Method and system for power consumption inspection process optimization based on machine learning

Abstract

The application discloses a power consumption inspection process optimization method and system based on machine learning, and belongs to the technical field of intelligent power operation and maintenance. The method comprises collecting power consumption inspection data, optimizing power consumption inspection data, power consumption inspection work quality evaluation, intelligent inspection scheduling and power consumption inspection process optimization. The application adopts an intelligent method combining a power consumption inspection work quality evaluation module and an intelligent inspection scheduling module, realizes adaptive matching of inspection resources and personnel capabilities, and improves overall inspection efficiency and accuracy. A bidirectional time convolution long short-term parallel model is used for power consumption work quality evaluation, power equipment inspection records are introduced as features, the evaluation results reflect the standardization degree of single inspection, and also reflect the contribution value of inspectors to the long-term safe operation of the power grid. A strategy evaluation enhanced network is used for intelligent inspection scheduling, dynamically balances task execution efficiency and inspection quality, effectively solves common problems in power grid scheduling, and improves the performance and efficiency of intelligent inspection scheduling.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent power operation and maintenance, and specifically refers to a power consumption inspection process optimization method and system based on machine learning. BACKGROUND

[0002] The power consumption inspection process optimization based on machine learning intelligently optimizes each link of the power consumption inspection process through machine learning technology, forms a data-driven intelligent inspection system, aims to reasonably allocate inspection tasks, improve power consumption inspection efficiency and quality, and reduce manual intervention, promotes the development of power operation and maintenance automation, and helps to improve the overall level of power management.

[0003] However, in the process of power consumption inspection process optimization, the power consumption inspection process is closely related to inspection scheduling, and traditional inspection scheduling is usually based on fixed rules or experience, without fully considering the actual work quality of inspectors, resulting in a disconnection between inspection quality and scheduling, high-risk areas may be assigned to low-efficiency personnel, and it is difficult to achieve proactive preventive inspection and affect the scheduling effect. In the process of power consumption inspection work quality evaluation, there are technical problems that manual scoring standards are not unified, resulting in large fluctuations in the scores of the same inspector in different assessment periods, simple weighted average method cannot distinguish the difference between sudden equipment defect handling and daily standard operation, and existing scoring system is difficult to balance real-time operation details and long-term work performance. In the process of intelligent inspection scheduling, there are technical problems that traditional manual scheduling relies on experience, resulting in high missing inspection rate of high-risk equipment, fixed scheduling mode cannot adapt to sudden power equipment failure, and single efficiency index orientation causes uneven inspection quality. SUMMARY

[0004] In view of the above, in order to overcome the defects of the prior art, the present application provides a power inspection process optimization method and system based on machine learning, which is aimed at the technical problem that in the power inspection process optimization process, the power inspection process is closely related to the inspection scheduling, the traditional inspection scheduling is usually based on fixed rules or experience, and the actual work quality of the inspector cannot be fully considered, resulting in disconnection between inspection quality and scheduling, high-risk areas may be allocated to low-efficiency personnel, and it is difficult to realize active preventive inspection and affect the scheduling effect. The present application creatively adopts an intelligent method combining power inspection work quality evaluation and intelligent inspection scheduling module, combines historical data and real-time evaluation results, realizes adaptive matching of inspection resources and personnel capacity, and improves the overall inspection efficiency and accuracy; for the technical problems in the power inspection work quality evaluation process, the artificial scoring standard is not unified, the same inspector has large score fluctuation in different assessment periods, the simple weighted average method cannot distinguish the difference between sudden equipment defect processing and daily normative operation, and the existing scoring system is difficult to balance real-time operation details and long-term work performance. The present application creatively adopts a bidirectional time convolution long-short term parallel model for power work quality evaluation, and innovatively introduces power equipment inspection records as features, so that the evaluation results reflect not only the standardization degree of single inspection, but also the contribution value of the inspector to the long-term safe operation of the power grid; for the technical problems in the intelligent inspection scheduling process, the traditional artificial scheduling relies on experience, resulting in high risk of missing inspection of high-risk equipment, the fixed scheduling mode cannot adapt to sudden power equipment failure, and the single efficiency index orientation causes uneven inspection quality. The present application creatively adopts a strategy evaluation enhanced network for intelligent inspection scheduling, dynamically balances task execution efficiency and inspection quality, effectively solves the common problems in power grid scheduling, and improves the performance and efficiency of intelligent inspection scheduling.

[0005] The technical scheme adopted by the present application is as follows: the power inspection process optimization method based on machine learning provided by the present application comprises the following steps:

[0006] Step S1: collecting power inspection data;

[0007] Step S2: optimizing power inspection data;

[0008] Step S3: power inspection work quality evaluation;

[0009] Step S4: intelligent inspection scheduling;

[0010] Step S5: power inspection process optimization.

[0011] Further, in step S1, the power inspection data is collected, which is used to collect the data required for power inspection process optimization, specifically to collect work quality evaluation related data and inspection scheduling related data, and obtain power inspection data;

[0012] The work quality evaluation related data includes inspector work data, power equipment inspection records and inspector inspection route;

[0013] The inspection scheduling related data includes inspector information, power equipment information, external environment information and task scheduling historical data.

[0014] Further, in step S2, the optimized power consumption inspection data is used to optimize the quality of power consumption inspection data, specifically by optimizing the power consumption inspection data to obtain power consumption inspection optimization data;

[0015] The optimization process includes data cleaning and data standardization;

[0016] The data cleaning includes deduplication, outlier removal and missing value filling;

[0017] The data standardization includes numerical data standardization, type data encoding and timestamp alignment.

[0018] Further, in step S3, the power consumption inspection work quality evaluation is used to evaluate the power consumption inspection work quality of the inspector, specifically by using a bidirectional time convolution long short-term parallel model to evaluate the power consumption work quality, to obtain a power consumption inspection work quality evaluation score;

[0019] The bidirectional time convolution long short-term parallel model includes a data input layer, a parallel feature analysis layer, a multi-level feature fusion layer and an evaluation output layer;

[0020] The parallel feature analysis layer includes a parallel bidirectional time convolution subnetwork and a long short-term memory subnetwork;

[0021] The bidirectional time convolution subnetwork is used to capture short-term trends to extract local features;

[0022] The long short-term memory subnetwork is used to capture long-term dependencies to extract global features;

[0023] The multi-level feature fusion layer is used to fuse local features and global features;

[0024] The power consumption inspection work quality evaluation includes the following steps:

[0025] Step S31: Data input layer construction, specifically by receiving work quality evaluation related data in the power consumption inspection optimization data through the data input layer to construct evaluation input data;

[0026] Step S32: parallel feature analysis layer construction, specifically, by setting a parallel bidirectional time convolution subnetwork and a long short-term memory subnetwork in the parallel feature analysis layer, the evaluation input data is subjected to feature analysis to obtain local short-term features and global long-term features of work quality, including the following steps:

[0027] Step S321: bidirectional time convolution subnetwork construction, specifically, three bidirectional dilated causal convolution layers are set in the bidirectional time convolution subnetwork, the convolution receptive field is expanded by setting different dilation rates in each bidirectional dilated causal convolution layer, and a residual connection layer and a batch normalization layer are set after each bidirectional dilated causal convolution layer to construct the bidirectional time convolution subnetwork, which extracts local features from the evaluation input data to generate local short-term features of work quality;

[0028] The dilation rates of the three bidirectional dilated causal convolution layers are set to 1, 2 and 4, respectively;

[0029] The bidirectional dilated causal convolution layer includes a forward dilated causal convolution layer, a backward dilated causal convolution layer and a feature splicing layer;

[0030] Step S322: long short-term memory subnetwork construction, specifically, input gates, forget gates and output gates are set in the long short-term memory subnetwork to extract global features from the evaluation input data to generate global long-term features of work quality;

[0031] Step S33: multi-level feature fusion layer construction, specifically, the multi-level feature fusion layer is used to splice the local short-term features of work quality and the global long-term features of work quality to generate comprehensive time sequence features of work quality;

[0032] Step S34: evaluation output layer construction, used for outputting model evaluation results, specifically, a dense layer is set in the evaluation output layer to evaluate the comprehensive time sequence features of work quality to obtain model evaluation results;

[0033] Step S35: bidirectional time convolution long short-term parallel model training, specifically, the data input layer construction, the parallel feature analysis layer construction, the multi-level feature fusion layer construction and the evaluation output layer construction are used to construct a bidirectional time convolution long short-term parallel model, and the model is trained to obtain an electricity inspection work quality evaluation model;

[0034] Step S36: electricity inspection work quality evaluation information generation, specifically, the electricity inspection work quality evaluation model is used to evaluate the electricity inspection work quality to obtain an electricity inspection work quality evaluation score.

[0035] Further, in step S4, the intelligent inspection scheduling is performed, specifically, the intelligent inspection scheduling is performed by using a policy evaluation enhanced network according to the power inspection work quality evaluation score and the power inspection optimization data, and a best inspection scheduling strategy is obtained, including the following steps.

[0036] Step S41: inspection scheduling modeling, specifically, the inspection scheduling is modeled as a Markov decision process, including the following steps.

[0037] Step S411: state space setting, specifically, the inspector information, the power equipment information and the external environment information are spliced into a state vector;

[0038] Step S412: action space setting, specifically, the inspection task is set as the action space, and the inspection task allocation is set as the action;

[0039] Step S413: reward function setting, used for measuring the pros and cons of the inspection scheduling strategy, specifically, the reward function is set according to the task execution efficiency and the power inspection work quality evaluation score, and the calculation formula is:

[0040] ;

[0041] In the formula, R z is the reward value of the zth time step, is the first reward weight, z is the second time step index, Eff z is the task execution efficiency reward of the zth time step, is the second reward weight, Qua z is the power inspection work quality evaluation score reward of the zth time step;

[0042] Step S42: policy evaluation enhanced network improvement, specifically, a policy network and an evaluation network are set in the policy evaluation network, and residual convolution layers are introduced into the policy network and the evaluation network respectively, used for relieving model overfitting, and the policy evaluation enhanced network is improved to obtain the policy evaluation enhanced network;

[0043] Step S43: intelligent inspection scheduling model construction, specifically, the policy evaluation enhanced network is parameter updated through model training, and the policy network and the evaluation network are optimized to obtain the intelligent inspection scheduling model, including the following steps.

[0044] Step S431: randomly initializing the policy network parameters and the evaluation network parameters;

[0045] Step S432: introducing KL divergence into the objective function, used for constraining the policy updating range, and the calculation formula is:

[0046] ;

[0047] In the formula, L ac is a target function, is an expectation calculation on a state distribution generated by a current inspection scheduling strategy, is a state distribution, is a discount factor, s z is a state at a zth time step, s z+1 is a state at a z+1th time step, is a state value function, used to calculate an average reward expectation in a state, is a KL divergence weight, is a KL divergence, is an old inspection scheduling strategy, is a new inspection scheduling strategy;

[0048] Step S433: global network parameter updating is performed by policy network gradient updating and evaluation network gradient updating;

[0049] Step S44: best inspection scheduling strategy generation, specifically, intelligent inspection scheduling is performed by an intelligent inspection scheduling model to obtain a best inspection scheduling strategy.

[0050] Further, in step S5, the power inspection process optimization, specifically, power inspection process optimization is performed by the power inspection work quality evaluation and the intelligent inspection scheduling, to generate power inspection process optimization reference data; the power inspection process optimization reference data includes a power inspection work quality evaluation score and a best inspection scheduling strategy.

[0051] The power inspection process optimization method and system based on machine learning provided by the application include a power inspection data collection module, a power inspection data optimization module, a power inspection work quality evaluation module, an intelligent inspection scheduling module, and a power inspection process optimization module.

[0052] The power inspection data collection module is configured to collect power inspection data, obtain power inspection data by collecting the power inspection data, and send the power inspection data to the power inspection data optimization module.

[0053] The power inspection data optimization module is configured to optimize power inspection data, obtain power inspection optimization data by optimizing the power inspection data, and send the power inspection optimization data to the power inspection work quality evaluation module.

[0054] The power inspection work quality evaluation module is configured to evaluate power inspection work quality, obtain a power inspection work quality evaluation score by evaluating the power inspection work quality, and send the power inspection work quality evaluation score to the intelligent inspection scheduling module and the power inspection process optimization module.

[0055] The intelligent inspection scheduling module is configured to intelligently schedule inspections, obtain an optimal inspection scheduling strategy through the electricity inspection process optimization module, and send the optimal inspection scheduling strategy to the electricity inspection process optimization module.

[0056] The electricity inspection process optimization module is configured to optimize the electricity inspection process and generate electricity inspection process optimization reference data.

[0057] The above-mentioned scheme has the following beneficial effects:

[0058] (1) In the process of electricity inspection process optimization, the electricity inspection process is closely related to the inspection scheduling, and the traditional inspection scheduling is usually based on fixed rules or experience, without fully considering the actual work quality of the inspectors, resulting in a disconnection between inspection quality and scheduling, high-risk areas may be assigned to low-efficiency personnel, and it is difficult to achieve proactive prevention inspection and affect the scheduling effect. The technical problem, the scheme creatively adopts an intelligent method combining electricity inspection work quality evaluation and intelligent inspection scheduling module, combines historical data and real-time evaluation results, realizes adaptive matching of inspection resources and personnel ability, and improves the overall inspection efficiency and accuracy.

[0059] (2) In the process of electricity inspection work quality evaluation, the artificial scoring standard is not unified, resulting in large fluctuations in the scores of the same inspector in different assessment periods, the simple weighted average method cannot distinguish the difference between sudden equipment defect handling and daily standard operation, and the existing scoring system cannot balance real-time operation details and long-term work performance. The technical problem, the scheme creatively adopts a bidirectional time convolution long-short term parallel model for electricity work quality evaluation, and innovatively introduces power equipment inspection records as features, so that the evaluation results reflect not only the standardization of single inspection, but also the contribution value of the inspector to the long-term safe operation of the power grid.

[0060] (3) In the process of intelligent inspection scheduling, the traditional manual scheduling relies on experience, resulting in high risk of missing inspection of high-risk equipment, fixed scheduling mode cannot adapt to sudden power equipment failure, and single efficiency index orientation causes uneven inspection quality. The technical problem, the scheme creatively adopts a strategy evaluation enhanced network for intelligent inspection scheduling, dynamically balances task execution efficiency and inspection quality, effectively solves the common problems in power grid scheduling, and improves the performance and efficiency of intelligent inspection scheduling. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 A flowchart of the electricity inspection process optimization method based on machine learning provided by the present application is shown.

[0062] Figure 2 A schematic diagram of the electricity inspection process optimization system based on machine learning provided by the present application is shown.

[0063] Figure 3 This is a flowchart illustrating step S3;

[0064] Figure 4 This is a flowchart illustrating step S4.

[0065] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0066] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0067] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0068] Example 1, see Figure 1 The present invention provides a machine learning-based method for optimizing electricity usage inspection processes, which includes the following steps:

[0069] Step S1: Collect electricity usage inspection data;

[0070] Step S2: Optimize electricity usage inspection data;

[0071] Step S3: Quality assessment of electricity inspection work;

[0072] Step S4: Intelligent inspection and scheduling;

[0073] Step S5: Optimize the electricity inspection process.

[0074] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the collection of electricity inspection data is used to collect data required for optimizing the electricity inspection process. Specifically, it involves collecting work quality assessment data and inspection scheduling data to obtain electricity inspection data.

[0075] The work quality assessment data includes inspector work data, power equipment inspection records, and inspector patrol routes.

[0076] The inspector work data includes the number of inspection tasks, the type of task, the difficulty of task, the quality of task, the inspection time and the work feedback;

[0077] The power equipment inspection record includes the type of power equipment, the fault history and the maintenance record;

[0078] The inspection scheduling related data includes the inspector information, the power equipment information, the external environment information and the task scheduling history data;

[0079] The inspector information includes the available time of inspector, the skill level and the inspection task;

[0080] The power equipment information includes the state of power equipment, the priority and the location;

[0081] The external environment information includes the weather data and the holiday data;

[0082] The task scheduling history data includes the execution time of inspection task and the execution efficiency of task.

[0083] Embodiment three, refer to Figure 1 The embodiment is based on the above-mentioned embodiment, in step S2, the optimization of power consumption inspection data is used to optimize the quality of power consumption inspection data, specifically, by optimizing the power consumption inspection data, the power consumption inspection optimization data is obtained;

[0084] The optimization processing includes data cleaning and data standardization;

[0085] The data cleaning includes deduplication, outlier elimination and missing value filling;

[0086] The data standardization includes numerical data standardization, type data encoding and timestamp alignment.

[0087] Embodiment four, refer to Figure 1 And Figure 3 The embodiment is based on the above-mentioned embodiment, in step S3, the power consumption inspection work quality evaluation is used to evaluate the power consumption inspection work quality of inspector, specifically, the bidirectional time convolution long short-term parallel model is used to evaluate the power consumption work quality, and the power consumption inspection work quality evaluation score is obtained;

[0088] The bidirectional time convolution long short-term parallel model includes data input layer, parallel feature analysis layer, multi-level feature fusion layer and evaluation output layer;

[0089] The parallel feature analysis layer includes parallel bidirectional time convolution subnetwork and long short-term memory subnetwork;

[0090] The bidirectional time convolution subnetwork is used to capture short-term trends to extract local features;

[0091] The long short-term memory subnetwork is used for capturing long-term dependencies to extract global features.

[0092] The multi-level feature fusion layer is used for fusing local features and global features, thereby enhancing the understanding ability of the model to data.

[0093] The power consumption inspection work quality evaluation includes the following steps:

[0094] Step S31: data input layer construction, specifically receiving work quality evaluation related data in the power consumption inspection optimization data through the data input layer, and constructing evaluation input data;

[0095] Step S32: parallel feature analysis layer construction, specifically setting parallel bidirectional time convolution subnetwork and long short-term memory subnetwork in the parallel feature analysis layer, performing feature analysis on the evaluation input data, obtaining work quality local short-term features and work quality global long-term features, including the following steps:

[0096] Step S321: bidirectional time convolution subnetwork construction, specifically setting three bidirectional dilated causal convolution layers in the bidirectional time convolution subnetwork, expanding the convolution receptive field by setting different dilation rates in each bidirectional dilated causal convolution layer, and setting a residual connection layer and a batch normalization layer after each bidirectional dilated causal convolution layer, constructing the bidirectional time convolution subnetwork, performing local feature extraction on the evaluation input data, and generating work quality local short-term features;

[0097] The dilation rates of the three bidirectional dilated causal convolution layers are set to 1, 2 and 4 respectively;

[0098] The bidirectional dilated causal convolution layer includes a forward dilated causal convolution layer, a backward dilated causal convolution layer and a feature concatenation layer, and the calculation formula is:

[0099] ;

[0100] In the formula, y fw (t) is the output feature of the forward dilated causal convolution layer at the t-th time step, t is the first time step index, k is the convolution kernel index, K is the convolution kernel size, w k is the k-th convolution kernel weight, d is the dilation rate, x(t-d∙k) is the evaluation input data at the t-d∙k time step, b fw is the forward dilated causal convolution layer bias term, y bw (t) is the output feature of the backward dilated causal convolution layer at the t-th time step, x(t+d∙k) is the evaluation input data at the t+d∙k time step, b bw is the backward dilated causal convolution layer bias term, y bitcn(t) is the output feature of the bidirectional inflation causal convolution layer at the t time step, and concat(·) is feature splicing;

[0101] Step S322: Long Short Term Memory subnetwork construction, specifically, input gate, forget gate and output gate are set in the Long Short Term Memory subnetwork, global feature extraction is performed on the evaluation input data, and global long-term features of work quality are generated;

[0102] Step S33: Multi-level feature fusion layer construction, specifically, the local short-term features of work quality and the global long-term features of work quality are spliced through the multi-level feature fusion layer to generate comprehensive time sequence features of work quality;

[0103] Step S34: Evaluation output layer construction, used for outputting model evaluation results, specifically, a dense layer is set in the evaluation output layer, the comprehensive time sequence features of work quality are evaluated, and model evaluation results are obtained;

[0104] Step S35: Bidirectional time convolution Long Short Term parallel model training, specifically, the data input layer construction, the parallel feature analysis layer construction, the multi-level feature fusion layer construction and the evaluation output layer construction are used to construct a bidirectional time convolution Long Short Term parallel model, and model training is performed to obtain a power inspection work quality evaluation model;

[0105] Step S36: Power inspection work quality evaluation information generation, specifically, power inspection work quality evaluation is performed through the power inspection work quality evaluation model to obtain a power inspection work quality evaluation score;

[0106] Through the above operations, in the process of optimizing the power inspection process, the power inspection process is closely related to the inspection scheduling, and the traditional inspection scheduling is usually based on fixed rules or experience, without fully considering the actual work quality of the inspectors, resulting in a disconnection between inspection quality and scheduling, high-risk areas may be assigned to low-efficiency personnel, and it is difficult to achieve active prevention inspection and affect the scheduling effect. The technical problem, the scheme creatively uses an intelligent method combining power inspection work quality evaluation and intelligent inspection scheduling module, combines historical data and real-time evaluation results, realizes adaptive matching of inspection resources and personnel capabilities, and improves overall inspection efficiency and accuracy.

[0107] Embodiment five, refer to Figure 1 and Figure 4 This embodiment is based on the above-mentioned embodiments, in step S4, the intelligent inspection scheduling, specifically, according to the power inspection work quality evaluation score and the power inspection optimization data, an intelligent inspection scheduling is performed by using a strategy evaluation enhanced network to obtain the best inspection scheduling strategy, including the following steps:

[0108] Step S41: patrol scheduling modeling, specifically modeling the patrol scheduling as a Markov decision process, including the following steps:

[0109] Step S411: state space setting, specifically concatenating inspector information, power equipment information and external environment information into a state vector;

[0110] Step S412: action space setting, specifically setting the inspection task as the action space and assigning the inspection task as the action;

[0111] Step S413: reward function setting, used to measure the pros and cons of the patrol scheduling strategy, specifically setting the reward function according to the task execution efficiency and electricity inspection work quality evaluation score, the calculation formula is:

[0112] ;

[0113] In the formula, R z is the reward value of the zth time step, is the first reward weight, z is the second time step index, Eff z is the task execution efficiency reward of the zth time step, is the second reward weight, Qua z is the electricity inspection work quality evaluation score reward of the zth time step;

[0114] Step S42: policy evaluation enhancement network improvement, specifically setting a policy network and an evaluation network in the policy evaluation network, and introducing a residual convolution layer in the policy network and the evaluation network respectively, used to relieve model overfitting, and performing policy evaluation enhancement network improvement to obtain a policy evaluation enhancement network;

[0115] Step S43: intelligent patrol scheduling model construction, specifically updating the parameters of the policy evaluation enhancement network through model training, optimizing the policy network and the evaluation network, and obtaining an intelligent patrol scheduling model, including the following steps:

[0116] Step S431: randomly initializing the parameters of the policy network and the evaluation network;

[0117] Step S432: introducing KL divergence in the objective function, used to constrain the policy update range, the calculation formula is:

[0118] ;

[0119] In the formula, L ac is the objective function, is the expected calculation of the state distribution generated by the current patrol scheduling strategy, is the state distribution, is the discount factor, s zis the state at the z-th time step, s z+1 is the state at the z+1-th time step, is the state value function, which is used to calculate the average reward expectation in the state, is the KL divergence weight, is the KL divergence, is the old inspection scheduling policy, is the new inspection scheduling policy;

[0120] Step S433: Perform global network parameter update by policy network gradient update and evaluation network gradient update;

[0121] The calculation formula of the policy network gradient update is:

[0122] ;

[0123] In the formula, is the update amount of the policy network parameter, is the policy network parameter, is the policy network learning rate, is the gradient of the policy function logarithm, which is used to measure the sensitivity of the current inspection scheduling policy to the policy network parameter, and log is the logarithmic function, is the policy function, which is used to represent the probability of taking action a z under state s z z is the action taken at the z-th time step, is the advantage function, which is used to measure the pros and cons of taking action a

[0124] The calculation formula of the advantage function is:

[0125] ;

[0126] In the formula, is the state action value function, which is used to calculate the expected cumulative reward of taking action a is the state value function, which is used to calculate the average reward expectation in the state;

[0127] The calculation formula of the evaluation network gradient update is

[0128] ;

[0129] In the formula, is the update amount of the evaluation network parameter, is the evaluation network parameter, is the evaluation network learning rate, is the gradient of the state value function with respect to the evaluation network parameter;

[0130] ​The calculation formula for performing global network parameter updating is:

[0131] ;

[0132] In the formula, is the updated policy network parameter, is the updated evaluation network parameter;

[0133] Step S44: generating the best inspection scheduling strategy, specifically, performing intelligent inspection scheduling through the intelligent inspection scheduling model to obtain the best inspection scheduling strategy;

[0134] Through the above operations, for the technical problems that in the power inspection work quality evaluation process, the artificial scoring standard is not unified, which leads to large score fluctuations of the same inspector in different assessment periods, the simple weighted average method cannot distinguish the difference between sudden equipment defect processing and daily standard operation, and the existing scoring system is difficult to balance real-time operation details and long-term work performance, the present scheme creatively uses a bidirectional time convolution long-short term parallel model for power work quality evaluation, and creatively introduces power equipment inspection records as features, so that the evaluation result reflects not only the standardization degree of single inspection, but also the contribution value of the inspector to the long-term safe operation of the power grid.

[0135] Embodiment six, referring to Figure 1 This embodiment is based on the above-mentioned embodiments, and in step S5, the power inspection process is optimized, specifically, the power inspection process is optimized through the power inspection work quality evaluation and the intelligent inspection scheduling to generate power inspection process optimization reference data; the power inspection process optimization reference data includes power inspection work quality evaluation scores and the best inspection scheduling strategy;

[0136] Through the above operations, for the technical problems that in the intelligent inspection scheduling process, the traditional artificial scheduling relies on experience, which leads to a high missing inspection rate of high-risk equipment, the fixed scheduling mode cannot adapt to sudden power equipment failures, and the single efficiency index orientation causes uneven inspection quality, the present scheme creatively uses a strategy evaluation enhanced network for intelligent inspection scheduling, dynamically balances task execution efficiency and inspection quality, effectively solves common problems in power grid scheduling, and improves the performance and efficiency of intelligent inspection scheduling.

[0137] Embodiment seven, referring to Figure 2 This embodiment is based on the above-mentioned embodiments, and the present application provides a power inspection process optimization method and system based on machine learning, which comprises: a power inspection data collection module, an optimized power inspection data module, a power inspection work quality evaluation module, an intelligent inspection scheduling module, and a power inspection process optimization module;

[0138] The collecting electricity inspection data module is configured to collect electricity inspection data, obtain electricity inspection data by collecting the electricity inspection data, and send the electricity inspection data to the optimizing electricity inspection data module.

[0139] The optimizing electricity inspection data module is configured to optimize electricity inspection data, obtain electricity inspection optimization data by optimizing the electricity inspection data, and send the electricity inspection optimization data to the electricity inspection work quality evaluation module.

[0140] The electricity inspection work quality evaluation module is configured to evaluate electricity inspection work quality, obtain an electricity inspection work quality evaluation score by evaluating the electricity inspection work quality, and send the electricity inspection work quality evaluation score to the intelligent patrol inspection scheduling module and the electricity inspection process optimization module.

[0141] The intelligent patrol inspection scheduling module is configured to intelligently schedule patrol inspection, obtain an optimal patrol inspection scheduling strategy by the electricity inspection process optimization module, and send the optimal patrol inspection scheduling strategy to the electricity inspection process optimization module.

[0142] The electricity inspection process optimization module is configured to optimize an electricity inspection process, and generate electricity inspection process optimization reference data.

[0143] It should be noted that, in this document, the terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply these entities or operations have any such actual relationship or order. Also, the terms "comprises", "comprising", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus that comprises a list of elements does not only include those elements, but also includes other elements not expressly listed, or inherent to such a process, method, article, or apparatus.

[0144] Although the embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application.

[0145] The above describes the present application and its embodiments, which are not restrictive, and the drawings only show one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person of ordinary skill in the art is inspired by it, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution can be designed, which should belong to the protection scope of the present application.

Claims

1. A method for optimizing electricity inspection processes based on machine learning, characterized in that: The method includes the following steps: Step S1: Collect electricity inspection data, which is used to collect the data required for optimizing the electricity inspection process and obtain electricity inspection data; Step S2: Optimize electricity inspection data to improve the quality of electricity inspection data. Specifically, this involves optimizing the electricity inspection data to obtain optimized electricity inspection data. Step S3: Evaluation of the quality of electricity inspection work. This is used to evaluate the quality of the inspector's electricity inspection work. Specifically, a bidirectional temporal convolutional long short-term parallel model is used to evaluate the quality of electricity inspection work and obtain an evaluation score for the quality of electricity inspection work. The bidirectional temporal convolutional long short-term memory parallel model includes a data input layer, a parallel feature analysis layer, a multi-level feature fusion layer, and an evaluation output layer. The parallel feature analysis layer includes parallel bidirectional temporal convolutional subnetworks and long short-term memory subnetworks. The bidirectional temporal convolutional subnetwork is used to capture short-term trends to extract local features. The long short-term memory subnetwork is used to capture long-term dependencies to extract global features. The multi-level feature fusion layer is used to fuse local and global features. The quality assessment of the electricity inspection work includes the following steps: Step S31: Data input layer construction, specifically, receiving work quality assessment-related data from the power consumption inspection and optimization data through the data input layer, and constructing assessment input data; Step S32: Parallel feature analysis layer construction, specifically, involves setting up parallel bidirectional temporal convolutional subnetworks and long short-term memory subnetworks in the parallel feature analysis layer to perform feature analysis on the evaluation input data, obtaining local short-term features and global long-term features of work quality, including the following steps: Step S321: Construction of a bidirectional temporal convolutional subnetwork. Specifically, three bidirectional dilated causal convolutional layers are set in the bidirectional temporal convolutional subnetwork. The receptive field of the convolution is expanded by setting different dilation rates in each bidirectional dilated causal convolutional layer. A residual connection layer and a batch normalization layer are set after each bidirectional dilated causal convolutional layer to construct a bidirectional temporal convolutional subnetwork. Local features are extracted from the evaluation input data to generate local short-term features of the working quality. The dilation rates of the three bidirectional dilated causal convolutional layers are set to 1, 2 and 4, respectively. The bidirectional dilated causal convolutional layer includes a forward dilated causal convolutional layer, a backward dilated causal convolutional layer and a feature concatenation layer. Step S322: Long Short-Term Memory Subnet Construction, specifically, setting up input gates, forget gates and output gates in the Long Short-Term Memory Subnet, extracting global features from the evaluation input data, and generating global long-term features of job quality; Step S33: Construction of multi-level feature fusion layer, specifically, the local short-term features of work quality and the global long-term features of work quality are spliced ​​together through multi-level feature fusion layer to generate comprehensive time-series features of work quality; Step S34: Construct the evaluation output layer to output the model evaluation results. Specifically, a dense layer is set in the evaluation output layer to evaluate the comprehensive time-series characteristics of work quality and obtain the model evaluation results. Step S35: Bidirectional temporal convolution long-short parallel model training, specifically, through the construction of the data input layer, the construction of the parallel feature analysis layer, the construction of the multi-level feature fusion layer and the construction of the evaluation output layer, a bidirectional temporal convolution long-short parallel model is constructed and trained to obtain the power consumption inspection work quality evaluation model; Step S36: Generation of electricity inspection work quality assessment information, specifically, the electricity inspection work quality assessment is carried out through the electricity inspection work quality assessment model to obtain the electricity inspection work quality assessment score; Step S4: Intelligent inspection scheduling, specifically, based on the quality assessment score of electricity inspection work and the optimization data of electricity inspection, a strategy evaluation enhancement network is used to carry out intelligent inspection scheduling to obtain the best inspection scheduling strategy; Step S5: Optimize the electricity inspection process.

2. The method for optimizing electricity consumption inspection process based on machine learning according to claim 1, characterized in that: In step S4, the intelligent inspection scheduling specifically involves using a strategy evaluation enhancement network to perform intelligent inspection scheduling based on the electricity inspection work quality assessment score and electricity inspection optimization data, to obtain the optimal inspection scheduling strategy, including the following steps: Step S41: Inspection scheduling modeling, specifically modeling the inspection scheduling as a Markov decision process, including the following steps: Step S411: State space setting, specifically, concatenating inspector information, power equipment information, and external environment information into a state vector; Step S412: Action space setting, specifically setting the inspection task as the action space and setting the inspection task assignment as an action; Step S413: Setting the reward function to measure the effectiveness of the inspection and scheduling strategy. Specifically, the reward function is set based on the task execution efficiency and the quality assessment score of the electricity inspection work. The calculation formula is as follows: ； In the formula, R z It is the reward value at time step z. Here, z is the first reward weight, z is the second time step index, and Eff is the second reward weight. z It is the task execution efficiency reward at time step z. It is the second reward weight, Qua z The reward is the quality assessment score for the electricity inspection work at time step z. Step S42: Improve the policy evaluation enhancement network. Specifically, set up a policy network and an evaluation network in the policy evaluation network, and introduce residual convolutional layers into the policy network and the evaluation network respectively to alleviate model overfitting. Improve the policy evaluation enhancement network to obtain the policy evaluation enhancement network. Step S43: Construction of the intelligent inspection and scheduling model, specifically, involves updating the parameters of the policy evaluation enhancement network through model training, optimizing the policy network and the evaluation network, and obtaining the intelligent inspection and scheduling model, including the following steps: Step S431: Randomly initialize policy network parameters and evaluate network parameters; Step S432: Introduce KL divergence into the objective function to constrain the policy update range. The calculation formula is as follows: ； In the formula, L ac It is the objective function. It involves calculating the expected value of the state distribution generated by the current inspection scheduling strategy. It is a state distribution. It is the discount factor, s z This is the state at time step z, s z+1 This is the state at time step z+1. It is a state-valued function used to calculate the average expected reward in a given state. It is the KL divergence weight. It is the KL divergence. It's the old inspection and dispatch strategy. This is a new inspection and dispatch strategy; Step S433: Update the global network parameters by updating the policy network gradient and evaluating the network gradient. Step S44: Optimal inspection scheduling strategy generation, specifically through intelligent inspection scheduling model to obtain the optimal inspection scheduling strategy.

3. The method for optimizing electricity consumption inspection process based on machine learning according to claim 2, characterized in that: In step S2, the optimized electricity inspection data is used to improve the quality of the electricity inspection data. Specifically, it involves optimizing the electricity inspection data to obtain optimized electricity inspection data. The optimization process includes data cleaning and data standardization. The data cleaning includes deduplication, outlier removal, and missing value imputation. The data standardization includes numerical data standardization, type data encoding, and timestamp alignment.

4. The method for optimizing electricity consumption inspection process based on machine learning according to claim 3, characterized in that: In step S5, the optimization of the electricity inspection process specifically involves optimizing the electricity inspection process through the electricity inspection work quality assessment and the intelligent inspection scheduling, and generating electricity inspection process optimization reference data; the electricity inspection process optimization reference data includes the electricity inspection work quality assessment score and the optimal inspection scheduling strategy.

5. The method for optimizing electricity consumption inspection process based on machine learning according to claim 4, characterized in that: In step S1, the collection of electricity inspection data is used to collect data required for optimizing the electricity inspection process. Specifically, it involves collecting work quality assessment data and inspection scheduling data to obtain electricity inspection data. The work quality assessment data includes inspector work data, power equipment inspection records, and inspector inspection routes. The inspection and scheduling data includes inspector information, power equipment information, external environment information, and historical task scheduling data.

6. The method for optimizing electricity consumption inspection process based on machine learning according to claim 5, characterized in that: It includes modules for collecting electricity inspection data, optimizing electricity inspection data, evaluating the quality of electricity inspection work, intelligent inspection scheduling, and optimizing the electricity inspection process.

7. A machine learning-based electricity inspection process optimization system, used to implement the machine learning-based electricity inspection process optimization method as described in any one of claims 1-6, characterized in that: The electricity inspection data collection module is used to collect electricity inspection data, obtain electricity inspection data by collecting electricity inspection data, and send the electricity inspection data to the electricity inspection data optimization module. The optimized electricity inspection data module is used to optimize electricity inspection data, obtain optimized electricity inspection data, and send the optimized electricity inspection data to the electricity inspection work quality evaluation module. The electricity inspection work quality assessment module is used to assess the quality of electricity inspection work. Through the assessment, an electricity inspection work quality assessment score is obtained, and the score is sent to the intelligent inspection scheduling module and the electricity inspection process optimization module. The intelligent inspection and scheduling module is used for intelligent inspection and scheduling. It obtains the optimal inspection and scheduling strategy through the electricity inspection process optimization module and sends the optimal inspection and scheduling strategy to the electricity inspection process optimization module. The electricity inspection process optimization module is used to optimize the electricity inspection process and generate reference data for the optimization of the electricity inspection process.

Images