Harmful gas detection text processing method and system based on big data

Abstract

The embodiment of the invention relates to the technical field of data processing, in particular to a harmful gas detection text processing method and system based on big data. The method comprises the following steps: firstly, acquiring to-be-analyzed gas sensing and monitoring text data of a target power production scene and pre-generated original gas sensing and monitoring text data corresponding to various original secondary abnormal state trend tags; converting the text data into semantic vector representation by utilizing a pre-debugged target text semantic mining model; carrying out key semantic attention processing on the semantic vector by utilizing a target semantic optimization model; and finally, comprehensively analyzing the processed semantic vector by using a target trend discrimination model and giving a harmful gas early warning output viewpoint. By means of the method, key semantic information related to gas state judgment can be effectively extracted, various influence factors can be comprehensively considered, accurate early warning judgment can be given, and therefore powerful technical support is provided for safety prevention and control in the power production process.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and more specifically, to a method and system for processing harmful gas detection texts based on big data. Background Art

[0002] With the rapid development of the power industry, gas safety monitoring in the power production process has become increasingly important. Timely and accurate monitoring and early warning of potential harmful gas states are of great significance for ensuring power production safety and preventing accidents. However, traditional gas monitoring methods usually rely on single sensor threshold judgment, unable to fully consider the complexity and diversity of gas state changes, resulting in limited accuracy and real-time performance of early warnings.

[0003] In recent years, with the continuous development of artificial intelligence and machine learning technologies, gas safety monitoring methods based on text semantic mining have gradually become a research hotspot. By analyzing and understanding the deep semantic information in gas sensing monitoring text data, this method can more accurately capture the changing trends of gas states, thus providing more reliable support for safety monitoring in the power production process.

[0004] However, there are still certain challenges in applying existing text semantic mining methods to gas safety monitoring. On the one hand, gas sensing monitoring text data in power production scenarios usually contains a large amount of noise and irrelevant information, and how to extract key semantic information related to gas state judgment from it is a difficult problem. On the other hand, the changing trends of gas states are usually affected by various factors, and how to comprehensively consider various factors and give accurate early warning judgments is also an important research direction. Summary of the Invention

[0005] To improve the technical problems existing in related technologies, this application provides a method and system for processing harmful gas detection texts based on big data.

[0006] In a first aspect, an embodiment of this application provides a method for processing harmful gas detection texts based on big data, which is applied to a detection text processing system. The method includes: Obtaining the text data of gas sensing monitoring to be analyzed in a target power production scenario, and obtaining the text data of each original gas sensing monitoring corresponding to the generation of various original secondary abnormal state trend labels; Inputting the text data of gas sensing monitoring to be analyzed into a target text semantic mining model that has completed pre-debugging to obtain corresponding basic sensing monitoring text semantic vectors, and inputting the text data of each original gas sensing monitoring into the target text semantic mining model to obtain corresponding reference sensing monitoring text semantic vectors; Using the target semantic optimization model that has completed pre - debugging, and based on the knowledge of each state trend element parsed from the gas sensing monitoring text debugging data for different trend tags during the pre - debugging process, perform key semantic attention processing on the basic sensing monitoring text semantic vector and each reference sensing monitoring text semantic vector respectively, to obtain the corresponding target sensing monitoring text semantic vector and each target sensing monitoring text semantic reference vector; Using the target trend discrimination model that has completed pre - debugging, and based on the semantic commonality score between the target sensing monitoring text semantic vector and each target sensing monitoring text semantic reference vector, determine the harmful gas warning output view corresponding to the gas sensing monitoring text data to be analyzed.

[0007] Combined with the first aspect, in a possible implementation manner of the first aspect, performing key semantic attention processing on the basic sensing monitoring text semantic vector based on the knowledge of each state trend element parsed from the gas sensing monitoring text debugging data for different trend tags during the pre - debugging process includes: Based on the knowledge of each state trend element parsed from the gas sensing monitoring text debugging data for different trend tags during the pre - debugging process, respectively determine the common description between the basic sensing monitoring text semantic vector and each state trend element knowledge; Based on the common description, determine the adaptability confidence between the basic sensing monitoring text semantic vector and each state trend element knowledge; Based on the adaptability confidence, determine the attention weight coefficient of each state trend element knowledge for each semantic distribution variable value in the basic sensing monitoring text semantic vector, and based on each attention weight coefficient, realize the key semantic attention optimization of the basic sensing monitoring text semantic vector.

[0008] Combined with the first aspect, in a possible implementation manner of the first aspect, the step of respectively determining the common description between the basic sensing monitoring text semantic vector and each state trend element knowledge based on the knowledge of each state trend element parsed from the gas sensing monitoring text debugging data for different trend tags during the pre - debugging process includes: Obtain the knowledge of each state trend element parsed from the gas sensing monitoring text debugging data for different trend tags during the pre - debugging process, perform feature mapping on each state trend element knowledge to obtain the state trend element mapping relationship network corresponding to each state trend element knowledge, and perform feature mapping on the basic sensing monitoring text semantic vector to obtain the corresponding monitoring text semantic mapping relationship network; Determine the relational network feature operation result between the state trend element mapping relational network and the monitoring text semantic mapping relational network, and obtain a commonality index list reflecting the common description between the basic sensing monitoring text semantic vector and each state trend element knowledge.

[0009] Combined with the first aspect, in a possible implementation manner of the first aspect, the method for determining the adaptability confidence between the basic sensing monitoring text semantic vector and each state trend element knowledge according to the common description, and determining the attention weight coefficient of each state trend element knowledge for each semantic distribution variable value in the basic sensing monitoring text semantic vector includes: Perform interval numerical mapping processing on the common variables of each list unit in the commonality index list, and obtain a confidence relationship spectrum characterizing the adaptability confidence between the basic sensing monitoring text semantic vector and each state trend element knowledge according to the common variables of each list unit after the interval numerical mapping processing; Determine the attention weight coefficient of each state trend element knowledge for the semantic vector value at each list unit in the basic sensing monitoring text semantic vector according to the relational network feature operation result between the knowledge vector relational network generated by each state trend element knowledge and the confidence relationship spectrum.

[0010] Combined with the first aspect, in a possible implementation manner of the first aspect, the method for optimizing the key semantic attention of the basic sensing monitoring text semantic vector according to each attention weight coefficient includes: Perform semantic vector value accumulation processing on the attention weight coefficient of each list unit and the semantic vector value of the corresponding list unit in the basic sensing monitoring text semantic vector to obtain the semantic vector value accumulation result corresponding to each list unit; Dynamically adjust the semantic vector value accumulation results corresponding to each list unit to achieve the focusing of the key semantics in the basic sensing monitoring text semantic vector.

[0011] Combined with the first aspect, in a possible implementation manner of the first aspect, the method for obtaining each original gas sensing monitoring text data corresponding to each original secondary abnormal state trend label in advance includes: In response to the task information of the monitoring platform system, determine each original secondary abnormal state trend label targeted by the task information, and respectively obtain each original gas sensing monitoring text data generated in advance for each original secondary abnormal state trend label.

[0012] In combination with the first aspect, in a possible implementation manner of the first aspect, pre-debugging the inspection text processing network including a text semantic mining model, a semantic optimization model, and a trend discrimination model includes: According to the shared debugging example set, performing cyclic pre-debugging on the inspection text processing network in multiple rounds of debugging phases until the set cumulative number of debugging times is reached, and in one debugging phase, perform the following processing: According to the gas sensing monitoring text debugging data group sampled from the local debugging example set, perform cyclic debugging on the inspection text processing network to be debugged for Yuci times to obtain an intermediate inspection text processing network, and determine the discrimination quality evaluation index of the intermediate inspection text processing network according to the gas sensing monitoring text evaluation data sampled from the local test example set, where the local debugging example set and the local test example set are included in the debugging example set; According to the discrimination quality evaluation index determined for the intermediate inspection text processing network in different debugging phases, determine the target inspection text processing network that meets the set debugging requirements from each intermediate inspection text processing network.

[0013] In combination with the first aspect, in a possible implementation manner of the first aspect, selecting gas sensing monitoring text debugging data from the local debugging example set includes: In the local debugging example set, for the preselected X types of secondary abnormal state trend labels, respectively select Y gas sensing monitoring text debugging data as a type of auxiliary debugging data, and respectively select Z non-overlapping gas sensing monitoring text debugging data as the to-be-forewarned debugging data; According to the sampled auxiliary debugging data and to-be-forewarned debugging data, create X*Z gas sensing monitoring text debugging data groups, where each gas sensing monitoring text debugging data group includes one to-be-forewarned debugging data corresponding to one type of secondary abnormal state trend label, where X, Y, and Z are positive integers.

[0014] In combination with the first aspect, in a possible implementation manner of the first aspect, the step of performing cyclic debugging on the inspection text processing network to be debugged for Yuci times according to the gas sensing monitoring text debugging data group sampled from the local debugging example set to obtain an intermediate inspection text processing network includes: Use the generated gas sensing monitoring text debugging data group to perform cyclic debugging on the inspection text processing network to be debugged for Yuci times, and improve the neural network parameters of the inspection text processing network according to the debugging error determined in each round of cyclic process; where in one round of cyclic debugging process, perform the following processing: Enter a gas sensing monitoring text debugging data group obtained into a text processing network to be debugged for inspection, obtain a harmful gas warning debugging view determined for the to-be-warned debugging data in the gas sensing monitoring text debugging data group, and determine a debugging error based on the warning view difference between the harmful gas warning debugging view and the corresponding harmful gas warning certification view.

[0015] Combined with the first aspect, in a possible implementation manner of the first aspect, the step of entering a gas sensing monitoring text debugging data group obtained into a text processing network to be debugged for inspection, and obtaining a harmful gas warning debugging view determined for the to-be-warned debugging data in the gas sensing monitoring text debugging data group includes: Enter the to-be-warned debugging data and various auxiliary debugging data included in a gas sensing monitoring text debugging data group into a text semantic mining model to be debugged, and obtain a sensing monitoring text semantic vector debugging sample and each auxiliary sensing monitoring text semantic vector debugging sample; Use a semantic optimization model to be debugged, and respectively perform key semantic attention processing on the sensing monitoring text semantic vector debugging sample and each auxiliary sensing monitoring text semantic vector debugging sample according to the state trend element knowledge parsed and generated for each trend keyword, to obtain a target sensing monitoring text semantic vector debugging sample and each target auxiliary sensing monitoring text semantic vector debugging sample; Use a trend discrimination model to be debugged, and determine a harmful gas warning output view corresponding to the to-be-warned debugging data according to the semantic commonality score between the target sensing monitoring text semantic vector debugging sample and each target auxiliary sensing monitoring text semantic vector debugging sample.

[0016] In a second aspect, the present application further provides a detection text processing system, including: a memory for storing program instructions and data; a processor for being coupled to the memory and executing the instructions in the memory to implement the method as described above.

[0017] In a third aspect, the present application further provides a computer storage medium containing instructions, which when executed on a processor, implement the method as described above.

[0018] This application proposes a harmful gas early warning method based on deep learning and text semantic mining. By combining advanced text processing technologies and machine learning models, this application can achieve efficient and accurate monitoring and early warning of gas states in power production scenarios. Specifically, this application first obtains the text data of gas sensing monitoring to be analyzed in the target power production scenario and the original gas sensing monitoring text data of the corresponding original secondary abnormal state trends generated in advance; then uses the pre-debugged target text semantic mining model to convert the text data into a semantic vector representation; then uses the target semantic optimization model to perform key semantic attention processing on the semantic vector; finally uses the target trend discrimination model to comprehensively analyze the processed semantic vector and give an output view of harmful gas early warning. Through this method, this application can not only effectively extract the key semantic information related to gas state judgment, but also comprehensively consider various influencing factors and give accurate early warning judgments, thus providing strong technical support for safety prevention and control in the power production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0020] Figure 1 is a schematic flowchart of a method for processing harmful gas detection text based on big data provided by an embodiment of the present application.

[0021] Figure 2 is a structural block diagram of a text processing system 300 for detection provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The following will describe the technical solutions in the present application in conjunction with the accompanying drawings.

[0023] Here, the exemplary embodiments will be described in detail, and the examples are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application.

[0024] It should be noted that the terms "first", "second", etc. in the specification of the present application and the above accompanying drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence.

[0025] The method embodiments provided by the embodiments of the present application can be executed in a detection text processing system, a computer device, or a similar computing device. Taking the operation on a detection text processing system as an example, the detection text processing system may include one or more processors (the processors may include, but are not limited to, processing devices such as a microprocessor MCU or a field programmable gate array FPGA), and a memory for storing data. Optionally, the above detection text processing system may further include a transmission device for communication functions. Those of ordinary skill in the art can understand that the above structure is only illustrative and does not limit the structure of the above detection text processing system. For example, the detection text processing system may further include more or fewer components than those shown above, or have a different configuration from those shown above.

[0026] The memory can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to a harmful gas detection text processing method based on big data in the embodiments of the present application. The processor executes various functional applications and data processing by running the computer program stored in the memory, that is, the above method is implemented. The memory may include a high-speed random access memory, and may further include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some instances, the memory may further include a memory remotely disposed relative to the processor, and these remote memories can be connected to the detection text processing system through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0027] The transmission device is used to receive or send data via a network. Specific examples of the above network may include a wireless network provided by a communication provider of the detection text processing system. In one instance, the transmission device includes a network interface controller (NIC for short), which can be connected to other network devices through a base station and thus communicate with the Internet. In one instance, the transmission device may be a radio frequency (RF for short) module, which is used to communicate with the Internet wirelessly.

[0028] Based on this, please refer to Figure 1 , Figure 1 is a schematic flowchart of a harmful gas detection text processing method based on big data provided by the embodiments of the present application. This method is applied to a detection text processing system and may further include step 110-step 140.

[0029] Step 110, obtain the gas sensing monitoring text data to be analyzed in the target power production scenario, and obtain the original gas sensing monitoring text data corresponding to each original secondary abnormal state trend label in advance.

[0030] In step 110, the target power production scenario refers to the specific power production environment where harmful gas monitoring and early warning are required. This environment can be a power plant, a substation, or any other facility related to power generation, transmission, and distribution. In these scenarios, various gases may be generated due to equipment operation, fuel combustion, or chemical reactions, including some harmful gases that may pose risks to human health or equipment operation. To ensure the safety of staff and the normal operation of equipment, continuous gas monitoring of these scenarios is necessary. For example, a coal-fired power plant is a typical target power production scenario. In this scenario, the combustion of coal generates a large amount of flue gas and waste gas, including harmful gases such as carbon dioxide, sulfides, and nitrides. The concentrations of these gases need to be monitored in real time to ensure that they do not exceed safety standards and have an adverse impact on the working environment and equipment performance of the power plant.

[0031] Gas sensing monitoring text data to be analyzed: It refers to the text-format data that is obtained in real time from gas sensors in the target power production scenario and transmitted to the detection text processing system. These data record the real-time concentrations, change trends, and other relevant information of various gases, and are the basis for harmful gas monitoring and early warning. Taking a coal-fired power plant as an example, the gas sensing monitoring text data to be analyzed may include sulfur dioxide concentration data, nitrogen oxide concentration data, and carbon monoxide concentration data in the flue gas. These data are sent to the detection text processing system in text form, and the system will analyze and process these data to determine whether the current gas state is abnormal or potentially hazardous.

[0032] Original secondary abnormal state trend labels: It refers to the labels formed after experts or systems mark and classify the change trends of gas concentrations during past gas monitoring. These labels reflect the change trends of gas concentrations in different situations, such as "concentration rising", "concentration falling", "concentration fluctuating", etc. They are used as reference bases for subsequent analysis and early warning. In the application scenario of a coal-fired power plant, the original secondary abnormal state trend labels may include "continuous increase in sulfur dioxide concentration", "sharp decrease in nitrogen oxide concentration", etc. These labels are formed based on historical monitoring data and expert knowledge, and they can help the detection text processing system better understand the current gas state and accurately predict future change trends.

[0033] The original gas sensing monitoring text data refers to the gas concentration data in text format obtained and saved by gas sensors during past monitoring processes. These data contain information on the concentration changes of various gases at different time periods and under different environmental conditions, and are important historical materials for harmful gas monitoring and early warning. In the application scenario of coal-fired power plants, the original gas sensing monitoring text data may include the concentration data of various harmful gases in flue gas over the past few months or years. These data are stored in a database and associated with corresponding original secondary abnormal state trend labels. When the detection text processing system needs to analyze new gas sensing monitoring data, it will compare and analyze these new data with the original data to find the commonalities and differences between them, and based on this, determine whether the current gas state is normal or whether there are potential hazards.

[0034] Specifically, in the power production scenario, ensuring safe production is of utmost importance. To achieve this goal, an effective gas monitoring system is essential. This system not only needs to monitor gas concentrations in real time but also has the ability to process and analyze the monitoring data in order to detect abnormal states in a timely manner and take corresponding measures. Below, two key aspects of this system will be introduced in detail: obtaining the gas sensing monitoring text data to be analyzed in the target power production scenario, and obtaining the original gas sensing monitoring text data generated corresponding to various original secondary abnormal state trend labels in advance.

[0035] First, the detection text processing system will obtain in real time the original monitoring data generated by gas sensors in the power production scenario. These data are usually presented in text form and contain information such as the concentration of various gases, the working status of the sensors, and the monitoring timestamps. To ensure the accuracy and integrity of the data, the system will preprocess these original text data, such as removing noise, correcting errors, and performing format conversion.

[0036] Next, the system will use predefined rules and algorithms to convert the preprocessed text data into a structured or semi-structured format for subsequent analysis and processing. This process may involve natural language processing techniques such as word segmentation, part-of-speech tagging, syntactic analysis, etc., as well as the assistance of domain-specific knowledge bases or dictionaries. Through the application of these techniques and tools, the system can more accurately extract the key information related to gas concentration changes in the text.

[0037] Meanwhile, to establish an effective abnormal state recognition mechanism, the system will define a series of secondary abnormal state trend tags in advance based on historical data and expert experience. These tags usually describe various possible abnormal scenarios, such as excessive gas concentration, abnormal concentration change rate, etc. For each tag, the system will generate a corresponding set of original gas sensing monitoring text data, which contains gas concentration changes and relevant context information that match the tag description.

[0038] After obtaining these original gas sensing monitoring text data, the system will perform similar preprocessing and structured transformation operations on them to ensure that they are consistent with the text data to be analyzed in the target power production scenario in terms of format and content. The purpose of doing this is to make it easier to find similarities or differences in subsequent analysis, thus providing a strong basis for judging whether the current gas state is abnormal.

[0039] Finally, by comparing and correlating the text data to be analyzed in the target power production scenario with the pre-generated original text data, the system can identify abnormalities or potential risks in the current gas state and trigger corresponding early warnings or control measures in a timely manner. All of this relies on the support of an efficient and reliable text processing system for detection. Through continuous monitoring and analysis of this system, the gas safety issues in the power production process are effectively guaranteed.

[0040] Step 120: Input the text data of the gas sensing monitoring to be analyzed into the target text semantic mining model that has completed pre-debugging to obtain the corresponding basic sensing monitoring text semantic vectors, and input each piece of the original gas sensing monitoring text data into the target text semantic mining model to obtain the corresponding reference sensing monitoring text semantic vectors.

[0041] In step 120, the target text semantic mining model that has completed pre - debugging is a machine - learning model that has been trained and optimized for extracting and analyzing semantic information in text data. In the gas monitoring application in the power production scenario, this model is used to process the text data generated by gas sensors, thereby converting this raw text data into a semantic vector form that is easier to analyze and understand. Pre - debugging is an important step before the application of this model. It involves using a large amount of historical gas sensing and monitoring text data to train the model and adjusting the parameters and structure of the model to ensure that the model can accurately extract the semantic information in the text. In this process, the model learns how to identify and understand keywords, phrases, and sentence structures related to gas concentration changes, as well as their associations and context information. Once the model has completed pre - debugging, it can be used to process new gas sensing and monitoring text data. By inputting this new data, the model can automatically extract the semantic information in the text and encode it into a vector form for subsequent analysis and processing. For example, in the power production scenario, the target text semantic mining model that has completed pre - debugging can receive a text data describing the change in sulfur dioxide concentration as input and then output a vector representing the semantic information of this text. This vector captures the key information about the change in sulfur dioxide concentration in the text, such as the specific value of the concentration, the change trend, and possible influencing factors, providing an important basis for subsequent harmful gas early warning.

[0042] The basic sensing and monitoring text semantic vector is a vector representation obtained by inputting the gas sensing and monitoring text data to be analyzed into the target text semantic mining model that has completed pre - debugging. This vector captures the key semantic information in the raw text data and is the basis for subsequent analysis and early warning. In the gas monitoring application in the power production scenario, the basic sensing and monitoring text semantic vector represents the current monitored gas state information. It contains real - time concentration data of various gases, change trends, and other key information related to the gas state. By encoding this information into a vector form, it is convenient for numerical calculation and analysis, thereby realizing automated gas state assessment and early warning. For example, if it is currently monitored that the sulfur dioxide concentration in the flue gas of a coal - fired power plant is continuously rising, then after inputting the corresponding text data into the target text semantic mining model that has completed pre - debugging, a basic sensing and monitoring text semantic vector representing this concentration rising trend can be obtained. This vector will be used as the input for subsequent analysis and early warning, helping the system better understand the current gas state and predict its future change trend.

[0043] The reference sensing monitoring text semantic vector is a vector representation obtained by inputting the original gas sensing monitoring text data into a target text semantic mining model that has completed pre - debugging. Different from the basic sensing monitoring text semantic vector, the reference vector is generated based on historical data. They represent different gas state trends and abnormal conditions and are used as a reference basis for subsequent analysis and early warning. In the gas monitoring application in the power production scenario, the reference sensing monitoring text semantic vector can help the system better understand the relationship between the currently monitored gas state and historical data. By comparing and analyzing the current basic sensing monitoring text semantic vector with various reference vectors, the system can identify abnormalities or potential hazards in the current gas state and trigger the corresponding early warning mechanism. For example, if the historical data records a situation where the sulfur dioxide concentration suddenly increased due to coal quality problems, then after inputting the corresponding original gas sensing monitoring text data into the target text semantic mining model that has completed pre - debugging, a reference sensing monitoring text semantic vector representing this abnormal situation can be obtained. When the system monitors a similar gas state change, it can compare and analyze the current basic vector with this reference vector to determine whether there is a similar abnormal situation and issue an early warning in a timely manner.

[0044] Specifically, in the power production scenario, gas monitoring is an important part of ensuring work safety and environmental quality. To achieve accurate gas state analysis and early warning, the detection text processing system plays a key role. This system is not only responsible for collecting and processing the original data from sensors, but also converting these data into informative semantic vectors through an advanced text semantic mining model, providing strong support for subsequent state trend analysis and early warning.

[0045] When the detection text processing system obtains the gas sensing monitoring text data to be analyzed in the target power production scenario, it will first perform necessary pre - processing on these data to ensure the quality and consistency of the data. The pre - processing may include steps such as removing noise, standardizing the format, and correcting errors. These operations help improve the accuracy of subsequent semantic analysis.

[0046] Next, the system inputs the processed text data to be analyzed into the target text semantic mining model that has completed pre - debugging. This model is specifically designed to understand and parse gas sensing monitoring text. It is trained with a large amount of historical data and expert knowledge and can accurately capture the semantic information in the text. When new text data is input into the model, the model will conduct in - depth semantic analysis on it, extract the key information related to the gas state, and encode this information into a basic sensing monitoring text semantic vector. This vector is a point in a high - dimensional space, and its position and direction reflect the semantic information about the gas state in the original text.

[0047] Meanwhile, the system will also input all the original gas sensing monitoring text data corresponding to the original secondary abnormal state trend labels into the same text semantic mining model. These original text data represent various known abnormal state trends, which are used to generate reference sensing monitoring text semantic vectors. These reference vectors are also in the high-dimensional space, and their positions and directions reflect the semantic features of the corresponding abnormal state trends.

[0048] By comparing the positional and distance relationships between the basic sensing monitoring text semantic vectors and each reference sensing monitoring text semantic vector, the system can judge the similarities and differences between the current gas state and the known abnormal state trends. This comparison and analysis provide an important basis for subsequent trend discrimination and early warning.

[0049] Generally speaking, the detection text processing system transforms the original gas sensing monitoring text data into informative semantic vectors by using an advanced text semantic mining model. These vectors not only provide strong support for subsequent state trend analysis, but also provide an important guarantee for gas safety monitoring and early warning in the power production process.

[0050] Step 130: Use the target semantic optimization model that has completed pre-debugging, and respectively perform key semantic attention processing on the basic sensing monitoring text semantic vector and each reference sensing monitoring text semantic vector according to each state trend element knowledge parsed from the gas sensing monitoring text debugging data for different trend labels during the pre-debugging process, to obtain the corresponding target sensing monitoring text semantic vector and each target sensing monitoring text semantic reference vector.

[0051] In step 130, the trend label is a short identifier or mark used to describe and classify the trend of gas concentration change. In the gas monitoring of the power production scenario, the trend label can help the staff or the automated system quickly identify and understand the current state of the gas and its possible development trend. These labels are usually defined based on historical data and expert knowledge, and are applied to real-time monitoring data and text for quick analysis and response. For example, in the gas monitoring of a coal-fired power plant, common trend labels may include "normal", "rising", "falling", "sharply rising", "fluctuating", etc. When the concentration of sulfur dioxide continuously rises within a short period of time and exceeds a certain threshold, this situation may be marked with the "sharply rising" trend label. These labels not only provide an intuitive understanding of the current gas state, but also trigger corresponding early warning or control measures.

[0052] Gas sensing monitoring text debugging data are text data used in the model pre-debugging stage, containing gas sensor monitoring results and relevant tagging information. These data are used to train and optimize the model to ensure that the model can accurately understand and analyze the semantic information in gas sensing monitoring texts. Debugging data usually include original gas concentration readings, corresponding timestamps, and trend tags or other relevant tags generated by experts or systems. In the power production scenario, gas sensing monitoring text debugging data may come from the monitoring results of multiple sensors over different time periods. After being preprocessed and tagged, these data are used to train text semantic mining models and semantic optimization models to help them learn how to extract key semantic information and understand the meaning behind different trend tags.

[0053] State trend element knowledge refers to the key elements and rules regarding gas state trends extracted and learned from gas sensing monitoring text debugging data during the model pre-debugging process. These knowledge include, but are not limited to, the change patterns of gas concentration, the association rules between trend tags and concentration changes, and various external factors that may affect the gas state. This knowledge is encoded into the model so that the model can more accurately identify and predict state trends when processing new gas sensing monitoring texts. In the application of coal-fired power plants, state trend element knowledge may include typical change patterns of sulfur dioxide concentration in different seasons or weather conditions, gas concentration change rules related to equipment operating status, etc. This knowledge helps the model more accurately judge whether the current gas state is normal or abnormal and predict future development trends.

[0054] Key semantic attention processing is a technique that emphasizes important semantic information and suppresses secondary information in text processing. In the gas monitoring application in the power production scenario, key semantic attention processing is used to enhance the model's sensitivity and accuracy to key information in gas sensing monitoring texts. By assigning different attention weights to different parts of the text, the model can pay more attention to information that is crucial for judging gas state trends, such as specific concentration values, change trends, or relevant influencing factors. For example, when processing a text describing the change in sulfur dioxide concentration, key semantic attention processing may make the model pay more attention to the specific numerical value and change trend description of the concentration, while relatively ignoring other less relevant information. The processed text vector will be more focused on representing the key information of sulfur dioxide concentration, thereby improving the accuracy of subsequent analysis and early warning.

[0055] The target sensing monitoring text semantic vector is a vector that represents the semantic information of the current gas sensing monitoring text to be analyzed after key semantic attention processing. This vector is generated in the pre-debugged target semantic optimization model, which captures the key semantic information about the gas state in the text and represents it in the form of a vector. This vector will be used as the input for the subsequent trend discrimination model to judge the state trend of the current gas and output the corresponding warning views. In the power production scenario, the target sensing monitoring text semantic vector represents the key semantic features of the current monitored gas state information. By inputting these features into the trend discrimination model, rapid analysis and warning response of the current gas state can be achieved.

[0056] The target sensing monitoring text semantic reference vector is a vector that represents the reference gas sensing monitoring text of historical or known abnormal state trends after key semantic attention processing. These vectors are generated in the same processing flow as the target sensing monitoring text and are used as the reference basis for the subsequent trend discrimination model. By comparing and analyzing the target sensing monitoring text semantic vector with each reference vector, the similarity and difference between the current gas state and the known abnormal state trends can be judged, and the corresponding warning views can be output. In the power production scenario, the target sensing monitoring text semantic reference vector represents the key semantic features of various known gas abnormal state trends. These features are used to compare and analyze with the target sensing monitoring text, helping the system to more accurately identify and predict the development trend of the current gas state and issue warnings or take control measures in a timely manner.

[0057] Specifically, in the gas monitoring and analysis process of power production, a core step is to use the pre-debugged target semantic optimization model to deeply process the collected text data. The purpose of this step is to more accurately capture the key semantic information in the text, so as to more accurately judge the state trend of the gas.

[0058] First of all, it is necessary to understand the role and function of the target semantic optimization model. This model is trained in the pre-debugging stage with a large number of gas sensing monitoring texts and corresponding state trend labels. During the training process, the model learns how to identify and understand the key semantic elements related to different state trends. These elements may include specific gas concentration thresholds, concentration change rates, durations, etc., which are crucial for judging the safety or potential risks of the gas.

[0059] When the detection text processing system obtains the basic sensing monitoring text semantic vector and each reference sensing monitoring text semantic vector, these vectors will be sent to the target semantic optimization model for further processing. The model will perform key semantic attention processing on each vector according to the knowledge of the state trend elements learned in the pre-debugging stage.

[0060] Key semantic attention processing is a mechanism that allows the model to place more focus on the semantic information that is most relevant to judging the trend of the gas state when processing text data. In this way, the model can ignore the noise and irrelevant information in the text and more accurately capture the key clues related to the current gas state or potential abnormal state.

[0061] During the processing, the target semantic optimization model reallocates the weights for each dimension in each vector. The dimensions that are more relevant to the state trend judgment will receive higher weights and thus play a greater role in subsequent analysis. This weight allocation is based on the knowledge learned by the model during the pre-debugging phase, ensuring the accuracy and effectiveness of the processing.

[0062] Finally, the basic sensing monitoring text semantic vector processed by key semantic attention is transformed into a target sensing monitoring text semantic vector, while the reference sensing monitoring text semantic vector is transformed into a target sensing monitoring text semantic reference vector. These vectors not only contain the key semantic information in the original text but also, through the optimization processing of the model, highlight the parts that are most relevant to the gas state trend judgment. This enables more accurate and reliable subsequent trend discrimination and warning output.

[0063] Step 140: Use the target trend discrimination model that has completed pre-debugging to determine the harmful gas warning output view corresponding to the gas sensing monitoring text data to be analyzed based on the semantic commonality score between the target sensing monitoring text semantic vector and each target sensing monitoring text semantic reference vector.

[0064] In step 140, the target trend discrimination model that has completed pre-debugging is a machine learning model specifically used for analyzing and judging the trend of the gas state. In the power production scenario, this model receives the gas sensing monitoring text semantic vector as input and uses the knowledge of state trend elements learned during the pre-debugging phase to accurately discriminate the state trend of the current gas. The pre-debugging process involves using a large amount of historical gas sensing monitoring text data and corresponding trend labels to train the model to ensure that it can accurately capture and identify the key semantic information in the text and judge the gas state trend based on this information. Once the model completes pre-debugging, it can be applied to real-time monitoring data to discriminate the trend of new gas sensing monitoring text. By inputting the target sensing monitoring text semantic vector, the model can output the judgment result of the current gas state trend, providing an important basis for harmful gas warning. This trend discrimination model plays a crucial role in power production safety. It can help staff promptly discover potential safety hazards and take corresponding control measures to ensure the smooth progress of the production process.

[0065] Semantic commonality scoring is a quantitative metric used to measure the semantic similarity or commonality between the semantic vectors of the target sensing and monitoring text and the reference vectors. In the gas monitoring application in the power production scenario, semantic commonality scoring is used to evaluate the similarity between the currently monitored gas state and the trends of historical or known abnormal states. The higher the score, the more similar the current gas state is semantically to the reference state trend, indicating the possible existence of similar abnormalities or risks. When calculating the semantic commonality score, metrics such as cosine similarity and Euclidean distance are usually used to compare the similarity between the semantic vectors of the target sensing and monitoring text and each reference vector. These scores help the trend discrimination model more accurately judge the state trend of the current gas and provide important references for harmful gas early warning. By combining the semantic commonality score and other relevant information, the system can more comprehensively evaluate the safety of the current gas state and issue early warnings or take control measures in a timely manner.

[0066] The output view of harmful gas early warning refers to the view or conclusion of warning about the possible harmful gas state in the power production scenario based on the analysis results of the trend discrimination model and information such as semantic commonality scoring. These views are obtained based on the accurate judgment of the current gas state trend and the reference comparison with the historical abnormal state trend. The early warning output views usually include different levels such as normal, attention, warning, and danger to indicate the safety risk level of the current gas state. For example, in the gas monitoring of a coal-fired power plant, if the trend discrimination model finds that the sulfur dioxide concentration continues to rise and exceeds the safety threshold, and at the same time has a high semantic commonality score with the known abnormal state trend, then the system may output a "danger" level early warning view and trigger corresponding emergency control measures, such as starting the emergency emission system or shutting down for inspection, etc. These early warning output views are crucial for ensuring power production safety, and they can help staff discover and respond to potential harmful gas risks in a timely manner.

[0067] Specifically, in the power production scenario, ensuring gas safety is of utmost importance. Therefore, an efficient and accurate harmful gas early warning system is essential. The core of this early warning system is to use advanced text processing technologies and machine learning models to deeply analyze the collected gas sensing and monitoring text data to determine potential harmful gas states and output early warning views in a timely manner.

[0068] After the detection text processing system has completed the key semantic attention processing of the semantic vectors of the basic sensing and monitoring text and the reference sensing and monitoring text, the next step is to use the pre-debugged target trend discrimination model to make the final early warning judgment. This model has been trained and optimized with a large amount of historical data and expert knowledge in the previous pre-debugging stage and has the ability to accurately judge the gas state trend.

[0069] During the early warning judgment process, the target trend discrimination model will first calculate the semantic commonality score between the target sensing and monitoring text semantic vector and each target sensing and monitoring text semantic reference vector. This score is a quantitative indicator used to measure the similarity or commonality degree between the current gas state and various known state trends. The calculation of the score may involve measurement methods such as cosine similarity and Euclidean distance, which can effectively compare the positions and directions of vectors in a high-dimensional space.

[0070] After obtaining the semantic commonality scores, the target trend discrimination model will determine the harmful gas early warning output view corresponding to the gas sensing and monitoring text data to be analyzed based on these scores. Specifically, the model will comprehensively consider the semantic commonality scores of each reference vector and the severity of the state trend labels corresponding to them. If the semantic commonality score of a certain reference vector is relatively high and the state trend label corresponding to it represents a dangerous or harmful state, then the model will tend to output a relatively severe early warning view.

[0071] In addition, other factors may also be considered during the model's judgment process, such as the absolute value of the current gas concentration, the concentration change rate, the duration, etc. These factors can provide more comprehensive information for the model to help it more accurately judge the safety state of the current gas.

[0072] Finally, the target trend discrimination model will output a harmful gas early warning view, which may be different levels such as "Normal", "Attention", "Warning", or "Danger". This view is based on a comprehensive analysis and evaluation of the current gas state and can provide an important basis for safety monitoring during the power production process. Once the output view reaches or exceeds the preset safety threshold, the system will trigger corresponding early warning mechanisms, such as sounding an alarm, notifying relevant personnel, or initiating emergency measures, etc., to ensure the safe progress of power production.

[0073] Combined with the above content, a complete example is introduced below.

[0074] During the power production process, in order to ensure the safety of the working environment and the normal operation of equipment, it is necessary to continuously monitor various generated gases. Especially for those harmful gases that may have an adverse impact on human health or equipment performance, timely monitoring and early warning are particularly important.

[0075] Multiple gas sensors are deployed at the power production site. These sensors can monitor the gas concentration in the environment in real time and send the monitoring results to the detection text processing system in the form of text data. The system first obtains the gas sensing monitoring text data to be analyzed from the target power production scenario. In addition, the system also obtains from the database the original gas sensing monitoring text data corresponding to each original secondary abnormal state trend label. These data have been marked with different abnormal state trends, such as "concentration rising", "concentration falling", or "concentration stable", etc. during the previous monitoring process, and they will be used as a reference for subsequent analysis.

[0076] The detection text processing system enters the newly obtained gas sensing monitoring text data to be analyzed into a target text semantic mining model that has completed pre-debugging. This model has been trained to extract the semantic information in the text data and convert it into a vector form for subsequent numerical calculation and analysis. Through the processing of the model, the system obtains the corresponding basic sensing monitoring text semantic vector. At the same time, the system also enters the previously obtained original gas sensing monitoring text data into the same model to generate the corresponding reference sensing monitoring text semantic vectors.

[0077] Next, the detection text processing system uses another target semantic optimization model that has completed pre-debugging to further process the basic sensing monitoring text semantic vector and each reference sensing monitoring text semantic vector generated previously. This model has learned the debugging data of the gas sensing monitoring text for different trend labels during the pre-debugging process and parsed the knowledge of each state trend element from it. Using this knowledge, the model can perform key semantic attention processing on the semantic vector, highlighting the semantic information that is crucial for judging the current gas state trend and suppressing the irrelevant or secondary information. After processing, the system obtains the corresponding target sensing monitoring text semantic vector and each target sensing monitoring text semantic reference vector.

[0078] Finally, the detection text processing system uses a target trend discrimination model that has completed pre-debugging to determine the harmful gas warning output view corresponding to the gas sensing monitoring text data to be analyzed according to the semantic commonality score between the target sensing monitoring text semantic vector and each target sensing monitoring text semantic reference vector. This score reflects the similarity between the currently monitored gas state and various known abnormal state trends. If the score exceeds the preset threshold, the system will trigger the corresponding warning mechanism, such as sounding an alarm, notifying relevant personnel, or taking automatic control measures, etc. to ensure the safe operation of power production.

[0079] The harmful gas warning system and method proposed in this application have shown significant beneficial effects in the power production scenario.

[0080] First, by obtaining the real-time gas sensing monitoring text data of the target power production scenario and combining it with the original gas sensing monitoring text data corresponding to each pre-generated original secondary abnormal state trend label, the present application provides a comprehensive and accurate data basis for gas safety monitoring in the power production process. This dual data acquisition mechanism ensures the comprehensiveness and real-time nature of the analysis, thus greatly improving the ability to identify potential harmful gas states.

[0081] Secondly, the present application innovatively introduces a text semantic mining model to transform the collected text data into a vector representation rich in semantic information. This step not only simplifies the subsequent data processing flow but also enhances the sensitivity and accuracy of the system to gas state changes by capturing the deep semantic information in the text.

[0082] Furthermore, the present application uses a pre-debugged target semantic optimization model to perform key semantic attention processing on the basic sensing monitoring text semantic vector and the reference sensing monitoring text semantic vector. This processing method enables the system to focus on the semantic information most relevant to the judgment of gas state trends, effectively filtering out the interference of noise and irrelevant information, and further improving the accuracy and reliability of early warning judgment.

[0083] Finally, through the comprehensive analysis of the target trend discrimination model, the present application can accurately determine the harmful gas early warning output view corresponding to the gas sensing monitoring text data to be analyzed based on the semantic similarity score between the target sensing monitoring text semantic vector and each reference vector. This innovative early warning mechanism not only realizes the real-time monitoring and early warning of harmful gas states but also provides strong technical support for safety prevention and control in the power production process.

[0084] In summary, the present application combines advanced text processing technologies and machine learning models to achieve efficient and accurate monitoring and early warning of gas states in power production scenarios. This innovative technical solution not only improves the safety of power production but also provides new ideas and methods for safety monitoring and early warning in related fields.

[0085] In an alternative embodiment, key semantic attention processing is performed on the basic sensing monitoring text semantic vector based on each state trend element knowledge parsed from the gas sensing monitoring text debugging data for different trend tags during the pre-debugging process, including: determining the common descriptions between the basic sensing monitoring text semantic vector and each state trend element knowledge respectively according to each state trend element knowledge parsed from the gas sensing monitoring text debugging data for different trend tags during the pre-debugging process; determining the adaptability confidence between the basic sensing monitoring text semantic vector and each state trend element knowledge according to the common descriptions; determining the attention weight coefficients of each state trend element knowledge for each semantic distribution variable value in the basic sensing monitoring text semantic vector according to the adaptability confidence, and optimizing the key semantics attention of the basic sensing monitoring text semantic vector according to each attention weight coefficient.

[0086] In an alternative embodiment, the process for the system to execute harmful gas early warning is as follows.

[0087] First, the system acquires the gas sensing monitoring text data to be analyzed in the target power production scenario. These data are obtained through real-time monitoring and reflect the state of the gas in the current power production environment. At the same time, the system also acquires the original gas sensing monitoring text data corresponding to each original secondary abnormal state trend tag in advance. These data are collected under different abnormal states and are used for providing reference and comparison.

[0088] Next, the system inputs the gas sensing monitoring text data to be analyzed into the target text semantic mining model that has completed pre-debugging. This model has been trained and learned extensively and already has the ability to extract semantic information from text data. Through the processing of the model, the text data to be analyzed is transformed into a basic sensing monitoring text semantic vector, which contains the key semantic information in the text. Similarly, each original gas sensing monitoring text data will also be input into this model and transformed into the corresponding reference sensing monitoring text semantic vectors.

[0089] Then, the system uses the target semantic optimization model that has completed pre-debugging to perform key semantic attention processing on the basic sensing monitoring text semantic vector. This processing process is carried out based on each state trend element knowledge parsed from the gas sensing monitoring text debugging data for different trend tags during the pre-debugging process. Specifically, the system first determines the common descriptions between the basic sensing monitoring text semantic vector and each state trend element knowledge according to these knowledge. These common descriptions reflect the similarity and relevance between the text vector and the state trend elements.

[0090] Next, the system determines the adaptability confidence between the basic sensing and monitoring text semantic vector and the knowledge of each state trend element based on these common descriptions. The adaptability confidence is a quantitative index used to measure the matching degree between the text vector and the state trend element. The higher the confidence, the more relevant the text vector is to the corresponding state trend element and the greater the contribution to the subsequent early warning judgment.

[0091] Finally, the system determines the attention weight coefficients of the knowledge of each state trend element for each semantic distribution variable value in the basic sensing and monitoring text semantic vector according to the adaptability confidence. These weight coefficients reflect the importance of different state trend elements in the early warning judgment. Then, the system performs key semantic attention optimization on the basic sensing and monitoring text semantic vector according to these weight coefficients to obtain the target sensing and monitoring text semantic vector. This vector not only retains the key semantic information in the original text but also highlights the part most relevant to the harmful gas state judgment.

[0092] Similarly, the system also performs similar processing on each reference sensing and monitoring text semantic vector to obtain each target sensing and monitoring text semantic reference vector. These reference vectors will be used for subsequent early warning judgment.

[0093] After the above processing is completed, the system uses the target trend discrimination model that has completed pre-debugging to perform the final early warning judgment. This model determines the harmful gas early warning output view corresponding to the gas sensing and monitoring text data to be analyzed according to the semantic commonality score between the target sensing and monitoring text semantic vector and each target sensing and monitoring text semantic reference vector. This view is based on a comprehensive analysis and evaluation of the current gas state and can provide an important basis for safety monitoring in the power production process.

[0094] From the description of the above embodiments, it can be seen that the beneficial effects of this application are as follows: By combining advanced text processing technologies and machine learning models, it realizes efficient and accurate monitoring and early warning of gas states in power production scenarios; Through key semantic attention processing, it highlights the semantic information most relevant to the harmful gas state judgment, improving the accuracy and reliability of early warning judgment; By comprehensively considering various influencing factors and giving accurate early warning judgments, it provides strong technical support for safety prevention and control in the power production process.

[0095] In another alternative embodiment, for each state trend element knowledge obtained by parsing the gas sensing monitoring text debugging data for different trend tags during the pre-debugging process, the common descriptions between the basic sensing monitoring text semantic vector and each state trend element knowledge are determined respectively, including: obtaining each state trend element knowledge obtained by parsing the gas sensing monitoring text debugging data for different trend tags during the pre-debugging process, and performing feature mapping on each state trend element knowledge to obtain the state trend element mapping relationship network corresponding to each state trend element knowledge, and performing feature mapping on the basic sensing monitoring text semantic vector to obtain the corresponding monitoring text semantic mapping relationship network; determining the relationship network feature operation result between the state trend element mapping relationship network and the monitoring text semantic mapping relationship network to obtain a commonality index list reflecting the common description between the basic sensing monitoring text semantic vector and each state trend element knowledge.

[0096] In another alternative embodiment, the technical solution for the system to perform key semantic attention processing is as follows.

[0097] First, the system acquires the gas sensing monitoring text debugging data for different trend tags during the pre-debugging process. This data contains the gas sensing monitoring text information collected under different abnormal states and is the basis for the system to learn and understand the state trend element knowledge.

[0098] Next, the system parses these debugging data to extract each state trend element knowledge. This knowledge reflects the characteristics and laws of the gas sensing monitoring text under different abnormal states and is of great significance for subsequent semantic analysis and early warning judgment.

[0099] Then, the system performs feature mapping on each extracted state trend element knowledge. Feature mapping is a process of transforming the original data into a form that is easier to process and analyze. Through feature mapping, the system can transform the state trend element knowledge into a state trend element mapping relationship network. This relationship network is a complex network structure that contains the association and interaction information between each state trend element.

[0100] At the same time, the system also performs feature mapping on the basic sensing monitoring text semantic vector. This process is similar to the processing of the state trend element knowledge, aiming to transform the text vector into a form that is easier to analyze and compare. Through feature mapping, the system can obtain the monitoring text semantic mapping relationship network. This relationship network reflects the relationship and interaction between each semantic distribution variable value in the text vector.

[0101] Next, the system will determine the result of the relational network feature operation between the state trend element mapping relational network and the monitoring text semantic mapping relational network. This process is achieved by comparing and analyzing the similarity and relevance between the two relational networks. The operation result reflects the common description between the basic sensing monitoring text semantic vector and each state trend element knowledge. These common descriptions are important bases for subsequent attention processing.

[0102] Finally, the system will obtain a list of commonality indices based on the result of the relational network feature operation. This list contains the commonality index values between the basic sensing monitoring text semantic vector and each state trend element knowledge. These values quantify the similarity and relevance degree between the text vector and the state trend elements, providing a basis for determining the subsequent attention weight coefficients.

[0103] After completing the above processing, the system can optimize the key semantic attention of the basic sensing monitoring text semantic vector according to the list of commonality indices. During the optimization process, the system will determine the attention weight coefficients of each state trend element knowledge for each semantic distribution variable value in the text vector according to the commonality index values. These weight coefficients reflect the importance degree of different state trend elements in the early warning judgment. Then, the system will perform weighted processing on the text vector according to these weight coefficients to obtain the target sensing monitoring text semantic vector. This vector not only retains the key semantic information in the original text but also highlights the part most relevant to the harmful gas state judgment.

[0104] From the description of the above embodiments, it can be seen that the beneficial effects of this application are as follows: accurate quantification of the common description between the basic sensing monitoring text semantic vector and each state trend element knowledge is achieved through technical means such as feature mapping and relational network feature operation; effective extraction and highlighting of key semantic information are realized through mechanisms such as the list of commonality indices and attention weight coefficients; the accuracy and reliability of early warning judgment are improved, providing strong support for safety monitoring in the power production process.

[0105] In some preferred technical solutions, based on the common description, determining the fitness confidence between the basic sensing monitoring text semantic vector and each state trend element knowledge, and based on the fitness confidence, determining the attention weight coefficient of each state trend element knowledge for each semantic distribution variable value in the basic sensing monitoring text semantic vector, includes: performing interval numerical mapping processing on the common variables of each list unit in the common index list, and based on the common variables of each list unit after the interval numerical mapping processing, obtaining a confidence relationship spectrum representing the fitness confidence between the basic sensing monitoring text semantic vector and each state trend element knowledge; determining the attention weight coefficient of each state trend element knowledge for the semantic vector values at each list unit in the basic sensing monitoring text semantic vector according to the result of the relationship network feature operation between the knowledge vector relationship network generated by each state trend element knowledge and the confidence relationship spectrum.

[0106] In some preferred technical solutions, the process by which the system determines the fitness confidence between the basic sensing monitoring text semantic vector and each state trend element knowledge, as well as the attention weight coefficient, is as follows.

[0107] First, the system processes each list unit in the common index list. Each list unit in the common index list contains the common variable value between the basic sensing monitoring text semantic vector and a certain state trend element knowledge. These common variable values are indicators quantifying the similarity and relevance between the two.

[0108] To facilitate subsequent calculations and processing, the system performs interval numerical mapping processing on these common variables. Interval numerical mapping processing is the process of mapping the original data into a specific numerical interval, which helps to eliminate the dimensional difference and outlier influence between data, making subsequent calculations more accurate and stable. Through the interval numerical mapping processing, the system can obtain a confidence relationship spectrum representing the fitness confidence between the basic sensing monitoring text semantic vector and each state trend element knowledge. This relationship spectrum is a continuous function or a set of discrete numerical values, reflecting the matching degree between the text vector and the state trend element.

[0109] Next, the system uses the knowledge vector relationship network generated by each state trend element knowledge for further analysis. The knowledge vector relationship network is a complex network structure containing the association information between each state trend element knowledge. Through this network structure, the system can comprehensively consider the influence of different state trend elements on the text vector.

[0110] Then, the system determines the attention weight coefficients of each state trend element knowledge for the semantic vector values at each list unit in the basic sensing monitoring text semantic vector based on the result of the relational network feature operation between the confidence relationship spectrum and the knowledge vector relational network. The result of the relational network feature operation is obtained by comparing and analyzing the similarity and relevance between the two relational networks. This result reflects the sensitivity and contribution degrees of different semantic distribution variable values in the text vector to each state trend element.

[0111] Specifically, if a certain state trend element is highly correlated with a certain semantic distribution variable value in the text vector, then the importance of this state trend element in the warning judgment will be relatively high, and the corresponding attention weight coefficient will also be relatively large. On the contrary, if a certain state trend element has a weak correlation with the semantic distribution variable value in the text vector, then its importance in the warning judgment will be relatively low, and the corresponding attention weight coefficient will also be relatively small.

[0112] Through the above processing process, the system can accurately determine the attention weight coefficients of each state trend element knowledge for each semantic distribution variable value in the basic sensing monitoring text semantic vector. These weight coefficients provide an important basis for the subsequent optimization of key semantic attention.

[0113] Generally speaking, the beneficial effects of these preferred technical solutions are as follows: The accurate quantification of the adaptability confidence and attention weight coefficients is achieved through technical means such as interval numerical mapping processing and relational network feature operation; The accuracy and reliability of the warning judgment are improved by comprehensively considering the influence of different state trend elements on the text vector; It provides strong support for the safety monitoring in the power production process.

[0114] Under some alternative design ideas, the realization of the key semantic attention optimization for the basic sensing monitoring text semantic vector based on each attention weight coefficient includes: performing semantic vector value accumulation processing on the attention weight coefficient of each list unit and the semantic vector value of the corresponding list unit in the basic sensing monitoring text semantic vector to obtain the semantic vector value accumulation result corresponding to each list unit; Dynamically adjusting the semantic vector value accumulation results corresponding to each list unit to achieve the focusing of the key semantics in the basic sensing monitoring text semantic vector.

[0115] Under some alternative design ideas, the process of the system optimizing the key semantic attention for the basic sensing monitoring text semantic vector is as follows.

[0116] First, the system obtains the attention weight coefficients of each list unit calculated previously. These attention weight coefficients reflect the importance degree of different state trend factor knowledge for each semantic distribution variable value in the semantic vector of the basic sensing and monitoring text.

[0117] Next, the system performs semantic vector value accumulation processing on the attention weight coefficients of each list unit and the semantic vector values of the corresponding list units in the semantic vector of the basic sensing and monitoring text. Semantic vector value accumulation processing is a process of multiplying and accumulating the attention weight coefficients with the corresponding semantic vector values, aiming to strengthen important semantic information and weaken unimportant semantic information. Through this process, the system can obtain the semantic vector value accumulation results corresponding to each list unit.

[0118] Then, the system dynamically adjusts the semantic vector value accumulation results corresponding to each list unit. The purpose of dynamic adjustment is to further highlight the key semantic information, making the optimized text vector more focused on the part most relevant to the harmful gas state judgment. The specific method of dynamic adjustment can be determined according to the actual application scenario and requirements. For example, smaller values in the accumulation results can be filtered out by setting thresholds, or the accumulation results can be normalized, etc.

[0119] After completing the above processing, the system can achieve the optimization of the key semantic attention of the semantic vector of the basic sensing and monitoring text. The optimized text vector not only retains the key semantic information in the original text but also highlights the part most relevant to the harmful gas state judgment, thereby improving the accuracy and reliability of the early warning judgment.

[0120] In summary, the beneficial effects of this replaceable design idea are as follows: The optimization of the key semantic attention of the semantic vector of the basic sensing and monitoring text is achieved through technical means such as semantic vector value accumulation processing and dynamic adjustment; important semantic information is strengthened and unimportant semantic information is weakened; the accuracy and reliability of the early warning judgment are improved; and strong support is provided for safety monitoring in the power production process. At the same time, this design idea also has flexibility and scalability and can be adjusted and optimized according to different application scenarios and requirements.

[0121] In some possible embodiments, the obtaining of the original gas sensing and monitoring text data corresponding to each original secondary abnormal state trend label in advance includes: in response to the task information of the monitoring platform system, determining each original secondary abnormal state trend label targeted by the task information, and respectively obtaining the original gas sensing and monitoring text data generated in advance for each of the original secondary abnormal state trend labels.

[0122] In some possible embodiments, the process by which the system obtains the original gas sensing monitoring text data corresponding to each original secondary abnormal state trend label in advance is as follows.

[0123] First, the system responds to the task information of the monitoring platform system. The monitoring platform system is a comprehensive platform integrating various monitoring functions and data analysis functions, and is used for real-time monitoring and early warning of the operating status of power equipment. When the monitoring platform system issues task information, the system receives this task information and parses and processes it.

[0124] The task information usually contains the original secondary abnormal state trend labels that need to be concerned about. The original secondary abnormal state trend labels are labels that refine and classify the possible abnormal states of power equipment, such as "too high temperature", "abnormal pressure", etc. These labels help the system to more accurately locate and analyze abnormal states.

[0125] After determining the original secondary abnormal state trend labels targeted by the task information, the system respectively obtains the original gas sensing monitoring text data generated in advance for these labels. These text data are obtained by real-time monitoring and recording of the gas components and concentrations in the operating environment of the power equipment through gas sensors during previous monitoring processes. They reflect the gas change characteristics of power equipment in different abnormal states and are important bases for subsequent semantic analysis and early warning judgment.

[0126] Specifically, the system can obtain these data by accessing the database or file server storing the original gas sensing monitoring text data. During the obtaining process, the system screens and classifies the text data according to the original secondary abnormal state trend labels to ensure that the obtained data is relevant, accurate, and reliable to the current task information.

[0127] After obtaining the original gas sensing monitoring text data, the system can use these data for subsequent semantic analysis and early warning judgment. By deeply mining and analyzing the text data, the system can discover the key information and rules hidden in it, providing strong support for the status monitoring and early warning of power equipment.

[0128] In summary, the beneficial effects of this embodiment are as follows: by responding to the task information of the monitoring platform system and determining the targeted original secondary abnormal state trend labels, the system can obtain the original gas sensing monitoring text data generated in advance in a targeted manner; these data reflect the gas change characteristics of power equipment in different abnormal states and provide an important basis for subsequent semantic analysis and early warning judgment; at the same time, this embodiment also has flexibility and scalability, and can obtain corresponding text data for analysis and processing according to different task information and abnormal state labels.

[0129] In an exemplary embodiment, pre-debugging is performed on an inspection text processing network including a text semantic mining model, a semantic optimization model, and a trend discrimination model, including: according to a shared debugging example set, performing cyclic pre-debugging on the inspection text processing network in multiple debugging phases until a set cumulative number of debugging times is reached, and in one debugging phase, performing the following processing: according to a gas sensing monitoring text debugging data group sampled from a local debugging example set, performing cyclic debugging on the inspection text processing network to be debugged for a certain number of times to obtain an intermediate inspection text processing network, and determining a discrimination quality evaluation index of the intermediate inspection text processing network according to gas sensing monitoring text evaluation data sampled from a local test example set, where the local debugging example set and the local test example set are included in the debugging example set; determining a target inspection text processing network that meets the set debugging requirements from each intermediate inspection text processing network according to the discrimination quality evaluation index determined for the intermediate inspection text processing network in different debugging phases.

[0130] In an exemplary embodiment, the system pre-debugs an inspection text processing network including a text semantic mining model, a semantic optimization model, and a trend discrimination model. This process is to ensure that the network can achieve the expected performance and accuracy when processing actual gas sensing monitoring text data.

[0131] First, the system performs cyclic pre-debugging on this inspection text processing network according to a shared debugging example set. This shared debugging example set contains a series of gas sensing monitoring text data for debugging and testing, and these data are representative and cover various possible abnormal state scenarios. In each debugging phase, the system makes careful adjustments and optimizations to this network to improve its performance and accuracy.

[0132] In each debugging phase, the system first samples a group of gas sensing monitoring text debugging data from a local debugging example set. This local debugging example set is a subset selected from the shared debugging example set, and its purpose is to more focusedly test and adjust a certain aspect or a specific function of the network. The system uses this sampled data to perform cyclic debugging on the current inspection text processing network for several times, and each debugging makes fine-tuning of some parameters or structures of the network, thereby obtaining an intermediate inspection text processing network.

[0133] Next, the system will sample a set of gas sensing monitoring text evaluation data from another local test sample set and use this set of data to evaluate the discrimination quality of the intermediate inspection text processing network just obtained. This local test sample set is also a subset selected from the shared debugging sample set but does not overlap with the local debugging sample set. The system will quantify the performance of this intermediate network through some evaluation metrics such as accuracy, recall, F1-score, etc.

[0134] The system will perform the above evaluation and processing for the intermediate inspection text processing network in each debugging stage. In this way, after completing the loop pre-debugging of all debugging stages, the system will obtain a series of intermediate inspection text processing networks with different performances and discrimination qualities.

[0135] Finally, the system will select a target inspection text processing network that meets the set debugging requirements from these intermediate networks according to the discrimination quality evaluation metrics determined for the intermediate inspection text processing networks in different debugging stages. This target network is the network that shows the best performance and discrimination quality in all debugging stages and will be used for subsequent actual gas sensing monitoring text processing tasks.

[0136] Through this multi-round debugging and loop optimization method, the system can effectively improve the performance and accuracy of the inspection text processing network when processing complex and variable gas sensing monitoring text data. This not only helps to improve the efficiency and reliability of power equipment status monitoring and early warning but also provides valuable reference for text processing tasks in other similar application scenarios.

[0137] Furthermore, select gas sensing monitoring text debugging data from the local debugging sample set, including: for each of the preselected X secondary abnormal state trend labels in the local debugging sample set, select Y gas sensing monitoring text debugging data as a type of auxiliary debugging data respectively, and select Z non-crossing gas sensing monitoring text debugging data as the debugging data to be warned; create X*Z gas sensing monitoring text debugging data groups according to the sampled auxiliary debugging data and debugging data to be warned, where each gas sensing monitoring text debugging data group includes one debugging data to be warned corresponding to one secondary abnormal state trend label, where X, Y, and Z are positive integers.

[0138] Furthermore, in terms of selecting gas sensing monitoring text debugging data, the system will perform more detailed and precise operations. First, in the local debugging sample set, the system will select according to the preselected X secondary abnormal state trend labels. These secondary abnormal state trend labels are a detailed classification of the possible abnormal states of power equipment, such as "too high temperature", "voltage abnormality", etc., and each label corresponds to a specific type of abnormal state.

[0139] For each secondary abnormal state trend label, the system will separately select Y gas sensing monitoring text debugging data from the local debugging sample set as auxiliary debugging data. These auxiliary debugging data are used to help the system better understand and analyze the characteristics and patterns of gas sensing monitoring texts in this type of abnormal state. At the same time, to ensure the diversity and representativeness of the data, the system will select non-overlapping, that is, Z independent gas sensing monitoring text debugging data as the debugging data to be warned. These debugging data to be warned are used to test the performance and accuracy of the system when facing actual warning tasks.

[0140] After selecting these data, the system will create X*Z gas sensing monitoring text debugging data groups based on the sampled auxiliary debugging data and debugging data to be warned. Each data group includes a debugging data to be warned corresponding to a secondary abnormal state trend label, and the auxiliary debugging data associated with this debugging data to be warned. In this way, each data group constitutes an independent debugging unit, which is used to carefully test and adjust the performance and accuracy of the system in the subsequent debugging stage.

[0141] Through such a data selection and organization method, the system can make more comprehensive and in-depth use of the data resources in the debugging sample set, improving the pertinence and efficiency of the debugging process. At the same time, this also provides a richer and more diverse test scenario for the subsequent debugging stage, helping the system better handle various complex abnormal state situations in actual applications.

[0142] Generally speaking, the beneficial effects of this careful data selection and organization method are as follows: improving the pertinence and efficiency of the debugging process; providing a richer and more diverse test scenario for the subsequent debugging stage; helping the system better handle various complex abnormal state situations in actual applications; and ultimately enhancing the overall performance and accuracy of the power equipment condition monitoring and warning system.

[0143] Further, based on the gas sensing monitoring text debugging data group sampled from the local debugging sample set, the inspection text processing network to be debugged is cyclically debugged for Yuci times to obtain an intermediate inspection text processing network, including: using the generated gas sensing monitoring text debugging data group to cyclically debug the inspection text processing network to be debugged for Yuci times, and improving the neural network parameters of the inspection text processing network according to the debugging errors determined in each round of the cycle; wherein, in one round of the cycle debugging process, the following processing is implemented: inputting an obtained gas sensing monitoring text debugging data group into the inspection text processing network to be debugged, obtaining a harmful gas warning debugging view determined for the to-be-warned debugging data in the gas sensing monitoring text debugging data group, and determining a debugging error according to the warning view difference between the harmful gas warning debugging view and the corresponding harmful gas warning certification view.

[0144] Further, when the system is being debugged, it will cyclically debug the inspection text processing network to be debugged based on the gas sensing monitoring text debugging data group sampled from the local debugging sample set. This process aims to improve the performance and accuracy of the inspection text processing network when processing gas sensing monitoring text data by continuously adjusting and optimizing the network parameters.

[0145] Specifically, the system will use the generated gas sensing monitoring text debugging data group to cyclically debug the inspection text processing network to be debugged for multiple times. In each cycle, the system will input a gas sensing monitoring text debugging data group into the inspection text processing network to be debugged, and through the operation and processing of the network, obtain a harmful gas warning debugging view corresponding to the to-be-warned debugging data in the data group. This harmful gas warning debugging view is the prediction and judgment made by the network based on the input data, and it reflects the network's understanding and cognition of the current data.

[0146] To evaluate the performance and accuracy of the network, the system will compare the harmful gas warning debugging view with the corresponding harmful gas warning certification view. The harmful gas warning certification view is a pre-determined view that is considered to be correct or true, and it can be expert judgment, actual observation results, or information provided by other reliable sources. By comparing the differences between these two views, the system can determine the debugging error, that is, the deviation or error of the network when processing the current data.

[0147] The debugging error is one of the important indicators for evaluating the network performance, and it reflects the possible defects and deficiencies of the network when dealing with actual problems. Therefore, after determining the debugging error, the system will improve and adjust the neural network parameters of the inspection text processing network according to the magnitude and direction of the error. This process aims to continuously optimize the network structure and parameter settings, reduce the debugging error, and improve the performance and accuracy of the network.

[0148] Through multiple rounds of loop debugging and parameter improvement, the system can finally obtain an intermediate inspection text processing network. This intermediate network has significantly improved in performance and accuracy compared to the initial network, and can better process gas sensing monitoring text data and make accurate early warning judgments. At the same time, since this process is carried out for the local debugging sample set, the obtained intermediate network also has stronger pertinence and adaptability, and can better handle the abnormal state early warning tasks in specific scenarios.

[0149] Generally speaking, the beneficial effects of this loop debugging and parameter improvement method are as follows: by continuously adjusting and optimizing the network parameter settings, the performance and accuracy of the inspection text processing network in processing gas sensing monitoring text data are improved; the obtained intermediate inspection text processing network has stronger pertinence and adaptability; it can better handle the abnormal state early warning tasks in specific scenarios; and it provides a more reliable and effective model basis for subsequent testing and application.

[0150] In the next step, inputting a gas sensing monitoring text debugging data group obtained into the inspection text processing network to be debugged to obtain a harmful gas early warning debugging view determined for the to-be-early-warned debugging data in the gas sensing monitoring text debugging data group includes: inputting the to-be-early-warned debugging data and various auxiliary debugging data included in a gas sensing monitoring text debugging data group into the text semantic mining model to be debugged to obtain a sensing monitoring text semantic vector debugging sample and each auxiliary sensing monitoring text semantic vector debugging sample; using the semantic optimization model to be debugged, and respectively performing key semantic attention processing on the sensing monitoring text semantic vector debugging sample and each auxiliary sensing monitoring text semantic vector debugging sample according to the state trend element knowledge parsed for each trend keyword to obtain a target sensing monitoring text semantic vector debugging sample and each target auxiliary sensing monitoring text semantic vector debugging sample; using the trend discrimination model to be debugged, and determining a harmful gas early warning output view corresponding to the to-be-early-warned debugging data according to the semantic commonality score between the target sensing monitoring text semantic vector debugging sample and each target auxiliary sensing monitoring text semantic vector debugging sample.

[0151] In the next step, the system will deeply process the gas sensing monitoring text debugging data group to generate corresponding harmful gas early warning debugging views. This process involves multiple models and detailed data processing.

[0152] First, the system inputs the warning data to be processed and various auxiliary debugging data included in a gas sensing monitoring text debugging data group into the text semantic mining model to be debugged. This model is specially trained to deeply understand the semantic content of text data and convert it into vector representations. In this way, the original text data is converted into sensing monitoring text semantic vector debugging examples and each auxiliary sensing monitoring text semantic vector debugging example. These vector examples capture the key information and semantic features in the text, providing a basis for subsequent processing.

[0153] Next, the system further processes these vector examples using the semantic optimization model to be debugged. This model has the ability to analyze the knowledge of state trend elements. It can identify trend keywords in the text and generate state trend element knowledge based on these keywords. This knowledge helps the system better understand the meaning and context of text data. During this process, the semantic optimization model performs key semantic attention processing on the sensing monitoring text semantic vector debugging examples and each auxiliary sensing monitoring text semantic vector debugging example. This means that the model focuses on the key information related to harmful gas warning and ignores irrelevant details. Through this processing method, the system obtains the target sensing monitoring text semantic vector debugging examples and each target auxiliary sensing monitoring text semantic vector debugging example. These target vector examples are more focused on the harmful gas warning task, improving the accuracy and efficiency of subsequent processing.

[0154] Finally, the system uses the trend discrimination model to be debugged to determine the harmful gas warning output view corresponding to the warning data to be processed. This model makes a judgment based on the semantic commonality scores between the target sensing monitoring text semantic vector debugging examples and each target auxiliary sensing monitoring text semantic vector debugging example. The semantic commonality score reflects the semantic similarity and relevance between different vector examples. By comparing these scores, the trend discrimination model can identify the information most relevant to harmful gas warning and generate the final harmful gas warning output view based on this. This view is a comprehensive understanding and judgment of the input text data, reflecting the system's perception and prediction of the current gas sensing monitoring situation.

[0155] In summary, this processing flow realizes the in-depth understanding and accurate warning of gas sensing monitoring text data by combining multiple models such as text semantic mining, semantic optimization, and trend discrimination. This not only improves the intelligent level of power equipment state monitoring but also provides strong support for ensuring industrial safety and environmental protection. At the same time, this technical solution also demonstrates the great potential and broad application prospects of artificial intelligence technology in processing complex industrial data.

[0156] In some independent embodiments, after using the target trend discrimination model that has completed pre-debugging to determine the harmful gas warning output view corresponding to the gas sensing monitoring text data to be analyzed based on the semantic commonality score between the target sensing monitoring text semantic vector and each target sensing monitoring text semantic reference vector, the method further includes: determining a power production emergency plan according to the harmful gas warning output view.

[0157] In some independent embodiments, the functions of the system are further expanded. It not only makes achievements in the processing of gas sensing monitoring text data and the generation of warning views, but also plays an important role in the formulation of power production emergency plans.

[0158] Specifically, when the system uses the target trend discrimination model that has completed pre-debugging to determine the harmful gas warning output view corresponding to the gas sensing monitoring text data to be analyzed based on the semantic commonality score between the target sensing monitoring text semantic vector and each target sensing monitoring text semantic reference vector, it does not stop there. Instead, the system will use these warning views as important inputs and further participate in the process of formulating power production emergency plans.

[0159] Power production emergency plans are a set of detailed plans and measures pre-developed by power enterprises to ensure personnel safety, equipment protection, and environmental protection in the face of potential or actual harmful gas leakage and other emergencies. These plans need to comprehensively consider various factors, including the types, concentrations, diffusion ranges, possible impacts, etc. of harmful gases, as well as the actual situation and emergency response capabilities of the enterprise.

[0160] In this process, the system will automatically or semi-automatically generate corresponding power production emergency plans according to the key information provided by the harmful gas warning output view, such as the types of harmful gases, concentration change trends, possible hazard levels, etc., combined with the actual situation and emergency response needs of the power enterprise. These plans may include aspects such as personnel evacuation routes, equipment shutdown sequences, emergency rescue measures, etc., to ensure a rapid and effective response in case of an emergency.

[0161] In addition, the system can also dynamically adjust and optimize the power production emergency plan according to the continuous update of real-time monitoring data and warning views. For example, when it is detected that the concentration of harmful gas exceeds the preset threshold, the system can automatically trigger the execution of the emergency plan and notify relevant personnel to take emergency measures to minimize potential safety risks and losses.

[0162] In summary, by combining the harmful gas warning output view with the formulation of power production emergency plans, the system not only improves the ability to identify and warn of potential safety risks but also provides strong guarantee for the safe production of power enterprises. The beneficial effects of this technical solution are to enhance the safety and reliability of the power production process, reduce the probability and impact degree of accidents, and at the same time improve the efficiency and accuracy of emergency response.

[0163] In the process of determining the power production emergency plan, it is crucial to formulate corresponding strategies according to the harmful gas warning output view. The following are the sub-steps of a detailed and creative technical solution for determining the power production emergency plan based on the harmful gas warning output view: (1) Warning level classification: Automatically classify the warning level according to the information in the harmful gas warning output view, such as the type, concentration, and diffusion rate of harmful gases. The warning levels include low, medium, high, and emergency, and each level corresponds to specific emergency response measures; (2) Plan template matching: Select the plan template corresponding to the determined warning level as the basis from multiple pre-stored power production emergency plan templates for different warning levels; (3) Key parameter adjustment: Automatically adjust the key parameters in the selected plan template according to the specific data provided by the harmful gas warning output view, such as gas concentration, wind direction and speed, including adjusting the evacuation radius, determining the priority of shutting down equipment, and calculating the required rescue resources; (4) Resource allocation optimization: Combine the actual resource situation of the power enterprise, such as personnel distribution, equipment status, and material reserves, to optimize the resource allocation in the plan to ensure the efficient use of limited resources in case of emergency; (5) Simulation exercise and evaluation: Use simulation software to conduct simulation exercises on the optimized plan, evaluate the effectiveness and feasibility of the plan, and make necessary adjustments and optimizations to the plan according to the exercise results; (6) Plan approval and release: Submit the plan that has undergone simulation exercises and evaluation to relevant management personnel for approval. After approval, officially release the plan and notify all relevant personnel; (7) Real-time update and monitoring: Real-time monitor the changes in harmful gases and the power production status. Once a new warning output view is generated, immediately update and adjust the plan to ensure that the plan matches the current safety risks.

[0164] The following is a detailed introduction to the above sub-steps.

[0165] (1) Early warning level classification: First, the system automatically classifies the early warning levels based on the information in the harmful gas early warning output view, such as the type, concentration, and diffusion rate of harmful gases. These levels can include low, medium, high, and emergency, etc., and each level corresponds to different emergency response measures.

[0166] (2) Pre - plan template matching: The system pre - stores multiple power production emergency plan templates for different early warning levels. Once the early warning level is determined, the system automatically matches the corresponding level of the plan template as a basis for further customization and adjustment.

[0167] (3) Key parameter adjustment: According to the specific data provided by the harmful gas early warning output view, such as gas concentration, wind direction and speed, etc., the system automatically adjusts the key parameters in the plan template. For example, adjusting the evacuation radius, determining the priority of shutting down equipment, calculating the required rescue resources, etc.

[0168] (4) Resource allocation optimization: The system combines the actual resource situation of the power enterprise, such as personnel distribution, equipment status, material reserves, etc., to optimize the resource allocation in the plan. This can ensure that limited resources can be utilized most efficiently in case of emergency.

[0169] Simulation exercise and evaluation: Before the plan is finalized, the system can use simulation software to conduct simulation exercises on the plan to evaluate the effectiveness and feasibility of the plan. Through simulation exercises, the system can discover potential problems in the plan and make corresponding adjustments and optimizations.

[0170] (5) Plan approval and release: The plan after simulation exercise and evaluation will be submitted to relevant management personnel for approval. After approval, the plan will be officially released and notified to all relevant personnel to ensure that they can act quickly according to the plan in case of emergency.

[0171] (6) Real - time update and monitoring: The system will monitor the changes in harmful gases and the power production status in real - time. Once a new early warning output view is generated, the system will immediately update and adjust the plan to ensure that the plan always matches the current safety risks.

[0172] Through the above sub - steps, the system can quickly and accurately determine the power production emergency plan according to the harmful gas early warning output view, providing a strong guarantee for the safe production of power enterprises. This technical solution not only improves the pertinence and practicality of the emergency plan, but also greatly enhances the efficiency and accuracy of emergency response.

[0173] Figure 2A structural block diagram of a detection text processing system 300 is shown, including: a memory 310 for storing program instructions and data; a processor 320 coupled to the memory 310 for executing the instructions in the memory 310 to implement the above method.

[0174] Further, a computer storage medium is provided, containing instructions that, when executed on a processor, implement the above method.

[0175] The foregoing is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A text processing method for detecting harmful gases based on big data, characterized in that, Applied to a text processing system for detection, the method includes: Obtaining the gas sensing monitoring text data to be analyzed for a target power production scenario, and obtaining each piece of original gas sensing monitoring text data corresponding to each original secondary abnormal state trend label in advance; Inputting the gas sensing monitoring text data to be analyzed into a target text semantic mining model that has completed pre-debugging to obtain corresponding basic sensing monitoring text semantic vectors, and inputting each piece of original gas sensing monitoring text data into the target text semantic mining model to obtain corresponding reference sensing monitoring text semantic vectors; Using a target semantic optimization model that has completed pre-debugging, and based on each state trend element knowledge parsed from the gas sensing monitoring text debugging data for different trend labels during the pre-debugging process, performing key semantic attention processing on the basic sensing monitoring text semantic vectors and each reference sensing monitoring text semantic vector respectively to obtain corresponding target sensing monitoring text semantic vectors and each target sensing monitoring text semantic reference vector; Using a target trend discrimination model that has completed pre-debugging, and based on the semantic commonality scores between the target sensing monitoring text semantic vectors and each target sensing monitoring text semantic reference vector, determining the harmful gas early warning output view corresponding to the gas sensing monitoring text data to be analyzed.

2. The method according to claim 1, characterized in that, Performing key semantic attention processing on the basic sensing monitoring text semantic vectors based on each state trend element knowledge parsed from the gas sensing monitoring text debugging data for different trend labels during the pre-debugging process, including: Based on each state trend element knowledge parsed from the gas sensing monitoring text debugging data for different trend labels during the pre-debugging process, respectively determining the common descriptions between the basic sensing monitoring text semantic vectors and each state trend element knowledge; Based on the common descriptions, determining the adaptability confidence levels between the basic sensing monitoring text semantic vectors and each state trend element knowledge; Based on the adaptability confidence levels, determining the attention weight coefficients of each state trend element knowledge for each semantic distribution variable value in the basic sensing monitoring text semantic vectors, and based on each attention weight coefficient, realizing the key semantic attention optimization of the basic sensing monitoring text semantic vectors.

3. The method according to claim 2, wherein The step of respectively determining the common descriptions between the basic sensing monitoring text semantic vectors and each state trend element knowledge based on each state trend element knowledge parsed from the gas sensing monitoring text debugging data for different trend labels during the pre-debugging process includes: Obtaining each state trend element knowledge parsed from the gas sensing monitoring text debugging data for different trend labels during the pre-debugging process, and performing feature mapping on each state trend element knowledge to obtain a state trend element mapping relationship network corresponding to each state trend element knowledge, and performing feature mapping on the basic sensing monitoring text semantic vectors to obtain a corresponding monitoring text semantic mapping relationship network; Determine the relationship network feature operation result between the state trend element mapping relationship network and the monitoring text semantic mapping relationship network, and obtain a list of commonality indices reflecting the common description between the basic sensor monitoring text semantic vector and various state trend element knowledge.

4. The method according to claim 3, characterized in that, Based on the common description, determine the adaptability confidence between the basic sensor monitoring text semantic vector and various state trend element knowledge, and based on the adaptability confidence, determine the attention weight coefficients of various state trend element knowledge for each semantic distribution variable value in the basic sensor monitoring text semantic vector, including: Perform interval numerical mapping processing on the common variables of each list unit in the commonality index list, and based on the common variables of each list unit after interval numerical mapping processing, obtain a confidence relationship spectrum characterizing the adaptability confidence between the basic sensor monitoring text semantic vector and various state trend element knowledge; Based on the relationship network feature operation result between the knowledge vector relationship network generated by various state trend element knowledge and the confidence relationship spectrum, determine the attention weight coefficients of various state trend element knowledge for the semantic vector values at each list unit in the basic sensor monitoring text semantic vector.

5. The method according to claim 2, characterized in that, Based on each attention weight coefficient, realize the key semantic attention optimization of the basic sensor monitoring text semantic vector, including: Perform semantic vector value accumulation processing on the attention weight coefficients of each list unit and the semantic vector values of the corresponding list unit in the basic sensor monitoring text semantic vector to obtain the semantic vector value accumulation result corresponding to each list unit; Dynamically adjust the semantic vector value accumulation results corresponding to each list unit to achieve the focusing of key semantics in the basic sensor monitoring text semantic vector.

6. The method according to claim 1, wherein Obtain each item of original gas sensor monitoring text data generated in advance corresponding to each original secondary abnormal state trend label, including: In response to the task information of the monitoring platform system, determine each original secondary abnormal state trend label targeted by the task information, and respectively obtain each item of original gas sensor monitoring text data generated in advance for each original secondary abnormal state trend label.

7. The method according to any one of claims 1 to 6, characterized in that, Pre-debug the inspection text processing network including a text semantic mining model, a semantic optimization model, and a trend discrimination model, including: Based on the shared debugging example set, perform cyclic pre-debugging in multiple debugging stages on the inspection text processing network until the set cumulative number of debugging times is reached, and in one debugging stage, perform the following processing: Based on the gas sensor monitoring text debugging data group sampled from the local debugging example set, perform cyclic debugging on the inspection text processing network to be debugged for the number of times of Yuci to obtain an intermediate inspection text processing network, and based on the gas sensor monitoring text evaluation data sampled from the local test example set, determine the discrimination quality evaluation index of the intermediate inspection text processing network, where the local debugging example set and the local test example set are included in the debugging example set; According to the discriminant quality evaluation index determined for the intermediate inspection text processing network in different debugging stages, the target inspection text processing network that meets the set debugging requirements is determined from each intermediate inspection text processing network.

8. The method according to claim 7, wherein Select gas sensing monitoring text debugging data from the local debugging sample set, including: For the preselected X types of secondary abnormal state trend labels in the local debugging sample set, respectively select Y gas sensing monitoring text debugging data as a type of auxiliary debugging data, and respectively select Z non-overlapping gas sensing monitoring text debugging data as the debugging data to be warned; According to the sampled auxiliary debugging data and the debugging data to be warned, create X*Z gas sensing monitoring text debugging data groups, where each gas sensing monitoring text debugging data group includes a debugging data to be warned corresponding to one secondary abnormal state trend label, where X, Y, and Z are positive integers; Among them, the loop debugging of the inspection text processing network to be debugged for Yuci times according to the gas sensing monitoring text debugging data group sampled from the local debugging sample set to obtain the intermediate inspection text processing network includes: Use the generated gas sensing monitoring text debugging data group to perform loop debugging of the inspection text processing network to be debugged for Yuci times, and improve the neural network parameters of the inspection text processing network according to the debugging errors determined in each round of loop; among them, in a round of loop debugging process, the following processing is implemented: Input a gas sensing monitoring text debugging data group obtained into the inspection text processing network to be debugged, obtain a harmful gas warning debugging view determined for the debugging data to be warned in the gas sensing monitoring text debugging data group, and determine the debugging error according to the warning view difference between the harmful gas warning debugging view and the corresponding harmful gas warning certification view; Among them, the input of a gas sensing monitoring text debugging data group obtained into the inspection text processing network to be debugged to obtain a harmful gas warning debugging view determined for the debugging data to be warned in the gas sensing monitoring text debugging data group includes: Input the debugging data to be warned and the auxiliary debugging data included in a gas sensing monitoring text debugging data group into the text semantic mining model to be debugged to obtain a sensing monitoring text semantic vector debugging sample and each auxiliary sensing monitoring text semantic vector debugging sample; Use the semantic optimization model to be debugged, and respectively perform key semantic attention processing on the sensing monitoring text semantic vector debugging sample and each auxiliary sensing monitoring text semantic vector debugging sample according to the state trend element knowledge parsed for each trend keyword to obtain the target sensing monitoring text semantic vector debugging sample and each target auxiliary sensing monitoring text semantic vector debugging sample; Use the trend discriminant model to be debugged to determine the harmful gas warning output view corresponding to the debugging data to be warned according to the semantic commonality score between the target sensing monitoring text semantic vector debugging sample and each target auxiliary sensing monitoring text semantic vector debugging sample.

9. A text processing system for detection, characterized in that, Including: A memory for storing program instructions and data; A processor for being coupled to the memory and executing the instructions in the memory to implement the method according to any one of claims 1-8.

10. A computer storage medium, characterized in that, Instructions that, when executed on a processor, implement the method according to any one of claims 1-8.

Images