Multi-dimensional operation risk assessment method based on big data analysis

Through big data analysis and artificial intelligence algorithms, a multi-dimensional operation risk assessment model is built, which solves the problem that traditional methods cannot effectively evaluate the risks of complex operations, achieves more accurate and timely risk identification and response, and improves operation safety.

CN119047855AInactive Publication Date: 2024-11-29STATE GRID GANSU ELECTRIC POWER CORP

Patent Information

Application Number
CN202411548784.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-01
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional operation risk assessment methods rely on limited experience and qualitative judgments, and cannot fully explore the interactions and dynamic changes between multi-dimensional factors, making it difficult to accurately identify and evaluate complex and diverse operation risks.

Method used

Using a multi-dimensional operation risk assessment method based on big data analysis, through data integration, cleaning and feature extraction, a multi-dimensional operation risk assessment model is constructed using improved support vector machine algorithm and iForest-RBF hybrid model, and automatically formulated through artificial intelligence algorithms. Control measures.

Benefits of technology

It improves the accuracy and timeliness of operation risk assessment, can more effectively identify and respond to emerging risks, and improves operation safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119047855A_ABST
    Figure CN119047855A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-dimensional operation risk assessment method based on big data analysis, and the method comprises the following steps: 1, fusing multi-source heterogeneous data into a unified data set through a data integration unit, and transmitting the data set to an industrial service terminal, 2, performing data cleaning, feature extraction, missing value processing, abnormal value processing and standardization operation on the data set; 3, the industrial service terminal classifies the processed data set through an improved support vector machine algorithm, and the classified data serve as a final set to be stored in a database, and 4, a multi-dimensional operation risk assessment model is established through an iForest-RBF hybrid model. And 5, automatically formulating management and control measures and optimizing a supervision strategy according to the comprehensive risk assessment result of the operation by using an artificial intelligence algorithm.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of network security perception and early warning analysis, and more specifically to a multi-dimensional operational risk assessment method based on big data analysis. Background Art

[0002] Traditional operational risk assessment methods often rely on limited experience, qualitative judgment or simple statistical analysis, and have many limitations. For example, experience-based assessment methods are highly subjective and easily limited by the knowledge and experience of the assessors; qualitative judgments often lack quantitative standards and precision; and simple statistical analysis may not be able to fully explore the hidden patterns and complex relationships in the data; in addition, the influencing factors of operational risk are multidimensional, including but not limited to the type of operation, the operating environment, the skill level of personnel, the condition of equipment, and management measures. Traditional methods are often unable to comprehensively consider these multidimensional factors and find it difficult to accurately capture the interactions and dynamic changes between them; at the same time, with the expansion of business and the increasing complexity of work processes, the manifestations of operational risks have become more diverse and complex. Traditional assessment methods are stretched to the limit in dealing with these challenges and are unable to identify and assess emerging risks in a timely and effective manner. To solve the above problems, a multidimensional operational risk assessment method based on big data analysis has emerged. This method evaluates various operational risks by constructing a multidimensional operational risk assessment model, and formulates measures and optimization strategies based on the assessment results. Summary of the invention

[0003] In view of the shortcomings of the above technologies, the present invention discloses a multi-dimensional operation risk assessment method based on big data analysis. The present invention adopts the following technical solutions: The multi-dimensional operation risk assessment method based on big data analysis includes the following steps: Step 1: The multi-source heterogeneous data are integrated into a unified data set through a data integration unit, and the data set is transmitted to the industrial service terminal. The multi-source heterogeneous data includes operation safety risk rating data, personnel big data information, power grid one-picture environmental meteorological information and equipment operation status data; Step 2: Perform data cleaning, feature extraction, missing value processing, outlier processing and standardization operations on the data set at the industrial service terminal; Step 3: The industrial service terminal classifies the data set processed in step 2 by using an improved support vector machine algorithm, and the classified data is stored in a database as a final set. The improved support vector machine algorithm runs in the industrial service terminal; Step 4: Establish a multi-dimensional operation risk assessment model through the iForest-RBF hybrid model. The final set of the industrial service terminal calls the database as the input variable of the multi-dimensional operation risk assessment model, and the comprehensive risk assessment result of the operation is used as the output variable; The iForest-RBF hybrid model includes an improved isolation forest iForest module, a feature fusion module, an RBF neural network module, a model training module, a data update module and a model evaluation module; the improved isolation forest iForest module uses an isolation forest algorithm based on the seagull algorithm to detect abnormal samples by calculating the abnormal score of each sample and comparing it with a threshold; the feature fusion module is used to fuse the abnormal sample with the processed feature sample as the input of the RBF neural network module; the RBF neural network module is used to output a risk assessment result; the model training module trains the RBF neural network through the data of the final set; and the data update module updates and optimizes the multidimensional The parameters of the multi-dimensional operation risk assessment model are at least weights, sample data standard deviations and learning rates. The model evaluation module uses the final set as a test set to evaluate the multi-dimensional operation risk assessment model; the output end of the improved isolation forest iForest module is connected to the input end of the feature fusion module, the output end of the feature fusion module is connected to the input end of the RBF neural network module, the output end of the RBF neural network module is connected to the input end of the model training module, and the output end of the model training module is respectively connected to the input end of the data update module and the input end of the model evaluation module; the output end of the data update module is connected to the input end of the RBF neural network module; Step 5: Utilize artificial intelligence algorithms to automatically formulate control measures and optimize inspection strategies based on the comprehensive risk assessment results of the operations; As a further technical solution of the present invention, the data integration unit includes a first data collection unit, a second data collection unit, a third data collection unit and a fourth data collection unit. The first data collection unit collects risk grading information of each professional operation through the internal risk management system of the enterprise. The risk grading information includes operation type, risk level and accident type. The operation type includes at least electrical operation, climbing operation, welding operation, lifting operation and excavation operation. The risk level is divided into a first risk level, a second risk level and a third risk level. The accident type includes at least electric shock accidents, high-altitude falling accidents, fire and explosion accidents, lifting injury accidents and collapse accidents. The second data collection unit obtains the personnel big data information through the personnel information management system. The personnel big data information includes at least the working years, violations and operation frequency of the team members; the third data collection unit obtains the environmental meteorological data of the operation area through the power grid geographic information system, and the environmental meteorological data includes at least temperature, humidity, wind speed, rainfall and lightning activity; the fourth data collection unit collects equipment operation status data through a PLC programmable logic controller.

[0004] As a further technical solution of the present invention, the risk levels are set as follows: the first risk level interval is 0-0.3, the second risk level interval is 0.3-0.6, and the third risk level interval is above 0.6.

[0005] As a further technical solution of the present invention, the working principle of the improved support vector machine algorithm is: (1) using the data set as a training set for the improved support vector machine algorithm; (2) Selecting Gaussian function as the kernel function of the support vector machine model; (3) training the support vector machine model using the training set; (4) Calculating the parameters of the support vector machine model by Lagrange multiplier method to determine the hyperplane, the parameters of the support vector machine model include penalty parameters, slack variables, energy parameters, a normal vector representing the direction of the hyperplane, and a displacement deviation term; (5) Classifying new data sets using the trained support vector machine model.

[0006] As a further technical solution of the present invention, the working principle of the improved isolation forest iForest module is: (1) Initialize the seagull population and the positions of the seagulls in the search space; (2) The seagull algorithm searches for the optimal segmentation value through its unique migration and predation behavior. The segmentation value for constructing the isolation tree is represented by the updated position of the seagull. The calculation formula for the position of the seagull is: ,in, Update the position for the seagull, P s ( t ) is the initial position of the seagull, l is the weight, A is the nonlinear convergence factor, p is the learning rate, t is the number of iterations, r is a random number between 0 and 1. P bs ( t ) is an individual seagull, i is an angle value within [0-2π], u and v is a constant defining the shape of the spiral, e is a natural mathematical constant, which is set to u =2, v =1; (3) The final set is a four-dimensional data set, where the four dimensions mean that the data has four characteristics: x1. x 2、 x 3 and x 4; Ten samples are extracted from the final set as sub-sampling sets, and for each feature of the sub-sampling set x 1. Contrast x 1 eigenvalue and the split value, put the records with eigenvalues ​​smaller than the split value in the left subtree, and vice versa in the right subtree, recursively repeat this process on each sub-sampling set, and build an isolation tree until the following stopping condition is met, the stopping condition is that the selected eigenvalues ​​of all samples are the same value or the number of samples in the node reaches 200; (4) Repeat the operation of (3), compare each feature with the segmentation value and record it to construct an isolation forest; (5) Calculate the anomaly score for each sample: , where s() is the abnormal score of each sample, i The sample points selected for anomaly score calculation, n is the sample size, for , H () is a harmonic number, h ( i ) is the path length of the sample point, and the path length is the number of edges required to reach the leaf node of the sample from the root node of the isolation tree, e ( h ( i )) is the average path length of the sample point in all trees; (6) The calculated anomaly score of each sample is compared with a threshold, and anomaly scores within the threshold range are determined as abnormal samples. The threshold is set to 0.4-0.6.

[0007] As a further technical solution of the present invention, the working principle of the RBF neural network module is: (1) Initialize the parameters of the RBF neural network, set the input layer, hidden layer and inference layer to contain 4 neurons each, and the output layer to contain 3 neurons. The 4 neurons corresponding to the input layer represent the 4 factors affecting the operation risk, and the 3 neurons in the output layer correspond to the 3 risk levels; (2) The data extracted from the final set is used as the node of the input layer, and the data is passed to the hidden layer; (3) The hidden layer performs nonlinear mapping on the output of the input layer, and the formula is: ,in, F ( ) is the radial basis function, l is the weight, is the vector of the sample's transverse coordinates, is the vector of sample longitudinal coordinates, is the vector of sample data attribute dimensions, n represents the sample size, s is the standard deviation of the sample data; (4) The inference layer further integrates the output of the hidden layer, and the output of each node of the inference layer is the product of the corresponding inputs of all nodes in the layer; (5) The output layer is a weighted sum of the data output by the inference layer, and the formula is: ,in, Y is the output of the output layer, and the output corresponds to three neurons, corresponding to three risk levels respectively. q is the number of neural units in the inference layer, l is the number of nodes, oh l For the l The connection weights of nodes and the output layer, β l is the basis function center, R l For the l nodes with the output of the inference layer.

[0008] As a further technical solution of the present invention, the artificial intelligence algorithm includes a data receiving module, an intelligent database module, a measure generation module, a strategy optimization module and a human-computer interaction module; the data receiving module is used to receive the output data of the multidimensional operation risk assessment model, the intelligent database module generates measure information and optimization strategy information by screening the information matching the output data of the multidimensional operation risk assessment model through the intelligent database, and outputs the generated information to the measure generation module and the strategy optimization module respectively, and the human-computer interaction module intuitively displays the measures and optimization strategies to the staff in the form of charts; the output end of the data receiving module is connected to the input end of the intelligent database module, the output end of the intelligent database module is connected to the input end of the measure generation module and the strategy optimization module, and the output end of the measure generation module and the strategy optimization module is connected to the input end of the human-computer interaction module.

[0009] The beneficial effects of the present invention are as follows: Different from conventional technologies, the present invention discloses a multidimensional operation risk assessment method based on big data analysis, which classifies the collected data through an improved vector machine algorithm, and constructs a multidimensional operation risk assessment model through an iForest-RBF hybrid model to output the risk level of the operation, and formulates measures and optimization strategies for the output risk level through an artificial intelligence algorithm to improve operation safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of the overall architecture of the multi-dimensional operation risk assessment method based on big data analysis of the present invention; Figure 2 It is a schematic diagram of the iForest-RBF hybrid model architecture in the multi-dimensional operation risk assessment method based on big data analysis of the present invention; Figure 3 It is a workflow diagram of the improved isolation forest iForest module in the multi-dimensional operation risk assessment method based on big data analysis in the present invention; Figure 4 It is a network topology diagram of the RBF neural network module in the multi-dimensional operation risk assessment method based on big data analysis of the present invention; Figure 5 This is a schematic diagram of the artificial intelligence algorithm architecture in the multi-dimensional operation risk assessment method based on big data analysis of the present invention. DETAILED DESCRIPTION

[0011] The multi-dimensional operation risk assessment method based on big data analysis includes the following steps: Step 1: The multi-source heterogeneous data are integrated into a unified data set through a data integration unit, and the data set is transmitted to the industrial service terminal. The multi-source heterogeneous data includes operation safety risk rating data, personnel big data information, power grid one-picture environmental meteorological information and equipment operation status data; Step 2: Perform data cleaning, feature extraction, missing value processing, outlier processing and standardization operations on the data set at the industrial service terminal; Step 3: The industrial service terminal classifies the data set processed in step 2 by using an improved support vector machine algorithm, and the classified data is stored in a database as a final set. The improved support vector machine algorithm runs in the industrial service terminal; Step 4: Establish a multi-dimensional operation risk assessment model through the iForest-RBF hybrid model. The final set of the industrial service terminal calls the database as the input variable of the multi-dimensional operation risk assessment model, and the comprehensive risk assessment result of the operation is used as the output variable; The iForest-RBF hybrid model includes an improved isolation forest iForest module, a feature fusion module, an RBF neural network module, a model training module, a data update module and a model evaluation module; the improved isolation forest iForest module uses an isolation forest algorithm based on the seagull algorithm to detect abnormal samples by calculating the abnormal score of each sample and comparing it with a threshold; the feature fusion module is used to fuse the abnormal sample with the processed feature sample as the input of the RBF neural network module; the RBF neural network module is used to output a risk assessment result; the model training module trains the RBF neural network through the data of the final set; and the data update module updates and optimizes the multidimensional The parameters of the multi-dimensional operation risk assessment model are at least weights, sample data standard deviations and learning rates. The model evaluation module uses the final set as a test set to evaluate the multi-dimensional operation risk assessment model; the output end of the improved isolation forest iForest module is connected to the input end of the feature fusion module, the output end of the feature fusion module is connected to the input end of the RBF neural network module, the output end of the RBF neural network module is connected to the input end of the model training module, and the output end of the model training module is respectively connected to the input end of the data update module and the input end of the model evaluation module; the output end of the data update module is connected to the input end of the RBF neural network module; Step 5: Use artificial intelligence algorithms to automatically formulate control measures and optimize inspection strategies based on the comprehensive risk assessment results of the operation; In the above embodiment, the data integration unit includes a first data collection unit, a second data collection unit, a third data collection unit and a fourth data collection unit. The first data collection unit collects risk grading information of each professional operation through the enterprise's internal risk management system. The risk grading information includes operation type, risk level and accident type. The operation type includes at least electrical operation, climbing operation, welding operation, lifting operation and excavation operation. The risk level is divided into a first risk level, a second risk level and a third risk level. The accident type includes at least electric shock accidents, high-altitude falling accidents, fire and explosion accidents, lifting injury accidents and collapse accidents. The second data collection unit obtains the personnel big data information through the personnel information management system (using tabular personnel information entry), and the personnel big data information includes at least the working years, violations and operation frequency of the team members; the third data collection unit obtains the environmental meteorological data of the operation area through the power grid geographic information system, and the environmental meteorological data includes at least temperature, humidity, wind speed, rainfall and lightning activity; the fourth data collection unit collects equipment operation status data through the PLC programmable logic controller.

[0012] In the above technical solution, An enterprise's internal risk management system refers to a set of frameworks and procedures designed to identify, assess, monitor and respond to various risks within an enterprise. The following are the main components and functions of an enterprise's internal risk management system: 1. Risk identification Internal Audit: The internal audit department reviews the company's business processes and identifies potential risks.

[0013] Employee feedback: Employees are encouraged to report possible risks, including operational risks, compliance risks, ethical risks, etc.

[0014] Process analysis: Conduct detailed analysis of the company’s business processes and identify risk points in the processes.

[0015] 2. Risk Assessment Qualitative analysis: Qualitative assessment of the likelihood and impact of risks.

[0016] Quantitative analysis: Use mathematical models or statistical methods to quantitatively assess risks.

[0017] Risk Matrix: Create a risk matrix to prioritize risks based on their likelihood and impact.

[0018] 3. Risk monitoring Continuous monitoring: Track changes in risks through regular risk assessment and monitoring.

[0019] Key Risk Indicators (KRIs): Key risk indicators are set to measure the effectiveness of risk management.

[0020] Reporting system: Establish a risk reporting system to ensure that risk information can be delivered to management in a timely manner.

[0021] 4. Risk Response Risk avoidance: Avoiding risks that may cause significant losses to the business.

[0022] Risk reduction: Taking steps to reduce the likelihood and impact of risks.

[0023] Risk transfer: The transfer of risk to a third party through insurance or other financial instruments.

[0024] Risk acceptance: For some low-risk or unavoidable risks, the company may choose to accept them.

[0025] 5. Risk Management Strategy Develop strategies: Develop corresponding risk management strategies based on risk assessment results.

[0026] Resource allocation: Allocate necessary resources to risk management activities, including human, material and financial resources.

[0027] Training and Communication: Provide risk management training to employees and ensure they understand risk management policies and procedures.

[0028] 6. Risk management tools and techniques Risk Management Software: Use risk management software to help identify, assess, and monitor risks.

[0029] Data analytics and artificial intelligence: Use data analytics tools and artificial intelligence techniques to predict and assess risks.

[0030] Compliance review: Ensure that the company's risk management practices comply with relevant laws, regulations and industry standards.

[0031] 7. Risk management culture Risk management awareness: Cultivate risk management awareness within the enterprise and make employees realize the importance of risk management.

[0032] Leadership support: Ensure senior leadership support and involvement in risk management.

[0033] Continuous Improvement: Continuously evaluate and improve risk management practices to adapt to changing circumstances.

[0034] By establishing an effective internal risk management system, companies can better identify and manage risks, thereby improving their overall performance and sustainability.

[0035] The power grid geographic information system (GIS) is a professional system that integrates geographic information system (Geographic Information System) technology and applies it to the power industry. It is a comprehensive information management system that is mainly used in the planning, design, construction, operation and maintenance of power systems. The following is the definition and main uses of the power grid GIS system: The power grid GIS system is a system that combines the geographic spatial information of the power system with attribute data and manages, analyzes and displays it through geographic information system software. It uses the spatial analysis, data management and visualization functions of GIS to provide an efficient information management tool for power companies. The system can help plan the location of new transmission lines and substations. Optimize the layout of power equipment, such as transformers, switch stations, etc. During the construction process, GIS can provide on-site geographic information to assist construction personnel in positioning and navigation. Track the location and status of equipment and plan maintenance routes. Real-time monitoring of the operating status of the power grid, including voltage, current, load, etc. In the event of a fault, quickly locate the fault point and improve the efficiency of emergency repairs.

[0036] Assess potential risks to the power grid, such as the impact of natural disasters such as floods and earthquakes. In emergency situations, GIS can provide key information to assist in emergency decision-making. Store and manage the geospatial data of the power grid, such as lines, equipment, geographic coordinates, etc. Combined with the spatial analysis function of GIS, in-depth analysis of the attribute data of the power grid is carried out. Help power companies quickly locate customers and provide more accurate services. Plan the power service network based on customer distribution. By combining the physical layout of the power system with geographic information, the power grid GIS system provides power companies with a powerful tool that helps improve the efficiency of power grid planning, construction, operation and maintenance, while enhancing the reliability and security of the power grid.

[0037] In the above embodiment, the first risk level interval is set to 0-0.3, the second risk level interval is set to 0.3-0.6, and the third risk level interval is set to above 0.6.

[0038] In the above embodiment, the working principle of the improved support vector machine algorithm is: (1) using the data set as a training set for the improved support vector machine algorithm; (2) Selecting Gaussian function as the kernel function of the support vector machine model; (3) training the support vector machine model using the training set; (4) calculating the parameters of the support vector machine model by Lagrange multiplier method to determine the hyperplane, wherein the parameters include a penalty parameter, a slack variable, an energy parameter, a normal vector representing the direction of the hyperplane, and a displacement deviation term; (5) Classifying new data sets using the trained support vector machine model.

[0039] In a specific embodiment, the Lagrange multiplier method is used to determine the support vector machine model hyperplane and construct the Lagrangian function, and the Lagrangian function formula is: ,in, is the normal vector of the hyperplane, i For the dimension, b is the displacement term of the hyperplane, α i is the Lagrange multiplier, x i for i dimensional vector, y i is the feature corresponding to the data point, c is the penalty parameter, g i is the slack variable, T represents the transposed identity, w irepresents the weight parameter, b i Indicates the update distance; The hyperplane equation is obtained by partial derivative calculation of the formula , and then classify the new data; In the above embodiment, the working principle of the improved isolation forest iForest module is: (1) Initialize the seagull population and the positions of the seagulls in the search space; (2) The seagull algorithm searches for the optimal segmentation value through its unique migration and predation behavior. The segmentation value for constructing the isolation tree is represented by the updated position of the seagull. The calculation formula for the position of the seagull is: ,in, Update the position for the seagull, P s ( t ) is the initial position of the seagull, l is the weight, A is the nonlinear convergence factor, p is the learning rate, t is the number of iterations, r is a random number between 0 and 1. P bs ( t ) is an individual seagull, i is an angle value within [0-2π], u and v is a constant defining the shape of the spiral, e is a natural mathematical constant, which is set to u =2, v =1; (3) The final set is a four-dimensional data set with four features: x 1. x 2、 x 3 and x 4; Ten samples are extracted from the final set as sub-sampling sets. For each feature of the sub-sampling set x 1. Contrast x 1 eigenvalue and the split value, put the records with eigenvalues ​​smaller than the split value in the left subtree, and vice versa in the right subtree, recursively repeat this process on each sub-sampling set, and build an isolation tree until the following stopping condition is met, the stopping condition is that the selected eigenvalues ​​of all samples are the same value or the number of samples in the node reaches 200; (4) Repeat the operation of (3), compare each feature with the segmentation value and record it to construct an isolation forest; (5) Calculate the anomaly score for each sample: , where s() is the abnormal score of each sample, i The sample points selected for anomaly score calculation, n is the sample size, for , H () is a harmonic number, h ( i ) is the path length of the sample point, and the path length is the number of edges required to reach the leaf node of the sample from the root node of the isolation tree, e ( h ( i )) is the average path length of the sample point in all trees; (6) The calculated anomaly score of each sample is compared with a threshold value, and anomaly scores within the threshold range are determined as abnormal samples. The threshold value is set to 0.4-0.6.

[0040] In a specific embodiment, the seagulls in the seagull algorithm continuously migrate and prey to find the optimal solution in the search space. The migration phase of the seagulls helps the algorithm to conduct global search to avoid falling into local optimality, and the predation phase conducts local search to further optimize the quality of the solution. In this way, the seagull optimization algorithm can balance the global exploration and local development capabilities to a certain extent to find the optimal solution to the problem. The updated position of the seagulls of the optimal solution is used as the segmentation value required to construct the isolation tree in the isolation forest algorithm, so that the multi-dimensional operation risk assessment model has higher accuracy. Then, by continuously iteratively constructing the isolation forest, the multi-dimensional operation risk assessment model can process a large amount of high-dimensional data and can still effectively identify abnormal points in high-dimensional space.

[0041] In the above embodiment, the working principle of the RBF neural network module is: (1) Initialize the parameters of the RBF neural network, set the input layer, hidden layer and inference layer, each containing 4 neurons, and the output layer containing 3 neurons. The 4 neurons corresponding to the input layer represent the 4 factors affecting the operation risk, and the 3 neurons in the output layer correspond to the 3 risk levels; (2) The data extracted from the final set is used as the node of the input layer, and the data is passed to the hidden layer; (3) The hidden layer performs nonlinear mapping on the output of the input layer, and the formula is: ,in, F () is the radial basis function, l is the weight, is the vector of the sample's transverse coordinates, is the vector of sample longitudinal coordinates, is the vector of sample data attribute dimensions, n represents the sample size, s is the standard deviation of the sample data; (4) The inference layer further integrates the output of the hidden layer, and the output of each node of the inference layer is the product of the corresponding inputs of all nodes in the layer; (5) The output layer is a weighted sum of the data output by the inference layer, and the formula is: ,in, Y is the output of the output layer, and the output corresponds to three neurons, corresponding to three risk levels respectively. q is the number of neural units in the inference layer, l is the number of nodes, oh l For the l The connection weights of nodes and the output layer, β l is the basis function center, R l For the l nodes with the output of the inference layer.

[0042] In a specific embodiment, the four factors corresponding to the four neurons in the input layer correspond to the four collection units of the data integration unit respectively, and the three risk levels corresponding to the three neurons in the output layer correspond to the set first risk level, second risk level and third risk level respectively.

[0043] In a specific embodiment, the RBF neural network model is trained by the model training module, and the connection weights, basis function centers and standard deviations of the RBF neural network model are used as training targets. According to the data of the final set, the training is performed by the spatial expression, and the formula of the spatial expression is: ,in, Q t+1 is the optimized data set of the RBF neural network model at time t+1, Q t is the optimized data set of the RBF neural network model at time t, W t+1 is the fitness of the RBF neural network model, l is the weight, and are the state transfer matrices, f is the fitness function, V t Indicates the amount of spatial index data; β Indicates the standard data volume of the optimized data set, and the parameters are trained through the calculated data.

[0044] In a specific embodiment, the working principle of the data updating module is to iteratively update the parameters of the multi-dimensional operation risk assessment model by Newton's method, and the formula is: ,in, d k is the objective function, x k+1 is the updated parameter value, x is the independent variable, is the Hessian matrix, p is the learning rate, is the inverse matrix of the Hessian matrix, x k Data attribute dimension k Update parameter value after update. g k Represents the learning rate after fusion in multiple dimensions; by calculating the updated parameter value, the parameters of the multi-dimensional operation risk assessment model are updated.

[0045] In a specific embodiment, the model evaluation module evaluates the performance of the multi-dimensional job risk assessment model through average accuracy, precision, recall and F1, and the output table is shown in Table 1: Table 1 Output table of model evaluation module

[0046] It can be seen from Table 1 that the multi-dimensional operation risk assessment model is continuously iterated, data is updated, and the more times of calculation, the stronger the performance of the model.

[0047] In the above embodiment, the artificial intelligence algorithm includes a data receiving module, an intelligent database module, a measure generation module, a strategy optimization module and a human-computer interaction module; the data receiving module is used to receive the output data of the multidimensional operation risk assessment model, the intelligent database module generates measure information and optimization strategy information by filtering the information matching the output data of the multidimensional operation risk assessment model through the intelligent database, and outputs the generated information to the measure generation module and the strategy optimization module respectively, and the human-computer interaction module intuitively displays the measures and optimization strategies to the staff in the form of charts; the output end of the data receiving module is connected to the input end of the intelligent database module, the output end of the intelligent database module is connected to the input end of the measure generation module and the strategy optimization module, and the output end of the measure generation module and the strategy optimization module is connected to the input end of the human-computer interaction module.

[0048] In a specific embodiment, the table output by the human-computer interaction module according to the risk level output by the multi-dimensional operation risk assessment model is shown in Table 2: Table 2 Output table of human-computer interaction module

[0049] Table 2 shows that different risk levels correspond to different response measures and optimized supervision strategies. These solutions show the efficiency of the multi-dimensional operation risk assessment model in processing high-dimensional data.

Claims

1. A multi-dimensional operational risk assessment method based on big data analysis, characterized by: The following steps are involved: Step 1: The multi-source heterogeneous data are integrated into a unified data set through a data integration unit, and the data set is transmitted to the industrial service terminal. The multi-source heterogeneous data includes operation safety risk rating data, personnel big data information, power grid one-picture environmental meteorological information and equipment operation status data; Step 2: Perform data cleaning, feature extraction, missing value processing, outlier processing and standardization operations on the data set at the industrial service terminal; Step 3: The industrial service terminal classifies the data set processed in step 2 by using an improved support vector machine algorithm, and the classified data is stored in a database as a final set; Step 4: Establish a multi-dimensional operation risk assessment model through the iForest-RBF hybrid model. The final set of the industrial service terminal calls the database as the input variable of the multi-dimensional operation risk assessment model, and the comprehensive risk assessment result of the operation is used as the output variable; The iForest-RBF hybrid model includes an improved isolation forest iForest module, a feature fusion module, an RBF neural network module, a model training module, a data update module and a model evaluation module; the improved isolation forest iForest module uses an isolation forest algorithm based on the seagull algorithm to detect abnormal samples by calculating the abnormal score of each sample and comparing it with a threshold; the feature fusion module is used to fuse the abnormal sample with the processed feature sample as the input of the RBF neural network module; the RBF neural network module is used to output a risk assessment result; the model training module trains the RBF neural network through the data of the final set; and the data update module updates and optimizes the multidimensional The parameters of the multi-dimensional operation risk assessment model are at least weights, sample data standard deviations and learning rates. The model evaluation module uses the final set as a test set to evaluate the multi-dimensional operation risk assessment model; the output end of the improved isolation forest iForest module is connected to the input end of the feature fusion module, the output end of the feature fusion module is connected to the input end of the RBF neural network module, the output end of the RBF neural network module is connected to the input end of the model training module, and the output end of the model training module is respectively connected to the input end of the data update module and the input end of the model evaluation module; the output end of the data update module is connected to the input end of the RBF neural network module; Step 5: Use artificial intelligence algorithms to automatically formulate control measures and optimize supervision strategies based on the comprehensive risk assessment results of the operations.

2. The multi-dimensional operation risk assessment method based on big data analysis according to claim 1 is characterized in that: The data integration unit includes a first data collection unit, a second data collection unit, a third data collection unit and a fourth data collection unit. The first data collection unit collects risk grading information of each professional operation through the enterprise's internal risk management system. The risk grading information includes operation type, risk level and accident type. The operation type includes at least electrical operation, climbing operation, welding operation, lifting operation and excavation operation. The risk level is divided into a first risk level, a second risk level and a third risk level. The accident type includes at least electric shock accidents, high-altitude falling accidents, fire and explosion accidents, lifting injury accidents and collapse accidents. The second data collection unit obtains the personnel big data information through the personnel information management system. The personnel big data information includes at least the working years, violations and operation frequency of the team members; the third data collection unit obtains the environmental meteorological data of the operation area through the power grid geographic information system. The environmental meteorological data includes at least temperature, humidity, wind speed, rainfall and lightning activity; the fourth data collection unit collects equipment operation status data through the PLC programmable logic controller.

3. The multi-dimensional operation risk assessment method based on big data analysis according to claim 2 is characterized in that: The risk levels are set as follows: the first risk level interval is 0-0.3, the second risk level interval is 0.3-0.6, and the third risk level interval is above 0.

6.

4. The multi-dimensional operation risk assessment method based on big data analysis according to claim 1 is characterized in that: The working principle of the improved support vector machine algorithm is: (1) using the data set as a training set for the improved support vector machine algorithm; (2) Selecting Gaussian function as the kernel function of the support vector machine model; (3) training the support vector machine model using the training set; (4) Calculating the parameters of the support vector machine model by Lagrange multiplier method to determine the hyperplane, the parameters of the support vector machine model include penalty parameters, slack variables, energy parameters, a normal vector representing the direction of the hyperplane, and a displacement deviation term; (5) Classifying new data sets using the trained support vector machine model.

5. The multi-dimensional operation risk assessment method based on big data analysis according to claim 1 is characterized in that: The working principle of the improved isolation forest iForest module is: (1) Initialize the seagull population and the positions of the seagulls in the search space; (2) The seagull algorithm searches for the optimal segmentation value through its unique migration and predation behavior. The segmentation value for constructing the isolation tree is represented by the updated position of the seagull. The calculation formula for the position of the seagull is: ,in, Update the position for the seagull, P s ( t ) is the initial position of the seagull, λ is the weight, A is the nonlinear convergence factor, p is the learning rate, t is the number of iterations, r is a random number between 0 and 1. P bs ( t ) is an individual seagull, θ is an angle value within [0-2π], u and v is a constant that defines the shape of the spiral, e is a natural mathematical constant, which is set to u =2, v =1; (3) The final set is a four-dimensional data set, that is, it has four features x 1. x 2、 x 3 and x 4; Ten samples are extracted from the final set as sub-sampling sets, and for each feature of the sub-sampling set x 1. Contrast x 1 eigenvalue and the split value, put the records with eigenvalues ​​smaller than the split value in the left subtree, and vice versa in the right subtree, recursively repeat this process on each sub-sampling set, and build an isolation tree until the following stopping condition is met, the stopping condition is that the selected eigenvalues ​​of all samples are the same value or the number of samples in the node reaches 200; (4) Repeat the operation of (3), compare each feature with the segmentation value and record it to construct an isolation forest; (5) Calculate the anomaly score for each sample: , where s() is the abnormal score of each sample, i The sample points selected for anomaly score calculation, n is the sample size, for , H () is a harmonic number, h ( i ) is the path length of the sample point, and the path length is the number of edges required to reach the leaf node of the sample from the root node of the isolation tree, e ( h ( i )) is the average path length of the sample point in all trees; (6) The calculated anomaly score of each sample is compared with a threshold, and anomaly scores within the threshold range are determined as abnormal samples. The threshold is set to 0.4-0.

6.

6. The multi-dimensional operation risk assessment method based on big data analysis according to claim 1 is characterized in that: The working principle of the RBF neural network module is: (1) Initialize the parameters of the RBF neural network, set the input layer, hidden layer and inference layer, each containing 4 neurons, and the output layer containing 3 neurons. The 4 neurons corresponding to the input layer represent the 4 factors affecting the operation risk, and the 3 neurons in the output layer correspond to the 3 risk levels; (2) The data extracted from the final set is used as the node of the input layer, and the data is passed to the hidden layer; (3) The hidden layer performs nonlinear mapping on the output of the input layer, and the formula is: ,in, F () is the radial basis function, λ is the weight, is the vector of the sample's transverse coordinates, is the vector of sample longitudinal coordinates, is the vector of sample data attribute dimensions, n represents the sample size, σ is the standard deviation of the sample data; The inference layer further integrates the output of the hidden layer, and the output of each node of the inference layer is the product of the corresponding inputs of all nodes in the layer; The output layer is a weighted sum of the data output by the inference layer, and the formula is: ,in, Y is the output of the output layer, and the output corresponds to three neurons, corresponding to three risk levels respectively. q is the number of neural units in the inference layer, l is the number of nodes, ω l For the l The connection weights of nodes and the output layer, β l is the basis function center, R l For the l nodes with the output of the inference layer.

7. The multi-dimensional operation risk assessment method based on big data analysis according to claim 1 is characterized in that: The artificial intelligence algorithm includes a data receiving module, an intelligent database module, a measure generation module, a strategy optimization module and a human-computer interaction module; the data receiving module is used to receive the output data of the multidimensional operation risk assessment model, the intelligent database module generates measure information and optimization strategy information by screening the information matching the output data of the multidimensional operation risk assessment model through the intelligent database, and outputs the generated information to the measure generation module and the strategy optimization module respectively, and the human-computer interaction module intuitively displays the measures and optimization strategies to the staff in the form of charts; the output end of the data receiving module is connected to the input end of the intelligent database module, the output end of the intelligent database module is connected to the input end of the measure generation module and the strategy optimization module, and the output end of the measure generation module and the strategy optimization module is connected to the input end of the human-computer interaction module.

Citation Information

Patent Citations

  • Method and device for assessing risk of electric power communication network

    CN102609778A

  • Power operation and maintenance data cleaning method based on isolation forest algorithm and neural network

    CN108776683A

  • Highway construction safety monitoring system

    CN117575334A

  • Optimization design method and system for geological storage parameters of carbon dioxide

    CN118036475A

  • Mine safety monitoring system based on AI

    CN118037047A

Cited By

  • Path planning method

    CN119227924A

  • Construction method of operator safety education training system

    CN120525124A

  • A construction method of a work personnel safety education and training system

    CN120525124B