A patrol inspector work evaluation method, system, electronic equipment and medium

Abstract

The present invention relates to the field of management evaluation technology, and in particular to a method, system, electronic device and medium for evaluating the work of inspectors. The method comprises the following steps: firstly obtaining the work data of inspectors and generating data samples; then constructing and training a deep reinforcement learning model, inputting the data samples into the constructed deep reinforcement learning model, completing reasoning through multi-step decision-making, and outputting evaluation weight results; then adjusting the evaluation weights, using the evaluation weights and the work data of inspectors, analyzing the work status of inspectors, and giving evaluation scores, thereby realizing a scientific and dynamic quantitative evaluation of information such as the work status and workload of inspectors, and being able to make optimal decisions scientifically and accurately according to different working environments.

Description

Technical Field

[0001] The present invention relates to the field of management evaluation technology, and in particular to a method, system, electronic equipment and medium for evaluating the work of inspectors. Background Art

[0002] Inspection work is a key link in ensuring the safe and stable operation of equipment. Its efficiency and quality are directly related to the production efficiency and economic benefits of the enterprise. The evaluation system that quantifies and guides the inspection work is particularly critical.

[0003] In recent years, with the rapid development of deep learning technology, more and more "deep learning + industrial application" projects have been implemented. In deep learning technology, reinforcement learning, as a cutting-edge technology, is widely used by the industry for its fast reasoning ability and powerful model capabilities, and has achieved good results in the fields of robot cruising and autonomous driving. Reinforcement learning converts industrial tasks into a continuous Markov decision process. Through the reward function of each decision step, it guides the intelligent agent to obtain the maximum task reward and ultimately achieve the task optimization goal. Applying reinforcement learning to the inspection work evaluation can achieve more accurate and scientific inspection work evaluation.

[0004] The work evaluation of inspectors involves multiple dimensions, such as the actual receipt of responsible inspection transactions, the actual receipt rate of inspections, the actual receipt of non-responsible inspection transactions, the number of inspection kilometers, the acceptance rate of inspection tasks, etc. These dimensions together constitute a comprehensive evaluation system for the work quality of inspectors. However, most of the existing inspection work evaluation methods are based on a single dimension or a simple weighted method, which is difficult to fully and objectively reflect the actual work performance of inspectors. Therefore, it is particularly important to develop a weight-adaptive inspection work evaluation method based on reinforcement learning, which can not only realize the intelligent management of inspection work, but also dynamically adjust the weight of evaluation indicators through reinforcement learning technology, provide inspectors with personalized and precise work feedback and guidance, and further improve the overall efficiency and quality of inspection work. Summary of the invention

[0005] In order to solve the above problems in the prior art, the present invention proposes a patrol inspector's work evaluation method, system, electronic device and medium. Based on the deep reinforcement learning model, the intelligent agent, environment, action space, state space and reward function of the reinforcement learning model are designed, and the evaluation weights are adjusted through multi-step decision-making to correct the evaluation model, so as to evaluate the patrol inspector's work.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a method for inspector work evaluation, comprising the following steps:

[0008] S1. Obtain inspector work data and generate data samples;

[0009] S2. Build and train a deep reinforcement learning model, input data samples into the built deep reinforcement learning model, complete reasoning through multi-step decision-making, and output evaluation weight results;

[0010] S3. Adjust the evaluation weights, use the evaluation weights and inspector work data to analyze the inspector's work status, and give evaluation scores.

[0011] As a further technical solution of the present invention, the specific process of step S1 is:

[0012] S11. Obtain the inspector's work data, which includes the inspector's actual receipt of responsible inspection transactions, inspection receipt rate, non-responsible inspection transaction receipt, invitation registration, recharge, inspection mileage, inspection ticket rate, overtime ticket rate, and inspection task acceptance rate, recorded as , , is the type of work data, and the work data is Standardization, get standardized working data ,in For the The mean of the class data, For the The standard deviation of the class data;

[0013] S12. Generate data samples based on the standardized working data, denoted as ;

[0014] S13. Obtain the evaluation weight of the previous cycle , , Indicates The weight of the previous period corresponding to the class work data, .

[0015] As a further technical solution of the present invention, the specific process of step S2 is:

[0016] S21. Build a deep reinforcement learning model and select A2C (Advantage Actor-Critic) as the reinforcement learning model algorithm. Specifically:

[0017] S211. Define a deep reinforcement learning model agent as a module for adjusting evaluation weights to achieve an optimal solution;

[0018] S212, abstract the evaluation process into a Markov decision process, and complete continuous decision-making by gradually adjusting the evaluation weights to maximize the overall return and ultimately approach the optimal solution;

[0019] S213. Design the state space of the deep reinforcement learning model as the current working data and evaluation weight, denoted as , ,in, is the current working data, For the The evaluation weight corresponding to the class work data. The initial evaluation weight is the evaluation weight of the previous cycle. ;

[0020] S214. Design the action space of the deep reinforcement learning model to adjust the increment of the evaluation weight, denoted as , ;

[0021] S215. Design the reward function of the deep reinforcement learning model, denoted as , the reward function should guide the agent to approach the actual optimal evaluation weight. The closer the evaluation weight of the current action is to the actual optimal evaluation weight, the greater the reward is, and the farther away it is, the smaller the reward is and the greater the penalty is. The degree of closeness is expressed by the Euclidean distance between the current evaluation weight and the actual optimal evaluation weight. to characterize;

[0022] ;

[0023] in, is the current time step, is the reward of the current time step, is the Euclidean distance between the current time step evaluation weight and the actual optimal evaluation weight, , is the current time step The evaluation weight corresponding to the class work data, For the The actual evaluation weight corresponding to the class work data, is the penalty factor, is the tolerance factor;

[0024] S22. Build a data set and set up an experience replay pool to train a deep reinforcement learning model. Specifically:

[0025] S221. Dataset construction, data set samples , ,in is the evaluation weight of the previous cycle, is the current working data, is the actual optimal evaluation weight, and the training set, validation set, and test set are divided into 70%, 15%, and 15% respectively;

[0026] S222. Determine the training configuration and learning rate of the deep reinforcement learning model Set to 0.001, batch training size Set to 64, return discount rate Set to 0.9, the maximum time step Set to 10, set the training epoch to 50, and adopt the offline training strategy by setting the experience replay pool. The upper limit of the experience replay pool size is set to 10000 and the lower limit is 100. Specifically, the initial state Set to the evaluation weight of the previous period in the data sample, , Indicates the initial state The evaluation weight corresponding to the class work data; from the time step Start iterative training, the model outputs the current time step The agent action at , , Represents the time step Time The increment of the adjustment evaluation weight corresponding to the class work data, the state is transferred to , get reward , by and Add up and calculate , ; For the time step When the reward is , with state transfer equation , , Represents the time step Time The evaluation weight corresponding to the class work data, Represents the time step Time The increment of the adjustment evaluation weight corresponding to the class work data is repeated in increments of time steps until the maximum time step , and obtain a four-tuple training trajectory, ; Put the four-tuple training trajectory into the experience replay pool. If the number of training trajectories in the experience replay pool exceeds the upper limit, the earliest training trajectory in the pool is removed from the experience replay pool and discarded. The experience replay pool is checked once every 100 training trajectories. If the number of training trajectories in the pool exceeds the lower limit, the trajectory data of the batch training size is randomly sampled from the pool, and the model is trained using the stochastic gradient descent algorithm to update the model weights.

[0027] S223, using the constructed data set, training is performed according to step S222 until the model converges, completing the model training, and saving the model;

[0028] S23, when the model is inferring, the action output of each time step takes the action with the largest probability value, and the final output of the model is the state of the last time step go through The result after function processing.

[0029] As a further technical solution of the present invention, when constructing the data set in step S221, historical data and expert annotation are used to construct it, and work data and corresponding evaluation weights are extracted from the historical data as annotation data; work data is randomly generated through a program, and experts in the field of inspector work evaluation are required to manually assign weights to the generated random work data, thereby completing the construction of the entire data set.

[0030] As a further technical solution of the present invention, the specific process of step S3 is:

[0031] S31. Based on the evaluation weight of the previous cycle And reinforcement learning model, calculate the evaluation weight of the current cycle If the evaluation weight data of the previous cycle is missing, the expert annotation is used as the evaluation weight of the previous cycle. ; Then the current working data and the evaluation weight input of the previous cycle In the trained learning model, continuous decision-making is performed through the reinforcement learning model to obtain the current cycle evaluation weight of the final output ;

[0032] S32, output the evaluation results, and convert the current working data According to the scoring table corresponding to each work data, get the scores corresponding to the current work data Each item is in percentage, and then each indicator is evaluated according to the current cycle evaluation weight obtained in step S31 Perform stratification and weighted calculation to obtain the weight score corresponding to each evaluation indicator : , where the subscript T represents transposition, and the operator represents matrix multiplication; the weighted total is The inspector’s work evaluation is realized based on the weighted scores of each evaluation indicator and the weighted total score.

[0033] In a second aspect, the present invention provides a patrol inspector work evaluation system, comprising:

[0034] A data acquisition module is used to obtain the inspector's work data to be processed;

[0035] The reinforcement decision module is used to build and train a deep reinforcement learning model, use the inspector data to make multi-step decisions, and obtain an adjustment plan for the evaluation weights;

[0036] The work evaluation module calculates the evaluation score and gives the evaluation result based on the adjustment plan of the evaluation weight and the inspector's work data.

[0037] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0038] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention effectively combines deep learning and reinforcement learning technology, solves the shortcomings of the traditional inspector work evaluation system with a single evaluation dimension and a solidified evaluation strategy, realizes a scientific and dynamic quantitative evaluation of the inspector's work status, workload and other information, and cleverly utilizes the decision-making ability of deep reinforcement learning to regard the inspector's work evaluation process as a continuous decision-making process, namely, a Markov decision process, and approximates the optimal solution through multi-step decision-making, so that the system can dynamically evaluate the inspector's work status more scientifically and accurately according to the current working environment. In addition, in order to make the reinforcement learning model decision more accurate, when constructing the reinforcement learning model, the present invention adopts an offline training strategy, which can make more full use of data samples and improve the decision-making ability of the model. Compared with the traditional inspector work evaluation method based on the weight matrix, the present invention adopts a deep learning model to adjust the evaluation weight, which has the advantage of more accurate dynamic evaluation. This is because the reinforcement learning model has good dynamic decision-making ability and can make the optimal decision scientifically and accurately according to different working environments. When the inspector's working environment changes and fluctuates greatly, this advantage is particularly significant, and it is more competitive in terms of evaluation scientificity and evaluation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0042] Figure 1 A flow chart of the inspector work evaluation method provided by the present invention.

[0043] Figure 2 A training diagram of the deep reinforcement learning model provided by the present invention.

[0044] Figure 3 This is a structural block diagram of the inspector work evaluation system provided by the present invention. DETAILED DESCRIPTION

[0045] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0046] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0047] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0048] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0049] Embodiment 1:

[0050] like Figure 1 As shown, this embodiment provides a method for inspector work evaluation, comprising the following steps:

[0051] S1. Data acquisition: obtain the inspector's work data and generate data samples; the specific process is as follows:

[0052] S11. Obtain the inspector's work data, which includes the inspector's actual receipt of responsible inspection transactions, inspection receipt rate, non-responsible inspection transaction receipt, invitation registration, recharge, inspection mileage, inspection ticket rate, overtime ticket rate, and inspection task acceptance rate, recorded as , , is the type of work data, and the work data is Standardization, get standardized working data ,in For the The mean of the class data, For the The standard deviation of the class data;

[0053] S12. Generate data samples based on the standardized working data, denoted as ;

[0054] S13. Obtain the evaluation weight of the previous cycle , , Indicates The weight of the previous period corresponding to the class work data, .

[0055] S2, Reinforcement decision: Build and train a deep reinforcement learning model, input data samples into the constructed deep reinforcement learning model, complete reasoning through multi-step decision-making, and output evaluation weight results; S21, build a deep reinforcement learning model, select A2C (Advantage Actor-Critic) as the reinforcement learning model algorithm, specifically:

[0056] S211. Define a deep reinforcement learning model agent as a module for adjusting evaluation weights to achieve an optimal solution;

[0057] S212, abstract the evaluation process into a Markov decision process, and complete continuous decision-making by gradually adjusting the evaluation weights to maximize the overall return and ultimately approach the optimal solution;

[0058] S213. Design the state space of the deep reinforcement learning model as the current working data and evaluation weight, denoted as , ,in, is the current working data, For the The evaluation weight corresponding to the class work data. The initial evaluation weight is the evaluation weight of the previous cycle. ;

[0059] S214. Design the action space of the deep reinforcement learning model to adjust the increment of the evaluation weight, denoted as , ;

[0060] S215. Design the reward function of the deep reinforcement learning model, denoted as , the reward function should guide the agent to approach the actual optimal evaluation weight. The closer the evaluation weight of the current action is to the actual optimal evaluation weight, the greater the reward is, and the farther away it is, the smaller the reward is and the greater the penalty is. The degree of closeness is expressed by the Euclidean distance between the current evaluation weight and the actual optimal evaluation weight. to characterize;

[0061] ;

[0062] in, is the current time step, is the reward of the current time step, is the Euclidean distance between the current time step evaluation weight and the actual optimal evaluation weight, , is the current time step The evaluation weight corresponding to the class work data, For the The actual evaluation weight corresponding to the class work data, is the penalty factor, is the tolerance factor;

[0063] S22, build a data set and set up an experience replay pool to train a deep reinforcement learning model, such as Figure 2 As shown, specifically:

[0064] S221. Dataset construction, data set samples , ,in is the evaluation weight of the previous cycle, is the current working data, The optimal evaluation weights are obtained, and the training set, validation set, and test set are divided into 70%, 15%, and 15% ratios. When constructing the data set, historical data and expert annotation are used to construct the data set. Work data and corresponding evaluation weights are extracted from historical data as annotation data. Work data are randomly generated through a program, and experts in the field of inspector work evaluation are required to manually assign weights to the generated random work data, thereby completing the construction of the entire data set.

[0065] S222. Determine the training configuration and learning rate of the deep reinforcement learning model Set to 0.001, batch training size Set to 64, return discount rate Set to 0.9, the maximum time step Set to 10, set the training epoch to 50, and adopt the offline training strategy by setting the experience replay pool. The upper limit of the experience replay pool size is set to 10000 and the lower limit is 100. Specifically, the initial state Set to the evaluation weight of the previous period in the data sample, , Indicates the initial state The evaluation weight corresponding to the class work data; from the time step Start iterative training, the model outputs the current time step The agent action at , , Represents the time step Time The increment of the adjustment evaluation weight corresponding to the class work data, the state is transferred to , get reward , by and Add up and calculate , ; For the time step When the reward is , with state transfer equation , , Represents the time step Time The evaluation weight corresponding to the class work data, Represents the time step Time The increment of the adjustment evaluation weight corresponding to the class work data is repeated in increments of time steps until the maximum time step , and obtain a four-tuple training trajectory, ; Put the four-tuple training trajectory into the experience replay pool. If the number of training trajectories in the experience replay pool exceeds the upper limit, the earliest training trajectory in the pool is removed from the experience replay pool and discarded. The experience replay pool is checked once every 100 training trajectories. If the number of training trajectories in the pool exceeds the lower limit, the trajectory data of the batch training size is randomly sampled from the pool, and the model is trained using the stochastic gradient descent algorithm to update the model weights.

[0066] S223, using the constructed data set, training is performed according to step S222 until the model converges, completing the model training, and saving the model;

[0067] S23, when the model is inferring, the action output of each time step takes the action with the largest probability value, and the final output of the model is the state of the last time step go through The result after function processing.

[0068] S3. Work evaluation: Adjust the evaluation weights, use the evaluation weights and inspector work data, analyze the inspector's work status, and give evaluation scores. The specific process is as follows:

[0069] S31. Based on the evaluation weight of the previous cycle And reinforcement learning model, calculate the evaluation weight of the current cycle If the evaluation weight data of the previous cycle is missing, the expert annotation is used as the evaluation weight of the previous cycle. ; Then the current working data and the evaluation weight input of the previous cycle In the trained learning model, continuous decision-making is performed through the reinforcement learning model to obtain the current cycle evaluation weight of the final output ;

[0070] S32, output the evaluation results, and convert the current working data According to the scoring table corresponding to each work data, get the scores corresponding to the current work data Each item is in percentage, and then each indicator is evaluated according to the current cycle evaluation weight obtained in step S31 Perform stratification and weighted calculation to obtain the weight score corresponding to each evaluation indicator : , where the subscript T represents transposition, and the operator represents matrix multiplication; the weighted total is The inspector’s work evaluation is realized based on the weighted scores of each evaluation indicator and the weighted total score.

[0071] Embodiment 2:

[0072] like Figure 3 As shown, this embodiment provides an inspector work evaluation system, including:

[0073] A data acquisition module is used to obtain the inspector's work data to be processed;

[0074] The reinforcement decision module is used to build and train a deep reinforcement learning model, use the inspector data to make multi-step decisions, and obtain an adjustment plan for the evaluation weights;

[0075] The work evaluation module calculates the evaluation score and gives the evaluation result based on the adjustment plan of the evaluation weight and the inspector's work data.

[0076] This embodiment uses the method described in Example 1 and the system described in Example 2 to evaluate the inspector's work. First, the inspector's current work data is obtained. and the evaluation weight data of the previous cycle :

[0077] ,

[0078] ;

[0079] Then and Input the enhanced decision module to get the predicted current evaluation weight:

[0080] ;

[0081] Then, through the work evaluation module, the current work data is scored according to the corresponding scoring table of each work data to obtain the corresponding scores: , stratify and weight according to weight W, and get the evaluation result:

[0082] Weighted scores , the weighted total score is 77.96, as shown in Table 1.

[0083] Table 1:

[0084]

[0085] In further embodiments, there is also provided:

[0086] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of simplicity, the detailed method content is not repeated here.

[0087] It should be clear that the processor in this embodiment is not limited to the central processing unit CPU, it can also be a general-purpose processor core, a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic device PLD, a discrete gate circuit or transistor logic component, a discrete hardware module, etc. In addition, the general-purpose processor can be a microprocessor MPU, or any conventional computing core.

[0088] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0089] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is completed.

[0090] The method in Example 1 can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.

[0091] Those skilled in the art will appreciate that the units, i.e., algorithm steps, of the various examples described in conjunction with this embodiment can be implemented in electronic hardware or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0092] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0093] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for evaluating the work of inspectors, characterized in that: The following steps are involved: S1. Obtain inspector work data and generate data samples; The specific process is: S11. Obtain the inspector's work data, which includes the inspector's actual receipt of responsible inspection transactions, inspection receipt rate, non-responsible inspection transaction receipt, invitation registration, recharge, inspection mileage, inspection ticket rate, overtime ticket rate, and inspection task acceptance rate, recorded as , , is the type of work data, and the work data is Standardization, get standardized working data ,in For the The mean of the class data, For the The standard deviation of the class data; S12. Generate data samples based on the standardized working data, denoted as ; S13. Obtain the evaluation weight of the previous cycle , , Indicates The weight of the previous period corresponding to the class work data, ; S2. Build and train a deep reinforcement learning model, input data samples into the built deep reinforcement learning model, complete reasoning through multi-step decision-making, and output evaluation weight results; the specific process is: S21. Build a deep reinforcement learning model and select A2C as the reinforcement learning model algorithm; S211. Define the deep reinforcement learning model agent as a module for adjusting the evaluation weight to achieve the optimal solution; specifically: S212, abstract the evaluation process into a Markov decision process, and complete continuous decision-making by gradually adjusting the evaluation weights to maximize the overall return and ultimately approach the optimal solution; S213. Design the state space of the deep reinforcement learning model as the current working data and evaluation weight, denoted as , ,in, is the current working data, For the The evaluation weight corresponding to the class work data. The initial evaluation weight is the evaluation weight of the previous cycle. ; S214. Design the action space of the deep reinforcement learning model to adjust the increment of the evaluation weight, denoted as , ; S215. Design the reward function of the deep reinforcement learning model, denoted as The closer the evaluation weight of the current action is to the actual optimal evaluation weight, the greater the reward is, and the farther away it is, the smaller the reward is and the greater the penalty is. The degree of closeness is expressed by the Euclidean distance between the current evaluation weight and the actual optimal evaluation weight. to characterize; ； in, is the current time step, is the reward of the current time step, is the Euclidean distance between the current time step evaluation weight and the actual optimal evaluation weight, , is the current time step The evaluation weight corresponding to the class work data, For the The actual evaluation weight corresponding to the class work data, is the penalty factor, is the tolerance factor; S22. Build a data set and set up an experience replay pool to train a deep reinforcement learning model; S23, deep reinforcement learning model reasoning, the action output for each time step is the action with the largest probability value, and the final output of the model is the state of the last time step go through The result after function processing; S3. Adjust the evaluation weights, use the evaluation weights and inspector work data to analyze the inspector's work status, and give evaluation scores.

2. The inspector work evaluation method according to claim 1 is characterized in that: The specific process of step S22 is: S221. Construction of data set, data set samples , ,in is the evaluation weight of the previous cycle, is the current working data, is the actual optimal evaluation weight, and the training set, validation set, and test set are divided into 70%, 15%, and 15% respectively; S222. Determine the training configuration and learning rate of the deep reinforcement learning model Set to 0.001, batch training size Set to 64, return discount rate Set to 0.9, the maximum time step Set to 10, set the training epoch to 50, and adopt the offline training strategy by setting the experience replay pool. The upper limit of the experience replay pool size is set to 10000 and the lower limit is 100. Specifically, the initial state Set to the evaluation weight of the previous period in the data sample, , Indicates the initial state The evaluation weight corresponding to the class work data; from the time step Start iterative training, the model outputs the current time step The agent action at , , Represents the time step Time The increment of the adjustment evaluation weight corresponding to the class work data, the state is transferred to , get reward , by and Add up and calculate , ; For the time step When the reward is , with state transfer equation , , Represents the time step Time The evaluation weight corresponding to the class work data, Represents the time step Time The increment of the adjustment evaluation weight corresponding to the class work data is repeated in increments of time steps until the maximum time step , and get a four-tuple training trajectory: ; Put the four-tuple training trajectory into the experience replay pool. If the number of training trajectories in the experience replay pool exceeds the upper limit, the earliest training trajectory in the pool will be removed from the experience replay pool and discarded. The experience replay pool will be checked every 100 training trajectories. If the number of training trajectories in the pool exceeds the lower limit, the trajectory data of the batch training size will be randomly sampled from the pool, and the stochastic gradient descent algorithm will be used to train the model and update the model weights.

3. The inspector work evaluation method according to claim 2 is characterized in that: When constructing the data set in step S221, historical data and expert annotation are used to construct it. Work data and corresponding evaluation weights are extracted from the historical data as annotation data; work data is randomly generated through a program, and experts in the field of inspector work evaluation are required to manually assign weights to the generated random work data, thereby completing the construction of the entire data set.

4. The inspector work evaluation method according to claim 3 is characterized in that: The specific process of step S3 is: S31. Based on the evaluation weight of the previous cycle And reinforcement learning model, calculate the evaluation weight of the current cycle If the evaluation weight data of the previous cycle is missing, the expert annotation is used as the evaluation weight of the previous cycle. ; Then the current working data and the evaluation weight input of the previous cycle In the trained learning model, continuous decision-making is performed through the reinforcement learning model to obtain the current cycle evaluation weight of the final output ; S32, output the evaluation results, and convert the current working data According to the scoring table corresponding to each work data, get the scores corresponding to the current work data Each item is in percentage, and then each indicator is evaluated according to the current cycle evaluation weight obtained in step S31 Perform stratification and weighted calculation to obtain the weight score corresponding to each evaluation indicator : , where the subscript T represents transposition, and the operator Represents matrix multiplication; The weighted total score is The inspector’s work evaluation is realized based on the weighted scores of each evaluation indicator and the weighted total score.

5. A patrol inspector work evaluation system, characterized in that: The method according to any one of claims 1 to 4 is performed, comprising: A data acquisition module is used to obtain the inspector's work data to be processed; The reinforcement decision module is used to build and train a deep reinforcement learning model, use the inspector data to make multi-step decisions, and obtain an adjustment plan for the evaluation weights; The work evaluation module calculates the evaluation score and gives the evaluation result based on the adjustment plan of the evaluation weight and the inspector's work data.

6. An electronic device, characterized in that: The invention comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 4 is completed.

7. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 4.

Images