Load transfer scheme intelligent recommendation method and device based on machine learning, and medium

Abstract

The invention discloses a load transfer scheme intelligent recommendation method and device based on machine learning and a medium, and the method comprises the steps: S01, obtaining feature parameters of a plurality of different load transfer schemes, and constructing a multi-dimensional feature set; step S02, calculating a performance score of the load transfer scheme to be evaluated; step S03, after every two load transfer schemes are combined, splicing the feature vectors to form a combined feature vector; step S04, determining a classification label of the classification model according to the size relationship of the performance scores of every two load transfer schemes; s05, training the classification model to obtain a transfer scheme classification model; and S06, obtaining various load transfer schemes to be recommended, inputting the various load transfer schemes into the trained classification model, sorting the schemes, and generating a recommendation result. The method has the advantages of being simple in implementation method, low in cost, high in intelligent degree, high in response speed and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of smart grids, and particularly to an intelligent recommendation method, device, and medium for load transfer schemes based on machine learning. Background Art

[0002] A load transfer scheme is a technical solution in a power system to transfer the load of one power user to another. By adjusting the load distribution, it realizes the balance and optimization of the power system. As a part of power dispatching, distribution network dispatching needs to perform accurate load forecasting and formulate load transfer schemes to ensure the stable operation of the power system. Therefore, formulating an efficient and reliable load transfer scheme is crucial for ensuring the safe and stable operation of the power grid.

[0003] Currently, the structure of the new power system is becoming more and more complex, and the uncertainty and dynamics of the system are further increasing. Therefore, the load transfer arrangement or path is not pre-fixed or completely determined during actual operation. Instead, the system needs to dynamically select and adjust the path of power transfer from one power supply line or device to another according to factors such as real-time demand, network status, and equipment conditions. Thus, an indeterminate load transfer scheme is formed. For the construction of an indeterminate load transfer scheme, there are mainly the following methods in the prior art: The first is to optimize the load transfer path based on the topological structure, so as to identify standby power sources and possible load transfer paths in the power grid to reduce network losses and improve the system recovery speed; the second is a multi-stage dynamic reconstruction model based on a snowflake-shaped distribution network, which constructs a flexible load transfer scheme with the goal of minimizing the load balance degree and active power loss; the third is an auxiliary decision-making method based on an expert system, which evaluates the economy and risk of the load transfer scheme by analyzing the number of switch operations, inrush current, and system security. However, the construction of the above-mentioned indeterminate load transfer scheme usually relies on empirical rules or static models, which not only have high computational complexity and low efficiency, but also are not flexible and accurate enough when dealing with large-scale and dynamically changing power system problems, and are difficult to meet the requirements of the new power system. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: aiming at the technical problems existing in the prior art, the present invention provides an intelligent recommendation method, device, and medium for load transfer schemes based on machine learning with a simple implementation method, low cost, high intelligence level, and fast response speed, which can accurately model and predict the dynamic changes of the power system, thereby improving the efficiency and accuracy of the load transfer scheme, and optimizing and enhancing the power grid resource allocation efficiency and corresponding speed.

[0005] To solve the above technical problems, the technical solution proposed by the present invention is: An intelligent recommendation method for load transfer schemes based on machine learning, the steps include: Step S01. Obtain the characteristic parameters of multiple different load transfer schemes, and construct a multi-dimensional feature set of the load transfer schemes. The characteristic parameters include line transfer index parameters and the change amount of the line transfer index parameters before and after transfer. Step S02. Calculate the performance scores of the load transfer schemes to be evaluated according to the characteristic parameters of each load transfer scheme in the multi-dimensional feature set of the load transfer schemes. The load transfer schemes include individual load transfer schemes and combined schemes, and the combined schemes are formed by combining two or more individual load transfer schemes. Step S03. Construct the feature vectors of each load transfer scheme according to the multi-dimensional feature set of the load transfer schemes, combine every two load transfer schemes to be evaluated and compared to form a combined sample, splice the feature vectors of the combined sample to form a combined feature vector, and form a feature database from all the combined feature vectors. Step S04. Determine the classification labels of all features in the feature database according to the magnitude relationship of the performance scores of every two load transfer schemes, and form a label database, where the difference greater than zero is set as the preferred class to train the model to distinguish the relative advantages and disadvantages between the schemes. Step S05. Construct a training database from the feature database and the label database, use the training database to train a classification model, and obtain a transfer scheme classification model after training. Step S06. Obtain multiple load transfer schemes to be recommended, input them into the trained transfer scheme classification model, and compare and rearrange all the schemes pairwise according to the classification results of the transfer scheme classification model.

[0006] Further, the line transfer index parameters include any one or more of line current value, number of important users on the line, number of dual-power users on the line, line load rate, and main transformer load rate on the line.

[0007] Further, in step S02, if the load transfer scheme to be evaluated is an individual load transfer scheme, calculating the performance score of the load transfer scheme to be evaluated includes: calculating the safety performance score using the post-transfer line load rate and main transformer load rate of the local line and the opposite line, and the main transformer load rate of the local line and the opposite line, calculating the reliability index score using any one or more of the number of important users, number of dual-power users, historical power outage duration, number of power outages, and number of affected dual-power users, calculating the cost index score using line loss rate, main transformer load balance, and line load balance, and comprehensively obtaining the performance score of the corresponding load transfer scheme by combining the safety performance score, reliability index score, and cost index score.

[0008] Further, in step S02, if the load transfer scheme to be evaluated is a combined scheme, calculating the performance score of the load transfer scheme to be evaluated includes: Check whether the load transfer scheme to be evaluated contains a sub-scheme that only involves isolation operations. If it exists, eliminate the sub-scheme involving isolation operations to form a processed combined scheme; Calculate the safety performance score, the reliability index score, and the cost index score of each sub-scheme in the processed combined scheme respectively. According to the scoring results of each sub-scheme, finally determine the safety performance score, the reliability index score, and the cost index score of the load transfer scheme to be evaluated.

[0009] Further, if the load transfer scheme to be evaluated is a single load transfer scheme, the calculation expression of the safety performance score is:

[0010] Wherein, represents the safety performance score, and are the normalized scores of the load rates of the local line and the main transformer respectively, and are the normalized scores of the load rates of the opposite line and the main transformer respectively; The calculation expression of the reliability index score is:

[0011] Wherein, represents the reliability index score, , , , , represent the normalized scores of the number of important users, the number of dual-power users, the historical power outage duration, the number of power outages, and the number of affected dual-power users respectively; The calculation expression of the cost index score is:

[0012] Wherein, is the cost index score, is the normalized score of the line loss rate, is the normalized score of the main transformer load balance, is the normalized score of the line load balance.

[0013] If the load transfer scheme to be evaluated is a combined scheme, the calculation expression of the safety performance score is:

[0014]

[0015] Among them, represents the safety performance score of the combination plan, is the sub - plan S i the information entropy weight in terms of security, represents the sub - plan S i 's safety performance score, represents the sub - plan S i 's information entropy, n represents the number of sub - plans in the combination plan; The calculation expression of the reliability index score is:

[0016]

[0017] Among them, represents the reliability index score of the combination plan, represents the sub - plan S i 's reliability index score, is the fuzzy adjustment coefficient used to control the balance between the maximum value and the mean value, is the sub - plan S i 's reliability membership degree, is the minimum value of the reliability index score, is the maximum value of the reliability index score; The calculation expression of the cost index score is:

[0018] Among them, represents the cost index score of the combination plan, represents the sub - plan S i 's cost index score, γ represents the preset coefficient.

[0019] Furthermore, in step S04, according to the magnitude relationship of the performance scores of every two load transfer plans, determine the classification labels of all features in the feature database, including: Calculate the comprehensive score difference between every two load transfer plans. The comprehensive score difference is the comprehensive score of the first load transfer plan minus the comprehensive score of the second load transfer plan among the two load transfer plans. The comprehensive score is calculated based on the safety performance score, reliability index score, and cost index score; If the scoring difference is greater than zero, the classification label is set to the first category for marking that the first load transfer scheme is superior to the second load transfer scheme; otherwise, the classification label is set to the second category for marking that the second load transfer scheme is superior to or equivalent to the first load transfer scheme.

[0020] Further, in step S05, a random forest classification model with an optimized construction structure based on the ascidian swarm algorithm is adopted to train a load transfer scheme classification model. The training process of the load transfer scheme classification model includes the following steps: Step S501. Initialize the random forest classification model to construct the basic framework of the load transfer scheme classification model. Step S502. Construct an ascidian population. Each individual in the ascidian population corresponds to a structural initialization scheme of a random forest. The structural initialization scheme defines the training sample subset and feature subset used by each sub-decision tree during the training process to initialize the construction structure of each decision tree in the forest. Step S503. Execute the ascidian swarm structure optimization strategy to iteratively update the position of each individual. In each iteration, the position of the individual is updated according to the classification performance of the random forest model constructed based on the corresponding structural scheme on the validation set. Step S504. When the set maximum number of iterations is reached, or when the change in the fitness function value is less than the convergence threshold for several consecutive rounds, terminate the iteration and output the currently fitness-optimal structural initialization scheme. Step S505. Apply the optimal structural initialization scheme obtained in step S504 to construct a random forest classification model and use the complete training set to complete the model training, finally obtaining a load transfer scheme classification model that meets the preset accuracy requirements.

[0021] Further, step S06 includes: Step S601. Use the scheme to be recommended as the current scheme set and randomly select a scheme from the current scheme set as the reference scheme. Step S602. Select a scheme other than the reference scheme from the current scheme set as the current comparison scheme. Step S603. Input the current comparison scheme and the reference scheme into the trained classification model to obtain the performance classification result between the current comparison scheme and the reference scheme. Select the next scheme other than the reference scheme from the current scheme set as the current comparison scheme, and return to step S602 until all schemes in the current scheme set have been compared with the reference scheme. Step S604. Divide the schemes with performance superior to the reference scheme in the current classification result into one group and the remaining schemes into another group. Use the two divided groups of schemes as the current scheme set respectively, and return to step S602 until all schemes in the scheme to be recommended have been compared to obtain the sorting result of the scheme to be recommended. Step S605. Output an intelligent recommendation result according to the sorting result of the to-be-recommended solution.

[0022] Step S606. Determine whether the current recommendation result is consistent with the actual working condition. If not, construct a feedback sample according to the actual superiority and inferiority relationship between the to-be-recommended solutions and add it to the training sample set to update the training sample set.

[0023] An intelligent recommendation device for load transfer solutions based on machine learning, including a processor and a memory. The memory is used to store a computer program, and the processor is used to execute the computer program to execute the method as described above.

[0024] A computer-readable storage medium storing a computer program, where the computer program, when executed by a processor, implements the above method.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows: By establishing a multi-dimensional feature set of load transfer solutions, calculating the performance scores of single solutions and combined solutions respectively based on the feature parameters of different load transfer solutions, combining each pair of load transfer solutions to form combined samples and splicing the feature vectors, a training database is formed to train a classifier model. At the same time, during the training process, classification labels are determined according to the performance scores of each solution, so that the classifier model can realize the superiority and inferiority comparison classification between single or combined load transfer solutions. When intelligent recommendation of load transfer solutions is required, multiple to-be-recommended load transfer solutions are classified using the trained classifier model, which can realize the intelligent recommendation of each load transfer solution, greatly improving the response speed and intelligence level of load transfer solutions, effectively realizing the optimization and improvement of the power grid resource allocation efficiency, and solving the problems of untimely response and unreasonable resource allocation in the current power grid dispatching process. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic flow chart of the implementation of the intelligent recommendation method for load transfer solutions based on machine learning in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] The present invention will be further described below in conjunction with the specification drawings and specific preferred embodiments, but the protection scope of the present invention is not limited thereby.

[0028] The present invention first establishes a multi-dimensional feature set of load transfer schemes through feature engineering, calculates the performance scores of single schemes and combined schemes respectively based on the feature parameters of different schemes, combines the load transfer schemes to be compared and evaluated in pairs to form training samples and splices the feature vectors to form a training database for training a classifier model. At the same time, classification labels are determined according to the performance scores of each scheme during the training process, so that the classifier model can realize the comparison and classification of the advantages and disadvantages between single or combined load transfer schemes. When intelligent recommendation of load transfer schemes is required, multiple load transfer schemes (alternative schemes) to be recommended are classified using the trained classifier model to sort each load transfer scheme, and intelligent recommendation of each load transfer scheme can be realized, greatly improving the response speed and intelligence level of the load transfer scheme, effectively realizing the optimization and improvement of the power grid resource allocation efficiency, and solving the problems of untimely response and unreasonable resource allocation in the current power grid dispatching process.

[0029] As Figure 1 shown, the steps of the intelligent recommendation method for load transfer schemes based on machine learning in this embodiment include: Step S01. Obtain the feature parameters of multiple different load transfer schemes, and construct a multi-dimensional feature set of load transfer schemes. The feature parameters include line transfer index parameters and the change amounts of line transfer index parameters before and after transfer.

[0030] In this embodiment, through feature engineering on the candidate load transfer scheme data, a multi-dimensional feature set is established as the input of the model, and then an accurate comprehensive score of the candidate scheme is realized. The line transfer index parameters include line current value, number of important users on the line, number of double-power users on the line, line load rate, main transformer load rate of the line, etc. By collecting the above line transfer index parameters and the change amounts before and after transfer, the characteristics of different load transfer schemes can be characterized comprehensively and accurately.

[0031] Specifically, data such as current values, load rates, and the number of important users before and after power transfer can be obtained, and their change amounts can be calculated. For example, subtracting the current values, load rates, etc. before and after power transfer and combining them to form the change characteristics of current and load rate before and after power transfer, thus forming a multi-dimensional feature set. As shown in Table 1, each feature in the multi-dimensional feature set of the constructed load transfer plan includes the index after power transfer (expected transferred load), the line current values before and after power transfer (including the change in the current value of the local line and the change in the current value of the opposite side line), the number of important users on the line before and after power transfer (including the change in the number of important users on the local line and the change amount of the number of important users on the opposite side line), the number of dual-power users on the line before and after power transfer (including the change in the number of dual-power users on the local line and the change in the number of dual-power users on the opposite side line), the line load rate before and after power transfer (including the change in the current line load rate and the change amount of the load rate of the opposite side line), the main transformer load rate of the line before and after power transfer (including the change in the current line main transformer load rate and the change in the main transformer load rate of the opposite side line), and other data. The other data includes the maximum current value of the current line in the most recent week, the maximum current value of the opposite side line in the most recent week, the number of distribution automation switches on the local line, the number of interconnection points on the local line, etc., which can be constructed to form a comprehensive and representative feature set.

[0032] Table 1: Feature Set of Load Transfer Plan

[0033] In a specific application embodiment, a feature set containing 22 features can be constructed for each candidate plan according to Table 1. The model uses the above features to comprehensively score each candidate plan, reflecting the comprehensive performance of the plan in terms of safety, reliability, economy, and business impact.

[0034] As an alternative implementation, the latest load transfer plan data can be obtained from a local database or a remote interface. The original data is in JSON format. To facilitate subsequent analysis and processing, the original JSON data can be converted into a data structure available within the system. The conversion process includes parsing the contents of the JSON data and then mapping them to a data model predefined by the system to unify the data format and ensure the accuracy and efficiency of subsequent processing.

[0035] As an alternative implementation, when obtaining feature data, it also includes performing preprocessing to obtain high-quality feature data. After completing the data conversion, a safety check is performed on each load transfer plan to ensure that the plan will not cause grid safety problems in actual applications. For example, during the safety check process, the following key parameters can be checked: Line load rate: Evaluate whether the load rate of the line exceeds the safety threshold after the implementation of the load transfer plan. An excessively high line load rate may lead to line overload, increase the risk of faults, and affect the safe operation of the power grid.

[0036] Main transformer load rate: Check the load change of the main transformer after the implementation of the plan to ensure that the load rate of the main transformer remains within the safe operating range. An excessively high load rate may accelerate equipment aging and even cause equipment failures.

[0037] Equipment operation years: Consider the operation years of equipment such as lines and main transformers, and evaluate their reliability and potential aging risks. Equipment with a longer operation year may require additional maintenance or early replacement to ensure safety.

[0038] Metering point integrity: Verify the information of the metering point and confirm whether there is a situation of "missing metering point". The data accuracy and integrity of the metering point are crucial for power metering, settlement, and user charging. Missing or inaccurate metering points may lead to economic losses and data analysis deviations.

[0039] By comprehensively checking the above key parameters, a power transfer plan with potential safety hazards can be identified. For example, if a certain plan causes the load rate of the line or main transformer to exceed the set safety threshold, it can be determined that the plan does not meet the safety requirements and needs to be adjusted or abandoned. Similarly, if it is identified that the equipment has been in operation for too long and there is an obvious aging risk, a prompt message can be output to prompt that the equipment needs to be repaired or replaced before implementing the plan.

[0040] Step S02. According to the characteristic parameters of each load transfer plan in the characteristic set of the load transfer plan, calculate the performance score of the load transfer plan to be evaluated. The load transfer plan includes a single load transfer plan and a combined plan, and the combined plan is formed by combining two or more single load transfer plans.

[0041] In this embodiment, the performance score of each load transfer plan is calculated by using the characteristic parameters to obtain the model label, so as to effectively quantify and measure the effectiveness of each transfer plan.

[0042] For the performance evaluation of a separate load transfer scheme, this embodiment considers three dimensions: safety, reliability, and cost. By using characteristic parameters, the safety, reliability, and cost indicators of different load transfer schemes are calculated to form a comprehensive score for load transfer. As an alternative implementation, calculating the performance score of each load transfer scheme includes: calculating the safety performance score using the post-transfer line load ratios of the local line and the opposite-side line, the main transformer load ratios, and the main transformer load ratios of the local line and the opposite-side line. That is, using the line load ratio, main transformer load ratio, etc. as safety indicators, and evaluating the safety performance of the transfer scheme using the line load ratios and main transformer load ratios before and after the transfer of the local and opposite-side lines; using reliability indicators such as the number of important users, the number of dual-power users, the historical power outage duration, the number of power outages, and the number of affected dual-power users for scoring. That is, using the number of important users, the number of dual-power users, the historical power outage duration, the number of power outages, and the number of affected dual-power users as reliability indicators, and evaluating the reliability of the transfer scheme based on the reliability indicators; using the line loss rate, main transformer load balance, and line load balance to calculate the cost indicator score. That is, taking the line loss rate, operating cost, etc. as cost indicators to evaluate the economy of the transfer scheme. Finally, the safety performance score, reliability indicator score, and cost indicator score are combined to obtain the performance score corresponding to the load transfer scheme.

[0043] Specifically, if the load transfer scheme to be evaluated is a separate load transfer scheme, in the safety indicator scoring, assume that the post-transfer line load ratios of the local side and the opposite side are respectively l 1 , l 2 and the main transformer load ratios of the local side and the opposite side are respectively l 3 , l 4 . The increase in the local load ratio and main transformer load ratio indicates an improvement in equipment utilization, so the score should be increased accordingly. Relatively, the increase in the opposite-side load ratio and main transformer load ratio may exacerbate the burden on the opposite-side system, so the score should be decreased accordingly. The safety performance score can be specifically obtained using the following calculation expression: (1) Wherein, represents the safety performance score, and are the normalized scores of the local line load ratio and main transformer load ratio respectively, indicating an improvement in equipment utilization, so they are directly included in the score. and are the normalized scores of the opposite-side line load ratio and main transformer load ratio respectively. Since the higher their values indicate an increased burden on the opposite-side system, they both need to be negated.

[0044] In the reliability index scoring, since power outage events have a negative impact on power supply reliability, the number of important users k 1 , the number of dual-power users k 2 , the historical power outage duration k 3 , the number of power outages k 4 and the number of affected dual-power users k 5 An increase in the index value will lead to a decrease in the scoring result. The reliability index scoring can be obtained according to the following calculation expression: (2) Among them, represents the reliability index scoring, , , , , respectively represent the normalized scores of the number of important users, the number of dual-power users, the historical power outage duration, the number of power outages, and the number of affected dual-power users. All reliability indexes need to be reversed to ensure that the higher the value, the better the performance.

[0045] In the cost index scoring, an increase in the line loss rate means a decrease in energy efficiency. Therefore, the worse the economy of the scheme, the lower the corresponding score. The line loss rate e can be used as a cost index for economic evaluation. The calculation expression of the cost index scoring can be expressed as: The calculation expression of the cost index scoring is: (3) Among them, represents the cost index scoring, is the normalized score of the line loss rate, the normalized score of the main transformer load balance, and the normalized score of the line load balance. The higher the line loss rate, the worse the economy, so it also needs to be reversed.

[0046] Furthermore, the above indexes are normalized using the maximum and minimum values, and the positive and negative indexes are processed separately to convert different indexes into a unified scoring system to ensure that all indexes are on the same evaluation scale. Among them, for the indexes positively correlated with the scoring, the formula for calculating the normalized score is: (4) Among them, S i represents the normalized score of the index i in a certain scoring, Xi is the original value of the corresponding index for the current solution, X min and X max are the minimum and maximum values of the index in the solution respectively.

[0047] For negative indexes, that is, the higher the index value, the worse the solution performance. The normalized score needs to be reversed to achieve the goal that the higher the index value, the lower the score. The reverse score calculation formula is 1 - S i .

[0048] The score of the final power transfer solution is the sum of the safety, reliability, and economy scores, that is: (5) Furthermore, other adjustment measures can also be considered during the scoring process. For example, if the power transfer solution causes important users to transfer out from this side, then 100 points (configurable) will be subtracted from the safety, reliability, and economy scores respectively.

[0049] As an alternative implementation method, for the performance scoring of a combined solution formed by two or more load transfer solutions, that is, the load transfer solution to be evaluated is a combined solution, the following steps can be adopted: Step S201. Check whether the load transfer solution to be evaluated contains sub-solutions that only involve isolation operations. If so, form a processed combined solution after removing the sub-solutions involving isolation operations; Step S202. If it is determined that there is a solution that is only an isolation operation in the load transfer solution to be evaluated, then form a processed combined solution after deleting the sub-solutions involving isolation operations; Step S203. Calculate the safety performance score, reliability index score, and the cost index score of each sub-solution in the processed combined solution respectively. Finally, determine the safety performance score, reliability index score, and cost index score of the load transfer solution to be evaluated according to the scoring results of each sub-solution; A combined solution usually consists of multiple sub-solutions, and some of the sub-solutions may only involve isolation operations (for example, the switch is only disconnected without closing operations). In the performance evaluation process of this embodiment, by excluding the sub-solutions that only involve isolation operations, it is possible to avoid the interference of the sub-solutions that only involve isolation operations on the overall score. For the evaluation of a combined solution, first screen out the sub-solutions that only involve isolation operations. After excluding the isolation operation sub-solutions, integrate the indicators of the remaining sub-solutions to construct the scoring object of the combined solution. By comparing the scoring indicators of all sub-solutions, determine the final score of the combined solution. For example, adopt the maximum impact principle, entropy weight dynamic adjustment, and non-linear normalization to calculate the final score of the combined solution.

[0050] Specifically, after excluding the isolation operation sub-scheme, the safety performance score of the combined scheme can be determined by applying the maximum impact principle, the reliability index score can be determined by entropy weight dynamic adjustment, and the cost index score can be determined by non-linear normalization, based on the safety performance score, reliability index score, and cost index of each load transfer scheme in the combined scheme; finally, the performance score of the combined scheme is obtained by comprehensively determining the safety performance score, reliability index score, and cost index score of the combined scheme.

[0051] As an alternative implementation, the calculation expressions for the corresponding index scores of the combined scheme are as follows: The calculation expression for the safety performance score can be: (6) (7) Wherein, represents the safety performance score of the combined scheme, is the information entropy weight of sub-scheme S i in terms of safety, represents the safety performance score of sub-scheme S i , represents the information entropy of sub-scheme S i , n represents the number of sub-schemes in the combined scheme.

[0052] The calculation expression for the reliability index score is: (8) (9) Wherein, represents the reliability index score of the combined scheme, represents the reliability index score of sub-scheme S i , is the fuzzy adjustment coefficient used to control the balance between the maximum value and the mean value, is the reliability membership degree of sub-scheme S i , is the minimum value of the reliability index score, is the maximum value of the reliability index score.

[0053] The calculation expression for the cost index score is: (10) Wherein, represents the cost index score of the combined scheme, Indicates a sub - solution S i The cost index score of, γ represents a preset coefficient. If γ is greater than 1, then more attention is paid to economic balance. Conversely, more attention is paid to the lowest - cost economic solution.

[0054] Finally, the S 安全 , S 可靠 , S 经济 are added together to obtain the total score.

[0055] For example, consider a combined solution consisting of three sub - solutions A, B, and C. Among them, sub - solutions A and B are complete transfer - supply operations, while sub - solution C only involves isolation operations (no load transfer). During the evaluation process of the combined solution, first, all sub - solutions are screened, and sub - solution C that only involves isolation operations is excluded to avoid its interference with the overall score. Subsequently, the scores of the remaining sub - solutions A and B are integrated. For each scoring index of the combined solution, a method combining the maximum - value principle and the information - entropy weighted mean is used for calculation to ensure that the final score of the combined solution can reflect the contribution of the best sub - solution and comprehensively consider the overall impact of all effective sub - solutions. For example, if the safety performance score of sub - solution A is 0.7 and the safety performance score of sub - solution B is 0.8, the safety performance score of the combined solution is calculated as follows: (11) where P A , P B are the weights calculated according to information entropy, which are used to balance the contributions of different sub - solutions in terms of safety to ensure the rationality of the score. Similarly, the reliability index score and economic score of the combined solution are calculated, and finally, the safety, reliability, and economic scores of the combined solution are added together to obtain the total score.

[0056] Step S03. Construct the feature vectors of each load transfer - supply solution according to the multi - dimensional feature set of the load transfer - supply solution, combine every two load transfer - supply solutions to form a combined sample, splice the feature vectors of the combined sample to form a combined feature vector, and form a training database from all combined feature vectors.

[0057] In this embodiment, by combining every two load transfer - supply solutions to form a combined sample, and then splicing the feature vectors of the combined sample to form a combined feature vector, combined samples for model training are generated to further evaluate and compare the advantages and disadvantages of different power transfer - supply solutions. The method of generating training samples by combining two - by - two solutions can enhance the model's learning ability of the relationship between the advantages and disadvantages of the solutions.

[0058] Specifically, assume that the set of characteristics of the power transfer plan obtained from step S1 is B ={ b 1 , b 2 ,…, b n}, n where is the number of power transfer plans; to meet the requirements of model training, each characteristic of the power transfer plan b i ={ z 1 , z 2 ,…, z 22} is reshaped into a feature vector of length 22; during the model construction process, to measure the advantages and disadvantages between plans, a pairwise combination method is used to generate a sample set b i , b j , where i and j represent plan 1 and plan 2 respectively. Through this permutation and combination method, all possible plan pairs can be generated. When processing each pair of plans (denoted as plan 1 and plan 2), a new combined feature vector is constructed, which is formed by concatenating the feature vectors of plan 1 and plan 2. Assume that the feature vector of plan 1 is b i ={ z i1 , z i2 ,…, z i22}, and the feature vector of plan 2 is B j ={ z j1 , z j2 ,…, z j22}. By concatenating these two feature vectors, the length of the final combined feature vector b ij will reach 44. Through this construction method, the differences between plans can be effectively captured, providing rich feature information for the model.

[0059] Step S04. Determine the classification labels of all features in the feature database according to the magnitude relationship of the performance scores of every two load transfer plans, and form a label database.

[0060] In the process of constructing the training database in this embodiment, the scoring difference between the two load transfer schemes in the combination scheme is used as the classification label, that is, the scoring difference is converted into the classification label, so that the model can effectively learn the relationship between the features and the advantages and disadvantages of the schemes. After training, the model can directly realize the discrimination of the performance advantages and disadvantages of different schemes.

[0061] As an alternative implementation, the steps for determining the classification labels of all features in the feature database according to the magnitude relationship of the performance scores of every two load transfer schemes can be as follows: Step S401. Calculate the comprehensive scoring difference between every two load transfer schemes. The comprehensive scoring difference is the comprehensive score of the first load transfer scheme in the two load transfer schemes minus the comprehensive score of the second load transfer scheme. The comprehensive score is calculated based on the safety performance score, the reliability index score, and the cost index score. Step S402. If the comprehensive scoring difference is greater than zero, the classification label is set to the first category, which is used to mark that the first load transfer scheme is superior to the second load transfer scheme; otherwise, the classification label is set to the second category, which is used to mark that the second load transfer scheme is superior to or equivalent to the first load transfer scheme.

[0062] Specifically, when the scoring difference is converted into the classification label during the training process, if the difference is greater than zero, it is set to the preferred category to train the model to discriminate the relative advantages and disadvantages between the schemes. For example, if the comprehensive score of Scheme 1 minus the comprehensive score of Scheme 2 is greater than 0, it is marked as Category 1 (indicating that Scheme 1 is superior to Scheme 2); otherwise, it is marked as Category 0 (indicating that Scheme 2 is superior to Scheme 1). Subsequently, the classifier is trained with the goal of maximizing the information gain or classification accuracy. After training, the model can directly realize the discrimination of the advantages and disadvantages of the schemes.

[0063] Step S05. Construct a training database from the feature database and the label database, and use the training database to train the classification model. After training, a load transfer scheme classification model is obtained. Preferably, in this embodiment, a random forest (RF) classification model based on the ascidian swarm optimization algorithm is adopted to train a load transfer scheme classification model to achieve efficient optimization of the load transfer scheme classification task. This model uses the swarm intelligence optimization strategy of the ascidian swarm optimization algorithm and can dynamically adjust the hyperparameters of the random forest model, thereby significantly improving the classification performance and accuracy of the model.

[0064] As an alternative implementation, the steps for training the random forest classification model based on the ascidian swarm optimization algorithm include: Step S501. Initialize the random forest classification model, including the number of decision trees n estimator, set basic parameters such as the maximum depth `max_depth` and the minimum number of samples in a leaf node `min_samples_split`; initialize the sea squirt population, and set the population size, the maximum number of iterations, and the fitness function.

[0065] Specifically, the default parameters can be set as follows: the number of decision trees (`n_estimators`) = 100, the maximum depth (`max_depth`) = None, and the minimum number of samples in a leaf node (`min_samples_split`) = 2. Initialize the sea squirt population, and set the population size (such as 30 sea squirts) and the maximum number of iterations (such as 100 times). At this stage, the Random Forest (RF) algorithm is used as the basic model, and parameters such as the initial number of decision trees, the maximum depth, and the minimum number of samples in a node are set to construct the initial structural framework of the model.

[0066] Step S502. Construct a sea squirt population. Each individual in the sea squirt population corresponds to a structural initialization scheme of a random forest. In this structural initialization scheme, the subset of training samples and the subset of features used by each sub-decision tree during the training process are defined. That is, by specifying the "training sub-samples" and "feature screening", the construction structure of each decision tree in the forest is initialized.

[0067] In this embodiment, a new "structural initialization scheme" is introduced based on step S501. The core idea is to encode the sample and feature selection process of each tree in the random forest as a vector, so as to have a clear and controllable scheme in the optimization stage, so that the construction process of the entire random forest model is no longer completely random, but can be optimized purposefully with the help of swarm intelligence algorithms, and finally improve the classification performance of the model.

[0068] Specifically, represent the structural initialization scheme of each random forest as a vector X i , which is composed of several sub-vectors spliced together. Each sub-vector corresponds to the construction details of a tree in the forest: use D (j) to represent the training samples selected by the j -th tree. In traditional random forests, each tree obtains a part of the samples through Bootstrap sampling. In this embodiment, the sampling result is represented as a vector D (j) , where each element represents the index of a selected sample. The feature mask vector is formed as , that is, it represents the feature mask of this tree. The length of this vector is d, representing the total number of features. In the vector, when the value at each position is 1, it indicates that the corresponding feature is used in the construction process of the tree, and when it is 0, it means it is not used. In the above way, the feature subset considered during the construction of each tree can be controlled, making the tree construction process more targeted. The form of the entire vector composition is:

[0069] Among them, is the number of trees in the forest, represents vector concatenation.

[0070] Initially, a set of the above structure initialization schemes is generated randomly to form a population of ascidians. In this population, each X i individual represents a candidate forest construction strategy. By explicitly encoding the construction process of each tree in the random forest, these structure initialization schemes can be searched during the optimization phase. Using the ascidian swarm algorithm to traverse different X i values, thus searching for a random forest construction strategy that can achieve the optimal classification performance on the validation set. By adopting this method, the traditional random construction method can be changed, making the model construction process change from "blind randomness" to "purposeful search", thereby improving the generalization ability and classification accuracy of the entire random forest.

[0071] Step S503. Execute the ascidian swarm structure optimization strategy to iteratively update the position of each individual. Among them, in each iteration, the position of the individual is updated according to the classification performance of the random forest model constructed based on the corresponding structure scheme of the individual on the validation set.

[0072] Specifically, in each iteration, based on the encirclement and cooperation behavior mechanism of the ascidian swarm algorithm, a corresponding random forest model is constructed according to the structure scheme described by each individual, and its classification performance is tested on the validation set. Then, the position of the individual is updated according to the classification performance of the random forest model constructed based on the corresponding structure scheme of the individual on the validation set. After multiple iterations, a better structure initialization scheme is continuously evolved. To ensure that the model can effectively distinguish positive and negative classes during the training process, the cross-entropy loss function can be used to measure the deviation between the predicted probability and the actual label. Its mathematical expression is: (12) Among them, N is the number of samples, y i is the true label (0 or 1), p i is the predicted probability of the model for sample i . By minimizing this loss function, the model can effectively adjust its parameters to improve the classification accuracy of the samples.

[0073] In this embodiment, a multi-index weighted fitness function can be used to reflect the overall performance of the model on the validation set. For example: (13) where ω 1 , ω 2 and λ are preset weights, represents the accuracy index, represents the F1 score calculated using precision and recall, and Overfitting Penalty represents the overfitting penalty index.

[0074] Based on the above fitness evaluation, using the core update mechanism of the ascidian swarm algorithm, the initialization schemes of each structure are continuously improved through global search. Specifically, the three individuals with the highest fitness are selected X α 、X β and X δ as guidance, and the update formula for updating the individual position is set as: (14)

[0075] where represents the position of the individual in the i-th iteration, represents the updated position of the individual after the i-th iteration, r 1 and r 2 are random numbers uniformly distributed, is a control factor that decreases with the number of iterations.

[0076] Through the above update mechanism, each ascidian individual continuously moves towards the global optimal direction in the population, so that the corresponding structure initialization scheme gradually evolves to be more suitable for the current data set, thereby improving the classification accuracy of the entire forest model on the validation set.

[0077] Step S504. When the set maximum number of iterations is reached, or when the change in the fitness function value is less than the convergence threshold for several consecutive rounds, terminate the iteration and output the structure initialization scheme with the optimal fitness at present.

[0078] Specifically, when the number of iterations reaches the preset maximum value or the change in the optimal fitness is lower than the threshold for several consecutive rounds, the iteration terminates. At this time, the structure initialization scheme corresponding to the individual with the highest fitness is selected as the final optimization result.

[0079] Step S505. Apply the optimal structure initialization scheme obtained in step S504 to construct a random forest classification model, and use the complete training set to complete the model training, and finally obtain a classifier that meets the accuracy requirements.

[0080] Specifically, use the optimal structure scheme in step S504 S ∗ Construct a complete random forest model, and perform final training on the model on the entire training set until the classification accuracy meets the preset requirements. Through this series of structure optimization and training processes, the obtained random forest classifier can not only overcome the uncertainty brought by random initialization in the initial construction stage, but also show better robustness and classification performance in various complex and high-dimensional feature data environments.

[0081] To verify the performance of the model, this embodiment further selects a test set to compare the advantages and disadvantages of the schemes. Assume that the test scheme set is , including the feature vectors of all test schemes, where g is a test scheme, and n t represents the total number of test schemes; randomly select two schemes from the test set G ( g i , g j ), where i ≠ j , splice the feature vectors of the two selected test schemes to form an input vector with a dimension of (1, 44) and pass it to the trained random forest model; the output of the model is the judgment of the advantages and disadvantages between the two schemes, with a dimension of (1, 1), and evaluate the two test schemes g i , g j according to the positive and negative size of the output difference.

[0082] This embodiment can provide an adequate learning basis for the model by using the diversity and richness of the combined samples, enabling it to better capture the subtle differences between different schemes. Furthermore, by generating a large number of combined samples, the generalization ability of the model can be greatly enhanced, thereby effectively improving the accuracy of evaluating new schemes. At the same time, using the random forest model for training can not only systematically evaluate and compare different power transfer schemes, but also provide more scientific and accurate support for the decision-making process, which helps to generate the required power transfer schemes more efficiently and accurately in the planning and operation of the power system.

[0083] Step S06. Obtain multiple load transfer schemes to be recommended, input them into the trained classification model, and compare and rearrange all the schemes pairwise based on the comparison result of advantages and disadvantages output by the model (the classification result of the transfer scheme classification model), so as to generate the recommended result of the load transfer scheme.

[0084] In this embodiment, multiple load transfer schemes to be recommended are input into the trained classification model. The classification model can respectively distinguish the advantages and disadvantages between the schemes, sort the load transfer schemes according to the judgment result, and finally obtain the sorted list of all the schemes for determining the optimal load transfer scheme, so as to assist the final scheme selection and decision-making. The scheme with the highest score can be preferentially selected for implementation to optimize the operation efficiency and reliability of the power grid. The recommended result determined in the above manner can not only ensure that the selected scheme is the best choice under the current situation, but also improve the transparency and interpretability of the decision-making process, thereby improving the operation efficiency and service quality of the power grid.

[0085] As an optional implementation manner, according to the scoring result, the quicksort method can be used to sort the load transfer schemes. The steps include: Step S601. Take the schemes to be recommended as the current scheme set, and randomly select a scheme from the current scheme set as the reference scheme; Step S602. Select a scheme other than the reference scheme from the current scheme set as the current comparison scheme; Step S603. Input the current comparison scheme and the reference scheme into the trained classification model to obtain the performance classification result between the current comparison scheme and the reference scheme. Select the next scheme other than the reference scheme from the current scheme set as the current comparison scheme, and return to step S502 until all the schemes in the current scheme set are compared with the reference scheme; Step S604. Divide the schemes with performance superior to the reference scheme in the current classification result into one group, and the remaining schemes into another group. Take the two divided groups of schemes as the current scheme set respectively, and return to step S502 until all the schemes to be recommended are compared, and obtain the sorting result of the schemes to be recommended; Step S605. Output the intelligent recommendation result according to the sorting result of the schemes to be recommended; Step S606. Judge whether the current recommendation result is consistent with the actual situation. If not, construct a feedback sample according to the actual advantages and disadvantages relationship between the schemes to be recommended and add it to the training sample set to update the training sample set.

[0086] Specifically, when there is a difference between the actual scheduling result and the model recommendation result, the system will construct feedback samples according to the input order of the plan pair and the actual superiority and inferiority relationship, and append them to the training sample library to participate in subsequent model updates. This feedback update process will continue until the model output is consistent with the actual scheduling result. Taking Plan A and Plan B as an example, if the model classifies the concatenated input [A, B] as label 1 (i.e., Plan A is better than B), but it is found that B is better in actual operation. At this time, the constructed sample input is [A, B], and the label is 0. This sample pair will be used as a set of feedback samples and appended to the training sample library to participate in subsequent model updates. By continuously iteratively adding such feedback samples, the model will gradually learn to conform to the actual situation and output judgment results that meet the actual operation requirements in subsequent classifications. Finally, through continuous sample appending and incremental training, the model will gradually correct the original sorting error until, when the same concatenated feature [A, B] is input, it can accurately output a label of 0, correctly judging that Plan B is better than Plan A, thus realizing the dynamic correction and continuous optimization of the recommendation logic.

[0087] In summary, in this embodiment, the data of historical load transfer plans (including single and combined plans) is first subjected to data cleaning, data verification, and data conversion. Through feature engineering, a typical feature set of load transfer is selected, including 22 types of index features before and after transfer. And safety, reliability, and cost and other scoring indicators are formed for different historical transfer plans as data labels, and a comprehensive score label is formed by adding the three scores. Among them, the combined plan is formed by combining multiple single plans for operation. After excluding the isolation plan, the maximum value of the scoring indicators of all single plans is selected to calculate the combined score. Subsequently, based on the existing load transfer plan data set, with the goal of maximizing information gain, a random forest model is constructed and trained. The input of the model is the concatenation of all features of the paired plans, a total of 44 features, and the output of the model is the comparison result of the superiority and inferiority of the two plans. Finally, for multiple load transfer plans to be recommended, the plans are combined in pairs and input into the trained random forest model to evaluate the superiority and inferiority relationship. Finally, all candidate plans are re-sorted to select the optimal plan, realizing the intelligent recommendation of suitable transfer plans. Through the above steps, it is possible to quickly evaluate and sort a large number of candidate plans, meet the needs of power grid real-time scheduling, and can evaluate and compare multiple feasible load transfer plan sets under different scenarios, and use machine learning algorithms to optimize and sort from the feasible plans to form different strategy recommendation sequences. Compared with the traditional weighted scoring algorithm, it can make the decision-making process more objective and reliable, reduce the interference of human factors, and thus reduce power outage events and improve power supply stability. In summary, compared with the traditional formulation of load transfer plans based on expert rules and mechanism models, the present invention can quickly and accurately formulate load transfer plans, reduce the human error rate, and at the same time has stronger scalability and can better adapt to the increasingly complex load changes of the power system.

[0088] This embodiment further provides an intelligent recommendation device for load transfer schemes based on machine learning, including a processor and a memory. The memory is used to store computer programs, and the processor is used to execute the computer programs to execute the method as described above.

[0089] It can be understood that the above method of this embodiment can be executed by a single device, such as a computer or a server, etc., or can also be applied to a distributed scenario where multiple devices cooperate with each other to complete. In the case of a distributed scenario, one of the multiple devices can only execute one or more steps of the above method of this embodiment, and the multiple devices interact with each other to complete the above method. The processor can be implemented in ways such as a general-purpose CPU, a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, etc., and is used to execute relevant programs to implement the above method of this embodiment. The memory can be implemented in forms such as a read-only memory ROM, a random access memory RAM, a static storage device, and a dynamic storage device, etc. The memory can store an operating system and other application programs. When implementing the above method of this embodiment through software or firmware, the relevant program codes are stored in the memory and are called and executed by the processor.

[0090] This embodiment further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, it implements the method as described above.

[0091] Those skilled in the art should understand that the above embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can be in the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes. The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1 a process or multiple processes and / or blocks Figure 1Apparatus for the functions specified in one or more boxes. These computer program instructions can also be stored in a computer-readable memory capable of guiding a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction apparatus, and the instruction apparatus implements the operations in the process Figure 1 one process or more processes and / or boxes Figure 1 Apparatus for the functions specified in one or more boxes. These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the process Figure 1 one process or more processes and / or boxes Figure 1 Apparatus for the functions specified in one or more boxes.

[0092] The above are only the preferred embodiments of the present invention and do not impose any formal limitations on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Therefore, any simple modifications, equivalent changes, and decorations made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention shall fall within the scope of the protection of the technical solution of the present invention.

Claims

1. A method for intelligently recommending load transfer schemes based on machine learning, characterized in that the steps include: Step S01. Acquire characteristic parameters of a plurality of different load transfer schemes, and construct a multi-dimensional characteristic set of load transfer schemes, wherein the characteristic parameters include line transfer index parameters and the change amount of the line transfer index parameters before and after the transfer; Step S02. Calculate the performance score of the load transfer scheme to be evaluated according to the characteristic parameters of each load transfer scheme in the multidimensional characteristic set of the load transfer scheme, wherein the load transfer scheme includes a single load transfer scheme and a combination scheme, wherein the combination scheme is formed by combining two or more single load transfer schemes; Step S03. Construct a feature vector of each load transfer scheme according to the multidimensional feature set of the load transfer scheme, and combine every two load transfer schemes to be evaluated and compared to form a combined sample, concatenate the feature vectors of the combined samples to form a combined feature vector, and form a feature database with all the combined feature vectors; Step S04. Determine the classification labels of all features in the feature database according to the size relationship between the performance scores of each two load transfer schemes to form a label database; Step S05. A training database is constructed by the feature database and the label database, and a classification model is trained using the training database. After the training is completed, a classification model for the supply transfer scheme is obtained; Step S06. Obtain multiple load transfer schemes to be recommended, input them into the trained transfer scheme classification model, compare and re-rank all schemes in pairs according to the classification results of the transfer scheme classification model, and obtain the load transfer scheme recommendation result.

2. The method for intelligently recommending load transfer schemes based on machine learning according to claim 1, characterized in that: The line transfer index parameters include any one or more of the line current value, the number of important line users, the number of line dual power users, the line load rate, and the line main transformer load rate.

3. The method for intelligently recommending load transfer schemes based on machine learning according to claim 1, characterized in that: In step S02, if the load transfer scheme to be evaluated is a separate load transfer scheme, calculating the performance score of the load transfer scheme to be evaluated includes: using the line load rate after transfer, the main transformer load rate and the main transformer load rate of the line on this side and the line on the opposite side to calculate the safety performance score; using any one or more of the number of important users, the number of dual power users, the historical power outage duration, the number of power outages and the number of dual power users affected to calculate the reliability index score; using the line loss rate, main transformer load balance and line load balance to calculate the cost index score; the safety performance score, reliability index score and cost index score are combined to obtain the performance score of the corresponding load transfer scheme.

4. The method for intelligently recommending load transfer schemes based on machine learning according to claim 3 is characterized in that: In step S02, if the load transfer scheme to be evaluated is a combination scheme, calculating the performance score of the load transfer scheme to be evaluated includes: Check whether the load transfer plan to be evaluated contains a sub-plan involving only isolation operation. If so, remove the sub-plan involving isolation operation to form a processed combined plan; The safety performance score, the reliability index score and the cost index score of each sub-scheme in the processed combined scheme are calculated respectively, and the safety performance score, reliability index score and cost index score of the load transfer scheme to be evaluated are finally determined according to the scoring results of each sub-scheme.

5. The method for intelligently recommending load transfer schemes based on machine learning according to claim 4 is characterized in that: If the load transfer scheme to be evaluated is a separate load transfer scheme, the calculation expression of the safety performance score is: in, represents the safety performance score, and are the normalized scores of the line load rate and the main transformer load rate on this side, and are the normalized scores of the opposite line load factor and the main transformer load factor respectively; The calculation expression of the reliability index score is: in, represents the reliability index score, , , , , They represent the number of important users, the number of dual-power users, the historical power outage duration, the number of power outages, and the normalized scores affecting the number of dual-power users; The calculation expression of the cost index score is: in, Score the cost indicator, is the normalized score of line loss rate, is the normalized score of the main variable load balancing, is the normalized score of line load balancing; If the load transfer scheme to be evaluated is a combination scheme, the calculation expression of the safety performance score is: in, represents the safety performance score of the combined solution, For sub-scheme S i The weight of information entropy on security, Represents a sub-scheme S i Safety performance rating, Represents a sub-scheme S i The information entropy of n Indicates the number of sub-schemes in the combined scheme; The calculation expression of reliability index score is: in, represents the reliability index score of the combined solution, Represents a sub-scheme S i The reliability index score of is the fuzzy adjustment coefficient used to control the balance between the maximum value and the mean value. For sub-scheme S i The reliability membership of is the minimum value of the reliability index score, The maximum value of the reliability index score; The calculation expression of cost indicator score is: in, represents the cost indicator score of the combination solution, Represents a sub-scheme S i The cost indicator score of is , and γ represents the preset coefficient.

6. The method for intelligently recommending a load transfer scheme based on machine learning according to any one of claims 1 to 5, characterized in that: In step S04, the classification labels of all features in the feature database are determined according to the performance score relationship between each two load transfer schemes, including: Calculate the difference in comprehensive scores between every two load transfer schemes, where the difference in comprehensive scores is the comprehensive score of the first load transfer scheme minus the comprehensive score of the second load transfer scheme, and the comprehensive score is calculated based on the safety performance score, the reliability index score, and the cost index score; If the comprehensive score difference is greater than zero, the classification label is set to the first category to mark that the first load transfer plan is better than the second load transfer plan; otherwise, the classification label is set to the second category to mark that the second load transfer plan is better than or equal to the first load transfer plan.

7. The method for intelligently recommending a load transfer scheme based on machine learning according to any one of claims 1 to 5, characterized in that: In step S05, a random forest classification model based on the optimization structure of the sea squirt group algorithm is used to train a transfer scheme classification model. The training process of the transfer scheme classification model includes the following steps: Step S501. Initialize the random forest classification model to construct a basic framework for forming a classification model for the power transfer scheme; Step S502. Construct a sea squirt population, wherein each individual in the sea squirt population corresponds to a structure initialization scheme of a random forest, wherein the structure initialization scheme defines a training sample subset and a feature subset used by each child decision tree in the training process to initialize the construction structure of each decision tree in the forest; Step S503. Execute the ascidian group structure optimization strategy to iteratively update the position of each individual, wherein in each round of iteration, the individual position is updated according to the classification performance of the random forest model constructed based on the individual corresponding structure scheme on the validation set; Step S504. When the maximum number of iterations is reached, or the fitness function value changes for several consecutive rounds are less than the convergence threshold, the iteration is terminated and the structure initialization scheme with the best current fitness is output; Step S505. Apply the optimal structure initialization scheme obtained in step S504 to construct a random forest classification model, and use the complete training set to complete the model training, and finally obtain a power transfer scheme classification model that meets the preset accuracy requirements.

8. The method for intelligently recommending a load transfer scheme based on machine learning according to any one of claims 1 to 5, characterized in that: Step S06 includes: Step S601. The solution to be recommended is used as the current solution set, and a solution is randomly selected from the current solution set as the benchmark solution; Step S602. Select a solution other than the baseline solution from the current solution set as the current comparison solution; Step S603. Input the current comparison scheme and the benchmark scheme into the trained classification model to obtain the performance classification result between the current comparison scheme and the benchmark scheme, select the next scheme other than the benchmark scheme from the current scheme set as the current comparison scheme, and return to step S602 until the comparison between all schemes in the current scheme set and the benchmark scheme is completed; Step S604: divide the solutions with better performance than the benchmark solutions in the current classification results into one group and the remaining solutions into another group. The two groups of solutions are respectively used as the current solution sets, and return to step S502 until all the solutions to be recommended are compared and the ranking results of the solutions to be recommended are obtained; Step S605. Output intelligent recommendation results according to the ranking results of the solutions to be recommended; Step S606: Determine whether the current recommendation result is consistent with the actual working condition. If not, construct a feedback sample based on the actual pros and cons relationship between the recommended solutions and add it to the training sample set to update the training sample set.

9. A device for intelligently recommending load transfer schemes based on machine learning, comprising a processor and a memory, wherein the memory is used to store a computer program, characterized in that: The processor is configured to execute the computer program to perform the method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images