Visual power distribution network investment strategy optimization method combined with machine learning

Abstract

The invention belongs to the technical field of power distribution, and particularly discloses a visual power distribution network investment strategy optimization method combined with machine learning, comprising the following steps: S1, establishing an overall investment strategy evaluation system; s2, obtaining investment data of a power grid company over the years, and performing data preprocessing; s3, performing power supply partition on each region according to load density, training historical investment data of different power supply partitions by using a TSO-XGBoost algorithm, and generating a black box of investment planning; s4, analyzing the generated black box by using an SHAP algorithm to obtain a visual investment contribution degree proportion; and S5, combining the obtained weight ratio with the investment model to obtain an optimal investment strategy meeting the investment demand. According to the visual power distribution network investment strategy optimization method combined with machine learning, the obtained black box is analyzed by using the SHAP algorithm, and the proportion of each visual investment contribution degree is obtained. Therefore, clear guidance can be provided when a power grid company needs accurate investment.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution, and in particular to a method for optimizing the visualization distribution network investment strategy combined with machine learning. Background Art

[0002] In recent years, in order to adapt to the rapid development of the economic society, Guangdong Power Grid has invested a large amount of funds to transform and upgrade urban and rural power grids, effectively supporting the rapid growth of electricity demand and the continuous improvement of power supply reliability and power quality. In the case of the continuous high level of investment scale, grass-roots units have different degrees of problems such as emphasizing investment and neglecting returns, emphasizing project establishment and neglecting management. How to establish a precise investment control system, optimize resource allocation, further improve the efficiency and effectiveness of investment, and promote the high-quality development of the company has become an urgent problem to be solved.

[0003] In order to improve the level of asset operation management, keep in line with international standards, and achieve the goal of comprehensive optimization of asset risk, efficiency, and cost, taking Guangdong Power Grid Corporation as an example, an asset management strategy of "investment is worth it, assets are fully utilized, risks are monitored, performance is improved, and management pursues excellence" is proposed, and the "Special Promotion Plan for Guangdong Power Grid Corporation to Comprehensively Promote the Whole Life Cycle Management of Assets" is issued, drawing on advanced asset management experience at home and abroad to achieve a qualitative leap in the company's asset management level. As the most front-end link of distribution network construction, the distribution network planning work should strengthen the scientific nature of its planning, fully reflect the role of planning in leading construction, and while meeting various technical dimension restrictive indicators, implement the concept of whole life cycle management of assets into the distribution network planning work to ensure the economy of the selection of planning projects and construction plans, and help the company achieve investment goals.

[0004] At present, the distribution network planning idea of the company is changing from problem orientation to reliability and target grid orientation. In the process of forming and optimizing planning projects, it is not only facing the current situation problems of the power grid, but turning to the more macroscopic logic of overall grid construction with reliability as the starting point, which puts forward higher requirements for the overall cost consideration of precise investment and project investment. At present, Guangdong Power Grid Corporation has put forward certain standardized ideas for the optimization of distribution network planning and construction projects: generally, projects are qualitatively classified into categories one to five according to the investment strategy first, and then the projects are optimized in combination with factors such as the improvement amplitude of indicators before and after project implementation and the weights of key technical indicators. However, when determining the index weights, the influence of subjectivity cannot be completely avoided, and the planning stages and situations in different regions are different, so this scheme cannot make an investment strategy adapted to local conditions in combination with local historical data.

[0005] Therefore, in view of the above problems, there is an urgent need for a visualization distribution network investment strategy planning method combined with the machine learning XGBoost (Extreme Gradient Boosting) algorithm.

[0006] Current approximate implementation solutions:

[0007] 1. Determine the index weights based on the Analytic Hierarchy Process (AHP)

[0008] The AHP method is applicable to problems involving multi-level indicators, complex dimensions, and many indicators that are difficult to quantify. This method is a decision-making method that constructs a reasonable multi-level analysis model based on a profound analysis of the internal relationships of multi-objective or multi-scheme problems, and uses mathematical formulas to calculate the weight values of all qualitative indicators at each level to solve complex decision-making problems with multiple objectives. Use AHP to determine the impact weights of investment projects on investment benefits, thereby forming investment plans.

[0009] 2. Determine the index weights based on the neural network model

[0010] This model can automatically identify and extract the non-linear relationships between complex investments and returns by learning a large amount of historical investment data, making the weight allocation more accurate. The neural network has strong adaptability and can process high-dimensional and multi-type data, so it is applicable to various decision-making scenarios. Using this method to analyze investment plans has good generalization ability. It still maintains high efficiency when facing large-scale investment data, thereby improving the flexibility and reliability of determining investment weights.

[0011] Disadvantages of the prior art:

[0012] 1) The weight assignment of each indicator in the Analytic Hierarchy Process model is too subjective, and the associations and characteristics between data are not fully considered. For example, the comparison of the relative importance between different indicators. This may lead to the results being affected by subjective views, so there may be certain uncertainties and biases. This method lacks objectivity and cannot reflect the "historical preferences" of investment strategies in different regions.

[0013] 2) The "black box" characteristic of the neural network model makes the decision-making process of the model not transparent enough, and it is difficult to explain the specific sources and meanings of the weights, thereby reducing the interpretability of the decision-making process. In addition, the overfitting problem may affect the generalization ability of the model, resulting in unsatisfactory performance on new data. Finally, the solution of the neural network model may be a "local optimal solution" rather than a global optimal solution, so it has limitations.

[0014] 3) The above several solutions generally use a unified prediction model to plan investment strategies for each region, and cannot reflect the investment differences. Summary of the Invention

[0015] The objective of the present invention is to provide a visualization-based distribution network investment strategy optimization method combined with machine learning. After training historical investment data using the TSO-XGBoost algorithm to obtain the "black box" of the investment strategy model, the SHAP (SHapley Additive exPlanations) algorithm is then used to obtain the visualized index weights, and an objective investment strategy combined with historical preferences is obtained by integrating the whole life cycle theory.

[0016] To achieve the above objective, the present invention provides a visualization-based distribution network investment strategy optimization method combined with machine learning, including the following steps:

[0017] S1. Establish an overall investment strategy evaluation system;

[0018] S2. Obtain the investment data of the power grid company over the years and perform data preprocessing;

[0019] S3. Divide each region into power supply zones according to the load density, and use the TSO-XGBoost algorithm to train the historical investment data of different power supply zones to generate the black box of the investment plan;

[0020] S4. Use the SHAP algorithm to analyze the generated black box to obtain the visualized proportion of investment contribution;

[0021] S5. Combine the obtained weight proportion with the investment model to obtain the optimal investment strategy that meets the investment requirements.

[0022] Preferably, in S1, key investment projects and investment returns are extracted from the investment habits, development prospects, and investment requirements of the power grid company over the years, and an index system for the investment evaluation model is established based on the power supply capacity, power quality, power supply reliability, and economic benefits of the power grid.

[0023] Preferably, in S2, the data preprocessing is specifically as follows:

[0024] According to the investment habits of the power grid company over the years, investment projects with the same or similar investment purposes are grouped into the same category;

[0025] The investment projects are classified into eleven categories according to the project purpose, namely, new outgoing lines of substations to meet the power supply of new loads, replacement of old and damaged equipment or lines, solution of safety hazards in medium and low voltage lines, improvement of medium voltage grids, solution of transformer overload, solution of transformer heavy load, solution of low voltage problems in distribution transformers, solution of medium voltage line overload, solution of medium voltage line heavy load, distribution automation projects, and new construction of distribution transformers to meet load requirements.

[0026] Preferably, in S3, regions with similar power supply situations are grouped into one category, and according to the load density differences of each region, different regions are divided into six categories: A+, A, B, C, D, and E according to the load density.

[0027] Preferably, the specific process of S3 is as follows:

[0028] Apply the machine learning XGBoost and use the tuna school optimization algorithm TSO to perform hyperparameter optimization on the model. The tuna school optimization algorithm TSO is used to simulate two foraging behaviors of the tuna school, namely spiral foraging and parabolic foraging. In the tuna school optimization algorithm TSO, each tuna individual represents a potential solution, and the entire group searches for the best solution through collaborative cooperation. The TSO tuna school optimization algorithm includes three stages: the search stage, the pursuit stage, and the attack stage;

[0029] Search stage: Tuna individuals perform random searches based on their own perception abilities and group information to explore the possible solution space;

[0030] Pursuit stage: When an individual discovers a better solution, other individuals follow the best individual to pursue and gradually approach the optimal solution;

[0031] Attack stage: When an individual discovers the best solution, the group will concentrate on attacking and finally obtain the optimal solution;

[0032] Use the gradient boosting tree algorithm to train multiple weak learners. Each weak learner is used to correct the error of the previous weak learner, and these weak learners are combined to form a powerful model;

[0033] For a given dataset with n samples and m features, which is the historical investment dataset of investment projects Use an additive model with the number K to predict the investment benefit output:

[0034]

[0035] Among them, f k represents a decision tree, K is the number of decision trees, x i is the investment input, and y is the model prediction output;

[0036] The objective function is the loss function plus the model complexity, and the expression is:

[0037]

[0038] Among them, the part before the plus sign is the loss function, and the part after the plus sign is the model complexity function;

[0039] When training the kth tree, the first k - 1 trees are all known, then the objective function at this time is rewritten as:

[0040]

[0041] Because the first k - 1 terms are known, then there is:

[0042]

[0043] Among them, y i is the true value, is the predicted value of the first k - 1 trees, and f k (x i ) is the predicted value of the kth tree, and Ω(f k ) is the complexity of the kth tree;

[0044] Thus, k trees are found to minimize the objective function;

[0045] For the h kinds of investment benefits in the sample data, h times of training are carried out to obtain h black boxes. The input data for each training is as follows:

[0046]

[0047] Among them, S h represents the set of the hth kind of investment return.

[0048] Preferably, the specific process of S4 is as follows:

[0049] In the XGBoost prediction model, for the jth feature X ij in the ith sample, the predicted investment benefit of the prediction model for this sample is y h , and the predicted benefit baseline of the whole model is y base , then the SHAP value follows the following equation:

[0050]

[0051] Among them, f h (X ij ) is the SHAP value of X ij , that is, the contribution value of the jth investment project in the ith sample to the predicted value y h of the h investment benefits; when f(X ij ) > 0, it means that this project has a positive effect on this investment benefit; otherwise, it means that it has a negative effect;

[0052] Conduct quantitative analysis on the SHAP values of each project:

[0053]

[0054] Among them, n is the number of samples, G(X ij ) is the investment amount of the jth project in the ith sample, and p hj is the unit average contribution degree of the jth type of investment project to the hth type of investment benefit.

[0055] Preferably, in S5, when making investment plans, the power grid company determines the investment requirements for each investment benefit based on the regulatory policies, economic status, and development directions of different investment regions, and calculates the investment amounts for each investment project through Equation (8):

[0056]

[0057] Therefore, the present invention adopts the above-mentioned visualization power distribution network investment strategy optimization method combined with machine learning, and the beneficial effects are as follows:

[0058] (1) The present invention establishes an investment evaluation system, and uses the TSO-XGBoost machine learning algorithm to train historical data to obtain a global optimal solution that meets historical preferences.

[0059] (2) The present invention uses the SHAP algorithm to analyze the obtained "black box" to obtain the visualized proportion of each investment contribution, so as to provide clear guidance when the power grid company needs precise investment.

[0060] (3) Due to differences in development stages, geographical attributes, load characteristics, and regional policies in each region, the impacts of investment strategies on each benefit also vary. The present invention proposes a method for regional division to customize unique investment planning methods for different regions.

[0061] Next, through the accompanying drawings and embodiments, the technical solutions of the present invention will be further described in detail. Description of the Drawings

[0062] Figure 1 is the overall flowchart of an embodiment of the visualization power distribution network investment strategy optimization method combined with machine learning of the present invention;

[0063] Figure 2 is the schematic diagram for constructing the investment strategy evaluation system of an embodiment of the visualization power distribution network investment strategy optimization method combined with machine learning of the present invention;

[0064] Figure 3 is the schematic diagram of the composition of the investment strategy evaluation system of an embodiment of the visualization power distribution network investment strategy optimization method combined with machine learning of the present invention;

[0065] Figure 4 is the schematic diagram of the TSO-XGBoost algorithm principle and hyperparameter optimization process of an embodiment of the visualization power distribution network investment strategy optimization method combined with machine learning of the present invention. Detailed Embodiment

[0066] The following further illustrates the technical solutions of the present invention through the accompanying drawings and embodiments.

[0067] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meanings as understood by those of ordinary skill in the field to which the present invention pertains.

[0068] A "black box" refers to a model based on data statistics and experience summary, usually established by analyzing and fitting existing data. This model generally does not consider the physical mechanisms or processes inside the system, but directly uses the correlations between data for prediction or simulation. The interpretability of the black box model is poor, and it is difficult to explain the physical or logical mechanisms behind the model. Therefore, in some cases, there may be limitations in understanding the system behavior and making predictions.

[0069] The present invention provides a method for optimizing the visualization distribution network investment strategy combined with machine learning, which is used to guide the distribution network investment strategy. First, in combination with the actual investment needs of the power grid company, an investment strategy evaluation model for evaluating investment returns is established. The investment benefits such as the improvement of the reliability benefits of the distribution network, the increased power supply benefits, and the low-voltage benefits are used as the objective function of the investment strategy. The historical investment data of the power grid company is obtained, and the development needs are predicted based on past experiences, geographical attributes, load characteristics, and future plans of the region, and the region is divided into partitions based on this. The historical data of different partitions is trained by the machine learning TSO-XGBoost algorithm to obtain the "black box" that can calculate the investment returns for each region. At the same time, the SHAP algorithm is used to visually present this "black box" to clarify the contribution ratio of various investments to different investment benefits. Finally, the optimal investment solution is obtained from the objective function, and the obtained optimal solution is used for the investment planning scheme.

[0070] As Figure 1 shown, a method for optimizing the visualization distribution network investment strategy combined with machine learning includes the following steps:

[0071] S1. Establish an overall investment strategy evaluation system, and extract key investment projects and investment returns from the power grid company's past investment habits, development prospects, and investment needs, as Figure 2 shown. Figure 2 In the left column in n are "investment projects": X1, X2, X3,... X n , and in the right column are "investment benefits": S1, S2, S3,... S n , which means that each investment project (from X1 to X n ) corresponds to an investment benefit (from S1 to S

[0072] This step makes a quantitative evaluation of the subsequent optimization effect and lays a foundation for the subsequent analysis and modeling. Figure 3 shown.

[0073] The power supply capacity includes the capacity-to-load ratio, the proportion of the expandable main transformer capacity, the line load rate ratio, and the main transformer heavy load ratio;

[0074] The power quality includes the voltage deviation index, the voltage fluctuation index, and the harmonic distortion index;

[0075] The power supply reliability includes the average power outage frequency of users, the average power outage time of users, the availability factor of high-voltage equipment, and the return on investment;

[0076] The economic benefits include the increased power supply per unit investment, the loss reduction per unit investment, the power supply per unit investment per unit distribution transformer capacity reduction, and the operation and maintenance cost per unit distribution transformer capacity.

[0077] S2. Obtain the investment data of the power grid company over the years and perform data preprocessing to provide reliable feature data input for the prediction model. The data preprocessing is specifically as follows:

[0078] According to the investment habits of the power grid company over the years, classify the investment projects with the same or similar investment purposes into the same category to reduce the input of feature quantities and simplify the calculation process;

[0079] Classify the investment projects into eleven categories according to the project purposes: "New outgoing lines of substations meet the power supply of new loads", "Replace old and damaged equipment or lines", "Solve the safety hazard problems existing in medium and low voltage lines", "Improve the medium voltage grid", "Solve the overloading of distribution transformers", "Solve the heavy load of distribution transformers", "Solve the problem of low voltage in the substation area", "Solve the overloading of medium voltage lines", "Solve the heavy load of medium voltage lines", "Distribution automation project", and "New substations meet the load demand".

[0080] S3. Divide each region into power supply zones according to the load density, and use the TSO-XGBoost algorithm to train the historical investment data of different power supply zones to generate the black box of the investment plan.

[0081] Since machine learning requires a large amount of data for training, and the problem faced is that if the investment plan is carried out for a certain region alone, the data volume is not enough to train an accurate investment model. Therefore, in this embodiment, combined with the concept of power supply zones, the regions with similar power supply situations are classified into one category to expand the data volume, so as to obtain a more accurate investment strategy model suitable for such regions.

[0082] The classification of domestic power supply regions can be considered from factors such as functional positioning, economic development level, electricity consumption level, load density, and user importance. Among them, the load density can best reflect the electricity consumption level and development needs of a region. Therefore, in this embodiment, according to the load density differences of each region, different regions are divided into six categories: A+, A, B, C, D, and E. The specific classification method is shown in Table 1, and the investment focuses of different zones are also different.

[0083] Table 1 Classification Diagram of Domestic Power Supply Areas

[0084]

[0085] In Table 1, σ is the load density of the power supply area type (unit: MW / km 2 ); the area of the power supply type area is generally not less than 5 km 2 ; when calculating the load density, the load of 110(66)kV dedicated lines, as well as the areas without effective power supply such as mountains, gobi, deserts, waters, and forests, should be deducted.

[0086] In this embodiment, it is necessary to use machine learning XGBoost and the tuna school optimization algorithm TSO to perform hyperparameter optimization on the model. The basic principle is as Figure 4 shown. First, analyze the historical investment data of different existing power supply areas, assign the optimal hyperparameter combination obtained by the tuna school optimization algorithm TSO to the XGBoost prediction model, perform error analysis on the prediction results, and use SHAP to perform interpretability analysis on the prediction model.

[0087] 1. The tuna school optimization algorithm TSO is mainly used to simulate two foraging behaviors of the tuna school, namely spiral foraging and parabolic foraging. The tuna school optimization algorithm has the characteristics of strong optimization ability, fast convergence speed, and few parameters to be adjusted. Therefore, it is proposed to use the tuna school optimization algorithm TSO to perform hyperparameter optimization on the XGBoost model. The tuna school optimization algorithm TSO is an optimization algorithm based on swarm intelligence.

[0088] In the tuna school optimization algorithm TSO, each tuna individual represents a potential solution, and the entire group cooperates to find the best solution through cooperation. The TSO tuna school optimization algorithm includes three stages: the search stage, the pursuit stage, and the attack stage;

[0089] Search stage: Tuna individuals perform random searches based on their own perception abilities and group information to explore the possible solution space;

[0090] Pursuit stage: When an individual discovers a better solution, other individuals follow the best individual to pursue and gradually approach the optimal solution;

[0091] Attack stage: When an individual discovers the best solution, the group will concentrate on attacking and finally obtain the optimal solution;

[0092] 2. Gradient boosting tree is an ensemble learning method that uses the gradient boosting tree algorithm to iteratively train multiple weak learners (decision trees). Each weak learner is used to correct the error of the previous weak learner. Finally, these weak learners are combined to form a powerful model.

[0093] For a given historical investment dataset of n samples and m features, which are investment projects Use an additive model with K trees to predict the investment benefit output:

[0094]

[0095] where f k represents a decision tree, K is the number of decision trees, x i is the investment input, and y is the model prediction output;

[0096] 3. XGBoost controls the model complexity by adding regularization terms to prevent overfitting. It uses L1 and L2 regularization and can also perform subsampling and column sampling to improve the model's generalization ability.

[0097] 4. XGBoost supports multiple loss functions, such as squared loss in regression problems and binary logistic loss in binary classification problems. By optimizing the loss function, XGBoost can better fit the data.

[0098] 5. The objective function is the loss function plus the model complexity, and the expression is:

[0099]

[0100] where the term before the plus sign is the loss function, and the term after the plus sign is the model complexity function;

[0101] When training the k-th tree, the first k - 1 trees are known, and the objective function (minimization) is rewritten as:

[0102]

[0103] Since the first k - 1 terms are known, we have:

[0104]

[0105] where y i is the true value, is the prediction of the first k - 1 trees, f k (x i ) is the prediction of the k-th tree, and Ω(f k ) is the complexity of the k-th tree.

[0106] Thus, find k trees to minimize the objective function.

[0107] For the h types of investment benefits in the sample data, perform h times of training to obtain h black boxes. The input data for each training is:

[0108]

[0109] Among them, S h represents the set of the h-th type of investment returns.

[0110] S4. Use the SHAP algorithm to analyze the generated black box, and obtain the visualized proportion of investment contribution.

[0111] In this embodiment, in order to obtain an interpretable optimal investment strategy model, SHAP (SHapley Additive exPlanations) is used to analyze the generated black box, so as to obtain the accurate contribution of different investment projects to each investment benefit, and provide guidance for the precise investment needs of the power grid company. The specific process is as follows:

[0112] In the XGBoost prediction model, for the j-th feature (i.e., the investment project) X in the i-th sample ij , the predicted investment benefit of the prediction model for this sample is y h , and the predicted benefit baseline of the entire model is y base , then the SHAP value follows the following equation:

[0113]

[0114] Among them, f h (X ij ) is the SHAP value of X ij , that is, the contribution value of the j-th investment project in the i-th sample to the predicted value y h of the h-th investment benefit; when f(X ij )>0, it indicates that this project has a positive effect on this investment benefit; otherwise, it indicates a negative effect.

[0115] The obtained SHAP value can reflect the influence effect of this investment project on the investment benefit. In order to further obtain a more accurate contribution degree, quantitative analysis is carried out on the SHAP values of each project:

[0116]

[0117] Among them, n is the number of samples, G(X ij ) is the investment amount of the j-th project in the i-th sample, and p hj is the unit average contribution degree of the j-th type of investment project to the h-th type of investment benefit.

[0118] S5. Combine the obtained weight ratio with the investment model to obtain the optimal investment strategy that meets the investment needs.

[0119] When making investment plans, power grid companies can determine the investment requirements for each investment benefit based on factors such as regulatory policies, economic status, and development directions in different investment regions. Among them, the investment amount of each investment project is calculated through Equation (8):

[0120]

[0121] By using the method presented in this embodiment, an accurate and interpretable investment plan can be provided for investors, thus achieving the purpose of investment planning.

[0122] Therefore, the present invention adopts the above-mentioned method for optimizing the visualization power distribution network investment strategy combined with machine learning. By combining SHAP with XGBoost, the prediction results of the XGBoost model in the investment benefit prediction task can be explained. SHAP can tell us how much each feature contributes to the prediction result of a certain sample, thereby helping investors and researchers understand the prediction process of the model and providing a relatively accurate investment planning guidance plan.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A visualization distribution network investment strategy optimization method combined with machine learning, characterized in that, It includes the following steps: S1. Establish an overall investment strategy evaluation system; S2. Obtain the investment data of the power grid company over the years and perform data preprocessing; S3. Divide each region into power supply zones according to the load density, and use the TSO-XGBoost algorithm to train the historical investment data of different power supply zones to generate a black box for investment planning; S4. Use the SHAP algorithm to analyze the generated black box to obtain the visualized proportion of investment contribution; S5. Combine the obtained weight ratio with the investment model to obtain the optimal investment strategy that meets the investment needs.

2. The visualization distribution network investment strategy optimization method combined with machine learning according to claim 1, characterized in that In S1, extract key investment projects and investment returns from the investment habits, development prospects and investment needs of the power grid company over the years, and establish an index system for the investment evaluation model according to the power supply capacity, power quality, power supply reliability and economic benefits of the power grid.

3. The visualization distribution network investment strategy optimization method combined with machine learning according to claim 2, wherein In S2, the data preprocessing is specifically as follows: According to the investment habits of the power grid company over the years, classify investment projects with the same or similar investment purposes into the same category; Classify investment projects into eleven categories according to the project purposes, namely, new outgoing lines of substations to meet the power supply of new loads, replacement of old and damaged equipment or lines, solution of safety hazards existing in medium and low voltage lines, improvement of medium voltage grids, solution of transformer overload, solution of transformer heavy load, solution of low voltage problems in distribution transformers, solution of medium voltage line overload, solution of medium voltage line heavy load, distribution automation projects, and new construction of distribution transformers to meet load requirements.

4. A visualization power distribution network investment strategy optimization method combined with machine learning according to claim 3, characterized in that In S3, classify regions with similar power supply situations into one category, and divide different regions into six categories of A+, A, B, C, D, and E according to the load density according to the load density differences of each region.

5. The visualization power distribution network investment strategy optimization method combined with machine learning according to claim 4, characterized in that The specific process of S3 is as follows: Apply the machine learning XGBoost and use the tuna school optimization algorithm TSO to perform hyperparameter optimization on the model. The tuna school optimization algorithm TSO is used to simulate two foraging behaviors of the tuna school, namely spiral foraging and parabolic foraging. In the tuna school optimization algorithm TSO, each tuna individual represents a potential solution, and the entire group searches for the best solution through cooperation. The TSO tuna school optimization algorithm includes three stages: the search stage, the pursuit stage, and the attack stage; Search stage: Tuna individuals perform random searches according to their own perception abilities and group information to explore the possible solution space; Pursuit stage: When an individual discovers a better solution, other individuals follow the best individual to pursue and gradually approach the optimal solution; Attack stage: When an individual discovers the best solution, the group will concentrate on attacking and finally obtain the optimal solution; Use the gradient boosting tree algorithm to train multiple weak learners, and each weak learner is used to correct the error of the previous weak learner. Combine these weak learners to form a powerful model; For a given historical investment dataset of investment projects with n samples and m features Use an additive model with K components to predict the output of investment benefits: Among them, f k represents a decision tree, K is the number of decision trees, x i is the investment input, and y is the model prediction output; The objective function is the loss function plus the model complexity, and the expression is: Among them, the part before the plus sign is the loss function, and the part after the plus sign is the model complexity function; When training the kth tree, the first k - 1 trees are all known, then the objective function at this time is rewritten as: Because the first k - 1 terms are known, then: where y i is the true value, is the predicted value of the first k - 1 trees, and f k (x i ) is the predicted value of the k-th tree, and Ω(f k ) is the complexity of the k-th tree; Find the kth tree from this to minimize the objective function; For the h kinds of investment benefits in the sample data, perform h times of training to obtain h black boxes. The input data for each training is: Among them, S h represents the set of the h-th type of investment returns.

6. The visualization method for optimizing the investment strategy of a distribution network combining machine learning according to claim 5, characterized in that, The specific process of S4 is as follows: In the XGBoost prediction model, for the j-th feature X in the i-th sample ij , the predicted investment benefit of the prediction model for this sample is y h , and the predicted benefit baseline of the entire model is y base , then the SHAP value follows the following equation: Among them, f h (X ij ) is the SHAP value of X ij , that is, the contribution value of the j-th investment project in the i-th sample to the predicted value y h of the h investment benefit; when f(X ij ) > 0, it indicates that the project has a positive effect on the investment benefit; otherwise, it indicates a negative effect; Conduct quantitative analysis on the SHAP values of each project: where n is the number of samples, G(X ij ) is the investment amount of the j-th item in the i-th sample, and p hj is the unit average contribution degree of the j-th type of investment project to the h-th type of investment benefit.

7. A visualization distribution network investment strategy optimization method combined with machine learning according to claim 6, characterized in that, In S5, when conducting investment planning, the power grid company determines the investment requirements for each investment benefit based on the regulatory policies, economic status, and development directions of different investment zones, and calculates the investment amount of each investment project through Equation (8):

Images