An intelligent audit method and system based on environmental impact assessment form data mining

Abstract

The present invention relates to an intelligent audit method and system based on data mining of environmental impact assessment forms, belonging to the field of artificial intelligence technology. The method comprises: first obtaining entity text data, constructing a legal knowledge map through model processing such as entity extraction and relationship extraction; then obtaining a quantitative analysis table, processing it with an extraction model, combining it with the legal knowledge map, and judging compliance through a decision tree; if the judgment result is non-compliant, generating a report and transmitting it to a human-computer interaction interface; if the judgment result is compliant, converting the evaluation form format, extracting information, and classifying feature codes to obtain basic and predicted data; preprocessing the predicted data and calculating the environmental impact prediction results through a prediction model, and finally transmitting the environmental impact prediction results to the human-computer interaction interface. Data mining and intelligent auditing of environmental impact assessment forms are realized.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to an intelligent audit method and system based on environmental impact assessment form data mining. Background Art

[0002] As environmental issues are increasingly valued, environmental impact assessment work has become increasingly important. As an important document for assessing the environmental impact of construction projects, the accuracy and efficiency of the environmental impact assessment form is crucial.

[0003] Traditional environmental impact assessment form review mainly relies on manual work, which has many disadvantages: (1) Inefficiency: Manual review requires a lot of time and effort to check the content of the evaluation form line by line and word by word. Especially for evaluation forms with long content and large amounts of data, the review speed is slow and it is difficult to meet the growing project review needs.

[0004] (2) High subjectivity: Different auditors may have different understandings and grasps of environmental laws and regulations, resulting in inconsistent audit standards and a lack of objectivity and fairness in the audit results.

[0005] (3) Key information may be missed: Manual review may miss some key information in the evaluation form due to fatigue, negligence, etc., making it impossible to fully and accurately assess the impact of the project on the environment.

[0006] In addition, as environmental laws and regulations are constantly updated and improved, the content of the laws and regulations is becoming increasingly complex, and manual audits are difficult to quickly and accurately grasp the latest regulatory requirements and apply them to audit work. At the same time, the environmental impact assessment form involves environmental impact data on air, water, sound, solid waste, etc. Traditional audit methods are difficult to deeply mine and effectively analyze these complex data, and cannot fully utilize the value of the data to assist decision-making. Therefore, there is an urgent need for an intelligent audit method to improve the efficiency and quality of environmental impact assessment form audits. Summary of the invention

[0007] In order to solve the above problems existing in the prior art, the present invention provides an intelligent audit method and system based on environmental impact assessment form data mining. The purpose of the present invention can be achieved through the following technical solutions: Acquire entity text data containing semantic information of environmental regulations, calculate an entity recognition result sequence through an entity extraction model according to the entity text data, calculate a relationship recognition result sequence through a relationship extraction model according to the entity recognition result sequence and the entity text data, calculate an entity similarity sequence according to the entity recognition result sequence through a cosine similarity calculation formula, clean the entity recognition result sequence according to the entity similarity sequence to obtain a cleaned entity sequence, construct a triple sequence with the cleaned entity sequence and the relationship recognition result sequence, and store the triple sequence in a graph database to obtain a regulatory knowledge graph; Obtain a quantitative analysis table covering the production management links based on environmental factors, evaluation and management processes, calculate a second entity recognition result sequence and a second relationship recognition result sequence through an extraction model according to the quantitative analysis table, obtain a mapping entity sequence through mapping according to the second entity recognition result sequence and the entity recognition result sequence, construct a second triple sequence according to the mapping entity sequence and the second relationship recognition result sequence, and perform compliance judgment through a decision tree according to the second triple sequence and the regulatory knowledge graph to obtain a judgment result, wherein the judgment result includes compliance and non-compliance; If the judgment result is non-compliant, a report is generated, and the judgment result and the report are transmitted to the human-computer interaction interface through the API interface; if the judgment result is compliant, the format conversion is performed according to the quantitative analysis table to obtain a conversion evaluation table, and information is extracted according to the conversion evaluation table through natural language processing technology to obtain key information data, and feature encoding is performed according to the key information data through unique hot encoding to obtain encoded data, and the encoded data is classified by a classifier to obtain basic information and environmental impact prediction data; Add a timestamp to the environmental impact prediction data to obtain time series prediction data, store the time series prediction data to obtain a prediction data sequence, calculate the time series data prediction result through a prediction model based on the prediction data sequence, calculate the environmental impact prediction result through a second prediction model based on the time series data prediction result and the basic information, and transmit the environmental impact prediction result to the human-computer interaction interface through the API interface.

[0008] Specifically, the specific calculation steps of the entity extraction model include: According to the entity text data, word segmentation and part-of-speech recognition are performed by a word segmenter to obtain a word sequence and a part-of-speech sequence; according to the entity text data, the word sequence and the part-of-speech sequence, encoding and serialization are performed to obtain a character input sequence, a word input sequence and a part-of-speech input sequence; according to the character input sequence, the word input sequence and the part-of-speech input sequence, an encoder is used to encode to obtain a character hidden state vector, a word hidden state vector and a part-of-speech hidden state vector; according to the character hidden state vector, a multi-head attention mechanism is used to fuse with the word hidden state vector and the part-of-speech hidden state vector to obtain a fused word representation and a fused part-of-speech representation; the character input sequence, the fused word representation and the fused part-of-speech representation are vector-concatenated to obtain a text semantic vector; The text semantic vector is calculated by a boundary learner to obtain a starting probability matrix and an ending probability matrix, a left boundary vector and a right boundary vector are calculated by a segment classifier according to the text semantic vector, the left boundary vector and the right boundary vector are spliced ​​to obtain a segment hidden state vector, a segment category probability is calculated by a normalization layer according to the segment hidden state vector, the segment category probabilities are integrated to obtain an entity category probability matrix, qualified segments are calculated by a diagnostic formula according to the starting probability matrix, the ending probability matrix, and the entity category probability matrix, an entity recognition result is calculated by a decoder according to the qualified segment, and the entity recognition result is stored to obtain the entity recognition result sequence; The fusion word representation expression is: , in, H 1 is the fusion word representation, Q 1 is the hidden state vector of the word, W 1 is the word mapping matrix, Q 2 is the hidden state vector of the word, W 2 is the word mapping matrix; The fused part-of-speech representation expression is: , in, H 2 To integrate part-of-speech representations, Q 3 is the part-of-speech hidden state vector, W 3 is the part-of-speech mapping matrix; The text semantic vector expression is: , in, H y is the text semantic vector,W is the adaptive weight parameter, p is the bias parameter; The expression of the diagnostic formula is: , in, P s i is the starting position probability of the fragment with starting position i in the starting probability matrix, P e j is the termination position probability of the segment with termination position j in the termination probability matrix, P ij t The entity category probability of the segment starting at position i and ending at position j in category t, a is the preset threshold, nspan The maximum span of the entity.

[0009] Specifically, the specific calculation steps of the relationship extraction model include: Calculate a context information vector using the ALBERT model according to the entity text data, connect the context information vector and the entity recognition result sequence to obtain an entity relationship connectome, calculate a relationship recognition result using a sigmoid function according to the entity relationship connectome, and store the relationship recognition result to obtain the relationship recognition result sequence; The entity relationship connector expression is: , Among them, R is the entity relationship connector, E1 is the element in the entity recognition result sequence, E2 is the element in the entity recognition result sequence, cls is the context information vector, and º represents the connection symbol; The relationship recognition result expression is: , Among them, R1 is the relationship recognition result, w is the weight matrix, and p is the bias parameter.

[0010] Specifically, the extraction model includes the entity extraction model and the relationship extraction model.

[0011] Specifically, the environmental impact prediction data includes atmospheric environmental impact data, water environmental impact data, acoustic environmental impact data, and solid waste environmental impact data.

[0012] Specifically, the basic information includes basic information on air pollution, basic information on water pollution, basic information on noise pollution, and basic information on solid waste.

[0013] Specifically, the specific calculation steps of the prediction model include: Filling missing values ​​according to the predicted data sequence by linear interpolation to obtain a filling sequence, presetting a moving window, intercepting the filling sequence according to the moving window to obtain a predicted intercepted segment, and calculating the predicted intercepted segment by a prediction algorithm to obtain the time series data prediction result; The prediction algorithm expression is: , in, H For the prediction results of time series data, k is the basic growth rate, b is the growth rate change vector, s(t) To predict the clip, t is the time variable, m To move the window size, c To adjust the intercept vector at the change point, n is a variable parameter, N is the Fourier order, a n , b n is the coefficient sequence, l is a variable parameter, L Gather for special moments, Z l （t） is a special time indicator function, K l is the influence of the special moment set on the predicted intercepted segment, d is the residual term.

[0014] An intelligent audit system based on environmental impact assessment form data mining, comprising: a construction module, a judgment module, an extraction module, and a prediction module; The construction module is used to obtain entity text data containing environmental regulatory semantic information, calculate an entity recognition result sequence through an entity extraction model based on the entity text data, calculate a relationship recognition result sequence through a relationship extraction model based on the entity recognition result sequence and the entity text data, calculate an entity similarity sequence through a cosine similarity calculation formula based on the entity recognition result sequence, clean the entity recognition result sequence based on the entity similarity sequence to obtain a cleaned entity sequence, construct a triple sequence with the cleaned entity sequence and the relationship recognition result sequence, and store the triple sequence in a graph database to obtain a regulatory knowledge graph; The judgment module is used to obtain a quantitative analysis table covering the production management links based on environmental factors, evaluation and management processes, calculate a second entity recognition result sequence and a second relationship recognition result sequence through an extraction model according to the quantitative analysis table, obtain a mapping entity sequence through mapping according to the second entity recognition result sequence and the entity recognition result sequence, construct a second triple sequence according to the mapping entity sequence and the second relationship recognition result sequence, and perform compliance judgment through a decision tree according to the second triple sequence and the regulatory knowledge graph to obtain a judgment result, wherein the judgment result includes compliance and non-compliance; The extraction module is used to generate a report if the judgment result is non-compliant, and transmit the judgment result and the report to the human-computer interaction interface through the API interface; if the judgment result is compliant, perform format conversion according to the quantitative analysis table to obtain a conversion evaluation table, perform information extraction according to the conversion evaluation table through natural language processing technology to obtain key information data, perform feature encoding according to the key information data through unique hot encoding to obtain encoded data, and classify according to the encoded data through a classifier to obtain basic information and environmental impact prediction data; The prediction module is used to add a timestamp to the environmental impact prediction data to obtain time series prediction data, store the time series prediction data to obtain a prediction data sequence, calculate the time series data prediction result through a prediction model based on the prediction data sequence, calculate the environmental impact prediction result through a second prediction model based on the time series data prediction result and the basic information, and transmit the environmental impact prediction result to the human-computer interaction interface through the API interface.

[0015] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the intelligent audit method based on environmental impact assessment form data mining as described above is implemented.

[0016] A storage medium containing computer executable instructions, which, when executed by a computer processor, are used to execute the intelligent audit method based on environmental impact assessment form data mining as described above.

[0017] The beneficial effects of the present invention are: (1) Improving audit efficiency: The present invention obtains entity text data and uses automated processing methods such as entity extraction models and relationship extraction models to quickly calculate the regulatory knowledge graph and perform intelligent analysis on the quantitative analysis table. It can automatically complete a series of processes from regulatory analysis to evaluation table review. Compared with traditional manual review, it greatly shortens the review time and significantly improves the review efficiency, which can meet the needs of rapid review of a large number of projects.

[0018] (2) Enhanced audit accuracy: Model-based entity recognition, relationship recognition, and compliance judgment, based on objective algorithms and data, avoid the problem of inconsistent audit standards caused by subjective factors in manual audits, ensure that the audit results are more objective and accurate, and be able to comprehensively and accurately identify possible non-compliance issues in the environmental impact assessment form.

[0019] (3) In-depth data mining and analysis: It can not only determine whether the environmental impact assessment form is compliant, but also, if compliant, perform format conversion, information extraction, feature coding and classification on the assessment form to obtain environmental impact prediction data such as atmospheric environmental impact data and water environmental impact data, as well as various basic information. Through further processing and analysis of these data, such as adding timestamps and building prediction models, time series prediction of environmental impacts can be achieved, providing more forward-looking and in-depth data support for environmental management and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0021] Figure 1 The present invention is a flowchart of an intelligent audit method based on environmental impact assessment form data mining. DETAILED DESCRIPTION

[0022] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0023] See also Figure 1 , an intelligent audit method based on environmental impact assessment form data mining, Acquire entity text data containing semantic information of environmental regulations, calculate an entity recognition result sequence through an entity extraction model according to the entity text data, calculate a relationship recognition result sequence through a relationship extraction model according to the entity recognition result sequence and the entity text data, calculate an entity similarity sequence according to the entity recognition result sequence through a cosine similarity calculation formula, clean the entity recognition result sequence according to the entity similarity sequence to obtain a cleaned entity sequence, construct a triple sequence with the cleaned entity sequence and the relationship recognition result sequence, and store the triple sequence in a graph database to obtain a regulatory knowledge graph; Obtain a quantitative analysis table covering the production management links based on environmental factors, evaluation and management processes, calculate a second entity recognition result sequence and a second relationship recognition result sequence through an extraction model according to the quantitative analysis table, obtain a mapping entity sequence through mapping according to the second entity recognition result sequence and the entity recognition result sequence, construct a second triple sequence according to the mapping entity sequence and the second relationship recognition result sequence, and perform compliance judgment through a decision tree according to the second triple sequence and the regulatory knowledge graph to obtain a judgment result, wherein the judgment result includes compliance and non-compliance; If the judgment result is non-compliant, a report is generated, and the judgment result and the report are transmitted to the human-computer interaction interface through the API interface; if the judgment result is compliant, the format conversion is performed according to the quantitative analysis table to obtain a conversion evaluation table, and information is extracted according to the conversion evaluation table through natural language processing technology to obtain key information data, and feature encoding is performed according to the key information data through unique hot encoding to obtain encoded data, and the encoded data is classified by a classifier to obtain basic information and environmental impact prediction data; Add a timestamp to the environmental impact prediction data to obtain time series prediction data, store the time series prediction data to obtain a prediction data sequence, calculate the time series data prediction result through a prediction model based on the prediction data sequence, calculate the environmental impact prediction result through a second prediction model based on the time series data prediction result and the basic information, and transmit the environmental impact prediction result to the human-computer interaction interface through the API interface.

[0024] In this embodiment, the triple expression is: , in, Y is a triple, X 1 is the entity recognition result, X 2 is the entity recognition result, a Identify results for the relationship.

[0025] It should be noted that both X1 and X2 are entity recognition results in the cleaned entity sequence, but their positions in the sequence are different.

[0026] Specifically, the specific calculation steps of the entity extraction model include: According to the entity text data, word segmentation and part-of-speech recognition are performed by a word segmenter to obtain a word sequence and a part-of-speech sequence; according to the entity text data, the word sequence and the part-of-speech sequence, encoding and serialization are performed to obtain a character input sequence, a word input sequence and a part-of-speech input sequence; according to the character input sequence, the word input sequence and the part-of-speech input sequence, an encoder is used to encode to obtain a character hidden state vector, a word hidden state vector and a part-of-speech hidden state vector; according to the character hidden state vector, a multi-head attention mechanism is used to fuse with the word hidden state vector and the part-of-speech hidden state vector to obtain a fused word representation and a fused part-of-speech representation; the character input sequence, the fused word representation and the fused part-of-speech representation are vector-concatenated to obtain a text semantic vector; The text semantic vector is calculated by a boundary learner to obtain a starting probability matrix and an ending probability matrix, a left boundary vector and a right boundary vector are calculated by a segment classifier according to the text semantic vector, the left boundary vector and the right boundary vector are spliced ​​to obtain a segment hidden state vector, a segment category probability is calculated by a normalization layer according to the segment hidden state vector, the segment category probabilities are integrated to obtain an entity category probability matrix, qualified segments are calculated by a diagnostic formula according to the starting probability matrix, the ending probability matrix, and the entity category probability matrix, an entity recognition result is calculated by a decoder according to the qualified segment, and the entity recognition result is stored to obtain the entity recognition result sequence; The fusion word representation expression is: , in, H 1 is the fusion word representation, Q 1 is the hidden state vector of the word, W 1 is the word mapping matrix, Q 2 is the hidden state vector of the word, W 2 is the word mapping matrix; The fused part-of-speech representation expression is: , in, H 2 To integrate part-of-speech representations, Q 3 is the part-of-speech hidden state vector, W 3 is the part-of-speech mapping matrix; The text semantic vector expression is: , in, H y is the text semantic vector,W is the adaptive weight parameter, p is the bias parameter; The expression of the diagnostic formula is: , in, P s i is the starting position probability of the fragment with starting position i in the starting probability matrix, P e j is the termination position probability of the segment with termination position j in the termination probability matrix, P ij t The entity category probability of the segment starting at position i and ending at position j in category t, a is the preset threshold, nspan The maximum span of the entity.

[0027] Specifically, the specific calculation steps of the relationship extraction model include: Calculate a context information vector using the ALBERT model according to the entity text data, connect the context information vector and the entity recognition result sequence to obtain an entity relationship connectome, calculate a relationship recognition result using a sigmoid function according to the entity relationship connectome, and store the relationship recognition result to obtain the relationship recognition result sequence; The entity relationship connector expression is: , Among them, R is the entity relationship connector, E1 is the element in the entity recognition result sequence, E2 is the element in the entity recognition result sequence, cls is the context information vector, and º represents the connection symbol; It should be noted that both E1 and E2 are elements in the entity recognition result sequence, but their positions in the sequence are different; The relationship recognition result expression is: , Among them, R1 is the relationship recognition result, w is the weight matrix, and p is the bias parameter.

[0028] Specifically, the extraction model includes the entity extraction model and the relationship extraction model.

[0029] Specifically, the environmental impact prediction data includes atmospheric environmental impact data, water environmental impact data, acoustic environmental impact data, and solid waste environmental impact data.

[0030] In this embodiment, the atmospheric environment impact data include: outlet flue gas temperature, flue gas generation concentration, flue gas generation amount, flue gas emission concentration, flue gas emission amount, flue gas emission rate, and wind speed; the water environment impact data include: water pollutant generation concentration, water pollutant generation amount, water pollutant emission concentration, water pollutant emission amount, wastewater flow, pH value, and chemical oxygen demand; the acoustic environment impact data include: noise sound pressure level and noise frequency; the solid waste environment impact data include solid waste generation amount, moisture content, calorific value, and toxic and harmful substance content.

[0031] Specifically, the basic information includes basic information on air pollution, basic information on water pollution, basic information on noise pollution, and basic information on solid waste.

[0032] In this embodiment, the basic information on air pollution includes: the location of air pollution emission sources, chimney height, chimney inner diameter, annual average temperature, annual precipitation, dominant wind direction, and atmospheric stability; the basic information on water pollution includes: the location of sewage discharge sources, sewage discharge methods, discharge destinations, types of sewage pollutants, the location of receiving water bodies, the names of receiving water bodies, the hydrological characteristics of receiving water bodies, and the current water quality of receiving water bodies; the basic information on sound pollution includes: the location of sound pollution sources, the number of sound pollution sources, the operation time of sound pollution sources, topography, ground coverage, and surrounding sound environment functional areas; the basic information on solid waste includes: the source of solid waste, the type of solid waste, the composition of solid waste, the scale of solid waste, the solid waste treatment capacity, and the solid waste treatment process.

[0033] Specifically, the specific calculation steps of the prediction model include: Filling missing values ​​according to the predicted data sequence by linear interpolation to obtain a filling sequence, presetting a moving window, intercepting the filling sequence according to the moving window to obtain a predicted intercepted segment, and calculating the predicted intercepted segment by a prediction algorithm to obtain the time series data prediction result; The prediction algorithm expression is: , in, H For the prediction results of time series data, k is the basic growth rate, b is the growth rate change vector, s(t) To predict the clip, t is the time variable, m To move the window size, c To adjust the intercept vector at the change point, n is a variable parameter, N is the Fourier order, a n , b nis the coefficient sequence, l is a variable parameter, L Gather for special moments, Z l （t） is a special time indicator function, K l is the influence of the special moment set on the predicted intercepted segment, d is the residual term.

[0034] Specifically, the second prediction model includes an atmospheric environment impact prediction model, a water environment impact prediction model, an acoustic environment impact prediction model, and a solid waste environment impact prediction model.

[0035] In this embodiment, the atmospheric environmental impact prediction model includes but is not limited to the AERMOD model, the ADMS model, and the CALPUFF model; the water environmental impact prediction model includes but is not limited to the QUAL2K model and the EFDC model; the acoustic environmental impact prediction model includes but is not limited to the Candna / A model and the soundPLAN model; the solid waste environmental impact prediction model includes but is not limited to the LEACHM model and the LandGEM model.

[0036] An intelligent audit system based on environmental impact assessment form data mining, comprising: a construction module, a judgment module, an extraction module, and a prediction module; The construction module is used to obtain entity text data containing environmental regulatory semantic information, calculate an entity recognition result sequence through an entity extraction model based on the entity text data, calculate a relationship recognition result sequence through a relationship extraction model based on the entity recognition result sequence and the entity text data, calculate an entity similarity sequence through a cosine similarity calculation formula based on the entity recognition result sequence, clean the entity recognition result sequence based on the entity similarity sequence to obtain a cleaned entity sequence, construct a triple sequence with the cleaned entity sequence and the relationship recognition result sequence, and store the triple sequence in a graph database to obtain a regulatory knowledge graph; The judgment module is used to obtain a quantitative analysis table covering the production management links based on environmental factors, evaluation and management processes, calculate a second entity recognition result sequence and a second relationship recognition result sequence through an extraction model according to the quantitative analysis table, obtain a mapping entity sequence through mapping according to the second entity recognition result sequence and the entity recognition result sequence, construct a second triple sequence according to the mapping entity sequence and the second relationship recognition result sequence, and perform compliance judgment through a decision tree according to the second triple sequence and the regulatory knowledge graph to obtain a judgment result, wherein the judgment result includes compliance and non-compliance; The extraction module is used to generate a report if the judgment result is non-compliant, and transmit the judgment result and the report to the human-computer interaction interface through the API interface; if the judgment result is compliant, perform format conversion according to the quantitative analysis table to obtain a conversion evaluation table, perform information extraction according to the conversion evaluation table through natural language processing technology to obtain key information data, perform feature encoding according to the key information data through unique hot encoding to obtain encoded data, and classify according to the encoded data through a classifier to obtain basic information and environmental impact prediction data; The prediction module is used to add a timestamp to the environmental impact prediction data to obtain time series prediction data, store the time series prediction data to obtain a prediction data sequence, calculate the time series data prediction result through a prediction model based on the prediction data sequence, calculate the environmental impact prediction result through a second prediction model based on the time series data prediction result and the basic information, and transmit the environmental impact prediction result to the human-computer interaction interface through the API interface.

[0037] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the intelligent audit method based on environmental impact assessment form data mining as described above is implemented.

[0038] A storage medium containing computer executable instructions, which, when executed by a computer processor, are used to execute the intelligent audit method based on environmental impact assessment form data mining as described above.

[0039] The computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.

[0040] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0041] The program code included on the computer readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination of the above. The computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or their combinations, and the programming language includes object-oriented programming languages-such as Java, Smalltalk, C++, and also includes conventional procedural programming languages-such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0042] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. An intelligent audit method based on environmental impact assessment form data mining, characterized in that: include: Acquire entity text data containing semantic information of environmental regulations, calculate an entity recognition result sequence through an entity extraction model according to the entity text data, calculate a relationship recognition result sequence through a relationship extraction model according to the entity recognition result sequence and the entity text data, calculate an entity similarity sequence according to the entity recognition result sequence through a cosine similarity calculation formula, clean the entity recognition result sequence according to the entity similarity sequence to obtain a cleaned entity sequence, construct a triple sequence with the cleaned entity sequence and the relationship recognition result sequence, and store the triple sequence in a graph database to obtain a regulatory knowledge graph; Obtain a quantitative analysis table covering the production management links based on environmental factors, evaluation and management processes, calculate a second entity recognition result sequence and a second relationship recognition result sequence through an extraction model according to the quantitative analysis table, obtain a mapping entity sequence through mapping according to the second entity recognition result sequence and the entity recognition result sequence, construct a second triple sequence according to the mapping entity sequence and the second relationship recognition result sequence, and perform compliance judgment through a decision tree according to the second triple sequence and the regulatory knowledge graph to obtain a judgment result, wherein the judgment result includes compliance and non-compliance; If the judgment result is non-compliant, a report is generated, and the judgment result and the report are transmitted to the human-computer interaction interface through the API interface; if the judgment result is compliant, the format conversion is performed according to the quantitative analysis table to obtain a conversion evaluation table, and information is extracted according to the conversion evaluation table through natural language processing technology to obtain key information data, and feature encoding is performed according to the key information data through unique hot encoding to obtain encoded data, and the encoded data is classified by a classifier to obtain basic information and environmental impact prediction data; Add a timestamp to the environmental impact prediction data to obtain time series prediction data, store the time series prediction data to obtain a prediction data sequence, calculate the time series data prediction result through a prediction model based on the prediction data sequence, calculate the environmental impact prediction result through a second prediction model based on the time series data prediction result and the basic information, and transmit the environmental impact prediction result to the human-computer interaction interface through the API interface.

2. The intelligent audit method based on environmental impact assessment form data mining according to claim 1 is characterized in that: The specific calculation steps of the entity extraction model include: According to the entity text data, word segmentation and part-of-speech recognition are performed by a word segmenter to obtain a word sequence and a part-of-speech sequence; according to the entity text data, the word sequence and the part-of-speech sequence, encoding and serialization are performed to obtain a character input sequence, a word input sequence and a part-of-speech input sequence; according to the character input sequence, the word input sequence and the part-of-speech input sequence, an encoder is used to encode to obtain a character hidden state vector, a word hidden state vector and a part-of-speech hidden state vector; according to the character hidden state vector, a multi-head attention mechanism is used to fuse with the word hidden state vector and the part-of-speech hidden state vector to obtain a fused word representation and a fused part-of-speech representation; the character input sequence, the fused word representation and the fused part-of-speech representation are vector-concatenated to obtain a text semantic vector; The text semantic vector is calculated by a boundary learner to obtain a starting probability matrix and an ending probability matrix, a left boundary vector and a right boundary vector are calculated by a segment classifier according to the text semantic vector, the left boundary vector and the right boundary vector are spliced ​​to obtain a segment hidden state vector, a segment category probability is calculated by a normalization layer according to the segment hidden state vector, the segment category probabilities are integrated to obtain an entity category probability matrix, qualified segments are calculated by a diagnostic formula according to the starting probability matrix, the ending probability matrix, and the entity category probability matrix, an entity recognition result is calculated by a decoder according to the qualified segment, and the entity recognition result is stored to obtain the entity recognition result sequence; The fusion word representation expression is: ， in, H 1 is the fusion word representation, Q 1 is the hidden state vector of the word, W 1 is the word mapping matrix, Q 2 is the hidden state vector of the word, W 2 is the word mapping matrix; The fused part-of-speech representation expression is: ， in, H 2 To integrate part-of-speech representations, Q 3 is the part-of-speech hidden state vector, W 3 is the part-of-speech mapping matrix; The text semantic vector expression is: ， in, H y is the text semantic vector, W is the adaptive weight parameter, p is the bias parameter; The expression of the diagnostic formula is: ， in, P s i is the starting position probability of the fragment with starting position i in the starting probability matrix, P e j is the termination position probability of the segment with termination position j in the termination probability matrix, P ij t The entity category probability of the segment starting at position i and ending at position j in category t, a is the preset threshold, nspan The maximum span of the entity.

3. The intelligent audit method based on environmental impact assessment form data mining according to claim 1 is characterized in that: The specific calculation steps of the relationship extraction model include: Calculate a context information vector using the ALBERT model according to the entity text data, connect the context information vector and the entity recognition result sequence to obtain an entity relationship connectome, calculate a relationship recognition result using a sigmoid function according to the entity relationship connectome, and store the relationship recognition result to obtain the relationship recognition result sequence; The entity relationship connector expression is: ， Among them, R is the entity relationship connector, E1 is the element in the entity recognition result sequence, E2 is the element in the entity recognition result sequence, cls is the context information vector, and º represents the connection symbol; The relationship recognition result expression is: ， Among them, R1 is the relationship recognition result, w is the weight matrix, and p is the bias parameter.

4. The intelligent audit method based on environmental impact assessment form data mining according to claim 1 is characterized in that: The extraction model includes the entity extraction model and the relationship extraction model.

5. The intelligent audit method based on environmental impact assessment form data mining according to claim 1 is characterized in that: The environmental impact prediction data includes atmospheric environmental impact data, water environmental impact data, acoustic environmental impact data, and solid waste environmental impact data.

6. The intelligent audit method based on environmental impact assessment form data mining according to claim 1 is characterized in that: The basic information includes basic information on air pollution, basic information on water pollution, basic information on noise pollution, and basic information on solid waste.

7. The intelligent audit method based on environmental impact assessment form data mining according to claim 1 is characterized in that: The specific calculation steps of the prediction model include: Filling missing values ​​according to the predicted data sequence by linear interpolation to obtain a filling sequence, presetting a moving window, intercepting the filling sequence according to the moving window to obtain a predicted intercepted segment, and calculating the predicted intercepted segment by a prediction algorithm to obtain the time series data prediction result; The prediction algorithm expression is: ， in, H For the prediction results of time series data, k is the basic growth rate, b is the growth rate change vector, s(t) To predict the clip, t is the time variable, m To move the window size, c To adjust the intercept vector at the change point, n is a variable parameter, N is the Fourier order, a n , b n is the coefficient sequence, l is a variable parameter, L Gather for special moments, Z l （t） is a special time indicator function, K l is the influence of the special moment set on the predicted intercepted segment, d is the residual term.

8. An intelligent audit system based on environmental impact assessment form data mining, characterized in that: include: Construction module, judgment module, extraction module, prediction module; The construction module is used to obtain entity text data containing environmental regulatory semantic information, calculate an entity recognition result sequence through an entity extraction model based on the entity text data, calculate a relationship recognition result sequence through a relationship extraction model based on the entity recognition result sequence and the entity text data, calculate an entity similarity sequence through a cosine similarity calculation formula based on the entity recognition result sequence, clean the entity recognition result sequence based on the entity similarity sequence to obtain a cleaned entity sequence, construct a triple sequence with the cleaned entity sequence and the relationship recognition result sequence, and store the triple sequence in a graph database to obtain a regulatory knowledge graph; The judgment module is used to obtain a quantitative analysis table covering the production management links based on environmental factors, evaluation and management processes, calculate a second entity recognition result sequence and a second relationship recognition result sequence through an extraction model according to the quantitative analysis table, obtain a mapping entity sequence through mapping according to the second entity recognition result sequence and the entity recognition result sequence, construct a second triple sequence according to the mapping entity sequence and the second relationship recognition result sequence, and perform compliance judgment through a decision tree according to the second triple sequence and the regulatory knowledge graph to obtain a judgment result, wherein the judgment result includes compliance and non-compliance; The extraction module is used to generate a report if the judgment result is non-compliant, and transmit the judgment result and the report to the human-computer interaction interface through the API interface; if the judgment result is compliant, perform format conversion according to the quantitative analysis table to obtain a conversion evaluation table, perform information extraction according to the conversion evaluation table through natural language processing technology to obtain key information data, perform feature encoding according to the key information data through unique hot encoding to obtain encoded data, and classify according to the encoded data through a classifier to obtain basic information and environmental impact prediction data; The prediction module is used to add a timestamp to the environmental impact prediction data to obtain time series prediction data, store the time series prediction data to obtain a prediction data sequence, calculate the time series data prediction result through a prediction model based on the prediction data sequence, calculate the environmental impact prediction result through a second prediction model based on the time series data prediction result and the basic information, and transmit the environmental impact prediction result to the human-computer interaction interface through the API interface.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the intelligent audit method based on environmental impact assessment form data mining as described in any one of claims 1-7 is implemented.

10. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions, when executed by a computer processor, are used to execute the intelligent audit method based on environmental impact assessment form data mining as described in any one of claims 1 to 7.

Images