indicates the estimated amount of resources used, C total Indicates the total amount of resources;

[0104] The ant colony algorithm is used to optimize the parameters of the autoregressive integrated moving average model in time series analysis. The specific steps include: defining the problem, clarifying the target to be optimized, minimizing the prediction error of the ARIMA model, adjusting the number of autoregressive terms p, the difference order d and the number of moving average terms q; initializing the ant colony algorithm, defining a certain number of ants, simulating their search in the parameter space, and initially allocating a certain amount of "pheromone" on each parameter value to affect the movement probability of the ants; the ants construct solutions, and each ant selects a parameter combination (p, d, q) according to the current parameter pheromone concentration and heuristic information; updating pheromones, updating pheromones according to the performance of the ants, when the ants complete a path, they will add pheromones to the path, good solutions (solutions with objective function values ​​close to the optimal solution, paths with higher pheromone concentrations, or a good balance between exploring new possibilities and using existing excellent paths) increase pheromones, and bad solutions (solutions with significantly poorer objective function values, paths with lower pheromone concentrations, or the algorithm falls into a local optimum) reduce pheromones. The expression of the ant colony algorithm is:

[0105]

[0106] Where, ΔT ij represents the amount of pheromone added on edge i to j, Q represents the strength of pheromone, L k represents the length of the path completed by ant k;

[0107] Repeat the above steps until the optimal solution has no significant improvement in several consecutive iterations, or the change of pheromone is less than the threshold, and finally select the parameter combination with the highest pheromone concentration as the optimal parameter;

[0108] For example: Figure 6 As shown, the substation branch line is decomposed into surrounding lines: for lines with 2-3 large branches, the large branches can be split and connected to surrounding lines with power supply margin for analysis; Figure 7As shown, for lines with a load rate higher than 70% in two years and with conditions for further development of the load, projects should be planned in time; the selected existing lines should meet the conditions for transfer; the two lines should avoid the same busbar and the same pole; small-section conductors are dilapidated or have safety hazards, or when it is difficult to connect the branch line to the main line, small-section lines should be selected for cascade transformation; when the branch line is predicted to be overloaded within two years due to the small cross-section, the transformation project should be planned in time, and the solution should be selected to comprehensively consider the number of households connected to the branch line, the power outage time of the transformation, the asset value and other factors; the cross-section of the overhead line secondary trunk line is 120mm 2 Consider that the branch line is 70mm 2 Consider that the cable branch line is 120mm 2 or 70mm 2 Consider: if the branch line is planned to be connected with other lines in the future, the branch line should be transformed into a conductor with the same cross-section as the main line; small-section conductors are dilapidated or have safety hazards, or when it is difficult to connect the branch line to the main line, choose to transform the small-section line in stages;

[0109] S7. Intelligent analysis of the causes of low voltage problems in the substation area: Through the principal component analysis of the deep learning analysis engine, the covariance matrix is ​​calculated, and the low voltage problems in the substation area caused by distribution transformers and low-voltage line problems are analyzed. Edge computing is used to process data at each node in the substation area, and data processing is moved closer to the data source. Edge devices are installed in the substation area to quickly collect and process real-time data on current, voltage, and load, check the monitoring data sent by the terminal, and monitor voltage and current fluctuations in real time. When voltage abnormalities are detected, the edge computing node can quickly analyze the data and issue an alarm in time. The formula for the edge computing is:

[0110]

[0111] Among them, R represents the comparative benefit, P AI is the prediction accuracy of the artificial intelligence model on edge computing, P cloud To calculate the prediction accuracy, P localThe baseline accuracy for local processing is obtained; the AI ​​interpretability technology LIME is used in the analysis process (LIME is an interpretability technology for explaining complex machine learning models. It approximates the behavior of complex models near specific inputs by building a local linear model to provide an explanation for a single prediction). The workflow of the AI ​​interpretability technology LIME is as follows: input data points; generate perturbation data (perturbation can be achieved by adding noise, randomly selecting feature values, etc.); predict the output of the perturbation data, use a complex model to predict the generated perturbation data, and record the prediction results of each perturbation instance; construct a weighted data set and assign weights to the perturbation data (the weights are based on the similarity between the perturbation data and the original instance; The closer the perturbation data is to the original instance, the higher the weight is, and the weight is usually calculated using an exponential decay function); train a local linear model, use the weighted perturbation data and the corresponding prediction results to train a simple interpretable model (this model aims to approximate the behavior of the complex model near the original instance; extract feature importance, extract the importance of features from the trained local linear model. The importance of a feature can be represented by the coefficient of the linear model. The larger the absolute value of the coefficient, the greater the influence of the feature on the prediction result); visualize the explanation (visualize the importance of the feature, usually using a bar chart or other graphical display, so that the user can intuitively understand which features have the greatest impact on the prediction results of the model); the expression for training the local linear model is:

[0112]

[0113] Where f represents a complex model and x′ i represents the generated neighboring samples, π(x′ i ) represents the weight of the generated neighboring samples, g represents a simple interpretable model, Ω(g) represents the complexity of the simple model, λ represents the regularization coefficient, and N represents the total number of neighboring samples;

[0114] For example: Figure 2As shown, the voltage data of the distribution transformer terminal of the metering automation system is used to determine whether there is a problem of low voltage at the head end of the substation; through the voltage data of the Class D voltage monitoring point of the voltage monitoring system, the voltage data of the smart meter of the metering automation system, the on-site voltage measurement data and the verification of low voltage complaints, it is determined whether there is a problem of low local voltage. Pay attention to the substation data with a voltage lower than 210V at the head end of the substation, a power supply radius of more than 500 meters, a load rate of more than 60%, a three-phase imbalance of more than 15%, a small wire diameter, and a power factor lower than 0.95; sort out the substations with potential low voltage risks and plan construction in advance; gear analysis: For the situation where the medium-voltage line voltage is qualified but the voltage at the head end of the substation is lower than 220V, check the voltage curve on the secondary side of the distribution transformer to obtain the maximum and minimum voltage values, calculate the voltage after the gear adjustment, and adjust the gear. If the original gear is "0%, "2.5%, "5%, "then adjust the gear to "-2.5%" or "-5%. After the gear adjustment, the secondary voltage of the distribution transformer is within the qualified range, and the load peak period is controlled at around 230V as much as possible; load analysis: if industrial load and residential load are mixed, split the industrial load and residential load, and evaluate whether the industrial load needs to be equipped with compensation devices for power quality management; by analyzing the main transformer voltage regulation mode, for the main transformer with no-load voltage regulation mode, if conditions permit, transform it into on-load voltage regulation mode; according to the principle of local reactive power balance in the 220kV area, configure reactive power compensation equipment of sufficient capacity in 110kV and 35kV substations. On this basis, the reactive power balance of the 220kV area during the summer peak load and the trough period of distributed power access should be calculated. If there is a lot of reactive power sent down in the 220kV area, the reactive power compensation capacity of the substation should be increased; AVC closed-loop control should be implemented for all 110kV and 220kV substations in the 220kV area, and the AVC setting should be adjusted according to the principle of "voltage optimization, local reactive power balance in the area". The reactive power compensation capacity of the medium-voltage distribution network substation should be configured according to the transformer load rate of 75% and the natural power factor of the load of 0.85, and the power factor on the medium-voltage side should be compensated to not less than 0.95. In practical applications, it can also be configured according to 20% to 40% of the transformer capacity; The low voltage problem in the substation area caused by low 10kV bus voltage is solved through the above analysis: By analyzing the main transformer voltage regulation mode, power factor, AVC setting and control strategy, the cause is located and corresponding measures are proposed.

[0115] Preferably, the visualization of the process of automatic analysis of solutions to typical problems in distribution network planning is as follows: the process of automatic allocation of solutions to typical problems in distribution network planning is solidified by a program using a JAVA workflow engine, each step of allocating solutions is recorded in detail in a database, and the automatic allocation process of solutions is checked to determine the operating status of power grid equipment.

[0116] Preferably, the data collection includes: collection of sensor data and operation log data.

[0117] Preferably, the data preprocessing includes: missing value processing, standardization and normalization, feature selection, principal component analysis and calculation of covariance matrix and eigenvalue and eigenvector solution. The missing value processing is specifically: processing using mean interpolation or linear interpolation. The expression of mean interpolation is: Among them, i represents the location of the missing data point, x i represents the value of the data point with missing data, n represents the number of samples, j represents the position of each non-missing data point in the data set, and x j represents the value of the jth data point in the data set; the standardized expression is: Wherein, z represents the standardized data value, x represents the original data value, μ represents the mean of the original data, and σ represents the standard deviation of the original data; the normalized expression is:

[0118] Wherein, x′ represents the normalized data value, and x represents the original data value; the feature selection is: using K-fold cross-validation ratio to select correlation features; the calculation formula of the covariance matrix is: Where n represents the number of samples, X T represents the transpose of matrix X, X represents the data matrix; the solution expression of the eigenvalue and the eigenvector is: det(∑-λI)=0, wherein Σ represents the covariance matrix, λ represents the eigenvalue, and Ι represents the unit matrix.

[0119] Preferably, the model selection and training includes: decision tree, support vector machine, regression analysis and central mean calculation of K-means clustering, and the expression of the decision tree is:

[0120] Where IG(D,A) represents the information gain obtained by partitioning feature A on data set D, IG represents information gain, D represents data set, A represents the feature used to partition data set, H represents entropy, H(D) represents the entropy of data set D, v represents the value of feature A, D v represents the subset of feature A in data set D whose value is v, H(D v ) represents the data subset D v The entropy of the support vector machine is expressed as: Among them, w represents the normal vector of the hyperplane, y i represents the label of the i-th sample, x i represents the feature vector of the i-th sample, w T x i Represents the weight vector w and the i-th sample feature vector xi The inner product of , b represents the intercept of the hyperplane; the expression of the regression analysis is: Among them, L represents the mean square error, N represents the number of samples, and y i represents the true value of the i-th sample, w represents the normal vector of the hyperplane, x i represents the eigenvector of the i-th sample, and b represents the intercept of the hyperplane; the expression for calculating the central mean of the K-means clustering is: Among them, μ j represents the centroid of the jth cluster, C j represents the set of all data points belonging to the jth cluster, x i Represents data cluster C j The i-th sample point of .

[0121] Preferably, the model evaluation and optimization includes the following steps: cross validation, hyperparameter adjustment, determination of evaluation index and mean square error optimization. The cross validation is specifically: dividing the data set into k parts, taking out one part as the test set in turn, and the remaining k-1 parts as the training set; the hyperparameter adjustment is specifically: using the grid search method to find the best hyperparameter combination in the specified search space; the expression of the evaluation index is:

[0122] Wherein, Accuracy represents the accuracy rate, TP represents the number of positive samples correctly predicted as positive by the model, TN represents the number of negative samples correctly predicted as negative by the model, FP represents the number of negative samples incorrectly predicted as positive by the model, and FN represents the number of positive samples incorrectly predicted as negative by the model; the expression of the mean square error is:

[0123] Among them, MSE represents mean square error, N represents the number of samples, and y i represents the true value of the i-th sample, Represents the predicted value of the i-th sample.

[0124] More preferably, in the mean square error expression The expression is: Where w represents the normal vector of the hyperplane, x i represents the feature vector of the i-th sample, and b represents the intercept of the hyperplane; the mean square error optimization is specifically: optimizing the model parameters w and b to minimize the mean square error.

[0125] Preferably, the anomaly detection and notification mechanism includes: performing problem diagnosis and analysis on the collected data, locating abnormal data and classifying problems; the expression of the anomaly detection algorithm is:

[0126]

[0127] Among them, Z i Represents data point x i The number of standard deviations from the population mean μ, x i represents the original data value, μ represents the mean of the original data, and σ represents the standard deviation of the original data;

[0128] If | Z i |>3, the data point is considered abnormal. The real-time warning is specifically to send the alarm information to the operation and maintenance personnel through API or internal notification system; the problem classification is specifically to classify the fault according to the severity and type of the problem.

[0129] Preferably, the continuous learning and feedback mechanism includes: online learning and feedback tuning, wherein the online learning specifically includes: updating the model in real time and incorporating new data; the feedback tuning specifically includes: using Q-learning or deep reinforcement learning methods to adjust the distribution network design strategy according to real-time feedback, and the reinforcement learning model formula is:

[0130]

[0131] Among them, Q(s, a) represents the estimated value of performing action a in state s, r represents the immediate reward, γ represents the discount factor, α represents the learning rate, s represents the current state, a represents the action in state s, s′ represents the new state after taking action a immediately, and a′ represents the optimal action.

[0132] Although the specific implementation of the present invention is introduced above in conjunction with the accompanying drawings, it does not limit the protection scope of the present invention. Those skilled in the art should understand that those non-creative modifications or equivalent replacements of the technical solution of the present invention, which do not depart from the protection scope of the technical solution, should be included in the scope of the claims of the present invention.

Claims

1. An algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence, characterized by: include: S1. Construct a standard library model for defining typical problems in distribution network planning. The typical problems in distribution network planning include four dimensions: problem causes, problem classification, investment strategies and grid indicators; S2. Build an artificial intelligence algorithm library to automatically analyze solutions to typical problems in distribution network planning: Based on the distribution network planning and network planning typical problem definition standard library model, solidify the automatic allocation of solutions to typical problems in distribution network planning, integrate different types of problem solutions, fields and algorithms into an algorithm library, where the fields are specific data fields related to problems and solutions; build an adaptive algorithm selection mechanism to associate typical problem types and algorithms in distribution network planning: Suppose there is a set P, representing all possible typical problems of distribution network planning, and a set R, representing all algorithms. The mapping relationship is defined as the relationship PR: PR=P×R, Where P×R represents the Cartesian product of the problem and the algorithm, PR represents the set of problem-algorithm pairs, and for each problem p∈P, there is a unique corresponding algorithm r∈R, expressed as f:P→R, where f(p)=R, indicating that the problem p is mapped to the algorithm r; The algorithm for automatically allocating the solution measures includes: genetic algorithm and particle swarm optimization algorithm. The application steps of the genetic algorithm are: initializing the population, randomly generating 2-4 sets of candidate solutions, each solution represents a different device location and topology configuration; performing fitness evaluation, and using the fitness function to evaluate the fitness of the population. The expression of the fitness function is: f(x)=α·Cost(x)+β·Reliability(x), Among them, α and β are weight coefficients, Cost represents cost, and Reliability represents reliability; Through crossover and mutation, individuals with higher fitness are selected for crossover and mutation to generate the next generation. The expression of crossover and mutation is: P mutated =P next +δ·N(0,σ), Among them, P next represents the parameter value of the next generation of individuals, P parent1 represents the parameter value of one of the parent individuals, P parent2 Represents the parameter value of another parent individual, p c represents the probability of selecting the first parent individual, P mutated represents the parameter value after mutation, δ represents the scalar coefficient, and N(0,σ) represents a normal distribution with a mean of 0 and a standard deviation of σ; The application steps of the particle swarm optimization algorithm are: initializing particles, each particle represents a topology configuration or fault recovery strategy, updating the speed of each particle, updating the position of each particle, updating the individual optimal position and the global optimal position, and achieving the global optimal position convergence to a smaller range of variation; the speed update formula of the particle swarm optimization algorithm is: Among them, v i (t+1) represents the velocity of the ith particle in the t+1th generation, v i (t) represents the velocity of the ith particle in the tth generation, ω represents the inertia weight, c1 and c2 represent acceleration constants, r1 and r2 represent random numbers uniformly distributed between [0,1], and pbest i represents the individual optimal position of the ith particle in the tth generation, gbest represents the global optimal position, and x i (t) represents the position of the i-th particle in the t-th generation; The position update formula of the particle swarm optimization algorithm is: Among them, x i (t+1) represents the position of the ith particle in the t+1th generation, x i (t) represents the position of the ith particle in the tth generation, v i (t+1) represents the velocity of the i-th particle in the t+1 generation; The expression for updating the individual optimal position of the particle swarm optimization algorithm is: Among them, p i (t+1) represents the individual optimal position of the i-th particle in the t+1th generation, p i (t) represents the individual optimal position of the i-th particle in the t-th generation, x i (t+1) represents the position of the i-th particle in the t+1 generation; The expression for updating the global optimal position of the particle swarm optimization algorithm is: Among them, g(t+1) represents the global optimal position in the t+1th generation, p i (t+1) represents the individual optimal position of the i-th particle in the t+1th generation; S3, automatic analysis and visualization of the solution process of typical problems in distribution network planning; S4, machine learning and deep learning analysis engine: data collection, data preprocessing, model selection and training, model evaluation and optimization, anomaly detection and notification mechanism and continuous learning and feedback mechanism. The data collection includes the collection of historical load data and equipment status data; S5. Intelligent fault location: Combined with deep learning models, the fault area can be quickly located by analyzing equipment data and historical fault records, and the specific location and impact range of the fault point can be displayed in a visual way; S6. Prediction of dynamic load problems based on artificial intelligence: Based on time series analysis, artificial intelligence models are used to predict the future load change trend of the distribution network. Time series analysis is used to analyze and model historical load data. The autoregressive integral moving average model is used to model load changes. The formula is: Y t =μ+φ1Y t-1 +φ2Y t-2 +...+φ p Y t-p +θ1∈ t-1 +θ2∈ t-2 +...+θ q ∈ t-q +∈ t , Among them, Y t represents the load at time t, μ represents the constant term, φ and θ both represent model parameters, and ∈t represents white noise; In the load forecasting task, the load forecasting model uses the following algorithm to optimize the resource usage of edge computing. The formula is: Among them, T predict It represents the time required for load forecasting, T total represents the total processing time, C used Indicates the estimated amount of resources used, C total Indicates the total amount of resources; The ant colony algorithm is used to optimize the parameters of the autoregressive integrated moving average model in time series analysis. The specific steps include: defining the problem, clarifying the target to be optimized, minimizing the prediction error of the ARIMA model, adjusting the number of autoregressive terms p, the difference order d and the number of moving average terms q; initializing the ant colony algorithm, defining a certain number of ants, simulating their search in the parameter space, initially allocating a certain amount of "pheromone" to each parameter value to affect the movement probability of the ants; the ants construct solutions, each ant selects a parameter combination (p, d, q) according to the current parameter pheromone concentration and heuristic information; updating pheromones, updating pheromones according to the performance of the ants, when the ants complete a path, they will add pheromones to the path, a good solution will increase pheromones, and a bad solution will reduce pheromones. The expression of the ant colony algorithm is: Where, ΔT ij represents the amount of pheromone added on edge i to j, Q represents the strength of pheromone, L k represents the length of the path completed by ant k; Repeat the above steps until the optimal solution has no significant improvement in several consecutive iterations, or the change of pheromone is less than the threshold, and finally select the parameter combination with the highest pheromone concentration as the optimal parameter; S7. Intelligent analysis of the causes of low voltage problems in the substation area: Through the principal component analysis of the deep learning analysis engine, the covariance matrix is ​​calculated, and the low voltage problems in the substation area caused by distribution transformers and low-voltage line problems are analyzed. Edge computing is used to process data at each node in the substation area, and data processing is moved closer to the data source. Edge devices are installed in the substation area to quickly collect and process real-time data on current, voltage, and load, check the monitoring data sent by the terminal, and monitor voltage and current fluctuations in real time. When voltage abnormalities are detected, the edge computing node can quickly analyze the data and issue an alarm in time. The formula for the edge computing is: Among them, R represents the comparative benefit, P AI is the prediction accuracy of the artificial intelligence model on edge computing, P cloud To calculate the prediction accuracy, P local Baseline accuracy for local processing; The artificial intelligence interpretability technology LIME is used in the analysis process. The workflow of the artificial intelligence interpretability technology LIME is as follows: input data points; generate perturbation data; predict the output of the perturbation data, use a complex model to predict the generated perturbation data, and record the prediction results of each perturbation instance; construct a weighted data set to assign weights to the perturbation data; train a local linear model, use the weighted perturbation data and the corresponding prediction results to train a simple interpretable model; visualize the explanation; the expression of the trained local linear model is: Where f represents a complex model and x′ i represents the generated neighboring samples, π(x′ i ) represents the weight of the generated neighboring samples, g represents a simple interpretable model, Ω(g) represents the complexity of the simple model, λ represents the regularization coefficient, and N represents the total number of neighboring samples; S8. Intelligent optimization analysis of fault problems: Through principal component analysis of the deep learning analysis engine, the blockchain smart contract technology is applied, and sensors and smart devices are deployed at the same time to collect data from the power system in real time, and these data are uploaded to the chain. The data obtained from monitoring is recorded in the blockchain, and the corresponding thresholds are set in the system. The system automatically judges and automatically executes the corresponding response measures. When a fault is triggered, the smart contract will automatically record the time, location, situation and handling process of the fault, ensuring that all information is uploaded to the chain and the execution of all contracts can be traced on the blockchain.

2. The algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence according to claim 1 is characterized in that: The visualization of the process of automatically analyzing solutions to typical problems in distribution network planning is specifically as follows: the process of automatically allocating solutions to typical problems in distribution network planning is solidified by using a JAVA workflow engine, each step of allocating solutions is recorded in detail in a database, and the automatic allocation process of solutions is checked to determine the operating status of power grid equipment.

3. The algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence according to claim 1 is characterized in that: The data collection includes: collection of sensor data and operation log data.

4. The algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence according to claim 1 is characterized in that: The data preprocessing includes: missing value processing, standardization and normalization, feature selection, principal component analysis and calculation of covariance matrix and eigenvalue and eigenvector solution. The missing value processing is specifically: processing using mean interpolation or linear interpolation. The expression of mean interpolation is: Among them, i represents the location of the missing data point, x i represents the value of the data point with missing data, n represents the number of samples, j represents the position of each non-missing data point in the data set, and x j represents the value of the jth data point in the data set; the standardized expression is: Wherein, z represents the standardized data value, x represents the original data value, μ represents the mean of the original data, and σ represents the standard deviation of the original data; the normalized expression is: Wherein, x′ represents the normalized data value, and x represents the original data value; the feature selection is: using K-fold cross-validation ratio to select correlation features; the calculation formula of the covariance matrix is: Where n represents the number of samples, x T represents the transpose of matrix X, X represents the data matrix; the solution expression for the eigenvalue and eigenvector is: det(Σ-λI)=0, wherein Σ represents the covariance matrix, λ represents the eigenvalue, and Ι represents the unit matrix.

5. The algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence according to claim 1 is characterized in that: The model selection and training includes: decision tree, support vector machine, regression analysis and central mean calculation of K-means clustering. The expression of the decision tree is: Where IG(D,A) represents the information gain obtained by partitioning feature A on data set D, IG represents information gain, D represents data set, A represents the feature used to partition data set, H represents entropy, H(D) represents the entropy of data set D, v represents the value of feature A, D v represents the subset of feature A in data set D whose value is v, H(D v ) represents the data subset D v The entropy of the support vector machine is expressed as: Among them, w represents the normal vector of the hyperplane, y i represents the label of the i-th sample, x i represents the feature vector of the i-th sample, w T x i Represents the weight vector w and the i-th sample feature vector x i The inner product of , b represents the intercept of the hyperplane; the expression of the regression analysis is: Among them, L represents the mean square error, N represents the number of samples, and y i represents the true value of the i-th sample, w represents the normal vector of the hyperplane, x i represents the eigenvector of the i-th sample, and b represents the intercept of the hyperplane; the expression for calculating the central mean of the K-means clustering is: Among them, μ j represents the centroid of the jth cluster, C j represents the set of all data points belonging to the jth cluster, x i Represents data cluster C j The i-th sample point of .

6. The algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence according to claim 1 is characterized in that: The model evaluation and optimization includes the following steps: cross validation, hyperparameter adjustment, determination of evaluation index and mean square error optimization. The cross validation is specifically: dividing the data set into k parts, taking out one part as the test set in turn, and the remaining k-1 parts as the training set; the hyperparameter adjustment is specifically: using the grid search method to find the best hyperparameter combination in the specified search space; the expression of the evaluation index is: Wherein, Accuracy represents the accuracy rate, TP represents the number of positive samples correctly predicted as positive by the model, TN represents the number of negative samples correctly predicted as negative by the model, FP represents the number of negative samples incorrectly predicted as positive by the model, and FN represents the number of positive samples incorrectly predicted as negative by the model; the expression of the mean square error is: Among them, MSE represents mean square error, N represents the number of samples, and y i represents the true value of the i-th sample, Represents the predicted value of the i-th sample.

7. The algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence according to claim 6 is characterized in that: The mean square error expression is The expression is: Where w represents the normal vector of the hyperplane, x i represents the feature vector of the i-th sample, and b represents the intercept of the hyperplane; the mean square error optimization is specifically: optimizing the model parameters w and b to minimize the mean square error.

8. The algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence according to claim 1 is characterized in that: The anomaly detection and notification mechanism includes: performing problem diagnosis and analysis on the collected data, locating abnormal data and classifying problems; the expression of the anomaly detection algorithm is: Among them, Z i Represents data point x i The number of standard deviations from the population mean μ, x i represents the original data value, μ represents the mean of the original data, and σ represents the standard deviation of the original data; If | Z i |>3, the data point is considered abnormal. The real-time warning is specifically to send the alarm information to the operation and maintenance personnel through API or internal notification system; the problem classification is specifically to classify the fault according to the severity and type of the problem.

9. The algorithm library for automatically allocating solutions to distribution network planning problems based on artificial intelligence according to claim 1 is characterized in that: The continuous learning and feedback mechanism includes: online learning and feedback tuning. The online learning is specifically: updating the model in real time and incorporating new data; the feedback tuning is specifically: using Q-learning or deep reinforcement learning methods to adjust the distribution network design strategy according to real-time feedback. The reinforcement learning model formula is: Among them, Q(s, a) represents the estimated value of performing action a in state s, r represents the immediate reward, γ represents the discount factor, α represents the learning rate, s represents the current state, a represents the action in state s, s′ represents the new state after taking action a immediately, and a′ represents the optimal action.

Images