Model weight information updating method and device, computer device, readable storage medium and program product

Abstract

The application relates to a model weight information updating method and device, computer equipment, a computer readable storage medium and a computer program product, which are applied to the technical field of big data, and the method comprises the following steps: acquiring model weight information of a plurality of model parameters in a pre-trained power risk assessment model; performing feature coding on the model weight information of the plurality of model parameters to obtain model weight coding features; performing a particle swarm construction step: constructing an initialized particle swarm according to the model weight coding features; evaluating the fitness of each particle in the initialized particle swarm to obtain particle fitness, wherein the particle fitness is used for representing the prediction accuracy of each model structure in the power risk assessment model; and performing iterative updating on the model weight information according to the particle fitness to obtain target model weight information. The method can improve the power risk assessment accuracy of a model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data, and in particular, relates to a model weight information updating method and device, computer equipment, a computer readable storage medium, and a computer program product. BACKGROUND

[0002] With the continuous development of economy, the demand for electricity is also increasing, and the corresponding power operation is also various and complex, such as three-dimensional crossing operation, continuous complex system engineering operation, etc., which leads to the existence of dangerous factors in the production process of power operation. Therefore, an electric power risk assessment method that can accurately assess the electric power risk is urgently needed.

[0003] In the traditional technology, the electric power risk of the power operation is assessed by using a pre-trained electric power risk assessment model. However, the electric power risk assessment accuracy of the pre-trained electric power risk assessment model depends on the number of training samples. Therefore, there may be a case that the electric power risk assessment accuracy of the trained electric power risk assessment model is low, which leads to low electric power risk assessment accuracy of the model. SUMMARY

[0004] Therefore, it is necessary to provide a model weight information updating method, device, computer equipment, computer readable storage medium, and computer program product that can improve the electric power risk assessment accuracy of the model to solve the above technical problems.

[0005] In a first aspect, the present application provides a model weight information updating method, comprising:

[0006] obtaining model weight information of a plurality of model parameters in a pre-trained electric power risk assessment model;

[0007] performing feature coding on the model weight information of the plurality of model parameters to obtain model weight coding features;

[0008] performing a particle swarm construction step: constructing an initialized particle swarm according to the model weight coding features;

[0009] evaluating the fitness of each particle in the initialized particle swarm to obtain particle fitness, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the electric power risk assessment model;

[0010] performing iterative updating on the model weight information according to the particle fitness to obtain target model weight information.

[0011] In one of the embodiments, the evaluating the fitness of each particle in the initialized particle swarm comprises: for each particle in the initialized particle swarm, evaluating the fitness of the particle according to a difference between output data of a model structure in the power risk evaluation model corresponding to the particle and corresponding real data, to obtain the particle fitness of the particle.

[0012] In one of the embodiments, the iteratively updating the model weight information according to the particle fitness to obtain target model weight information comprises: if the particle fitness of the current iteration does not meet a preset fitness condition, obtaining a random number in a preset value range, adjusting the model weight information of the current iteration according to the random number, and returning to perform the particle swarm construction step; if the particle fitness of the current iteration meets the preset fitness condition, determining the model weight information of the current iteration as the target model weight information.

[0013] In one of the embodiments, the adjusting the model weight information of the current iteration according to the random number comprises: obtaining an iteration design parameter; determining global particle information according to the particle fitness of each particle in the particle swarm of the current iteration; and adjusting the model weight information of the current iteration according to the global particle information, the random number and the iteration design parameter.

[0014] In one of the embodiments, after the iteratively updating the model weight information according to the particle fitness to obtain the target model weight information, the method further comprises: adjusting the power risk evaluation model according to the target model weight information to obtain a target power risk evaluation model; if power operation information of a power operation to be evaluated is detected, extracting features of the power operation information through the target power risk evaluation model to obtain power operation features; and evaluating a power safety risk of the power operation to be evaluated according to the power operation features to obtain a power risk evaluation result.

[0015] In one of the embodiments, the power operation information comprises a plurality of power operation data belonging to at least one category that changes over time; and the extracting features of the power operation information through the target power risk evaluation model to obtain power operation features comprises: for power operation data belonging to each category, obtaining influence information of the power operation data on the power safety risk evaluation of the power operation to be evaluated; normalizing the power operation data according to the influence information of the power operation data on the power safety risk evaluation of the power operation to be evaluated to obtain power operation normalized data; and extracting features of the power operation normalized data through the target power risk evaluation model to obtain the power operation features.

[0016] In a second aspect, the present application also provides a model weight information updating device, comprising:

[0017] a obtaining module, configured to obtain model weight information of a plurality of model parameters in a pre-trained power risk assessment model;

[0018] an encoding module, configured to perform feature encoding on the model weight information of the plurality of model parameters to obtain model weight encoding features;

[0019] a constructing module, configured to perform a particle swarm constructing step, i.e., constructing an initialized particle swarm according to the model weight encoding features;

[0020] an evaluating module, configured to evaluate the fitness of each particle in the initialized particle swarm to obtain particle fitness, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the power risk assessment model;

[0021] an updating module, configured to perform iterative updating on the model weight information according to the particle fitness to obtain target model weight information.

[0022] In a third aspect, the present application also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor realizes the following steps when executing the computer program:

[0023] obtaining model weight information of a plurality of model parameters in a pre-trained power risk assessment model;

[0024] performing feature encoding on the model weight information of the plurality of model parameters to obtain model weight encoding features;

[0025] performing a particle swarm constructing step, i.e., constructing an initialized particle swarm according to the model weight encoding features;

[0026] evaluating the fitness of each particle in the initialized particle swarm to obtain particle fitness, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the power risk assessment model;

[0027] performing iterative updating on the model weight information according to the particle fitness to obtain target model weight information.

[0028] In a fourth aspect, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program realizes the following steps when executed by a processor:

[0029] obtaining model weight information of a plurality of model parameters in a pre-trained power risk assessment model;

[0030] feature encode the model weight information of the plurality of model parameters to obtain model weight encoded features;

[0031] perform a particle swarm construction step of constructing an initialized particle swarm according to the model weight encoded features;

[0032] evaluate the fitness of each particle in the initialized particle swarm to obtain particle fitness, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the power risk assessment model;

[0033] update the model weight information iteratively according to the particle fitness to obtain target model weight information.

[0034] In a fifth aspect, the present application also provides a computer program product comprising a computer program which, when executed by a processor, implements the following steps:

[0035] obtain model weight information of a plurality of model parameters in a pre-trained power risk assessment model;

[0036] feature encode the model weight information of the plurality of model parameters to obtain model weight encoded features;

[0037] perform a particle swarm construction step of constructing an initialized particle swarm according to the model weight encoded features;

[0038] evaluate the fitness of each particle in the initialized particle swarm to obtain particle fitness, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the power risk assessment model;

[0039] update the model weight information iteratively according to the particle fitness to obtain target model weight information.

[0040] The model weight information updating method, device, computer device, computer readable storage medium and computer program product can obtain model weight information of a plurality of model parameters in a pre-trained power risk assessment model, perform feature coding on the model weight information of the plurality of model parameters to obtain model weight coding features, perform a particle swarm construction step, that is, constructing an initialized particle swarm according to the model weight coding features, evaluate the fitness of each particle in the initialized particle swarm to obtain particle fitness, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the power risk assessment model, and iteratively update the model weight information according to the particle fitness to obtain target model weight information. The model prediction effect of the power risk assessment model is refined to the prediction accuracy of each model structure, and the particle fitness representing the prediction accuracy of the model structure is used as the basis for updating the model weight information, so that in the case that the power risk assessment accuracy of the trained power risk assessment model is low, the power risk assessment accuracy of the power risk assessment model using the target model weight information as the basis for updating the model weight information can be improved by iteratively updating the model weight information. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other related drawings can be obtained by those skilled in the art without creative labor.

[0042] Figure 1 An application environment diagram of the model weight information updating method in an embodiment;

[0043] Figure 2 A flowchart of the model weight information updating method in an embodiment;

[0044] Figure 3 A flowchart of the step of obtaining the target model weight information by iteratively optimizing the model weight information according to the weight fusion information through the particle swarm algorithm in an embodiment;

[0045] Figure 4 A structure diagram of the target power risk assessment result in an embodiment;

[0046] Figure 5 A structure block diagram of the model weight information updating device in an embodiment;

[0047] Figure 6 An internal structure diagram of the computer device in an embodiment. DETAILED DESCRIPTION

[0048] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application.

[0049] It should be noted that the information (including but not limited to model weight information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the acquisition, transmission, storage, use and processing of the relevant data comply with the relevant provisions of national laws and regulations. The content pushed to the user (for example, power risk assessment results, etc.) can be refused or can be conveniently refused by the user. In the embodiments of the present application, some software, components, models, etc. may be mentioned, which are industry existing solutions, and they should be considered as exemplary, and the purpose is only to illustrate the feasibility of the implementation of the technical solutions of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0050] The model weight information updating method provided by the embodiments of the present application can be applied in an application environment as shown in Figure 1 . The terminal 102 communicates with the server 104 through a network. The data storage system can store the data required to be processed by the server 104. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers. The server 104 obtains the model weight information of the pre-trained power risk assessment model, wherein the model weight information is used to represent the weight of the model parameters in the power risk assessment model; the model weight information is fused to obtain weight fusion information, and the model weight information is iteratively optimized through a particle swarm algorithm according to the weight fusion information to obtain target model weight information. The terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart television, a smart air conditioner, a smart vehicle device, a projection device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 104 can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0051] In an exemplary embodiment, as shown in Figure 2 , a model weight information updating method is provided, and the method is applied inFigure 1 The server 104 in the power risk assessment system 100 is taken as an example to illustrate the method, including the following steps 202 to 210. Among them:

[0052] In step 202, the model weight information of the plurality of model parameters in the pre-trained power risk assessment model is obtained.

[0053] The plurality of model parameters in step 202 are input parameters of each model structure in the power risk assessment model.

[0054] For example, step 202 includes querying a preset database to obtain the model weight information of the pre-trained power risk assessment model.

[0055] As an example, querying the preset database to obtain the model weight information of the pre-trained power risk assessment model includes that the preset database includes model configuration information of the power risk assessment model, wherein the model configuration information includes structure parameters of each model structure in the power risk assessment model, and weights of input parameters of each model structure in the power risk assessment model, and the weights of the input parameters of each model structure in the power risk assessment model in the model configuration information of the power risk assessment model are determined as the model weight information of the pre-trained power risk assessment model.

[0056] In step 204, the model weight information of the plurality of model parameters is feature encoded to obtain a model weight encoding feature.

[0057] For example, step 204 includes feature encoding the model weight information of the plurality of model parameters according to the position of the model structure corresponding to the model weight information in the power risk assessment model to obtain the model weight encoding feature.

[0058] As an example, feature encoding the model weight information of the plurality of model parameters according to the position of the model structure corresponding to the model weight information in the power risk assessment model to obtain the model weight encoding feature includes that the model weight encoding feature includes a weight encoding matrix, the model weight information of the plurality of model parameters includes weights of input parameters of each model structure in the power risk assessment model, and the weights of the input parameters of each model structure in the power risk assessment model are feature encoded according to the position of each model structure in the power risk assessment model in the power risk assessment model to obtain the weight encoding matrix.

[0059] The model structure in the power risk assessment model includes but is not limited to a convolutional layer, a pooling layer, and a fully connected layer.

[0060] Furthermore, based on the position of each model structure in the power risk assessment model, the weights of the input parameters of each model structure in the power risk assessment model are feature-encoded to obtain a weight encoding matrix. This includes: performing particle swarm encoding and model structure position encoding on the weights of the input parameters of each model structure in the power risk assessment model based on their position in the model, to obtain the encoding information of the weights of the input parameters of each model structure in the power risk assessment model; and fusing the weights of the input parameters of each model structure in the power risk assessment model based on the encoding information of the weights of the input parameters of each model structure in the power risk assessment model to obtain a weight encoding matrix.

[0061] Optionally, based on the encoding information of the weights of the input parameters of each model structure in the power risk assessment model, the weights of the input parameters of each model structure in the power risk assessment model are fused to obtain a weight encoding matrix, which can be expressed by the formula:

[0062] P = [ w 11 1 c 1 … w kL 1 cn … w 11 2 c 1 … w kL 2 cn … … … … … w 11 Mc 1 … w kL Mcn … w 11 1 p 1 … … w kL 1 pn w 11 2 p 1 … … w kL 2 pn … … … … w 11 Mp 1 … … w kL Mpn ]

[0063] in, For the weight encoding matrix, The weights of the first input data to the first output data of the first convolutional layer belonging to the first particle swarm species. For the first particle swarm species The first convolutional layer The input data for the th _ ... The weights of each output data point The weights of the first input data to the first output data in the first pooling layer belonging to the first particle swarm species. For the first particle swarm species The first pooling layer The input data for the th _ ... The weights of each output data point The weights of the first input data to the first output data of the first convolutional layer belonging to the second particle swarm are given. For the second particle swarm species The first convolutional layer The input data for the th _ ... The weights of each output data point The weights of the first input data to the first output data in the first pooling layer belonging to the second particle swarm species. For the second particle swarm species The first pooling layer The input data for the th _ ... a weight of the first input data of the first convolutional layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first pooling layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first convolutional layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first convolutional layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first convolutional layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first convolutional layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first convolutional layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first pooling layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first pooling layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first pooling layer belonging to the first particle swarm for the first output data. a weight of the first input data of the first pooling layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first pooling layer belonging to the first particle swarm for the first output data, a weight of the first input data of the first pooling layer belonging to the first particle swarm for the first output data. a weight of the first input data of the first pooling layer belonging to the first particle swarm for the first output data. a weight of the first input data of the first pooling layer belonging to the first particle swarm for the first output data.

[0064] In step 206, a particle swarm construction step is performed: constructing an initialized particle swarm according to the model weight coding features.

[0065] For example, step 206 includes determining particle swarm information of the initialized particle swarm, and constructing the initialized particle swarm according to the weight fusion information and the particle swarm information, wherein the particle swarm information includes particle velocity.

[0066] For example, constructing the initialized particle swarm according to the weight fusion information and the particle swarm information includes: determining each weight in the weight fusion information as each particle in the initialized particle swarm, and determining the particle velocity as the movement speed of each particle in the initialized particle swarm.

[0067] In step 208, the fitness of each particle in the initialized particle swarm is evaluated to obtain particle fitness, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the power risk assessment model.

[0068] For example, step 208 includes: for each particle in the initialized particle swarm, evaluating the fitness of the particle according to the difference between the output data of the model structure in the power risk assessment model corresponding to the particle and the corresponding real data, to obtain the particle fitness of the particle.

[0069] Furthermore, based on the difference between the output data of the model structure in the power risk assessment model corresponding to the particle and the corresponding real data, the fitness of the particle is evaluated to obtain the particle fitness. This includes: acquiring multiple training samples, wherein each training sample includes the input data and real data of the model structure in the power risk assessment model corresponding to the particle; mapping the input data of the multiple training samples to the output data of the model structure in the power risk assessment model corresponding to the particle, respectively, through the model structure in the power risk assessment model corresponding to the particle; and evaluating the particle fitness based on the difference between the output data of the model structure in the power risk assessment model corresponding to the multiple training samples and the corresponding real data to obtain the particle fitness.

[0070] Optionally, the fitness of a particle is evaluated based on the difference between the output data of the model structure in the power risk assessment model corresponding to multiple training samples and the corresponding real data. The particle fitness can be expressed by the formula:

[0071] E = 1 n ∑ j = 1 n [ ∑ k = 1 a ( t jk − y jk ) 2 ]

[0072] in, For particle fitness, For the sample size, For the number of model structures, For the first In the nth sample The real data corresponding to each model structure For the first In the nth sample Output data for each model structure.

[0073] Step 210: Based on the particle fitness, iteratively update the model weight information to obtain the target model weight information.

[0074] Exemplarily, the step 210 comprises: traversing a size relationship between the particle fitness of each particle in the particle swarm and the particle fitness of the optimal particle in the particle swarm, and for each particle in the particle swarm, if the particle fitness of the particle is greater than the particle fitness of the optimal particle in the particle swarm, the particle fitness of the particle is determined as the particle fitness of the optimal particle in the particle swarm; after the traversal, if the particle fitness of the optimal particle in the particle swarm is greater than the particle fitness of the optimal particle in the particle swarm in the last iteration, the particle fitness of the optimal particle in the particle swarm in the current iteration is determined as the particle fitness of the optimal particle in the particle swarm in the current iteration, if the particle fitness of the optimal particle in the particle swarm is not greater than the particle fitness of the optimal particle in the particle swarm in the last iteration, the particle fitness of the optimal particle in the particle swarm in the last iteration is determined as the particle fitness of the optimal particle in the particle swarm in the current iteration; if the particle fitness in the current iteration does not meet a preset fitness condition, the model weight information in the current iteration is adjusted, and the particle swarm construction step is executed again; if the particle fitness in the current iteration meets the preset fitness condition, the model weight information in the current iteration is determined as the target model weight information.

[0075] As an embodiment, the adjustment of the model weight information in the current iteration comprises: obtaining a random number in a preset value range, and adjusting the model weight information in the current iteration according to the random number.

[0076] Further, the adjustment of the model weight information in the current iteration according to the random number comprises: obtaining an iteration design parameter, and determining global particle information, wherein the iteration design parameter is a parameter set on demand, and can also be a shrinkage expansion coefficient; and adjusting the model weight information in the current iteration according to the global particle information, the random number and the iteration design parameter.

[0077] As an embodiment, the determination of the global particle information comprises: determining an average value of the optimal particles in the particle swarm in all iterations as the global particle information.

[0078] Further, the determination of the global particle information comprises: determining a ratio between the sum of the optimal particles in the particle swarm in all iterations and the number of iterations as the global particle information.

[0079] Wherein, the value of the particle is a corresponding weight value, for example, is the value of a particle in the first particle swarm.

[0080] Optionally, the ratio between the sum of the optimal particles in the particle swarm in all iterations and the number of iterations is determined as the global particle information, which can be expressed by a formula as follows:

[0081]

[0082] wherein, is global particle information, is iteration number, is the optimal particle in the particle swarm of the th iteration.

[0083] As an embodiment, the model weight information of the current iteration is adjusted according to the global particle information, the random number and the iteration design parameter, including: the random number includes a first random number and a second random number satisfying a standard normal distribution, and a third random number satisfying a uniform probability distribution, the local particle information is determined according to the first random number, the second random number and the particle of the current iteration; the particle adjustment information is determined according to the local particle information, the global particle information, the iteration design parameter, the first random number and the third random number; the model weight information of the current iteration is adjusted according to the particle adjustment information.

[0084] As an embodiment, the local particle information is determined according to the first random number, the second random number and the particle of the current iteration, including: the product between the first random number and a particle of the current iteration is determined as a first parameter, the product between the second random number and the optimal particle of the current iteration is determined as a second parameter, the sum between the first random number and the second random number is determined as a third parameter, and the ratio between the sum of the first parameter and the second parameter and the third parameter is determined as the local particle information.

[0085] Optionally, the ratio between the sum of the first parameter and the second parameter and the third parameter is determined as the local particle information, which can be expressed by a formula as follows:

[0086]

[0087] wherein, is local particle information, is the first random number, is a particle of the current iteration, is the second random number, is the optimal particle of the current iteration.

[0088] As an embodiment, the particle adjustment information is determined according to the local particle information, the global particle information, the iteration design parameter, the first random number and the third random number, including: in a case where the third random number is greater than or equal to a preset value, mapping the local particle information, the global particle information, the iteration design parameter, the first random number, model weight information of a current iteration and the third random number into the particle adjustment information by a first mapping function; in a case where the third random number is less than the preset value, mapping the local particle information, the global particle information, the iteration design parameter, the first random number, the model weight information of the current iteration and the third random number into the particle adjustment information by a second mapping function, wherein the preset value can be 0.5, or can be other values, which are not limited herein.

[0089] Optionally, in a case where the preset value is 0.5, the first mapping function and the second mapping function can be expressed by a formula as follows:

[0090]

[0091] wherein, is the particle adjustment information, is the iteration design parameter, is the model weight information of the current iteration, is the third random number.

[0092] As an embodiment, the model weight information of the current iteration is adjusted according to the particle adjustment information, including: determining the particle adjustment information as information after adjusting the model weight information of the current iteration.

[0093] In this way, the model weight information is adjusted by the Gaussian quantum behavior, the random sequence is replaced by the Gaussian mutation operator, and the particle mutation and iteration optimization are guided, wherein the particle adjustment is iterated by the wave function, so that the model weight information is prevented from converging to a local optimum too early in the particle swarm algorithm, and the model prediction accuracy of the target power risk assessment model under the target model weight information obtained by adjustment is ensured.

[0094] In the model weight information updating method, the model weight information of the plurality of model parameters in the pre-trained power risk assessment model is obtained; the model weight information of the plurality of model parameters is feature coded to obtain model weight coding features; a particle swarm construction step is performed: an initialized particle swarm is constructed according to the model weight coding features; the fitness of each particle in the initialized particle swarm is evaluated to obtain particle fitness, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the power risk assessment model; the model weight information is iteratively updated according to the particle fitness to obtain target model weight information, the model prediction effect of the power risk assessment model is refined to the prediction accuracy of each model structure, and the particle fitness representing the prediction accuracy of the model structure is used as the basis for updating the model weight information, so that in the case that the power risk assessment accuracy of the trained power risk assessment model is low, the power risk assessment accuracy of the power risk assessment model using the target model weight information as the basis for updating the model weight information can be improved by iteratively updating the model weight information.

[0095] In one exemplary embodiment, as shown in Figure 3 After the target model weight information is obtained by iteratively optimizing the model weight information according to the weight fusion information through the particle swarm algorithm, steps 302 to 306 are included. Among them:

[0096] Step 302: Adjusting the power risk assessment model according to the target model weight information to obtain a target power risk assessment model.

[0097] Exemplarily, step 302 includes: adjusting the weight of the output parameter of the corresponding model structure in the power risk assessment model according to the target model weight information to obtain the target power risk assessment model.

[0098] Step 304: If the power operation information of the power operation to be evaluated is detected, the power operation information is feature extracted through the target power risk assessment model to obtain power operation features.

[0099] The power operation to be evaluated in step 304 is a power operation waiting for risk assessment, which can be set by a user as needed, or in response to a preset condition. Further, the preset condition can be a preset interval time. In each preset interval time, the power operation to be evaluated is randomly selected from unassessed power operations. Alternatively, the power operation to be evaluated is determined as a power operation that has been separated from the completion time of the last power operation risk assessment by a preset interval time. The preset condition can also be the operation type of the power operation. The unassessed power operation whose operation type represents a high risk coefficient higher than a preset risk coefficient threshold is determined as the power operation to be evaluated. The preset risk coefficient threshold can be set by a user as needed or can be an experience value, which is not limited herein.

[0100] The power operation information in step 304 includes power operation data of at least one of a personnel information category, an operation construction category, an operation environment category, and an operation management system category. The power operation data belonging to the personnel information category includes, but is not limited to, the number of production personnel, the number of safety monitoring personnel, the technical level of personnel, the safety awareness of personnel, and the health level of personnel. The power operation data belonging to the operation construction category includes, but is not limited to, the power supply area, the proportion of insulated lines, the power grid automation coverage rate, safety protection tools, and operation monitoring videos. The power operation data belonging to the operation environment category includes, but is not limited to, operation environment terrain information, operation weather information, operation area dryness information, and operation area natural disaster information. The power operation data belonging to the operation management system category includes, but is not limited to, risk management system, risk management agency information, risk management measures, violation management mechanism, and operation risk assessment mechanism.

[0101] For example, step 304 includes: if the power operation information of the power operation to be evaluated is detected, performing feature extraction on the power operation information through the feature extraction layer in the target power risk assessment model to obtain power operation features.

[0102] As an example, the feature extraction on the power operation information through the feature extraction layer in the target power risk assessment model to obtain power operation features includes: the power operation information includes a plurality of power operation data belonging to at least one category that changes with time sequence; for the power operation data belonging to each category, obtaining influence information of the power operation data on the power safety risk assessment of the power operation to be evaluated; performing normalization processing on the power operation data according to the influence information of the power operation data on the power safety risk assessment of the power operation to be evaluated to obtain power operation normalized data; and performing feature extraction on the power operation normalized data through the feature extraction layer in the target power risk assessment model to obtain power operation features.

[0103] The influence information of the power operation data on the power safety risk assessment of the to-be-evaluated power operation represents a positive influence state or a negative influence state of the power operation data on the power safety risk assessment of the to-be-evaluated power operation. When the power operation data on the power safety risk assessment of the to-be-evaluated power operation is in the positive influence state, it is indicated that the greater the power operation data is, the lower the power safety risk of the to-be-evaluated power operation is. When the power operation data on the power safety risk assessment of the to-be-evaluated power operation is in the negative influence state, it is indicated that the greater the power operation data is, the higher the power safety risk of the to-be-evaluated power operation is.

[0104] Further, the power operation data is normalized according to the influence information of the power operation data on the power safety risk assessment of the to-be-evaluated power operation to obtain power operation normalized data, including: obtaining a data maximum value and a data minimum value in each data of the same category as the power operation data, and performing normalization on the power operation data according to the influence state represented by the influence information of the power operation data on the power safety risk assessment of the to-be-evaluated power operation, the data maximum value and the data minimum value, to obtain the power operation normalized data.

[0105] As an embodiment, the power operation data is normalized according to the influence state represented by the influence information of the power operation data on the power safety risk assessment of the to-be-evaluated power operation, the data maximum value and the data minimum value, to obtain the power operation normalized data, including: if the influence state represented by the influence information of the power operation data on the power safety risk assessment of the to-be-evaluated power operation is a positive influence state, then the ratio of the difference between the power operation data and the data minimum value to the difference between the data maximum value and the data minimum value is determined as the power operation normalized data; if the influence state represented by the influence information of the power operation data on the power safety risk assessment of the to-be-evaluated power operation is a negative influence state, then the ratio of the difference between the data maximum value and the power operation data to the difference between the data maximum value and the data minimum value is determined as the power operation normalized data.

[0106] Optionally, the ratio of the difference between the power operation data and the data minimum value to the difference between the data maximum value and the data minimum value is determined as the power operation normalized data, which can be expressed by a formula as follows:

[0107]

[0108] wherein, the power operation normalized data is, the power operation data is, the data minimum value is, the data maximum value is.

[0109] Optionally, a ratio of a difference between the data maximum value and the power operation data, and a difference between the data maximum value and the data minimum value, is determined as the power operation normalized data, which can be expressed by a formula as follows:

[0110]

[0111] In this way, the power operation data is normalized according to the influence state represented by the influence information of the power safety risk assessment of the to-be-evaluated power operation, so that the data processing scale of the target power risk assessment model can be ensured to be small, that is, the power risk assessment efficiency of the model is improved, and the power operation normalized data after the normalization processing can represent the influence characteristics of the power operation data on the power safety risk assessment, so that the power risk assessment accuracy of the model is improved. In summary, the power risk assessment efficiency and the power risk assessment accuracy of the model are considered.

[0112] In step 306, the power safety risk of the to-be-evaluated power operation is evaluated according to the power operation characteristics, and a power risk assessment result is obtained.

[0113] For example, step 306 includes mapping the power operation characteristics to the power risk assessment result by the full connection layer in the target power risk assessment model.

[0114] As an embodiment, referring to Figure 4 The target power risk assessment model includes a convolution layer, a pooling layer, and a full connection layer as shown, the power operation information of the to-be-evaluated power operation is taken as the input of the target power risk assessment model, the power operation information is sequentially processed by the convolution layer and the pooling layer of the target power risk assessment model, and the power operation characteristics are obtained, and the power operation characteristics are mapped to the power risk assessment result by the full connection layer.

[0115] In this embodiment, the target power risk assessment model is obtained by adjusting the power risk assessment model according to the target model weight information; if the power operation information of the to-be-evaluated power operation is detected, the power operation information is feature extracted by the target power risk assessment model, and the power operation characteristics are obtained; the power safety risk of the to-be-evaluated power operation is evaluated according to the power operation characteristics, and the power risk assessment result is obtained; the power safety risk of the to-be-evaluated power operation is evaluated by the target power risk assessment model after the weight update according to the power operation information of the to-be-evaluated power operation, and the accuracy of the power safety risk assessment is improved.

[0116] As a detailed embodiment, model weight information of a plurality of model parameters in a pre-trained power risk assessment model is acquired; the model weight information of the plurality of model parameters is feature-encoded to obtain model weight encoding features; a particle swarm construction step is executed: an initialized particle swarm is constructed according to the model weight encoding features; for each particle in the initialized particle swarm, the fitness of the particle is evaluated according to the difference between the output data of the model structure in the power risk assessment model corresponding to the particle and the corresponding true data, to obtain the particle fitness; if the particle fitness of the current iteration does not meet a preset fitness condition, a random number in a preset value range is acquired, an iteration design parameter is acquired, and global particle information is determined; the model weight information of the current iteration is adjusted according to the global particle information, the random number and the iteration design parameter, and the particle swarm construction step is returned to be executed; if the particle fitness of the current iteration meets the preset fitness condition, the model weight information of the current iteration is determined as target model weight information.

[0117] Further, the target power risk assessment model is adjusted according to the target model weight information; for the power operation data belonging to each category, influence information of the power operation data on the power safety risk assessment of the to-be-evaluated power operation is acquired; the power operation data is normalized according to the influence information of the power operation data on the power safety risk assessment of the to-be-evaluated power operation, to obtain power operation normalized data; the power operation normalized data is feature-extracted through the target power risk assessment model, to obtain power operation features; the power safety risk of the to-be-evaluated power operation is evaluated according to the power operation features, to obtain a power risk assessment result.

[0118] In this way, by acquiring model weight information of a plurality of model parameters in a pre-trained power risk assessment model; feature-encoding the model weight information of the plurality of model parameters to obtain model weight encoding features; executing a particle swarm construction step: constructing an initialized particle swarm according to the model weight encoding features; evaluating the fitness of each particle in the initialized particle swarm to obtain particle fitness, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the power risk assessment model; iteratively updating the model weight information according to the particle fitness to obtain target model weight information, refining the model prediction effect of the power risk assessment model to the prediction accuracy of each model structure, taking the particle fitness representing the prediction accuracy of the model structure as the basis for updating the model weight information, thereby improving the power risk assessment accuracy of the power risk assessment model taking the target model weight information as the basis for updating the model weight information, in the case that the power risk assessment accuracy of the power risk assessment model obtained by training is low.

[0119] Further, the model weight information is adjusted by Gaussian quantum behavior, a Gaussian mutation operator is used to replace a random sequence to guide particle mutation and iterative optimization, the particle adjustment is iterated by using a wave function, so that the model weight information is prevented from converging to a local optimum too early in the particle swarm optimization, and therefore the model prediction accuracy of the target power risk assessment model obtained by adjusting the target model weight information is ensured.

[0120] It should be understood that, although each step in the flowchart involved in the above embodiments is shown in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in the above embodiments can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or steps or stages in other steps.

[0121] Based on the same inventive concept, the embodiments of the present application also provide a model weight information updating device for implementing the above-mentioned model weight information updating method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more model weight information updating device embodiments provided below can refer to the limitations of the model weight information updating method in the above text, which will not be repeated here.

[0122] In one exemplary embodiment, as shown in Figure 5 A model weight information updating device 500 is provided, comprising: an acquisition module 502, an encoding module 504, a construction module 506, an evaluation module 508 and an updating module 510, wherein:

[0123] The acquisition module 502 is configured to acquire model weight information of a plurality of model parameters in a pre-trained power risk assessment model;

[0124] The encoding module 504 is configured to perform feature encoding on the model weight information of the plurality of model parameters to obtain model weight encoding features;

[0125] The construction module 506 is configured to perform a particle swarm construction step: constructing an initialized particle swarm according to the model weight encoding features;

[0126] The evaluation module 508 is configured to evaluate the fitness of each particle in the initialized particle swarm, to obtain particle fitness, where the particle fitness is used to represent the prediction accuracy of the model structure in the power risk assessment model.

[0127] The updating module 510 is configured to update the model weight information iteratively according to the particle fitness, to obtain target model weight information.

[0128] In one of the embodiments, the evaluation module 508 is further configured to, for each particle in the initialized particle swarm, evaluate the fitness of the particle according to the difference between the output data of the model structure in the power risk assessment model corresponding to the particle and the corresponding real data, to obtain the particle fitness of the particle.

[0129] In one of the embodiments, the updating module 510 is further configured to, if the particle fitness of the current iteration does not meet a preset fitness condition, obtain a random number in a preset value range, adjust the model weight information of the current iteration according to the random number, and return to perform the particle swarm construction step; if the particle fitness of the current iteration meets the preset fitness condition, determine the model weight information of the current iteration as the target model weight information.

[0130] In one of the embodiments, the updating module 510 is further configured to obtain an iteration design parameter, and determine global particle information; and adjust the model weight information of the current iteration according to the global particle information, the random number and the iteration design parameter.

[0131] In one of the embodiments, after the model weight information is iteratively optimized by the particle swarm algorithm according to the weight fusion information to obtain the target model weight information, the device further comprises an adjusting module configured to adjust the power risk assessment model according to the target model weight information, to obtain a target power risk assessment model; if the power operation information of the power operation to be evaluated is detected, extract features of the power operation information by using the target power risk assessment model, to obtain power operation features; and evaluate the power safety risk of the power operation to be evaluated according to the power operation features, to obtain a power risk assessment result.

[0132] In one of the embodiments, the power operation information comprises a plurality of power operation data belonging to at least one category and changing with time sequence; the adjusting module is further configured to, for the power operation data belonging to each category, obtain influence information of the power operation data on the power safety risk assessment of the power operation to be evaluated; normalize the power operation data according to the influence information of the power operation data on the power safety risk assessment of the power operation to be evaluated, to obtain power operation normalized data; and extract features of the power operation normalized data by using the target power risk assessment model, to obtain the power operation features.

[0133] Each module in the model weight information updating apparatus can be implemented by software, hardware, or a combination thereof. The modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in a computer device in software form, so as to be invoked and executed by the processor to perform operations corresponding to the modules.

[0134] In an example embodiment, a computer device, which can be a terminal, is provided, and an internal structure diagram of the computer device can be as shown in Figure 6 The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to perform wired or wireless communication with external terminals. The wireless communication can be achieved through WIFI, mobile cellular network, near field communication (NFC), or other technologies. The computer program is executed by the processor to implement a model weight information updating method. The display unit of the computer device is configured to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball, or touchpad arranged on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0135] Those skilled in the art can understand that Figure 6 The structure shown in the above description is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. Specifically, the computer device can include more or fewer components than those shown in the diagram, or combine certain components, or have a different arrangement of components.

[0136] In an example embodiment, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0137] In an embodiment, a computer readable storage medium is provided, having stored thereon a computer program, which, when executed by a processor, implements the steps of any of the above method embodiments.

[0138] In an embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the steps of any of the above method embodiments.

[0139] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the computer program can include the processes of the above-mentioned embodiments. Any reference to a memory, a database or other medium used in the embodiments provided in the present application can include at least one of a non-volatile memory and a volatile memory. The non-volatile memory can include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical storage, a high-density embedded non-volatile memory, a resistive random access memory (ReRAM), a magnetoresistive random access memory (MRAM), a ferroelectric random access memory (FRAM), a phase change memory (PCM), a graphene memory, etc. The volatile memory can include a random access memory (RAM) or an external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0140] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, any combination of these technical features is deemed to be within the scope of the present application.

[0141] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for updating model weight information, characterized in that, The method comprises: obtaining model weight information of a plurality of model parameters in a pre-trained power risk assessment model; characteristic coding is performed on the model weight information of the plurality of model parameters to obtain model weight coding features, wherein the model weight coding features comprise a weight coding matrix, and parameters in the weight coding matrix represent the weight of each input data corresponding to each model structure in each particle swarm with respect to each output data, the model structure comprising at least one of a convolutional layer and a pooling layer; performing a particle swarm construction step: constructing an initialized particle swarm according to the model weight coding features; for each particle in the initialized particle swarm, evaluating the fitness of the particle according to the difference between the output data of the model structure corresponding to the particle in the power risk assessment model and the corresponding true data, to obtain the particle fitness, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the power risk assessment model; if the particle fitness of the current iteration does not meet the preset fitness condition, obtaining a random number in a preset value range, obtaining an iteration design parameter, and determining global particle information; the random number comprises a first random number and a second random number satisfying a standard normal distribution, and a third random number satisfying a uniform probability distribution, the local particle information is determined according to the first random number, the second random number and the particle of the current iteration; in the case that the third random number is greater than or equal to a preset value, the local particle information, the global particle information, the iteration design parameter, the first random number, the model weight information of the current iteration and the third random number are mapped into particle adjustment information through a first mapping function; in the case that the third random number is less than the preset value, the local particle information, the global particle information, the iteration design parameter, the first random number, the model weight information of the current iteration and the third random number are mapped into particle adjustment information through a second mapping function, wherein the first mapping function and the second mapping function are wave functions; the model weight information of the current iteration is adjusted according to the particle adjustment information, and the particle swarm construction step is returned to be executed; if the particle fitness of the current iteration meets the preset fitness condition, the model weight information of the current iteration is determined as target model weight information.

2. The method of claim 1, wherein, After the target model weight information is obtained by iteratively updating the model weight information according to the particle fitness, the method further comprises: adjusting the power risk assessment model according to the target model weight information to obtain a target power risk assessment model; if the power operation information of the power operation to be evaluated is detected, the power operation information is subjected to feature extraction through the target power risk assessment model to obtain power operation features; the power safety risk of the power operation to be evaluated is evaluated according to the power operation features to obtain a power risk assessment result.

3. The method of claim 2, wherein, The power operation information comprises a plurality of power operation data belonging to at least one category that changes over time. The power operation information is subjected to feature extraction through the target power risk assessment model to obtain power operation features, including: For power operation data belonging to each category, obtain influence information of the power operation data on power safety risk assessment of the to-be-evaluated power operation; According to the influence information of the power operation data on power safety risk assessment of the to-be-evaluated power operation, the power operation data is subjected to normalization processing to obtain power operation normalized data; The power operation normalized data is subjected to feature extraction through the target power risk assessment model to obtain power operation features.

4. A model weight information updating apparatus characterized by comprising: The device comprises: An acquisition module is configured to acquire model weight information of a plurality of model parameters in a pre-trained power risk assessment model, wherein the model weight information is used to represent the weight of each input data corresponding to each output data of a model structure belonging to each particle swarm species, and the model structure comprises at least one of a convolutional layer and a pooling layer. An encoding module is configured to perform feature encoding on the model weight information of the plurality of model parameters to obtain model weight encoding features, wherein the model weight encoding features comprise a weight encoding matrix, and parameters in the weight encoding matrix represent the weight of each input data corresponding to each output data of a model structure belonging to each particle swarm species, and the model structure comprises at least one of a convolutional layer and a pooling layer. A construction module is configured to perform a particle swarm construction step: constructing an initialized particle swarm according to the model weight encoding features. An evaluation module is configured to, for each particle in the initialized particle swarm, evaluate the fitness of the particle according to the difference between the output data of the model structure corresponding to the particle in the power risk assessment model and the corresponding true data, to obtain the particle fitness of the particle, wherein the particle fitness is used to represent the prediction accuracy of each model structure in the power risk assessment model. The updating module is configured to: if the particle fitness of the current iteration does not meet the preset fitness condition, obtain a random number in a preset value range, obtain an iteration design parameter, and determine global particle information; the random number includes a first random number satisfying a standard normal distribution, a second random number, and a third random number satisfying a uniform probability distribution; the local particle information is determined according to the first random number, the second random number, and the particle of the current iteration; in a case where the third random number is greater than or equal to a preset value, the local particle information, the global particle information, the iteration design parameter, the first random number, model weight information of the current iteration, and the third random number are mapped into particle adjustment information by using a first mapping function; in a case where the third random number is less than the preset value, the local particle information, the global particle information, the iteration design parameter, the first random number, model weight information of the current iteration, and the third random number are mapped into particle adjustment information by using a second mapping function, wherein the first mapping function and the second mapping function are wave functions; the model weight information of the current iteration is adjusted according to the particle adjustment information, and the particle swarm construction step is returned to be executed; if the particle fitness of the current iteration meets the preset fitness condition, the model weight information of the current iteration is determined as target model weight information.

5. The apparatus of claim 4, wherein, After the target model weight information is obtained by iteratively optimizing the model weight information according to the weight fusion information by using the particle swarm algorithm, the device further includes an adjusting module configured to adjust the power risk assessment model according to the target model weight information to obtain a target power risk assessment model; if power operation information of a power operation to be assessed is detected, the power operation information is subjected to feature extraction by using the target power risk assessment model to obtain power operation features; and the power safety risk of the power operation to be assessed is assessed according to the power operation features to obtain a power risk assessment result.

6. The apparatus of claim 5, wherein, The power operation information includes a plurality of power operation data belonging to at least one category and varying with time sequences; the adjusting module is further configured to, for the power operation data belonging to each category, obtain influence information of the power operation data on the power safety risk assessment of the power operation to be assessed; The power operation data is subjected to normalization processing according to the influence information of the power operation data on the power safety risk assessment of the power operation to be assessed to obtain power operation normalized data; and the power operation normalized data is subjected to feature extraction by using the target power risk assessment model to obtain power operation features. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6 when the computer program is executed by the processor. The processor implements the steps of the method of any one of claims 1 to 3 when executing the computer program.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 3.

9. A computer program product comprising a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 3. The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 3.

Images