Distribution network robot performance evaluation method, system, medium and equipment

Abstract

The invention discloses a distribution network robot performance evaluation method and system, a medium and equipment. The method comprises the following steps: acquiring historical multi-dimensional operation data of a distribution network robot; calculating each feature and an importance value corresponding to the feature according to the historical multi-dimensional operation data, and screening out a plurality of key factors according to the importance values; according to the plurality of key factors, constructing an original performance evaluation model; optimizing the original performance evaluation model to obtain a final performance evaluation model; and outputting a performance evaluation report according to the final performance evaluation model and the current operation data of the distribution network robot. According to the invention, the comprehensive performance of the distribution network robot in a complex environment is accurately and dynamically evaluated by collecting and processing comprehensive multi-dimensional historical operation data and constructing and optimizing the evaluation model.

Description

Technical Field

[0001] The present invention belongs to the field of power system engineering and relates to a method, system, medium and device for evaluating the performance of distribution network robots. Background Art

[0002] With the continuous improvement of the intelligence and automation levels of power systems, the safety, stability and efficiency of power distribution networks have become the focus of attention of power enterprises. As an important tool for realizing the intelligence of power distribution networks, distribution network robots have been widely used in the fields of power system operation, maintenance and fault handling. These robots have functions such as remote control, autonomous navigation, and intelligent detection, and can perform various tasks in various complex environments, such as line inspection, equipment detection, fault repair and switch operation, etc., thereby effectively reducing the risks of manual operations and improving work efficiency.

[0003] However, most of the existing methods for evaluating the performance of distribution network robots rely on empirical judgment or single evaluation indicators, lacking comprehensive consideration of multi-dimensional factors. Traditional methods often only focus on the performance of robots under ideal working conditions, and fail to fully consider the complex environmental factors in actual applications (such as temperature, humidity, wind speed, precipitation), the operating status of power systems (such as voltage fluctuations, harmonic interference, short-circuit faults), as well as the impacts of mechanical, operation and control factors on the performance of robots. In addition, with the development of artificial intelligence and machine learning technologies, it is difficult for existing evaluation methods to make full use of these advanced technologies to improve the intelligence level of evaluation, resulting in inaccurate or limited evaluation results and unable to meet the growing application requirements. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present application provides a method, system, medium and device for evaluating the performance of distribution network robots, which realizes accurate evaluation of the performance of distribution network robots in complex working environments.

[0005] To achieve the above object, in the first aspect, the present invention provides a method for evaluating the performance of a distribution network robot, including:

[0006] Obtaining historical multi-dimensional operation data of the distribution network robot; wherein, the historical multi-dimensional operation data includes: environmental data, power system data, operation and control data, mechanical data and energy data;

[0007] Calculating each feature and its corresponding importance value according to the historical multi-dimensional operation data, and screening out several key factors according to the importance value; wherein, corresponding weights are assigned to the key factors;

[0008] Constructing an original performance evaluation model according to the several key factors;

[0009] Optimize the original performance evaluation model to obtain the final performance evaluation model;

[0010] Output a performance evaluation report according to the final performance evaluation model and the current operation data of the distribution network robot.

[0011] Compared with the prior art, the embodiments of the present application have the following beneficial effects: By collecting, processing, constructing, and optimizing the evaluation model for comprehensive multi-dimensional historical operation data, the comprehensive performance of the distribution network robot in a complex environment can be accurately and dynamically evaluated, thereby improving the efficiency and reliability of power system maintenance work.

[0012] In some embodiments of the first aspect of the present application, calculating, according to the historical multi-dimensional operation data, an importance value corresponding to each feature, and screening out several key factors according to the importance value, includes:

[0013] According to the preset random forest algorithm, on each decision tree, calculate the contribution of each feature in the historical multi-dimensional operation data to the prediction result when splitting nodes as the importance value of the feature;

[0014] Sort the importance values from high to low, and use the features corresponding to the several importance values ranked in the front as the key factors.

[0015] Compared with the prior art, the above embodiments have the following beneficial effects: Using the random forest algorithm to calculate the importance value of each feature in the historical multi-dimensional operation data can effectively identify the key factors that have a significant impact on the performance of the distribution network robot; By sorting and selecting the features with higher importance as the key factors, when constructing the performance evaluation model later, it is more focused on the factors that really affect the performance, reducing the interference of irrelevant variables and improving the accuracy and robustness of the model.

[0016] In some embodiments of the first aspect of the present application, calculating, according to the historical multi-dimensional operation data, an importance value corresponding to each feature, and screening out several key factors, further includes:

[0017] Construct a Bayesian network according to the preset Bayesian network algorithm and the key factors; where the Bayesian network includes: a set of factor nodes and a set of directed edges between the factors, each factor node corresponds to a key factor, and each directed edge represents the conditional dependence relationship between the key factors;

[0018] Calculate the weight corresponding to each key factor according to the marginal probability distribution of the Bayesian network.

[0019] Compared with the prior art, the above embodiments have the following beneficial effects: By defining the conditional dependence relationship between key factors through a Bayesian network and calculating the weights of key factors based on the marginal probability distribution, this method introduces a method of probability theory to quantify the influence degree of each factor, which can not only describe the complex interaction between key factors more precisely, but also improve the adaptability of the subsequent constructed model to uncertainty.

[0020] In some embodiments of the first aspect of the present application, constructing the original performance evaluation model according to the preset model construction algorithm and the key factors includes:

[0021] Taking the key factors and their corresponding weights as independent variables of a preset weighted linear regression model, and taking the historical true evaluation scores corresponding to the historical multi-dimensional operation data as the dependent variable of the preset weighted linear regression model, and iteratively training to obtain a weighted linear regression model;

[0022] Taking the key factors and weights as input data of a preset original neural network model, and taking the historical true evaluation scores as output data of the original neural network model, and iteratively training to obtain a neural network model;

[0023] Taking the weighted linear regression model and the neural network model as the original performance evaluation model.

[0024] Compared with the prior art, the above embodiments have the following beneficial effects: Applying the key factors and their corresponding weights to the training of the weighted linear regression model and the neural network model respectively to construct the original performance evaluation model; By combining these two different modeling methods and evaluating by combining the outputs of the two models in the subsequent stage, the simplicity and intuitiveness of the linear model and the powerful expression ability of the non-linear model (i.e., neural network) are fully utilized, so that the evaluation model can handle both simple linear relationships and complex non-linear scenarios, providing a strong guarantee for the performance evaluation of the distribution network robot in diverse tasks.

[0025] In some embodiments of the first aspect of the present application, optimizing the original performance evaluation model to obtain the final performance evaluation model includes:

[0026] Optimizing the weighted linear regression model and the neural network model according to a preset stochastic gradient descent algorithm to obtain the final weighted linear regression model and the final neural network model;

[0027] Taking the final weighted linear regression model and the final neural network model as the final performance evaluation model.

[0028] Compared with the prior art, the above embodiments have the following beneficial effects: The random gradient descent algorithm is used to optimize the original performance evaluation model. In this process, the parameters of the weighted linear regression model and the neural network model are dynamically adjusted to minimize the loss function, achieving the maximization of the model performance; by continuously updating the model parameters to adapt to new multi-dimensional operation data, it is ensured that the evaluation model is always in the optimal state, ensuring that even when the working environment of the distribution network robot changes, its performance can be evaluated in a timely and accurate manner, enhancing the flexibility and response speed of the system.

[0029] In some embodiments of the first aspect of the present application, the random gradient descent algorithm is as follows:

[0030] θ = [β0, β1, ···, β j , w0, w1, ···, w j , ∈, W1, W2, b1, b2];

[0031] Among them, β0 represents the constant term of the weighted linear regression model, β1, ···, β j represents the linear regression coefficients of the weighted linear regression model, w j represents the weight of the jth key factor in the weighted linear regression model, ∈ represents the error term of the weighted linear regression model, W1 and W2 represent the weight matrices of the neural network model, b1 and b2 represent the bias vectors of the neural network model, X t and Y t represent the new multi-dimensional operation data and the corresponding true evaluation scores, η is the learning rate, L(Y t , f(X t ; θ t )) is the loss function; f(X t ; θ t ) represents the comprehensive output after weighted adjustment of the outputs of the weighted linear regression model and the neural network model;

[0032] Compared with the prior art, the above embodiments have the following beneficial effects: Integrating the regression coefficients of the linear model, the weights of the key factors, and the parameters of the neural network into a unified parameter system, realizing the collaborative optimization and dynamic weight allocation of the linear model and the non-linear model; forcing the two types of models to share the same objective function when updating through joint gradient calculation, not only retaining the ability of the linear model to capture the global trend, but also using the neural network to fit complex non-linear relationships, and enhancing the robustness against data drift by dynamically adjusting the weights of the key factors, breaking through the limitations of traditional static weights or independent model optimization, enabling the evaluation model to achieve accurate and adaptive performance evaluation in the complex and changeable scenarios of the power system.

[0033] In some embodiments of the first aspect of the present application, the acquisition of historical multi-dimensional operation data of the distribution network robot includes:

[0034] Collect the historical multi-dimensional original operation data of the distribution network robot;

[0035] According to a preset preprocessing algorithm, process the historical multi-dimensional original operation data to obtain historical multi-dimensional operation data; wherein, the preprocessing includes any one or more combinations of the following: outlier removal and data normalization.

[0036] Compared with the prior art, the above embodiments have the following beneficial effects: By preprocessing the collected historical multi-dimensional original operation data, including steps such as outlier removal and data normalization, the quality of the data used for evaluation is ensured, errors caused by data quality problems are avoided, and the credibility of model construction and evaluation is improved.

[0037] In the second aspect, the present invention also provides a performance evaluation system for a distribution network robot, including: a data acquisition module, a key factor screening module, an original model construction module, a model optimization module, and a result output module;

[0038] Among them, the data acquisition module is used to acquire the historical multi-dimensional operation data of the distribution network robot; wherein, the historical multi-dimensional operation data includes: environmental data, power system data, operation and control data, mechanical data, and energy data;

[0039] The key factor screening module is used to calculate each feature and its corresponding importance value according to the historical multi-dimensional operation data, and screen out several key factors according to the importance value; wherein, corresponding weights are assigned to the key factors;

[0040] The original model construction module is used to construct an original performance evaluation model according to the several key factors;

[0041] The model optimization module is used to optimize the original performance evaluation model to obtain a final performance evaluation model;

[0042] The result output module is used to output a performance evaluation report according to the final performance evaluation model and the current operation data of the distribution network robot.

[0043] Compared with the prior art, the embodiments of the present application have the following beneficial effects: By collecting, processing, constructing, and optimizing the evaluation model for comprehensive multi-dimensional historical operation data, the comprehensive performance of the distribution network robot in a complex environment is accurately and dynamically evaluated, thereby improving the efficiency and reliability of power system maintenance work.

[0044] In a third aspect, the present invention further provides a performance evaluation device for a distribution network robot, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the computer program is loaded into the processor, the steps of the performance evaluation method for a distribution network robot are implemented.

[0045] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the performance evaluation method for a distribution network robot are implemented. Description of the Drawings

[0046] Figure 1 : A flowchart of a performance evaluation method for a distribution network robot provided in some embodiments of the present invention.

[0047] Figure 2 : A structural diagram of a performance evaluation system for a distribution network robot provided in some embodiments of the present invention.

[0048] Figure 3 : A structural diagram of a performance evaluation device for a distribution network robot provided in some embodiments of the present invention. Detailed Embodiments

[0049] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0050] Embodiment 1:

[0051] Please refer to Figure 1 , a performance evaluation method for a distribution network robot provided in an embodiment of the present invention, including steps S1 to S5:

[0052] Step S1: Obtain historical multi-dimensional operation data of the distribution network robot.

[0053] Among them, the historical multi-dimensional operation data includes: environmental data, power system data, operation and control data, mechanical data, and energy data; specifically, the environmental data may include: data such as temperature, humidity, wind speed, and precipitation; the power system data may include: data such as voltage fluctuations, harmonic interference, short-circuit fault frequency, and load changes of the distribution network robot; the operation and control data may include: data such as friction force, mechanical wear, and vibration intensity of the distribution network robot; the mechanical data may include: data such as control algorithm accuracy, operation strategy efficiency, communication delay, and communication reliability of the distribution network robot; the energy data may include: data such as battery power, battery health status, and energy management strategy of the distribution network robot.

[0054] Preferably, in some embodiments of the present application, step S1 can be implemented through the following preferred implementation manners, including steps S11 to S12, and the specific steps are as follows:

[0055] S11: Collect the historical multi-dimensional original operation data of the distribution network robot;

[0056] S12: Process the historical multi-dimensional original operation data according to a preset preprocessing algorithm to obtain historical multi-dimensional operation data; among them, the preprocessing includes any one or more combinations of the following: outlier removal and data normalization.

[0057] Further, the outlier processing can be implemented through the median absolute deviation (MAD) algorithm, and the algorithm is as follows: MAD = median(|x i - median(x)|); if |x i - median(x)| > λ·MAD, it is regarded as an outlier, where x represents a single data set in the historical multi-dimensional original operation data, x i represents the i-th data in a single data set, and λ is a constant.

[0058] Further, the data normalization can be implemented through the min-max normalization method, and the algorithm is as follows:

[0059] Where x i ′ represents the normalized value, and min(x) and max(x) represent the minimum and maximum values in the data set x.

[0060] In this embodiment, step S1 preprocesses the collected historical multi-dimensional original operation data, including steps such as outlier removal and data normalization, ensuring the data quality for evaluation, avoiding incorrect conclusions caused by data quality problems, and improving the credibility of model construction and evaluation.

[0061] Step S2: Calculate the importance value corresponding to each feature based on the historical multi-dimensional operation data, and screen out several key factors according to the importance value; wherein, corresponding weights are assigned to the key factors.

[0062] Preferably, in some embodiments of the present application, step S2 can be implemented through the following preferred implementation manners, including steps S21 to S24, and the specific steps are as follows:

[0063] S21: According to the preset random forest algorithm, on each decision tree, calculate the contribution of each feature in the historical multi-dimensional operation data to the prediction result when splitting nodes as the importance value of the feature, and the specific implementation is as follows:

[0064] Define the importance value of feature f j as: where N t is the total number of decision trees, and ΔI t (f j ) is the information gain of feature f j in the t-th tree.

[0065] In addition, in specific embodiments, the random forest algorithm is only one algorithm selection, and other algorithms can also be selected for implementation, such as the gradient boosting tree (GBDT) algorithm, the LASSO regression algorithm or other similar algorithms, which are not limited here.

[0066] S22: Sort the importance values from high to low, and use the features corresponding to the top several importance values as key factors.

[0067] In specific implementation, the top K features can be used as key factors, where K can be determined by cross-validation.

[0068] In steps S21 - S22, using the random forest algorithm to calculate the importance value of each feature in the historical multi-dimensional operation data can effectively identify the key factors that have a significant impact on the performance of the distribution network robot; by sorting and selecting the features with higher importance as key factors, it makes the subsequent construction of the performance evaluation model more focused on the factors that truly affect the performance, reduces the interference of irrelevant variables, and improves the accuracy and robustness of the model.

[0069] S23: Construct a Bayesian network according to the preset Bayesian network algorithm and the key factors. The Bayesian network is defined as G = (V, E), where V is the set of factor nodes, E is the set of directed edges between factors, each factor node corresponds to a key factor, and each directed edge represents the conditional dependence relationship between key factors. The parameters of the Bayesian network can be calculated by maximum likelihood estimation, Bayesian estimation or other methods, which belong to the prior art and will not be elaborated here.

[0070] S24: Calculate the weight w corresponding to each key factor according to the marginal probability distribution of the Bayesian network. j , where the calculation formula can be as follows: w j = P(X j | parents(X j ))), where parents(X j ) represents the set of parent nodes of the key factor X j .

[0071] In steps S23 - S24, the conditional dependence relationship between key factors is defined through the Bayesian network, and the weights of key factors are calculated based on the marginal probability distribution. This method introduces the method of probability theory to quantify the influence degree of each factor, which can not only describe the complex interaction between key factors more precisely, but also improve the adaptability of the subsequent constructed model to uncertainty.

[0072] Step S3: Construct an original performance evaluation model according to the key factors.

[0073] Preferably, in some embodiments of the present application, step S3 can be implemented by the following preferred implementation method, including steps S31 to S32, and the specific steps are as follows:

[0074] S31: Use the key factors and their corresponding weights as the independent variables of a preset weighted linear regression model, and use the historical true evaluation scores corresponding to the historical multi-dimensional operation data as the dependent variable of the preset weighted linear regression model, and iteratively train to obtain a weighted linear regression model.

[0075] In specific implementation, the weighted linear regression model can be as follows:

[0076] where β0 represents the constant term, β j represents the linear regression coefficient, w j represents the weight of the jth key factor, ∈ represents the error term, and Y represents the historical true evaluation score corresponding to the historical multi-dimensional operation data.

[0077] S32: Use the key factors and weights as the input data of a preset original neural network model, and use the historical true evaluation scores as the output data of the original neural network model, and iteratively train to obtain a neural network model.

[0078] Among them, the output of the neural network can be:

[0079] Y = σ(W2·σ(W1·X + b1) + b2); where W1 and W2 represent weight matrices, b1 and b2 represent bias vectors, X represents all key factors, and σ is an activation function.

[0080] After steps S31 and S32, the weighted linear regression model and the neural network model can be used as the original performance evaluation model. When actually used, we can adjust the results according to the specific situation of the current running data. For example, if the environmental factors are relatively simple, only select the result of the regression model as the final result; if the environmental factors are complex, only select the output of the neural network as the final output result, or perform weighted fusion on the two results according to the specific situation.

[0081] In this embodiment, step S3 applies the key factors and their corresponding weights to the training of the weighted linear regression model and the neural network model respectively to construct the original performance evaluation model; by combining these two different modeling methods and evaluating by combining the outputs of the two models in the follow-up, it makes full use of the simplicity and intuitiveness of the linear model and the powerful expression ability of the non-linear model (i.e., the neural network), enabling the evaluation model to handle both simple linear relationships and complex non-linear scenarios, providing a strong guarantee for the performance evaluation of the distribution network robot in diverse tasks.

[0082] Step S4: Optimize the original performance evaluation model to obtain the final performance evaluation model.

[0083] Preferably, in some embodiments of the present application, step S4 can be implemented by the following preferred implementation:

[0084] Optimize the weighted linear regression model and the neural network model according to the preset stochastic gradient descent algorithm to obtain the final weighted linear regression model and the final neural network model;

[0085] Furthermore, the stochastic gradient descent algorithm can be as follows:

[0086] θ = [β0, β1, ···, β j , w0, w1, ···, w j , ∈, W1, W2, b1, b2];

[0087] Among them, β0 represents the constant term of the weighted linear regression model, β1, ···, β j represents the linear regression coefficients of the weighted linear regression model, w j represents the weight of the j-th key factor in the weighted linear regression model, ∈ represents the error term of the weighted linear regression model, W1 and W2 represent the weight matrices of the neural network model, b1 and b2 represent the bias vectors of the neural network model, X t and Y t represent the new multi-dimensional operation data and the corresponding true evaluation scores, η is the learning rate, L(Y t , f(X t ; θ t )) is the loss function; f(X t ; θ t ) represents the comprehensive output after weighted adjustment of the outputs of the weighted linear regression model and the neural network model.

[0088] After the optimization is completed, the final weighted linear regression model and the final neural network model are used as the final performance evaluation model.

[0089] In this embodiment, step S4 uses the stochastic gradient descent algorithm to optimize the original performance evaluation model. This process dynamically adjusts the parameters of the weighted linear regression model and the neural network model, aims to minimize the loss function, and realizes the maximization of the model performance; by continuously updating the model parameters to adapt to the new multi-dimensional operation data, it ensures that the evaluation model is always in the optimal state, and ensures that even when the working environment of the distribution network robot changes, its performance can be evaluated timely and accurately, enhancing the flexibility and response speed of the system.

[0090] In addition, in the specific stochastic gradient descent algorithm of step S4 above, the regression coefficients of the linear model, the weights of the key factors, and the parameters of the neural network are integrated into a unified parameter system, realizing the collaborative optimization and dynamic weight allocation of the linear model and the non-linear model; by forcing the two types of models to share the same objective function during the update through joint gradient calculation, it not only retains the ability of the linear model to capture the global trend, but also uses the neural network to fit complex non-linear relationships, and also enhances the robustness against data drift by dynamically adjusting the weights of the key factors, breaking through the limitations of traditional static weights or independent model optimization, enabling the evaluation model to achieve accurate and adaptive performance evaluation in the complex and changeable scenarios of the power system.

[0091] Step S5: Output a performance evaluation report according to the final performance evaluation model and the current operation data of the distribution network robot.

[0092] Similarly, in actual use, we can adjust the results according to the specific situation of the current operation data. For example, if the environmental factors are relatively simple, only the results of the regression model are selected as the final results; if the environmental factors are complex, only the output of the neural network is selected as the final output result, or the two results are weighted and fused according to the specific situation to output the performance evaluation score and report. The report can include the performance analysis of the distribution network robot in each dimension, the weight distribution of key factors, performance improvement suggestions, and the prediction of future performance trends.

[0093] In this embodiment, step S5 realizes the accurate evaluation of the comprehensive performance of the distribution network robot in different complex environments by fusing the outputs of the two models, improving the reliability of the results.

[0094] In summary, compared with the prior art, the above embodiments of the present application have the following beneficial effects: By collecting, processing, constructing, and optimizing the evaluation model for comprehensive multi-dimensional historical operation data, the accurate and dynamic evaluation of the comprehensive performance of the distribution network robot in complex environments is realized, thereby improving the efficiency and reliability of the power system maintenance work.

[0095] Embodiment 2:

[0096] Please refer to Figure 2 , based on the same inventive concept, a distribution network robot performance evaluation system disclosed in an embodiment of the present invention includes: a data collection module M1, a key factor screening module M2, an original model construction module M3, a model optimization module M4, and a result output module M5;

[0097] Among them, the data collection module M1 is used to obtain the historical multi-dimensional operation data of the distribution network robot; among them, the historical multi-dimensional operation data includes: environmental data, power system data, operation and control data, mechanical data, and energy data.

[0098] The data collection module M1 includes: an original data collection unit and a preprocessing unit;

[0099] Among them, the original data collection unit is used to collect the historical multi-dimensional original operation data of the distribution network robot;

[0100] The preprocessing unit is used to process the historical multi-dimensional original operation data according to a preset preprocessing algorithm to obtain historical multi-dimensional operation data; among them, the preprocessing includes any one or more combinations of the following: outlier removal and data normalization.

[0101] In this embodiment, the data acquisition module M1 preprocesses the collected historical multi-dimensional raw operation data, including steps such as outlier removal and data normalization, ensuring the data quality for evaluation, avoiding incorrect conclusions caused by data quality issues, and improving the credibility of model construction and evaluation.

[0102] The key factor screening module M2 is used to calculate, based on the historical multi-dimensional operation data, the importance value corresponding to each feature, and screen out several key factors according to the importance value; among them, corresponding weights are assigned to the key factors.

[0103] The key factor screening module M2 includes: an importance value calculation unit and a sorting unit;

[0104] Among them, the importance value calculation unit is used to calculate, based on the preset random forest algorithm, on each decision tree, the contribution of each feature in the historical multi-dimensional operation data to the prediction result at the splitting node as the importance value of the feature;

[0105] The sorting unit is used to sort the importance values from high to low, and take the features corresponding to the several importance values ranked in the front as key factors.

[0106] In this embodiment, the key factor screening module M2 uses the random forest algorithm to calculate the importance value of each feature in the historical multi-dimensional operation data, which can effectively identify the key factors that have a significant impact on the performance of the distribution network robot; by sorting and selecting the features with higher importance as key factors, when constructing the performance evaluation model later, it is more focused on the factors that really affect the performance, reducing the interference of irrelevant variables and improving the accuracy and robustness of the model.

[0107] Further, the key factor screening module M2 further includes: a Bayesian network construction unit and a weight calculation unit.

[0108] The Bayesian network construction unit is used to construct a Bayesian network based on the preset Bayesian network algorithm and the key factors; where the Bayesian network includes: a set of factor nodes and a set of directed edges between factors, each factor node corresponds to a key factor, and each directed edge represents the conditional dependence relationship between key factors;

[0109] The weight calculation unit is used to calculate the weight w corresponding to each key factor according to the marginal probability distribution of the Bayesian network j , and the calculation formula is as follows: w j = P(X j | parents(X j )), where, parents(Xj ) represents the key factor X j and the set of its parent nodes.

[0110] In this embodiment, the key factor screening module M2 defines the conditional dependence relationships between key factors through a Bayesian network and calculates the weights of key factors based on the marginal probability distribution. This method introduces the method of probability theory to quantify the influence degree of each factor, which can not only describe the complex interaction between key factors more precisely, but also improve the adaptability of the subsequently constructed model to uncertainty.

[0111] The original model construction module M3 is used to construct an original performance evaluation model according to the several key factors.

[0112] The original model construction module M3 includes: a regression model construction unit and a neural network construction unit;

[0113] Among them, the regression model construction unit is used to take the key factor and its corresponding weight as the independent variables of a preset weighted linear regression model, and take the historical true evaluation score corresponding to the historical multi-dimensional operation data as the dependent variable of the preset weighted linear regression model, and iteratively train to obtain a weighted linear regression model;

[0114] The neural network construction unit is used to take the key factor and weight as the input data of a preset original neural network model, and take the historical true evaluation score as the output data of the original neural network model, and iteratively train to obtain a neural network model;

[0115] The weighted linear regression model and the neural network model are used as the original performance evaluation model.

[0116] In this embodiment, the original model construction module M3 applies the key factor and its corresponding weight to the training of the weighted linear regression model and the neural network model respectively to construct an original performance evaluation model; by combining these two different modeling methods and combining the outputs of the two models for evaluation in the follow-up, it makes full use of the simplicity and intuitiveness of the linear model and the powerful expression ability of the non-linear model (i.e., the neural network), so that the evaluation model can handle both simple linear relationships and complex non-linear scenarios, providing a strong guarantee for the performance evaluation of the distribution network robot in diverse tasks.

[0117] The model optimization module M4 is used to optimize the original performance evaluation model to obtain a final performance evaluation model;

[0118] Furthermore, the model optimization module M4 can be implemented through the following preferred implementation manners:

[0119] Optimize the weighted linear regression model and the neural network model according to the preset stochastic gradient descent algorithm to obtain the final weighted linear regression model and the final neural network model;

[0120] Further, the stochastic gradient descent algorithm is as follows:

[0121] θ = [β0, β1, ···, β j , w0, w1, ···, w j , ∈, W1, W2, b1, b2];

[0122] Among them, β0 represents the constant term of the weighted linear regression model, β1, ···, β j represents the linear regression coefficients of the weighted linear regression model, w j represents the weight of the j-th key factor in the weighted linear regression model, ∈ represents the error term of the weighted linear regression model, W1 and W2 represent the weight matrices of the neural network model, b1 and b2 represent the bias vectors of the neural network model, X t and Y t represent the new multi-dimensional operation data and the corresponding true evaluation scores, η is the learning rate, L(Y t , f(X t ; θ t )) is the loss function; f(X t ; θ t ) represents the comprehensive output after weighted adjustment of the outputs of the weighted linear regression model and the neural network model;

[0123] Use the final weighted linear regression model and the final neural network model as the final performance evaluation model.

[0124] In this implementation, the model optimization module M4 uses the stochastic gradient descent algorithm to optimize the original performance evaluation model. This process dynamically adjusts the parameters of the weighted linear regression model and the neural network model, aiming to minimize the loss function, and realizes the maximization of the model performance; by continuously updating the model parameters to adapt to the new multi-dimensional operation data, it ensures that the evaluation model is always in the optimal state, ensuring that even when the working environment of the distribution network robot changes, its performance can be evaluated in a timely and accurate manner, enhancing the flexibility and response speed of the system.

[0125] The result output module M5 is used to output a performance evaluation report according to the final performance evaluation model and the current operation data of the distribution network robot.

[0126] Preferably, in a specific implementation, the result output module M5 can perform weighted adjustment on the results of the two models according to the preset rules and the specific situation of the current operation data, and output the final evaluation score and the evaluation report.

[0127] In this embodiment, the result output module M5 realizes the accurate evaluation of the comprehensive performance of the distribution network robot in different complex environments by fusing the outputs of the two models, improving the reliability of the results.

[0128] In summary, compared with the prior art, the embodiments of the present application have the following beneficial effects: by collecting, processing, constructing, and optimizing the evaluation model for comprehensive multi-dimensional historical operation data, the accurate and dynamic evaluation of the comprehensive performance of the distribution network robot in complex environments is realized, thereby improving the efficiency and reliability of the power system maintenance work.

[0129] The division of the above-described modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules can be combined or integrated into another system.

[0130] Embodiment 3:

[0131] Figure 3 The structure diagram of a distribution network robot performance evaluation device of the present application is shown. As Figure 3 shown, the distribution network robot performance evaluation device may include: a processor N1, a memory N2, a data interface N3, and a communication bus N4.

[0132] Among them: the processor N1, the memory N2, and the data interface N3 complete mutual communication through the communication bus N4; the data interface N3 is used for data communication with other devices such as input devices or output devices; the processor N1 is used to execute the program N5, and specifically can execute the relevant steps in the above-described embodiment of a distribution network robot performance evaluation method.

[0133] Specifically, the program N5 may include program code, and the program code includes computer-executable instructions.

[0134] The processor N1 may be a central processing unit CPU, or a specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application. One or more processors included in the distribution network robot performance evaluation device may be of the same type of processor, such as one or more CPUs, or may be of different types of processors, such as one or more CPUs and one or more ASICs.

[0135] The memory N2 is used to store the program N5. The memory N2 may include a high-speed RAM memory and may also include non-volatile memory, such as at least one disk memory.

[0136] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Additionally, the embodiments of the present application are not directed to any particular programming language.

[0137] Embodiment 4:

[0138] The embodiments of the present invention further provide a computer-readable storage medium storing at least one executable instruction, which, when running on a distribution network robot performance evaluation device / system, causes the distribution network robot performance evaluation device / system to execute the distribution network robot performance evaluation method in any of the above method embodiments.

[0139] In the specification provided herein, a large number of specific details are set forth. It will be understood, however, that the embodiments of the present application may be practiced without these specific details. Similarly, in order to streamline the present application and assist in understanding one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present application, the various features of the embodiments of the present application are sometimes grouped together in a single embodiment, figure, or description thereof. Among them, the claims following the specific implementation mode are hereby expressly incorporated into the specific implementation mode, where each claim itself serves as a separate embodiment of the present application.

[0140] Those skilled in the art can understand that the modules in the devices in the embodiments can be adaptively changed and disposed in one or more devices different from the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and in addition, they can be divided into multiple sub-modules or sub-units or sub-components. Except that at least some of such features and / or processes or units are mutually exclusive.

Claims

1. A performance evaluation method for distribution network robots, characterized in that Including: Obtaining historical multi-dimensional operation data of a distribution network robot; wherein, the historical multi-dimensional operation data includes: environmental data, power system data, operation and control data, mechanical data, and energy data; Calculating each feature and its corresponding importance value according to the historical multi-dimensional operation data, and screening out several key factors according to the importance value; wherein, corresponding weights are assigned to the key factors; Constructing an original performance evaluation model according to the several key factors; Optimizing the original performance evaluation model to obtain a final performance evaluation model; Outputting a performance evaluation report according to the final performance evaluation model and the current operation data of the distribution network robot.

2. The performance evaluation method of a distribution network robot according to claim 1, wherein The calculating each feature and its corresponding importance value according to the historical multi-dimensional operation data, and screening out several key factors according to the importance value includes: Calculating the contribution of each feature in the historical multi-dimensional operation data to the prediction result when splitting nodes on each decision tree as the importance value of the feature according to a preset random forest algorithm; Sorting the importance values from high to low, and taking the features corresponding to several importance values ranked in the front as key factors.

3. The performance evaluation method of a distribution network robot according to claim 2, characterized in that The calculating each feature and its corresponding importance value according to the historical multi-dimensional operation data, and screening out several key factors further includes: Constructing a Bayesian network according to a preset Bayesian network algorithm and the key factors; wherein the Bayesian network includes: a set of factor nodes and a set of directed edges between the factors, each factor node corresponds to a key factor, and each directed edge represents a conditional dependence relationship between the key factors; Calculating the weight corresponding to each key factor according to the marginal probability distribution of the Bayesian network.

4. The performance evaluation method of a distribution network robot according to claim 3, characterized in that, The constructing an original performance evaluation model according to a preset model construction algorithm and the key factors includes: Taking the key factors and their corresponding weights as independent variables of a preset weighted linear regression model, and taking the historical true evaluation score corresponding to the historical multi-dimensional operation data as the dependent variable of the preset weighted linear regression model, and iteratively training to obtain a weighted linear regression model; Taking the key factors and weights as input data of a preset original neural network model, and taking the historical true evaluation score as output data of the original neural network model, and iteratively training to obtain a neural network model; Taking the weighted linear regression model and the neural network model as the original performance evaluation model.

5. The performance evaluation method of a distribution network robot according to claim 4, characterized in that, The optimizing the original performance evaluation model to obtain a final performance evaluation model includes: Optimizing the weighted linear regression model and the neural network model according to a preset stochastic gradient descent algorithm to obtain a final weighted linear regression model and a final neural network model; Taking the final weighted linear regression model and the final neural network model as the final performance evaluation model.

6. The performance evaluation method of a distribution network robot according to claim 5, wherein, The stochastic gradient descent algorithm is as follows: θ = [β0, β1, ···, β j , w0, w1, ···, w j , ∈, W1, W2, b1, b2]; Among them, β0 represents the constant term of the weighted linear regression model, β1, ···, β j represents the linear regression coefficients of the weighted linear regression model, w j represents the weight of the j-th key factor in the weighted linear regression model, ∈ represents the error term of the weighted linear regression model, W1 and W2 represent the weight matrices of the neural network model, b1 and b2 represent the bias vectors of the neural network model, X t and Y t represent the new multi-dimensional operation data and the corresponding true evaluation scores, η is the learning rate, L(Y t , f(X t ; θ t )) is the loss function; f(X t ; θ t ) represents the comprehensive output after weighted adjustment of the outputs of the weighted linear regression model and the neural network model.

7. A method for evaluating the performance of a distribution network robot according to any one of claims 1 to 6, characterized in that, The obtaining historical multi-dimensional operation data of the distribution network robot includes: Collecting historical multi-dimensional original operation data of the distribution network robot; Process the historical multi-dimensional original operation data according to a preset preprocessing algorithm to obtain historical multi-dimensional operation data; wherein, the preprocessing includes any one or more combinations of the following: outlier removal and data normalization.

8. A performance evaluation system for distribution network robots, characterized in that, It includes: a data acquisition module, a key factor screening module, an original model construction module, a model optimization module, and a result output module; wherein, the data acquisition module is used to obtain the historical multi-dimensional operation data of the distribution network robot; wherein, the historical multi-dimensional operation data includes: environmental data, power system data, operation and control data, mechanical data, and energy data; the key factor screening module is used to calculate each feature and its corresponding importance value according to the historical multi-dimensional operation data, and screen out several key factors according to the importance value; wherein, the key factors are assigned corresponding weights; the original model construction module is used to construct an original performance evaluation model according to the several key factors; the model optimization module is used to optimize the original performance evaluation model to obtain a final performance evaluation model; the result output module is used to output a performance evaluation report according to the final performance evaluation model and the current operation data of the distribution network robot.

9. A performance evaluation device for a distribution network robot, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the computer program is loaded into the processor, it implements the steps of a method for evaluating the performance of a distribution network robot according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of a method for evaluating the performance of a distribution network robot according to any one of claims 1-7.

Images