Emergency disposal process intelligent redisk analysis method based on data mining

Abstract

The invention belongs to the technical field of emergency disposal, and particularly relates to an emergency disposal process intelligent redisk analysis method based on data mining. The method comprises the following steps: firstly, collecting multi-source data in an emergency disposal process, wherein the multi-source data comprises environment, operation and resource allocation data; secondly, converting the unstructured data to enable the unstructured data to become analyzable structured data; feature extraction is carried out by using an improved deep belief network DBN; and finally, inputting the feature vectors into a decision evaluation model, evaluating the reasonability of an emergency disposal decision, analyzing the aspects of personnel operation, resource allocation and response time, finding out problems and optimizable links, and presenting the problems and optimizable links in a visual mode. The data value can be effectively mined, the power grid emergency disposal efficiency and accuracy are improved, and support is provided for safe and stable operation of the power grid.

Description

Technical Field

[0001] The present invention belongs to the technical field of emergency response, and specifically relates to an intelligent review and analysis method for emergency response processes based on data mining. Background Art

[0002] During the operation of the power grid, various emergencies may occur, such as failures, disasters, etc. These events pose a serious threat to the safe and stable operation of the power grid. In order to improve the emergency response capability and disposal effect of the power grid, it is necessary to accurately re-analyze the emergency disposal process of the power grid. However, the existing power grid emergency disposal re-analysis method mainly relies on manual experience and simple data statistics, and has problems such as insufficient analysis depth and low efficiency, and it is difficult to fully mine the potential information and patterns in the data. With the continuous expansion of the scale of the power grid and the improvement of the level of intelligence, the amount of data generated in the emergency disposal process of the power grid has increased dramatically, and the data types have become more complex and diverse, including structured, semi-structured and unstructured data. How to effectively process and analyze these massive data and extract valuable information from them has become an important challenge facing the current field of power grid emergency disposal. Therefore, the present invention proposes an intelligent re-analysis method for the emergency disposal process of the power grid based on data mining, which aims to solve the above problems and provide strong support for the safe and stable operation of the power grid. Summary of the invention

[0003] In view of the technical problems existing in power grid emergency disposal, the present invention proposes a method for intelligent replay and analysis of power grid emergency disposal process based on data mining, which has reasonable design, simple method, strong theoretical basis and can realize.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is: comprising the following steps:

[0005] S1. First, use monitoring equipment and related information systems to collect multi-source data during emergency response;

[0006] S2. Secondly, the unstructured data of the collected multi-source data is transformed into analyzable structured data using text extraction technology in natural language processing technology;

[0007] S3, then use the improved deep belief network DBN to extract features from the preprocessed data to obtain key feature vectors related to the emergency response process;

[0008] S4. Finally, the extracted feature vector is input into the decision evaluation model obtained based on reinforcement learning training to evaluate the rationality of each decision-making link in the emergency response process. Based on the output results, the personnel operation, resource allocation, and response time in the emergency response process are analyzed to find out the existing problems and links that can be optimized.

[0009] Preferably, the multi-source data includes environmental data when the emergency event occurs, operational data of the handling personnel, and emergency resource allocation data, and the data format includes structured data and unstructured data.

[0010] Preferably, the implementation process of converting the text into analyzable structured data using the text extraction technology in natural language processing technology in step S2 is as follows:

[0011] S21, first remove the noise information in the text, correct the errors in the text, divide the long text into paragraphs, further process each paragraph into sentences, and divide the text into independent sentences to facilitate subsequent analysis;

[0012] S22, secondly, determine the entity type, collect the labeled emergency response related data as a training set, and use the conditional random field CRF to train the named entity recognition model;

[0013] S23, then input the unstructured text data to be converted into the trained named entity recognition model, the model recognizes various entities in the text, and labels the type of each entity;

[0014] S24, then determine the relationship type between different entities in the emergency response process, use a support vector machine to automatically extract entity relationships through a training model, and perform entity relationship extraction on the text after named entity recognition, determine the relationship between different entities, and express these relationships in a structured form;

[0015] S25. Finally, the entities and relationships are organized according to the pre-set structured data format, and the accuracy and reliability are tested.

[0016] Preferably, in step S3, an improved deep belief network DBN is used to extract features from the preprocessed data. The DBN is a deep neural network formed by stacking multiple restricted Boltzmann machines RBM. The specific implementation process is as follows:

[0017] S31. First, RBM consists of a visible layer v and a hidden layer h, v = (v1, v2, ..., v nv )h=(h1,h2,...,h nh ), where n v ,h v are the number of neurons in the visible layer and the hidden layer respectively, and the energy function is defined as where a i ,b j are the biases of the visible layer and the hidden layer, respectively, and W ij is the weight connecting the visible layer and the hidden layer;

[0018] S32, then the weight update rule of adaptive learning rate is adopted, and the learning rate is η t , the weight at time t is updated to in is the expectation calculated based on the original data, It is the expectation calculated based on the reconstructed data, and uses the new activation function Leaky ReLU function to enhance the nonlinear expression ability of the network and avoid the problem of neuron death;

[0019] S33, then stacking multiple improved RBMs to form a deep belief network, the visible layer of the first layer RBM receives the preprocessed data, and its hidden layer output is used as the visible layer input of the second layer RBM, and so on;

[0020] S34, then use unsupervised learning method to train RBM layer by layer, and train RBM by maximizing the log-likelihood function. The maximized log-likelihood function is Where P(v) is the data distribution. For each layer of RBM, the energy function is minimized by the contrast divergence algorithm, and the weights and biases are updated. After k steps of sampling, the reconstructed visible layer vector v~ is obtained, and then the weights and biases are updated. Where ε is the learning rate;

[0021] S35. Next, after the pre-training is completed, the entire deep belief network is regarded as a deep neural network, and the back propagation algorithm is used to fine-tune the entire network to optimize the performance of the network;

[0022] S36. Finally, the preprocessed new data is input into the trained deep belief network, and the output of the high-level hidden layer is extracted as the key feature vector through the forward propagation process.

[0023] Preferably, the specific operations of evaluating the rationality of each decision-making link in the emergency response process, outputting the results and analyzing them in step S4 are:

[0024] S41, firstly, the extracted feature vector is expressed as F = (f1, f2, ..., f n ), where n is the dimension of the feature vector. The emergency response process is modeled as a reinforcement learning environment including the state space S, the action space A, the incentive function R and the transition probability P. The state space can include various emergency response situations, which are composed of the feature vector and other related information, namely s∈S. The action space is the decision that can be taken in the emergency response process, namely a∈A.

[0025] S42, Next, introduce a reinforcement learning agent, which selects actions in the environment according to the current state. At time step t, the agent is in state s t , select an action at , the environment is based on actions and transition probabilities P(s t+1 |s t ,a t )Transfer to the next state s t+1 , and give the agent a reward r t =R(s t ,a t );

[0026] S43, then the value function is estimated, the state value function V Γ (s) represents the expected cumulative reward obtained by following strategy Γ starting from state s, i.e. Where γ is the discount factor, k is the index of the time step; the action value function Q Γ (s,a) represents the expected cumulative reward obtained by following strategy Γ after taking action a in state s, that is,

[0027] S44, then use the policy gradient algorithm to improve the value function Where θ is the policy parameter, J(θ) is the expected cumulative reward, and Γ θ It is a strategy with parameter θ, which is optimized in the direction of increasing the cumulative reward by continuously adjusting θ;

[0028] S45. Finally, according to the action value function value output by the model, the personnel operation, resource allocation and response time in the emergency response process are analyzed.

[0029] Preferably, the step S4 identifies existing problems and links that can be optimized and presents them in a visual manner, generating charts and graphs to intuitively display the advantages, disadvantages and improvement suggestions of the emergency response process.

[0030] Compared with the prior art, the advantages and positive effects of the present invention are that the natural language processing technology is used to convert unstructured data into structured data, thereby improving the availability and analysis efficiency of data. The improved deep belief network DBN can more effectively extract key feature vectors related to the emergency response process, and enhance the accuracy and depth of feature extraction. The decision evaluation model based on reinforcement learning training can more accurately evaluate the rationality of each decision-making link in the emergency response process, providing a strong basis for optimization. DETAILED DESCRIPTION

[0031] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0032] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0033] Embodiment, in the operation process of modern power grids, emergencies frequently occur, such as equipment failures, natural disasters, etc. These events pose a serious threat to the safe operation of the power grid. As the scale of the power grid continues to expand, the amount and complexity of data generated during emergency response also increase. Existing emergency replay analysis mostly relies on manual operation and simple data statistics, with low analysis efficiency and insufficient depth, making it difficult to fully explore the potential patterns in the data. Therefore, an intelligent replay analysis method based on data mining is proposed to improve the efficiency and accuracy of emergency response.

[0034] First of all, in the process of emergency response, the data sources are complex and diverse, and multi-source data needs to be collected through monitoring equipment and information systems. Multi-source data includes: Environmental data: such as temperature, humidity, wind speed, etc. These data are crucial to the judgment of the background of the event. Operation data: record the operation logs, dispatch instructions and operation execution results of the disposal personnel. Resource allocation data: including the use of emergency materials, and the allocation records of personnel and vehicles. In specific implementation, the above-mentioned environmental data is automatically collected through on-site sensor equipment, and at the same time connected to the operating platform through the system interface, to collect the operation data and allocation data of the disposal personnel in real time, which will serve as the basis for subsequent analysis.

[0035] Secondly, considering that the collected multi-source data contains structured and unstructured data, it is necessary to process them uniformly so that the data can be easily stored and used. The text extraction technology in natural language processing technology is used to convert it into analyzable structured data. The implementation process is as follows: first, remove the noise information in the text, correct the errors in the text, divide the long text into paragraphs, further process each paragraph into sentences, and divide the text into independent sentences to facilitate subsequent analysis; secondly, determine the entity type, collect the labeled emergency response related data as the training set, and use the conditional random field CRF to train the named entity recognition model; then input the unstructured text data to be converted into the trained named entity recognition model, the model recognizes various entities in the text, and labels each entity with its type; then determine the relationship type between different entities in the emergency response process, use the support vector machine to automatically extract the entity relationship through the training model, and extract the entity relationship of the text after the named entity recognition, determine the relationship between different entities, and express these relationships in a structured form; finally, organize the entities and relationships according to the pre-set structured data format, and detect the accuracy and reliability.

[0036] After data preprocessing, the improved deep belief network DBN is used to extract features from the preprocessed data. DBN is composed of multiple restricted Boltzmann machines (RBMs) stacked together. The network is trained layer by layer through unsupervised learning to extract key feature vectors closely related to the emergency response process. First, RBM consists of a visible layer v and a hidden layer h. where n v ,h v are the number of neurons in the visible layer and the hidden layer respectively, and the energy function is defined as where a i ,b j are the biases of the visible layer and the hidden layer, respectively, and W ij is the weight connecting the visible layer and the hidden layer; secondly, the weight update rule of adaptive learning rate is adopted, and the learning rate is η t , the weight at time t is updated to in is the expectation calculated based on the original data, It is the expectation calculated based on the reconstructed data, and uses the new activation function Leaky ReLU function to enhance the nonlinear expression ability of the network and avoid the problem of neuron death; then multiple improved RBMs are stacked to form a deep belief network, and the visible layer of the first layer RBM receives the preprocessed data, and its hidden layer output is used as the visible layer input of the second layer RBM, and so on; then the unsupervised learning method is used to train the RBM layer by layer, and the RBM is trained by maximizing the log-likelihood function, and the maximized log-likelihood function is Where P(v) is the data distribution. For each layer of RBM, the energy function is minimized by the contrast divergence algorithm, and the weights and biases are updated. After k steps of sampling, the reconstructed visible layer vector v~ is obtained, and then the weights and biases are updated. Where ε is the learning rate; next, after the pre-training is completed, the entire deep belief network is regarded as a deep neural network, and the back-propagation algorithm is used to fine-tune the entire network to optimize the performance of the network; finally, the pre-processed new data is input into the trained deep belief network, and the output of the high-level hidden layer is extracted as the key feature vector through the forward propagation process.

[0037] In order to make a more reasonable evaluation of the decision-making process, this paper models the emergency response process as a reinforcement learning environment. The ultimate goal of this model is to optimize the decision chain through action selection, improve the overall response efficiency, and perform a visual display. The specific implementation is to first express the extracted feature vector as F = (f1, f2, ..., f n), where n is the dimension of the feature vector. The emergency response process is modeled as a reinforcement learning environment including state space S, action space A, incentive function R and transition probability P. The state space can include various emergency response situations, which are composed of feature vectors and other related information, namely s∈S. The action space is the decision that can be taken in the emergency response process, namely a∈A. Secondly, a reinforcement learning agent is introduced. The agent selects actions according to the current state in the environment. At time step t, the agent is in state s t , select an action a t , the environment is based on actions and transition probabilities P(s t+1 |s t ,a t )Transfer to the next state s t+1 , and give the agent a reward r t =R(s t ,a t ); then the value function is estimated, the state value function V Γ (s) represents the expected cumulative reward obtained by following strategy Γ starting from state s, i.e. Where γ is the discount factor, k is the index of the time step; the action value function Q Γ (s,a) represents the expected cumulative reward obtained by following strategy Γ after taking action a in state s, that is,

[0038] Then use the policy gradient algorithm to improve the value function Where θ is the policy parameter, J(θ) is the expected cumulative reward, and Γ θ is a strategy with parameter θ. By continuously adjusting θ, the strategy is optimized in the direction of increasing the cumulative reward. Finally, according to the action value function value output by the model, the personnel operation, resource allocation, and response time in the emergency response process are analyzed. Through the visualization of the final results, managers can clearly see the bottleneck links of the long response time of operators and untimely resource allocation in the disposal process, providing strong support for the safe and stable operation of the power grid.

[0039] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent review and analysis method for emergency response process based on data mining, characterized in that: The following steps are involved: S1. First, use monitoring equipment and related information systems to collect multi-source data during emergency response; S2. Secondly, the unstructured data of the collected multi-source data is transformed into analyzable structured data using text extraction technology in natural language processing technology; S3, then use the improved deep belief network DBN to extract features from the preprocessed data to obtain key feature vectors related to the emergency response process; S4. Finally, the extracted feature vector is input into the decision evaluation model obtained based on reinforcement learning training to evaluate the rationality of each decision-making link in the emergency response process. Based on the output results, the personnel operation, resource allocation, and response time in the emergency response process are analyzed to find out the existing problems and links that can be optimized.

2. According to claim 1, a method for intelligent replay analysis of emergency response process based on data mining is characterized in that: The multi-source data includes environmental data when an emergency occurs, operational data of response personnel, and emergency resource allocation data, and the data format includes structured data and unstructured data.

3. According to claim 1, the intelligent replay analysis method for emergency response process based on data mining is characterized in that: The implementation process of converting the text into analyzable structured data using the text extraction technology in natural language processing technology in step S2 is as follows: S21, first remove the noise information in the text, correct the errors in the text, divide the long text into paragraphs, further process each paragraph into sentences, and divide the text into independent sentences to facilitate subsequent analysis; S22, secondly, determine the entity type, collect the labeled emergency response related data as a training set, and use the conditional random field CRF to train the named entity recognition model; S23, then input the unstructured text data to be converted into the trained named entity recognition model, the model recognizes various entities in the text, and labels the type of each entity; S24, then determine the relationship type between different entities in the emergency response process, use a support vector machine to automatically extract entity relationships through a training model, and perform entity relationship extraction on the text after named entity recognition, determine the relationship between different entities, and express these relationships in a structured form; S25. Finally, the entities and relationships are organized according to the pre-set structured data format, and the accuracy and reliability are tested.

4. According to claim 1, the intelligent replay analysis method for emergency response process based on data mining is characterized in that: In step S3, the improved deep belief network DBN is used to extract features from the preprocessed data. The DBN is a deep neural network composed of multiple restricted Boltzmann machines RBM stacked together. The specific implementation process is as follows: S31. First, RBM consists of a visible layer v and a hidden layer h. Where n v ,h v are the number of neurons in the visible layer and the hidden layer respectively, and the energy function is defined as where a i ,b j are the biases of the visible layer and the hidden layer, respectively, and W ij is the weight connecting the visible layer and the hidden layer; S32, then the weight update rule of adaptive learning rate is adopted, and the learning rate is η t , the weight at time t is updated to in is the expectation calculated based on the original data, It is the expectation calculated based on the reconstructed data, and uses the new activation function Leaky ReLU function to enhance the nonlinear expression ability of the network and avoid the problem of neuron death; S33, then stacking multiple improved RBMs to form a deep belief network, the visible layer of the first layer RBM receives the preprocessed data, and its hidden layer output is used as the visible layer input of the second layer RBM, and so on; S34, then use unsupervised learning method to train RBM layer by layer, and train RBM by maximizing the log-likelihood function. The maximized log-likelihood function is Where P(v) is the data distribution. For each layer of RBM, the energy function is minimized by the contrast divergence algorithm, the weights and biases are updated, and the reconstructed visible layer vector is obtained after k-step sampling. Then update the weights and biases Where ε is the learning rate; S35. Next, after the pre-training is completed, the entire deep belief network is regarded as a deep neural network, and the back propagation algorithm is used to fine-tune the entire network to optimize the performance of the network; S36. Finally, the preprocessed new data is input into the trained deep belief network, and the output of the high-level hidden layer is extracted as the key feature vector through the forward propagation process.

5. According to claim 1, the intelligent replay analysis method for emergency response process based on data mining is characterized in that: The specific operations of evaluating the rationality of each decision-making link in the emergency response process, outputting the results and analyzing them in step S4 are as follows: S41. First, the extracted feature vector is expressed as F = (f1, f2, ..., f n ), where n is the dimension of the feature vector. The emergency response process is modeled as a reinforcement learning environment including the state space S, the action space A, the incentive function R and the transition probability P. The state space can include various emergency response situations, which are composed of the feature vector and other related information, namely s∈S. The action space is the decision that can be taken in the emergency response process, namely a∈A. S42, Next, introduce a reinforcement learning agent, which selects actions in the environment according to the current state. At time step t, the agent is in state s t , select an action a t , the environment is based on actions and transition probabilities P(s t+1 |s t ,a t )Transfer to the next state s t+1 , and gives the agent a reward r t =R(s t ,a t ); S43, then the value function is estimated, the state value function V Γ (s) represents the expected cumulative reward obtained by following strategy Γ starting from state s, i.e. Where γ is the discount factor, k is the index of the time step; the action value function Q Γ (s,a) represents the expected cumulative reward obtained by following strategy Γ after taking action a in state s, that is, S44, then use the policy gradient algorithm to improve the value function Where θ is the policy parameter, J(θ) is the expected cumulative reward, and Γ θ It is a strategy with parameter θ, which is optimized in the direction of increasing the cumulative reward by continuously adjusting θ; S45. Finally, according to the action value function value output by the model, the personnel operation, resource allocation and response time in the emergency response process are analyzed.

6. According to claim 1, a method for intelligent replay analysis of emergency response process based on data mining is characterized in that: The step S4 identifies existing problems and links that can be optimized and presents them in a visual manner, generating charts and graphs to intuitively display the advantages, disadvantages and improvement suggestions in the emergency response process.

Images