Agent intelligent customer service and anomaly detection system

Abstract

The invention provides an Agent intelligent customer service and anomaly detection system, and belongs to the field of natural language processing, the Agent intelligent customer service and anomaly detection system is composed of a knowledge industry large model, three platform components and N large model application scenes, and Agent-based intelligent customer service and general industry-oriented anomaly detection application are constructed on the basis of the knowledge industry large model, the three platform components and the N large model application scenes. Through AI algorithms such as a large model technology, an NLP technology and anomaly detection, the method is used for different roles such as front-line customer service personnel, operation and maintenance personnel and data analysis personnel. According to the invention, complex industry challenges can be better handled, and wide application of the AI technology in actual services is promoted.

Description

Technical Field

[0001] The present invention relates to the fields of natural language processing, image recognition and recommendation systems, and in particular to an Agent intelligent customer service and anomaly detection system. Background Art

[0002] At present, the use of large models lacks a unified framework. In the field of artificial intelligence, especially in natural language processing, image recognition, recommendation systems and other applications, there is a lack of a unified theoretical or technical framework to guide the development and application of large models. Operators also lack unified management when doing field segmentation and upper-level applications based on industry large models. At the manual customer service level, there are problems such as peak service pressure, repetitive problem handling, and the balance between service standardization and personalized needs. There is an urgent need for intelligent methods to replace manual services to achieve the goal of reducing costs and increasing efficiency.

[0003] Existing technical solutions

[0004] 1. At the level of intelligent customer service, operators can automate customer service by using large language models, such as automatically answering common questions, handling user complaints, and providing technical support. This approach can significantly improve service efficiency and reduce labor costs. At the same time, using large models to analyze network data can predict network congestion, optimize resource allocation, and improve network efficiency. For example, by analyzing user distribution and traffic patterns, network capacity can be adjusted in advance to ensure service quality.

[0005] 2. In terms of business applications, traditional anomaly detection algorithms are mainly divided into statistical-based methods, proximity-based methods, and model-based methods. These methods have their own advantages in dealing with anomaly detection problems and are suitable for different data types and scenarios. They include statistical-based methods, proximity-based methods, model-based methods, etc.

[0006] Disadvantages and shortcomings of existing technical solutions

[0007] 1. Lack of a unified framework for large models. Development of general AI frameworks, such as multimodal learning, transfer learning, and federated learning. These frameworks aim to provide a general approach to building and deploying large models to improve development efficiency, reduce costs, and ensure the interpretability and security of models.

[0008] 2. Traditional anomaly detection algorithms are still very effective in many practical applications, especially when the data volume is small and the feature engineering is relatively simple. However, with the advent of the big data era, data has become more complex and large, and traditional methods may encounter performance bottlenecks when processing high-dimensional data, large-scale data, and complex data types (such as text, images, and audio). Therefore, researchers are also constantly exploring and developing new anomaly detection technologies, such as deep learning-based methods, to meet these challenges. Summary of the invention

[0009] In order to solve the above technical problems, the present invention provides an Agent intelligent customer service and anomaly detection system.

[0010] The technical solution of the present invention is:

[0011] An Agent intelligent customer service and anomaly detection system includes: a knowledge industry big model, platform components, and big model application scenarios, wherein:

[0012] Industry models, combining industry-specific data and knowledge to provide professional and accurate solutions;

[0013] The platform components include:

[0014] Knowledge base management component: used to manage and maintain the industry knowledge base to ensure the accuracy and update of data;

[0015] Agent component: realizes automatic processing of complex tasks through intelligent agent technology;

[0016] Robot interaction component: provides a human-computer interaction interface, allowing users to easily communicate and operate with the AI ​​system;

[0017] Model application scenarios, covering application requirements, including:

[0018] Intelligent question and answer: realize efficient information query and answer functions;

[0019] Intelligent customer service: answer user questions and provide reasonable suggestions through multiple rounds of dialogue;

[0020] Intelligent reporting: Based on Text2SQL technology, it provides intelligent data report query and presentation;

[0021] Anomaly detection: Provides a general anomaly detection algorithm model to support anomaly detection applications in other scenarios.

[0022] Furthermore,

[0023] Agent components include perception and memory, autonomous planning, tools, and action functions.

[0024] Perception and memory: used by the system super-intelligence to obtain various original and high-dimensional abstract perception results; memory for storing and managing user information to optimize performance under user personalized needs;

[0025] Autonomous planning: Based on implicit or explicit user demands, the system can autonomously break down subtasks and arrange subtask chains to form closed-loop goals. At the same time, the system can iteratively reflect on the execution results of subtasks and autonomously improve planning and orchestration capabilities.

[0026] Tools: A collection of tools and services that can be called, including system tools and third-party services. Based on interface definition specifications, it enables the retrieval and distribution of massive tools and services.

[0027] Action: Execute the subtasks according to the plan to achieve the closed-loop task objectives.

[0028] Agent execution includes the following:

[0029] Execution Agent: Before each step needs to be executed, it uses the output of all previous executions to decide the next execution;

[0030] Plan and execute agents: decide the complete sequence of operations in advance and then execute all operations without updating the plan;

[0031] The user tells the Agent the key information, and the Agent decides which tools to use. In fact, there is an object LLM in the Agent, and the final decision on which tools to use is made by the LLM (ChatGLM) in the Agent.

[0032] In response to user complaints, the Agent automatically selects tools to complete the following:

[0033] Determine the complaint type based on the content of the user's complaint;

[0034] Call the tool to obtain relevant database and knowledge base content and output the root cause of the complaint;

[0035] Generate intelligent speech based on complaint root cause analysis;

[0036] Generate complaint JSON and send it to the dispatch system or early warning system;

[0037] Write complaint document analysis reports and organize them to form a new knowledge base.

[0038] Based on the "ChatGLM3" general model, using the LangChain framework architecture, build an intelligent complaint customer service Agent model.

[0039] Furthermore,

[0040] Anomaly detection includes single-indicator anomaly quantification and multi-indicator anomaly quantification with dynamic weights.

[0041] Quantification of single indicator anomaly

[0042] Given an index X = x1, x2, x3, ...x n , input the current x n+1 , quantify the degree of deviation of the input value, that is, the abnormal index score

[0043] 1) Detect data anomalies based on algorithms such as isolation forest

[0044] Parameter setting: contamination: float in (0., 0.5),

[0045] optional(default=0.1)

[0046] Remove the abnormal data and get m normal data, m <n。

[0047] X=x1,x2,x3,x4,...x m

[0048] 2) Calculate the mean and standard deviation of m values

[0049] Mean:

[0050]

[0051] Standard Deviation:

[0052]

[0053] 3) Calculate the score of a single indicator

[0054] Enter the new x value and the calculation formula is as follows:

[0055] y -score =10 F(x；μ，σ)

[0056] in, If: F(x; μ, σ) < 0.5, then:

[0057] F(x;μ,σ)=1-F(x;μ,σ)

[0058] 4) Score interpretation

[0059] The value range of y_score is [3.16, 10], where 10 to the power of 0.5 is 3.16 and 10 to the power of 1 is 10. Anomaly indicator scores greater than the specified value are highly abnormal and require special attention.

[0060] Multi-index anomaly quantification with dynamic weights

[0061] Assume that the data set X consists of n data samples. Based on this data set, m indicators are defined, which can be expressed in mathematical language as follows:

[0062] X1, X2, X3, ...X m

[0063] Among them, X i =x1,x2,...x n ;

[0064] 1) Data normalization

[0065] Use the maximum and minimum values ​​for normalization and map the index of X to the interval [0,100]. The formula is as follows:

[0066]

[0067] 2) Calculate data volatility

[0068] The volatility of each X is replaced by the standard deviation of X. The larger the standard deviation, the more information the indicator can reflect, and the more weight is given to this indicator. The formula is as follows:

[0069]

[0070] 3) Calculate data contribution

[0071] The larger the linear positive correlation coefficient between indicators, the smaller the weight is; the higher the contribution, the higher the weight is.

[0072]

[0073] Where rij represents the correlation coefficient between the i-th indicator and the j-th indicator, and this value can be obtained from the covariance matrix of the indicator;

[0074] 4) Calculate the information carrying capacity

[0075] Information carrying capacity = volatility * contribution, that is

[0076] C j =S j *R j

[0077] 5) Calculate indicator weights

[0078] The weight of each indicator is the contribution ratio of each indicator information carrying capacity Cj, that is,

[0079]

[0080] If the current values ​​of X1, X2, and X3 give anomaly scores of E1, E2, and E3, then the overall anomaly perception

[0081] Y=W1*E1+W2*E2+W3*E3

[0082] In this way, the quantification of abnormal indicators and the dynamic allocation of weights of different indicators are completed, and finally a Y value is obtained to represent the final score of the entire abnormality.

[0083] The present invention uses Agent big model technology and proposes a method based on the mutual integration of RAG and Agent for the field of intelligent customer service. Customer service questions are decomposed and planned, and the Agent calls external methods to generate answers.

[0084] The present invention proposes a quantization algorithm based on outliers and a dynamic weight calculation algorithm, and integrates the advantages of the two algorithms to construct an outlier detection algorithm for any indicator.

[0085] The beneficial effects of the present invention are

[0086] 1. A "1+3+N" large-model basic platform architecture was proposed, which provides a powerful and flexible solution for various industries by integrating industry large models, platform components and multiple application scenarios.

[0087] 2. On the above platform, an intelligent customer service based on a large model agent was built, which has natural language processing capabilities, machine learning capabilities, big data analysis, and RAG-based knowledge base retrieval technology.

[0088] 3. This solution proposes a model technology for data anomaly detection and analysis in general fields. It proposes a quantization algorithm based on outliers and a dynamic weight calculation algorithm, and combines the advantages of the two algorithms to construct an outlier detection algorithm for any indicator. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 This is a schematic diagram of the 1+3+N platform architecture;

[0090] Figure 2 It is a schematic diagram of the Agent execution process;

[0091] Figure 3 This is a schematic diagram of the intelligent complaint customer service Agent system architecture;

[0092] Figure 4 It is a schematic diagram of the RAG knowledge retrieval process;

[0093] Figure 5 It is a flow chart of intelligent complaint customer service agent. DETAILED DESCRIPTION

[0094] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0095] The present invention provides an Agent intelligent customer service and anomaly detection system. The architecture is an AI large model application platform with high scalability and adaptability. Its structure consists of 1 knowledge industry large model, 3 platform components, and N large model application scenarios. On this basis, Agent-based intelligent customer service and anomaly detection applications for general industries are constructed.

[0096] 1. 1+3+N platform architecture

[0097] The "1+3+N" platform architecture is a highly scalable and adaptable AI large model application platform. Its structure consists of the following three levels:

[0098] 1 knowledge industry big model, 3 platform components, N big model application scenarios

[0099] This architectural design achieves modularity and flexibility of the platform through a hierarchical approach, with each layer having its own unique functions and roles. Figure 1 shown.

[0100] 1 Knowledge Industry Enhanced Industry Model

[0101] The core of the platform is a big model for a specific industry, such as the telecommunications industry. This big model combines industry-specific data and knowledge to provide professional and accurate solutions.

[0102] The role of industry models: By embedding the expertise and data of a specific industry, the application effect of the model in that field can be improved. For example, the telecommunications industry model can process complex data from telecommunications operator systems, provide intelligent question-and-answer customer service, and detect anomalies in general indicators.

[0103] 3 Platform Components

[0104] The middle layer of the platform consists of three key components: knowledge base management component, agent component, and robot interaction component.

[0105] These components provide powerful functional support, enabling the platform to efficiently manage and utilize knowledge bases, conduct intelligent agent interactions, and achieve automated operations through robotics technology.

[0106] Knowledge base management component: used to manage and maintain the industry knowledge base to ensure the accuracy and update of data. The knowledge base is the basis of the large model and contains a large amount of industry information and data.

[0107] Agent component: Through intelligent agent technology, complex tasks can be automatically processed. For example, AI agents can automatically handle customer service requests, analyze data, and provide decision support.

[0108] Robot interaction components: Provide a human-computer interaction interface so that users can easily communicate and operate with the AI ​​system. This can include voice assistants, chatbots, etc. to improve user experience and interactivity.

[0109] N large model application scenarios

[0110] The outermost layer is N large model application scenarios, which cover a wide range of application needs, including but not limited to:

[0111] Intelligent Q&A: Realize efficient information query and answer functions to improve customer service quality.

[0112] Intelligent customer service: answer user questions and provide reasonable suggestions through multiple rounds of dialogue.

[0113] Intelligent reporting: Based on Text2SQL technology, it provides intelligent data report query and presentation.

[0114] Anomaly detection: Provides a general anomaly detection algorithm model to support anomaly detection applications in other scenarios.

[0115] 2. Agent intelligent customer service design

[0116] Agent Process

[0117] Regardless of the agent model, its core functions are the same, including perception and memory, autonomous planning, tools, and actions.

[0118] Perception and memory: The perception module is mainly used by the system super-intelligent agent to obtain various original and high-dimensional abstract perception results; the memory module is the memory of user information stored and managed in the system super-intelligent agent to optimize the performance under the user's personalized needs.

[0119] Autonomous planning: Based on implicit or explicit user demands, the system can autonomously break down subtasks and arrange subtask chains to form closed-loop goals. At the same time, it can iteratively reflect on the execution results of subtasks and autonomously improve planning and orchestration capabilities.

[0120] Tools: A collection of tools and services that can be called, including system tools (such as calendars, calculators, web page retrieval, etc.) and third-party services (such as meta-services, domain agents). Based on good interface definition specifications, the system super-intelligence can achieve efficient retrieval and distribution of massive tools and services.

[0121] Action: Execute the subtasks broken down according to the plan (including calling relevant tools) to close the task objectives.

[0122] The basic Agent execution process of the intelligent complaint customer service is as follows: Figure 2 shown.

[0123] Execution Agent: Before each step needs to be executed, it uses the output of all previous executions to decide the next execution

[0124] Plan and Execute Agent: Decide the complete sequence of operations in advance, then execute all operations without updating the plan

[0125] The user tells the Agent the key information, and the Agent decides which tools to use. In fact, there is an object LLM in the Agent, and the final decision on which tools to use is made by the LLM (ChatGLM) in the Agent.

[0126] Program Objectives

[0127] Intelligent complaint customer service agent, in response to user complaints, automatically selects tools to complete the following:

[0128] Determine the complaint type based on the content of the user's complaint

[0129] Call the tool to obtain relevant database and knowledge base content and output the root cause of the complaint

[0130] Generate intelligent speech based on complaint root cause analysis

[0131] Generate complaint JSON and send it to the dispatch system or early warning system

[0132] Write complaint document analysis reports and organize them into a new knowledge base

[0133] Based on the "ChatGLM3" general model, using the LangChain framework architecture, we build an intelligent complaint customer service Agent model. Figure 3 shown.

[0134] RAG knowledge retrieval process Figure 4 shown.

[0135] Intelligent complaint customer service agent automatically selects and uses the following tools to answer user questions, such as Figure 5 shown.

[0136] Anomaly detection algorithm design

[0137] Quantification of single indicator anomaly

[0138] This scheme provides a general and quantifiable anomaly detection algorithm. Given an indicator X = x1, x2, x3, ... x n , input the current x n+1 , quantify the degree of deviation of the input value, that is, the anomaly indicator score.

[0139] 1. Detect data anomalies based on algorithms such as isolation forest

[0140] Parameter setting: contamination: float in (0., 0.5),

[0141] optional(default=0.1)

[0142] Remove the abnormal data and get m normal data, m <n。

[0143] X=x1,x2,x3,x4,...x m

[0144] 2. Calculate the mean and standard deviation of m values

[0145] Mean:

[0146]

[0147] Standard Deviation:

[0148]

[0149] 3. Calculate the score of a single indicator

[0150] Enter the new x value and the calculation formula is as follows:

[0151] y -score =10 F(x；μ，σ)

[0152] in, If: F(x; μ, σ) < 0.5, then:

[0153] F(x;μ,σ)=1-F(x;μ,σ)

[0154] 4. Score interpretation

[0155] The value range of y_score is [3.16, 10], 10 to the power of 0.5 is 3.16, 10 to the power of 1 is 10. The reference value of the abnormality score is:

[0156]

[0157] Anomaly index scores greater than 7.94 are highly abnormal and require special attention.

[0158] Multi-index anomaly quantification with dynamic weights

[0159] Assume that the data set X consists of n data samples. Based on this data set, m indicators are defined, which can be expressed in mathematical language as follows:

[0160] X1, X2, X3, ...X m

[0161] Among them, X i =x1,x2,...x n .

[0162] 1. Data normalization

[0163] Normalization is to transform the data by scaling and translating it, and mapping the data to the interval [0,1]. The change here is essentially a linear change. This solution uses the maximum and minimum values ​​for normalization, and maps the index of X to the interval [0,100]. The formula is as follows:

[0164]

[0165] 2. Calculate data volatility

[0166] The volatility of each X is replaced by the standard deviation of X. The core idea is: the larger the standard deviation, the more information the indicator can reflect, and the more weight should be given to this indicator. The formula is as follows:

[0167]

[0168] 3. Calculate data contribution

[0169] The correlation coefficient is often used to measure the degree of linear correlation between two variables. The core idea is: the larger the linear positive correlation coefficient between indicators, the smaller the contribution, and the smaller the weight should be given.

[0170] The higher the contribution, the higher the weight given.

[0171]

[0172] Where rij represents the correlation coefficient between the ith indicator and the jth indicator, and this value can be obtained from the covariance matrix of the indicator.

[0173] 4. Calculate the information carrying capacity

[0174] Information carrying capacity = volatility * contribution, that is

[0175] C j =S j *R j

[0176] 5. Calculate indicator weights

[0177] The weight of each indicator is the contribution ratio of each indicator information carrying capacity Cj, that is,

[0178]

[0179] If the current values ​​of X1, X2, and X3 give anomaly scores of E1, E2, and E3, then the overall anomaly perception

[0180] Y=W1*E1+W2*E2+W3*E3

[0181] In this way, the abnormal indicators are quantified and the weights of different indicators are dynamically allocated, and finally a Y value is obtained to represent the final score of the entire abnormality. For example, the Y value can be used to represent the overall abnormal user score when different indicators are abnormal. The final abnormal user score is Y.

[0182] Taking the modeling process as an example, outlier detection is performed based on two basic indicators, "packet loss rate" and "latency". The process is as follows.

[0183] 1) Obtain historical data of packet loss rate and latency.

[0184] 2) Using the isolation forest algorithm, set the anomaly rate of historical data to 2% for clearing.

[0185] 3) Based on the “abnormality quantification algorithm”, calculate the abnormality of each indicator.

[0186] 4) Based on the “dynamic weight calculation algorithm”, calculate the weights of the above two indicators’ contributions to outliers.

[0187] 5) The weight is added to the abnormality in step 3 to calculate the final user abnormality score.

[0188] 6) Set a certain confidence level (such as 90%) and output an exception list.

[0189] The technical advantages of the present invention are as follows:

[0190] 1. The "1+3+N" large model framework provides technical and directional guidance for subsequent research and development based on large models. With the continuous advancement of AI technology, this platform architecture will surely be applied and promoted in more fields. Through this innovative architecture design, it can better cope with complex industry challenges and promote the widespread application of AI technology in actual business.

[0191] 2. Agent intelligent customer service improves work efficiency. Intelligent customer service can provide services 24 hours a day and handle a large number of repetitive issues, thereby freeing up manual customer service and allowing them to focus on more complex and personalized service needs. It can reduce the company's dependence on manual customer service, thereby reducing labor costs. At the same time, the intelligent customer service system can handle multiple requests at the same time, improve work efficiency and reduce operating costs.

[0192] 3. General anomaly detection algorithms can identify data points, events or patterns that do not conform to normal behavior patterns. The social benefits of this technology are broad and far-reaching. It can empower the communications industry, financial industry, network security, social services, business analysis and other fields.

[0193] The above description is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. An Agent intelligent customer service and anomaly detection system, It is characterized in that include: Knowledge industry big models, platform components, and big model application scenarios, among which: Industry models, combining industry-specific data and knowledge to provide professional and accurate solutions; The platform components include: Knowledge base management component: used to manage and maintain the industry knowledge base to ensure the accuracy and update of data; Agent component: realizes automatic processing of complex tasks through intelligent agent technology; Robot interaction component: provides a human-computer interaction interface, allowing users to easily communicate and operate with the AI ​​system; Model application scenarios, covering application requirements, including: Intelligent question and answer: realize information query and answer functions; Intelligent customer service: answer user questions and provide reasonable suggestions through dialogue; Intelligent reporting: Based on Text2SQL technology, it provides intelligent data report query and presentation; Anomaly detection: Provides a general anomaly detection algorithm model to support anomaly detection applications in other scenarios.

2. The system according to claim 1, characterized in that Agent components include perception and memory, autonomous planning, tools, and action functions. Perception and memory: used by the system super-intelligence to obtain various original and high-dimensional abstract perception results; memory for storing and managing user information to optimize performance under user personalized needs; Autonomous planning: Based on the implicit or explicit demands of users, autonomously disassemble subtasks and arrange subtask chains to form closed-loop goals; At the same time, it can conduct iterative reflection based on the execution results of subtasks and independently improve planning and scheduling capabilities; Tools: A collection of tools and services that can be called, including system tools and third-party services. Based on interface definition specifications, it enables the retrieval and distribution of massive tools and services. Action: Execute the subtasks according to the plan to achieve the closed-loop task objectives.

3. The system according to claim 1, characterized in that Agent execution includes the following: Execution Agent: Before each step needs to be executed, it uses the output of all previous executions to decide the next execution; Plan and execute agents: decide the complete sequence of operations in advance and then execute all operations without updating the plan; The user tells the Agent the key information, and the Agent decides which tools to use. In fact, there is an object LLM in the Agent, and the final decision on which tools to use is made by the LLM (ChatGLM) in the Agent.

4. The system according to claim 1, characterized in that In response to user complaints, the Agent automatically selects tools to complete the following: Determine the complaint type based on the content of the user's complaint; Call the tool to obtain relevant database and knowledge base content and output the root cause of the complaint; Generate intelligent speech based on complaint root cause analysis; Generate complaint JSON and send it to the dispatch system or early warning system; Write complaint document analysis reports and organize them to form a new knowledge base.

5. The system according to claim 4, characterized in that Based on the ChatGLM3 general model and using the LangChain framework, we build an intelligent complaint customer service Agent model.

6. The system according to claim 1, characterized in that Anomaly detection includes single-indicator anomaly quantification and multi-indicator anomaly quantification with dynamic weights.

7. The system according to claim 6, characterized in that Quantification of single indicator anomaly Given an index X = x1, x2, x3, ...x n , input the current x n+1 , quantify the degree of deviation of the input value, that is, the abnormal index score 1) Detect data anomalies based on algorithms such as isolation forest Parameter settings: contamination: float in (0., 0.5), optional(default=0.1) Remove duplicate abnormal data and get m normal data, m<n X=x1,x2,x3,x4,...x m 2) Calculate the mean and standard deviation of m values Mean: Standard Deviation: 3) Calculate the score of a single indicator Enter the new x value and the calculation formula is as follows: y -s c the re =10 F (x;μ,σ) in, If: F(x; μ, σ) < 0.5, then: F(x;μ,σ)=1-F(x;μ,σ) 4) Score interpretation y The value range of _score is [3.16, 10]. 10 to the power of 0.5 is 3.16, and 10 to the power of 1 is 10. Anomaly indicator scores greater than the specified value are highly abnormal and require special attention.

8. The system according to claim 6, characterized in that Multi-index anomaly quantification with dynamic weights Assume that the data set X consists of n data samples. Based on this data set, m indicators are defined, which can be expressed in mathematical language as follows: X1,X2,X3,...X m Among them, X i =x1,x2,...x n ; 1) Data normalization Use the maximum and minimum values ​​for normalization and map the index of X to the interval [0, 100]. The formula is as follows: 2) Calculate data volatility The volatility of each X is replaced by the standard deviation of X. The larger the standard deviation, the more information the indicator can reflect, and the more weight is given to this indicator. The formula is as follows: 3) Calculate data contribution The larger the linear positive correlation coefficient between indicators, the smaller the weight is; the higher the contribution, the higher the weight is. Where rij represents the correlation coefficient between the i-th indicator and the j-th indicator, and this value can be obtained from the covariance matrix of the indicator; 4) Calculate the information carrying capacity Information carrying capacity = volatility * contribution, that is Cj=sj*Rj 5) Calculate indicator weights The weight of each indicator is the contribution ratio of each indicator information carrying capacity Cj, that is, If the current values ​​of X1, X2, and X3 give anomaly scores of E1, E2, and E3, then the overall anomaly perception Y=W1*E1+W2*E2+W3*E3 In this way, the quantification of abnormal indicators and the dynamic allocation of weights of different indicators are completed, and finally a Y value is obtained to represent the final score of the entire abnormality.

Images