Temperature and Humidity Monitoring and Early Warning System for Station Equipment Rooms

Abstract

The present invention discloses a temperature and humidity monitoring and early warning system for a station equipment room, which relates to the technical field of data processing. The system includes: collecting the geographical environment characteristics of the target station equipment room, and obtaining a monitoring background feature set according to the collection results; obtaining a set of abnormal temperature and humidity monitoring information of the sample station equipment room; obtaining a set of screened sample abnormal information; obtaining a set of screened sample temperature and humidity feature groups; fusing the newly added sample abnormal information set with the screened sample abnormal information set to obtain a fused sample abnormal information set; obtaining the trained monitoring and early warning network layer; using the monitoring and early warning network layer to perform abnormal analysis on the target temperature and humidity monitoring data for temperature and humidity monitoring and early warning. It solves the technical problems in the prior art of ignoring background features and having a slow abnormal recognition, which leads to low accuracy of temperature and humidity monitoring and a lag in monitoring results, and achieves the technical effect of improving the accuracy and timeliness of temperature and humidity monitoring and early warning.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and specifically to a temperature and humidity monitoring and early warning system for station equipment rooms. Background Art

[0002] With the rapid development of modern transportation networks, stations and their equipment rooms, as important facilities, undertake a large number of operation and management tasks. To ensure the normal operation of equipment and the safety of passengers and staff, environmental monitoring of station equipment rooms becomes particularly important, especially the monitoring of temperature and humidity, which is of great significance for the normal operation of equipment and avoiding equipment failures caused by environmental factors. However, traditional temperature and humidity monitoring mostly relies on real-time data collection by a single sensor and simple threshold alarm methods, and there are still many deficiencies in the recognition and early warning of temperature and humidity anomalies in complex environments. For example, the environmental conditions in station equipment rooms change frequently, the sensor data is huge and complex, and it is difficult to effectively extract important information only by traditional methods, especially abnormal temperature and humidity changes and potential hidden dangers; it is slow to respond to sudden abnormal changes and environmental changes, resulting in untimely early warnings; existing methods ignore the comprehensive analysis of background features, lack effective modeling and mining of specific environmental factors in station equipment rooms, and the geographical, environmental, usage conditions, etc. of different station equipment rooms vary, resulting in inaccurate early warnings.

[0003] Therefore, in the current related technologies, there are technical problems of ignoring background features and being slow in abnormal recognition, which in turn lead to low accuracy of temperature and humidity monitoring and lagging monitoring results. Summary of the Invention

[0004] By providing a temperature and humidity monitoring and early warning system for station equipment rooms, this application solves the technical problems in the prior art of ignoring background features and being slow in abnormal recognition, which in turn lead to low accuracy of temperature and humidity monitoring and lagging monitoring results, and achieves the technical effect of improving the accuracy and real-time performance of temperature and humidity monitoring and early warning.

[0005] This application provides a temperature and humidity monitoring and early warning system for a station equipment room. The system includes: a monitoring background feature set obtaining module, configured to collect the geographical environment features of a target station equipment room, and obtain a monitoring background feature set according to the collection result; an abnormal temperature and humidity monitoring information set obtaining module, configured to use the monitoring background feature set as a label to perform frequent item mining on big data, and obtain an abnormal temperature and humidity monitoring information set of a sample station equipment room; a sample quality verification and screening module, configured to use an NLI filter in combination with the monitoring background feature set to perform sample quality verification and screening on the abnormal temperature and humidity monitoring information set of the sample station equipment room, and obtain a screened sample abnormal information set; a temperature and humidity feature extraction module, configured to perform temperature and humidity feature extraction on the screened sample abnormal information set, and obtain a screened sample temperature and humidity feature group set; a semantic difference analysis module, configured to perform semantic difference analysis on the screened sample temperature and humidity feature group set, and perform combination and paraphrasing on the screened sample abnormal information set according to the analysis result, obtain a new sample abnormal information set, and fuse the new sample abnormal information set with the screened sample abnormal information set to obtain a fused sample abnormal information set; a monitoring and early warning network layer obtaining module, configured to perform generative adversarial training on the monitoring and early warning network layer based on the fused sample abnormal information set, and obtain the trained monitoring and early warning network layer; a temperature and humidity monitoring and early warning module, configured to use a sensor array to perform temperature and humidity monitoring on the target station equipment room, obtain target temperature and humidity monitoring data, and use the monitoring and early warning network layer to perform abnormal analysis on the target temperature and humidity monitoring data, and perform temperature and humidity monitoring and early warning according to the analysis result.

[0006] In a possible implementation, using an NLI filter in combination with the monitoring background feature set to perform sample quality verification and screening on the abnormal temperature and humidity monitoring information set of the sample station equipment room, and obtaining a screened sample abnormal information set, the following processing is further performed: generating hypothesis data and data labels based on the monitoring background feature set, obtaining a hypothesis data set and a data label set, where the data labels include contradiction, entailment, and neutral, and each hypothesis data corresponds to a data label; training an NLI recognizer constructed based on the BERT model using the hypothesis data set and the data label set until the training converges, and obtaining a trained NLI filter; inputting the monitoring background feature set and the abnormal temperature and humidity monitoring information set of the sample station equipment room into the NLI filter to obtain a sample data label set; performing sample quality verification and screening on the abnormal temperature and humidity monitoring information set of the sample station equipment room based on the sample data label set, and obtaining a screened sample abnormal information set.

[0007] In a possible implementation, based on the sample data label set, sample quality verification and screening are performed on the sample abnormal temperature and humidity monitoring information set of the sample station equipment room to obtain a screened sample abnormal information set, and the following processing is also performed: when the sample data label in the sample data label set is "implication", the corresponding sample abnormal temperature and humidity monitoring information of the sample station equipment room is added to the screened sample abnormal information set; when the sample data label in the sample data label set is "contradiction", the corresponding sample abnormal temperature and humidity monitoring information of the sample station equipment room is excluded; when the sample data label in the sample data label set is "neutral", it is sent to the sample quality verification personnel for manual verification, and when the manual verification passes, the corresponding sample abnormal temperature and humidity monitoring information of the sample station equipment room is added to the screened sample abnormal information set.

[0008] In a possible implementation, semantic difference analysis is performed on the screened sample temperature and humidity feature group set, and based on the analysis result, combination and paraphrasing are performed on the screened sample abnormal information set to obtain a new sample abnormal information set, and the following processing is also performed: calculate the semantic difference degree between any one screened sample temperature and humidity feature group in the screened sample temperature and humidity feature group set and the remaining screened sample temperature and humidity feature groups to obtain a screened sample temperature and humidity feature group semantic difference degree set; with a preset semantic difference degree set as a constraint, use the screened sample temperature and humidity feature group semantic difference degrees in the screened sample temperature and humidity feature group semantic difference degree set that are lower than the preset semantic difference degree set as new templates to obtain a new template set; perform combination and paraphrasing on the new template set to obtain the new sample abnormal information set.

[0009] In a possible implementation, combination and paraphrasing are performed on the new template set to obtain the new sample abnormal information set, and the following processing is also performed: randomly combine the new template set according to a preset combination method to obtain a new combined sample abnormal information set, where the preset combination method is to perform information intersection between any one new template and at least two new templates; traverse and extract the keywords of the new template set to obtain a new template keyword group set, where each new template keyword group corresponds to a new template; perform random synonym conversion on the new template keyword group set, and integrate the new template set according to the conversion result to obtain a new paraphrased sample abnormal information set; use the new combined sample abnormal information set and the new paraphrased sample abnormal information set as the new sample abnormal information set.

[0010] In a possible implementation, based on the fused sample anomaly information set, generative adversarial training of the monitoring and early warning network layer is performed to obtain the trained monitoring and early warning network layer, and the following processing is also executed: dividing the fused sample anomaly information set into a training set and a validation set according to a preset ratio; constructing an initial monitoring and early warning network layer based on the training set; validating the initial monitoring and early warning network layer based on the validation set, and if the validation passes, obtaining the trained monitoring and early warning network layer.

[0011] In a possible implementation, after obtaining the trained monitoring and early warning network layer, the following processing is also executed: traversing the training set to extract feature vectors, and performing mean processing on the extraction results to obtain concentrated feature vectors; introducing random noise vectors, and combining the concentrated feature vectors to input into a generator to generate interference samples, obtaining an interference sample set; using the training set and the interference sample set to train a discriminator to obtain discrimination results, and obtaining a discrimination loss coefficient based on the discrimination results; using the discrimination loss coefficient to perform iterative update of network parameters for the discriminator and the generator until a preset number of iterations is satisfied, and combining the generator and the discriminator with updated network parameters to obtain an initially trained initial monitoring and early warning network layer.

[0012] In a possible implementation, after extracting the temperature and humidity features of the filtered sample anomaly information set to obtain a set of filtered sample temperature and humidity feature groups, the following processing is also executed: pre-constructing a feature extraction network layer based on a convolutional neural network; using the feature extraction network layer to perform feature analysis on the filtered sample anomaly information set to obtain the set of sample temperature and humidity feature groups.

[0013] It is intended to propose a temperature and humidity monitoring and early warning system for a station equipment room through this application, collect the geographical environment features of the target station equipment room, obtain a set of monitoring background features according to the collection results; obtain a set of abnormal temperature and humidity monitoring information for the sample station equipment room; obtain a set of filtered sample anomaly information; obtain a set of filtered sample temperature and humidity feature groups; fuse the new sample anomaly information set with the filtered sample anomaly information set to obtain a fused sample anomaly information set; obtain the trained monitoring and early warning network layer; use the monitoring and early warning network layer to perform anomaly analysis on the target temperature and humidity monitoring data for temperature and humidity monitoring and early warning. This solves the technical problems in the prior art of ignoring background features and having a slow anomaly recognition, which leads to low accuracy and lagging monitoring results in temperature and humidity monitoring, and achieves the technical effect of improving the accuracy and real-time performance of temperature and humidity monitoring and early warning. Description of the Drawings

[0014] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the accompanying drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the operations described above or below do not necessarily need to be executed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more steps can be removed from these processes.

[0015] Figure 1 Schematic structural diagram of the temperature and humidity monitoring and early warning system for the station equipment room provided by the embodiment of the present application;

[0016] Figure 2 Schematic diagram of the execution process of the sample quality verification and screening module in the temperature and humidity monitoring and early warning system for the station equipment room provided by the embodiment of the present application.

[0017] Explanation of reference numerals: Monitoring background feature set acquisition module 10, Abnormal temperature and humidity monitoring information set acquisition module 20, Sample quality verification and screening module 30, Temperature and humidity feature extraction module 40, Semantic difference analysis module 50, Monitoring and early warning network layer acquisition module 60, Temperature and humidity monitoring and early warning module 70. Detailed implementation manners

[0018] The above description is only an overview of the technical solutions of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following specifically gives the detailed implementation manners of the present application.

[0019] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0020] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "first" and "second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, system, product, or server that includes a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or modules not clearly listed or inherent to these processes, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application.

[0021] An embodiment of the present application provides a temperature and humidity monitoring and early warning system for a station equipment room, as Figure 1 shown, the system includes:

[0022] A monitoring background feature set acquisition module 10, configured to collect the geographical environment features of a target station equipment room and obtain a monitoring background feature set according to the collection result.

[0023] Preferably, collecting the geographical environment features of the target station equipment room means obtaining the geographical and environmental data of the target station equipment room and establishing a monitoring background feature set. Specifically, the specific geographical information of the target station equipment room is collected, usually including the geographical location, such as the specific location of the station, the geographical area where it is located (e.g., city, terrain, climate zone, etc.); the building structure, the structure, floor, isolation degree, and orientation of the building where the equipment room is located; the surrounding environment, the surrounding climate conditions (such as temperate, tropical climate, humidity, etc.), the natural environment (such as wind speed, precipitation, etc.), and the pedestrian flow (the passenger flow density of the station), etc. Then, based on the collected geographical and environmental features, they are integrated to form a monitoring background feature set, including environmental temperature and humidity features, and may also include the equipment features in the room (such as the operating status, power consumption, etc. of air conditioners, ventilation equipment, power equipment, etc.), so as to obtain an environmental feature database, which serves as the basis for the monitoring system to analyze temperature and humidity changes and identify abnormalities.

[0024] An abnormal temperature and humidity monitoring information set acquisition module 20, configured to perform frequent item mining on big data with the monitoring background feature set as a label to obtain an abnormal temperature and humidity monitoring information set of a sample station equipment room.

[0025] Preferably, in the temperature and humidity monitoring and early warning system, background features are used as markers, and the frequent item mining technology is adopted to extract abnormal temperature and humidity data samples from large-scale monitoring data. Specifically, the monitoring background feature set is used as a comparison label, providing a standardized environmental background for subsequent data analysis. For example, under different environmental conditions, the same temperature and humidity may have different normal ranges. By combining the collected monitoring data with these background features, the data can be classified and labeled, and then frequent item mining is performed on the big data. Among them, frequent item mining is a technology in data mining, usually used to extract items or patterns that often appear simultaneously from a large amount of data, that is, by analyzing the temperature and humidity monitoring values in the big data, identifying which temperature and humidity change patterns frequently appear in the sample station equipment rooms. For example, a certain specific temperature and humidity value may frequently appear at a specific time and geographical environment. By comparing with the background feature set, the temperature and humidity combinations that should not frequently appear under normal circumstances are mined, forming an abnormal temperature and humidity monitoring information set, which represents the temperature and humidity conditions that may indicate equipment failures, environmental anomalies, or improper equipment operation. Moreover, these abnormal monitoring information comes from different station equipment rooms, representing the abnormal situations that may occur under different environmental conditions in each equipment room. The abnormal temperature and humidity monitoring information set contains all abnormal temperature and humidity data obtained based on data mining technology. By using these abnormal information, it is judged whether there are potential failures or safety hazards in the equipment room.

[0026] The sample quality verification and screening module 30 is used to utilize the NLI filter to combine with the monitoring background feature set to perform sample quality verification and screening on the abnormal temperature and humidity monitoring information set of the sample station equipment room, and obtain a screened sample abnormal information set.

[0027] Preferably, by using an NLI (Non-Linear Integration) filter and combining the extracted set of monitoring background features, the abnormal temperature and humidity data mined from big data is subjected to quality verification and screening to obtain a set of abnormal information with relatively high quality. Among them, the NLI filter is a non-linear integration technology, usually used for the fusion and screening of multi-data sources, capable of performing quality verification and noise reduction on data, filtering out unqualified, abnormal or irrelevant samples. The NLI filter can handle non-linear data relationships. Especially for the changes in temperature and humidity data in a complex environment, the NLI filter can perform more accurate screening of data based on non-linear methods. Through the NLI filter, data from different sources (such as temperature and humidity data, equipment room background features, historical data, etc.) can be combined to comprehensively evaluate the reliability of abnormal data, thereby improving the accuracy of screening. Specifically, during the screening process of abnormal data, the NLI filter does not work in isolation, but combines with the set of monitoring background features for data verification. Guided by the background features, the NLI filter can more effectively distinguish which temperature and humidity anomalies are real system problems and which are just invalid anomalies caused by environmental differences or data acquisition errors, and then determine which temperature and humidity data belong to reasonable fluctuations in a specific room and which belong to potential anomalies.

[0028] Preferably, after obtaining a large set of abnormal temperature and humidity monitoring information, the NLI filter and the set of background features are used together to screen these data to ensure that only abnormal samples with relatively high quality are retained. The specific screening may include noise filtering to remove the incorrect data caused by data transmission errors, equipment failures or other external factors, and to prevent these invalid data from affecting subsequent analysis; the NLI filter can screen out the truly warning abnormal data by comparing with the set of background features. For data that highly does not match the background conditions (for example, the temperature and humidity exceed the normal range or fluctuate frequently), the NLI filter will determine it as abnormal and retain it in the screening set; finally, a set of abnormal information of the screened samples is obtained, which contains the temperature and humidity abnormal information screened from a large amount of monitoring data, verified in quality and with real warning significance. The abnormal data in the set of abnormal information of the screened samples is of high quality and high reliability, and the abnormal data of each station equipment room will be screened according to its specific environmental background to ensure the personalization and pertinence of the data and improve the accuracy of temperature and humidity monitoring.

[0029] The temperature and humidity feature extraction module 40 is used to extract the temperature and humidity features from the set of abnormal information of the screened samples to obtain a set of temperature and humidity feature groups of the screened samples.

[0030] Preferably, important features related to temperature and humidity are extracted from the screened sample anomaly information set to form a set containing temperature and humidity features, namely, the screened sample temperature and humidity feature group set. Specifically, temperature and humidity feature extraction is to identify and extract key temperature and humidity data features from the screened abnormal samples to convert the original temperature and humidity data into effective features that can be used for analysis and modeling. The extracted temperature and humidity features may include average temperature and humidity (average temperature and humidity over a period of time), fluctuation range (fluctuation range or standard deviation of temperature and humidity data, reflecting the severity of temperature and humidity changes), peak and valley values (identifying extreme temperature and humidity values in the data), change rate (rate of change of temperature and humidity), time window features (dividing temperature and humidity data into different time windows, such as hours, days, weeks), and trend features (extracting the long-term trend of temperature and humidity based on time series data and judging its rising or falling direction). These features are integrated into the screened sample temperature and humidity feature group set, which contains the key temperature and humidity features corresponding to all screened abnormal data samples. Among them, each screened sample temperature and humidity feature group set includes a temperature feature and a humidity feature, such as high temperature and high humidity, low temperature and high humidity, etc. Through the characterized temperature and humidity data, the model can more accurately identify potential anomalies and faults and issue accurate warnings.

[0031] The semantic difference analysis module 50 is used to perform semantic difference analysis on the screened sample temperature and humidity feature group set, and combine and paraphrase the screened sample anomaly information set according to the analysis results to obtain a new sample anomaly information set, and fuse the new sample anomaly information set with the screened sample anomaly information set to obtain a fused sample anomaly information set.

[0032] Preferably, semantic difference analysis, integration, paraphrasing, and optimization are performed on different temperature and humidity features of the screened sample temperature and humidity feature group set to form a fused anomaly information set, namely, the fused sample anomaly information set, which is used to improve the accuracy and comprehensiveness of the monitoring system. Specifically, semantic difference analysis explores the differences between different samples from the semantic level of the data. There may be some minor differences in the data, but these differences often represent different equipment states, environmental changes, or potential problems. The differences at the semantic level mean that among different temperature and humidity features, some changes may have a substantial impact, while some differences may be noise or external environmental interference. For example, a relatively fast rate of temperature change may indicate equipment failure, while a slight temperature fluctuation may just be a normal change in environmental factors. Through semantic difference analysis, the essential differences in temperature and humidity changes between different samples can be identified and distinguished, further improving the accuracy of anomaly recognition.

[0033] Preferably, after the semantic difference analysis is completed, the screened sample abnormal information set is combined and paraphrased according to the analysis results, that is, according to the analysis results of different samples, new abnormal samples are synthesized to more comprehensively identify potential problems or faults. Specifically, the key features or abnormal patterns of multiple abnormal samples are combined to generate new samples. For example, if multiple samples show similar abnormalities (such as a sudden increase in temperature and a large change in humidity) during certain time periods, these samples are combined into a more representative abnormal pattern, and then the originally scattered abnormal information is transformed into a structured and more understandable abnormal description. For example, the original large change in temperature and humidity range is paraphrased as a potential fault that the device may have overheating or excessive humidity, and then a new sample abnormal information set is obtained, which is a new abnormal pattern or sample formed by difference analysis and combination on the basis of the original screened samples, and can better reflect the actual changes and potential problems of the device environment, and is more representative and accurate than the original sample set; finally, the new sample abnormal information set is fused with the screened sample abnormal information set to generate a complete abnormal data set that includes multiple abnormal patterns and covers different device environments and operating states, that is, the fused sample abnormal information set, to improve the accuracy of abnormal detection, increase the ability to identify multiple potential problems, and improve the accuracy and timeliness of early warning.

[0034] The monitoring and early warning network layer obtaining module 60 is used to perform generative adversarial training on the monitoring and early warning network layer based on the fused sample abnormal information set to obtain the trained monitoring and early warning network layer.

[0035] Preferably, generative adversarial training on the monitoring and early warning network layer is performed based on the fused sample abnormal information set. Among them, the generative adversarial network (GAN) is a deep learning model that generates high-quality data through adversarial training (adversarial learning), including two parts: a generator and a discriminator. The generator generates samples that are as close as possible to real data according to the input random noise or existing data, that is, generates similar abnormal temperature and humidity data according to the characteristics of the fused samples to simulate potential abnormal patterns. The discriminator is used to distinguish whether the input data is real (that is, from actual monitoring data) or fake data generated by the generator, that is, to judge whether a certain temperature and humidity data belongs to an abnormal pattern or is fake data generated by the system; the training process of the generative adversarial network is an adversarial training process between the generator and the discriminator. The generator continuously improves the quality of the generated data, and the discriminator continuously improves the ability to identify the generated data. Eventually, the generator can generate samples that are very close to real abnormal data, and the discriminator can accurately judge the authenticity of the data.

[0036] Preferably, the obtained set of abnormal information of the fusion samples is used as training data and input into the GAN model. The generator will attempt to generate new samples that appear similar to the real abnormal temperature and humidity data based on this abnormal data. The discriminator classifies the input data into two categories: "real" or "fake". The goal is to correctly identify whether the data is real monitoring data or fake data generated by the generator. Through adversarial training, the discriminator continuously improves the accuracy of identification. The goal of the generator is to make the discriminator unable to distinguish the generated data from the real data, while the goal of the discriminator is to try to identify the fake data. Finally, through multiple rounds of adversarial training, the abnormal data generated by the generator will be more real, and the discriminator will also become more accurate, thus obtaining a monitoring and warning network layer for analyzing temperature and humidity data and predicting in real time whether there are abnormalities, and being able to provide accurate abnormal detection and warning.

[0037] The temperature and humidity monitoring and warning module 70 is used to monitor the temperature and humidity of the target station equipment room by using a sensor array, obtain the target temperature and humidity monitoring data, and perform abnormal analysis on the target temperature and humidity monitoring data by using the monitoring and warning network layer, and conduct temperature and humidity monitoring and warning according to the analysis results.

[0038] Preferably, in the target station equipment room, the temperature and humidity data are collected in real time through a sensor array, and these data are analyzed by the trained monitoring and warning network layer to identify potential temperature and humidity abnormalities in real time and give warnings. Specifically, the sensor array consists of multiple temperature and humidity sensors and is used to monitor the changes in temperature and humidity in the target station equipment room in real time. Through the sensor array, the temperature and humidity data can be obtained at different room positions or different heights (such as the ground and the ceiling), and the temperature and humidity data at each position in the room can be obtained more comprehensively and accurately, that is, the target temperature and humidity monitoring data, thereby improving the coverage and accuracy of monitoring. Then, the monitoring and warning network layer is used to perform abnormal analysis on the target temperature and humidity monitoring data, that is, to judge whether these data exceed the normal range or show abnormal patterns, including judging whether the temperature and humidity exceed the standard, sudden changes (for example, the temperature rises sharply within a few minutes), and continuous trends (long-term abnormal changes in temperature and humidity), so as to identify any temperature and humidity abnormal phenomena that may affect the normal operation of the equipment or cause potential safety hazards. Finally, temperature and humidity monitoring and warning are carried out according to the analysis results. For example, when a temperature and humidity abnormality is detected, warning notifications are sent to the management personnel through various means (such as text messages, emails, visual interfaces, etc.); an abnormal report is generated, which details the time, location, temperature and humidity data, etc. of the abnormality occurrence to help the staff analyze the cause of the fault; it is linked with the automated control system of the station (such as air conditioners, ventilation equipment, etc.) to automatically adjust the temperature and humidity to avoid overheating or overhumidification of the equipment; ensure that the environment of the equipment room is adjusted in a timely manner to avoid equipment failures or other safety problems caused by temperature and humidity abnormalities.

[0039] The temperature and humidity monitoring and early warning system for the station equipment room according to the embodiments of the present invention is used to solve the technical problems existing in the prior art, such as ignoring background features and being slow in abnormal identification, which in turn lead to low accuracy of temperature and humidity monitoring and lagging monitoring results, and achieves the technical effect of improving the accuracy and real-time performance of temperature and humidity monitoring and early warning. The temperature and humidity monitoring and early warning system for the station equipment room includes: a monitoring background feature set acquisition module 10, an abnormal temperature and humidity monitoring information set acquisition module 20, a sample quality verification and screening module 30, a temperature and humidity feature extraction module 40, a semantic difference analysis module 50, a monitoring and early warning network layer acquisition module 60, and a temperature and humidity monitoring and early warning module 70.

[0040] Next, the specific configuration of the sample quality verification and screening module 30 will be described in detail. As Figure 2 shown, the sample quality verification and screening module 30 may further include: generating hypothesis data and data labels based on the monitoring background feature set to obtain a hypothesis data set and a data label set, where the data labels include contradiction, implication, and neutral, and each hypothesis data corresponds to a data label; using the hypothesis data set and the data label set to train an NLI recognizer constructed based on the BERT model until the training converges to obtain a trained NLI filter; inputting the monitoring background feature set and the sample station equipment room abnormal temperature and humidity monitoring information set into the NLI filter to obtain a sample data label set; performing sample quality verification and screening on the sample station equipment room abnormal temperature and humidity monitoring information set based on the sample data label set to obtain a screened sample abnormal information set.

[0041] Preferably, according to the background features, some possible temperature and humidity data, i.e., hypothesis data, are generated. These data do not necessarily represent the monitoring data in the real environment but are used as part of the training data. Each hypothesis data will be labeled with a data label representing the relationship between the data and the target data, such as contradiction, neutral, implication, etc. Then, the hypothesis data set and the data label set are used to train an NLI recognizer constructed based on the BERT model. Among them, BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language model based on Transformer, which is usually used for natural language processing tasks. It can understand the context information in the text and judge the relationship between different sentences, that is, the BERT model is used to understand the relationship between the hypothesis data and the real data and judge whether they have a contradictory, implicative, or neutral relationship. Through a large number of hypothesis data and data labels, the model will learn how to judge whether the temperature and humidity monitoring data has the above three relationships. The training process continues until the model can accurately classify when processing data, and finally a trained NLI filter is obtained.

[0042] Preferably, after training is completed, the NLI filter is used to analyze the actual sample station equipment room abnormal temperature and humidity monitoring information set. Specifically, the monitoring background feature set and the abnormal monitoring information set are input into the trained NLI filter. The NLI filter will generate a data label for each input abnormal monitoring data sample. The label indicates whether the data is abnormal and the nature of the abnormality (contradiction, implication or neutrality), which helps to distinguish the abnormalities in the data from normal data; then, sample quality verification and screening are performed based on the sample data label set. The sample data is screened according to the data label set to determine which abnormal data is reliable and which data needs to be excluded, that is, those samples marked as contradictory or implicit are selected as high-quality abnormal data, and the data marked as neutral are excluded. Finally, after screening and quality verification by the NLI filter, a screened sample abnormal information set is obtained, which has high quality and reliability and can be used for subsequent analysis, training or real-time monitoring, thereby improving the accuracy of abnormality identification and early warning effect.

[0043] The specific configuration of the sample quality verification and screening module 30 will be described in detail below. The sample quality verification and screening module 30 may further include: when the sample data label in the sample data label set is implied, the corresponding abnormal temperature and humidity monitoring information of the equipment room of the sample station is added into the screening sample abnormal information set; when the sample data label in the sample data label set is contradictory, the corresponding abnormal temperature and humidity monitoring information of the equipment room of the sample station is eliminated; when the sample data label in the sample data label set is neutral, it is sent to the sample quality verification personnel for manual verification, and when the manual verification passes, the corresponding abnormal temperature and humidity monitoring information of the equipment room of the sample station is added into the screening sample abnormal information set.

[0044] Preferably, when the sample data label is "implication", it indicates that the relationship between the abnormal data and the monitoring background feature set meets the expectations, that is, the abnormal changes in the data may occur under known environmental conditions, and these changes are reasonable and possible. This data sample can be regarded as valid abnormal information and should be retained. Add these data to the screened sample abnormal information set; when the sample data label is "contradiction", it indicates that there is an obvious conflict or inconsistency between the abnormal data and the background feature set, usually indicating that this data sample does not conform to the expected abnormal pattern, or is caused by incorrect device monitoring, data noise, or other external interferences. Since the samples with the "contradiction" label may be inaccurate or do not represent real abnormalities, these data samples will be excluded to maintain the high quality of the data set; when the sample data label is "neutral", it means that this data sample cannot be clearly judged as abnormal or normal under the current analysis model. Its relationship with the background feature set neither constitutes an obvious abnormality nor completely conforms to the known abnormal pattern, being in a fuzzy state. It is impossible to simply rely on the model to judge whether it is valid abnormal data. Send these data to the sample quality verification personnel for manual verification, that is, conduct manual analysis based on these data samples to judge whether they are valid abnormal data. For example, check whether there are special factors in the monitoring environment that cause these data samples to be unclear; further verify whether the data is reasonable according to the specific temperature and humidity change patterns and the device operation conditions; if the manual verification personnel confirm that the data is valid abnormal information, add it to the screened sample abnormal information set to ensure the quality and accuracy of the final abnormal information set, so as to improve the accuracy and reliability of the monitoring and warning system.

[0045] Next, the specific configuration of the semantic difference analysis module 50 will be described in detail. The semantic difference analysis module 50 may further include: calculating the semantic difference degree between any one of the screened sample temperature and humidity feature groups in the screened sample temperature and humidity feature group set and the remaining screened sample temperature and humidity feature groups to obtain a screened sample temperature and humidity feature group semantic difference degree set; using the preset semantic difference degree set as a constraint, taking the screened sample information corresponding to the screened sample temperature and humidity feature group semantic difference degree lower than the preset semantic difference degree set in the screened sample temperature and humidity feature group semantic difference degree set as a new template to obtain a new template set; combining and paraphrasing the new template set to obtain the new sample abnormal information set.

[0046] Preferably, the screened sample temperature and humidity feature group set includes multiple screened abnormal temperature and humidity feature groups. Select any one of the screened sample temperature and humidity feature groups, and calculate the semantic difference degree between this feature group and other screened sample temperature and humidity feature groups, that is, by means of cosine similarity, Euclidean distance, etc., compare each pair of screened sample feature groups, calculate the semantic difference degree between them, and quantify the difference between two sample features in temperature and humidity data. Among them, a high semantic difference degree indicates that the two samples are quite different in temperature and humidity features, which may represent two different types of abnormalities; a low semantic difference degree indicates that the temperature and humidity features between the two samples are very similar, which may represent similar abnormal situations and belong to the same type of abnormality. Thus, a semantic difference degree set of screened sample temperature and humidity feature groups is obtained, which contains the set of difference degrees between all screened sample temperature and humidity feature groups. Through this set, it can be found which samples have similar temperature and humidity features and which are quite different.

[0047] Preferably, use the preset semantic difference degree set for constrained screening. Among them, the preset semantic difference degree set is a predefined threshold set used to specify the acceptable maximum difference degree range. Using this set as a constraint condition, screen the semantic difference degree of the screened samples, and regard the samples with a semantic difference degree lower than the preset semantic difference degree threshold as new templates, representing potential abnormal types. Then, combine and paraphrase the selected new template set, that is, "fuse" and re-express these similar abnormal samples to expand the samples. Specifically, merge or integrate multiple similar sample features to form a new abnormal pattern template, which helps to extract common features, simplify abnormal information, and enhance the universality of the pattern. Through paraphrasing, multiple abnormal samples are described as a concise and representative abnormal pattern or information, and finally, a new sample abnormal information set is obtained, which represents some specific types of abnormal patterns, provides new templates for the subsequent abnormal recognition system, and helps the system to more accurately identify and warn potential equipment abnormalities in the subsequent monitoring process.

[0048] Next, the specific configuration of the semantic difference analysis module 50 will be further described in detail. The semantic difference analysis module 50 may further include: randomly combining the new template set according to a preset combination method to obtain a new combined sample abnormal information set, where the preset combination method is to perform information intersection between any one new template and at least two new templates; traversing and extracting the keywords of the new template set to obtain a new template keyword group set, where each new template keyword group corresponds to a new template; randomly converting the synonyms of the new template keyword group set, and integrating the new template set according to the conversion result to obtain a new paraphrased sample abnormal information set; using the new combined sample abnormal information set and the new paraphrased sample abnormal information set as the new sample abnormal information set.

[0049] Preferably, the newly added template set is randomly combined according to a preset combination method, that is, each newly added template is randomly cross-combined with at least two other newly added templates to generate multiple new abnormal samples, which represent the intersection or comprehensive result of multiple abnormal patterns. Through random combination, different template features will be combined together to form new abnormal information, which can enrich the abnormal detection model, capture more potential abnormal patterns, and enhance the system's recognition ability; by traversing all newly added templates, the keywords in all templates are extracted and organized into a keyword group set. Each newly added template will have a corresponding keyword group, which represents the core features such as the temperature and humidity changes of the template. By extracting keywords, the core features of each template can be more accurately identified and understood, providing a basis for subsequent synonym conversion and information integration.

[0050] Preferably, synonym replacement is performed on the keyword group set of the newly added templates. Synonym conversion means replacing some keywords in the template with words having the same or similar meanings. The purpose is to generate the same abnormal pattern with different expressions. For example, "temperature increase" is converted to "temperature rise", and "humidity too high" is converted to "humidity too large". Random conversion means performing conversion by randomly selecting synonyms to avoid all templates being too single or repetitive in expression, providing multiple different expressions for each abnormal template, enabling the model to learn more variant abnormal patterns, and improving its recognition and adaptation ability; combining the abnormal information set of the newly added combined samples (generated by random combination) and the abnormal information set of the newly added paraphrased samples (the set integrated through synonym conversion) together to form the abnormal information set of the newly added samples, which contains abnormal information samples generated by different methods, covering the combined abnormal patterns and the diverse expressions generated by paraphrasing. Through this combination and integration, the system generates an abnormal sample set with high diversity and representativeness for training and optimizing the model, helping the system better identify and respond to different types of temperature and humidity abnormalities, and enhancing the system's recognition and prediction ability.

[0051] Next, the specific configuration of the acquisition module 60 of the monitoring and early warning network layer will be described in detail. The acquisition module 60 of the monitoring and early warning network layer may further include: dividing the fused sample abnormal information set into a training set and a validation set according to a preset ratio; constructing an initial monitoring and early warning network layer based on the training set; validating the initial monitoring and early warning network layer based on the validation set, and if the validation passes, obtaining the trained monitoring and early warning network layer.

[0052] Preferably, the set of abnormal information of the fusion samples is divided into a training set and a validation set according to a preset ratio. Usually, 70% is used for training and 30% is used for validation. The training set contains most of the data and is used to train the model so that the model can learn the relationship between the input data and the target output. The validation set is used to evaluate the performance of the model during the training process to check whether the model is overfitting or overly dependent on the training data. An initial monitoring and warning network layer is constructed based on a deep learning model (such as a neural network), and the sample data in the training set is used to train it to learn how to identify abnormal patterns in the temperature and humidity monitoring data. Usually, it includes an input layer (receiving the temperature and humidity data samples as inputs), a hidden layer (processing the data through multiple neurons and activation functions to extract the features of abnormal patterns), and an output layer (outputting the determination result of abnormal or normal according to the training samples, such as whether there is temperature and humidity abnormality). Then, the validation set is used to test and validate the initial monitoring and warning network layer to evaluate the accuracy and stability of the initial model. When the initial monitoring and warning network layer meets the preset performance standard on the validation set, it is considered that the validation is passed, and the trained monitoring and warning network layer is obtained, indicating that it has learned how to identify and warn of temperature and humidity abnormal information, ensuring that accurate abnormal detection and warning of the temperature and humidity data in the equipment room can be carried out based on the trained network layer.

[0053] Next, the specific configuration of the monitoring and warning network layer obtaining module 60 will be further described in detail. The monitoring and warning network layer obtaining module 60 may further include: traversing the training set to extract feature vectors, and performing mean processing on the extraction results to obtain concentrated feature vectors; introducing random noise vectors, combining the concentrated feature vectors and inputting them into a generator to generate interference samples to obtain an interference sample set; using the training set and the interference sample set to train a discriminator to obtain discrimination results, and obtaining a discrimination loss coefficient based on the discrimination results; using the discrimination loss coefficient to iteratively update the network parameters of the discriminator and the generator until the preset number of iterations is met, and combining the generator and the discriminator with the updated network parameters to obtain the initially trained initial monitoring and warning network layer.

[0054] Preferably, all sample data in the training set are traversed, and the feature vectors of each sample are extracted, which can reflect the core features of the sample and may include information such as temperature, humidity change patterns, abnormal peaks, etc. For example, for each sample, features such as the fluctuation range of temperature and the change rate of humidity may be extracted. For all the feature vectors in the training set, mean processing is performed, that is, the feature vectors of all samples are averaged to obtain a global average eigenvalue, called the concentrated feature vector, which represents the average feature of the entire training set. Then, a random noise vector is introduced to provide diverse inputs for the generator, so as to generate different samples and avoid the generator always generating similar samples. Specifically, the generated random noise vector is combined with the obtained concentrated feature vector and input into the generator to generate new interference samples. These samples may be different from the real samples in the training set and have a certain degree of randomness, thereby expanding the training set data, enabling the model to better learn various abnormal patterns and having stronger generalization ability.

[0055] Preferably, the task of the discriminator is to determine whether the input sample is a real sample (from the training set) or a generated sample (from the generator). Specifically, the discriminator is trained using the training set and the interference sample set together to learn how to distinguish samples from real training data and interference samples generated by the generator. The output result of the discriminator is a judgment on whether the input sample is real, usually returning a probability value indicating the likelihood that the input sample is a real sample, and a discriminant loss coefficient is obtained based on the discriminant result. The loss coefficient is usually calculated using the cross-entropy loss function based on the difference between the output result of the discriminator and the real label (i.e., whether the sample is real). Using the calculated discriminant loss coefficient, the parameters of the generator and the discriminator are iteratively updated using the backpropagation algorithm. That is, the generator adjusts its parameters according to the feedback of the discriminator to make the generated samples more realistic, and the discriminator optimizes its parameters according to the difference between the generated samples and the real samples to better distinguish real and fake samples. This continues until the preset number of iterations is reached. The generator and the discriminator with updated network parameters are combined to obtain the initially trained monitoring and warning network layer, which can effectively identify and warn of temperature and humidity anomalies and help realize an intelligent monitoring and warning system.

[0056] Next, the specific configuration of the temperature and humidity feature extraction module 40 will be described in detail. The temperature and humidity feature extraction module 40 may further include: a feature extraction network layer pre-constructed based on a convolutional neural network; and using the feature extraction network layer to perform feature analysis on the screened sample abnormal information set to obtain the sample temperature and humidity feature group set.

[0057] Preferably, a convolutional neural network (CNN) is used for feature extraction and to analyze the screened sample anomaly information set, and finally a temperature and humidity feature group set is obtained. Among them, the convolutional neural network is a deep learning algorithm, especially suitable for processing image data and time series data, that is, for extracting potential features in temperature and humidity data. Specifically, according to the known anomaly data and background features, a feature extraction network layer is pre-constructed. This network layer includes several convolutional layers (for extracting local features) and pooling layers (for feature dimensionality reduction and abstraction), and can automatically identify meaningful features from the data. Then, the feature extraction network layer is used to perform feature analysis on the screened sample anomaly information set, that is, to analyze each data sample in the screened sample anomaly information set, extract the key features of each sample, and through the action of the convolutional neural network, identify patterns related to temperature and humidity anomalies from these samples, such as sudden changes in temperature and humidity, long-term trend changes, periodic fluctuations, etc., and then obtain a sample temperature and humidity feature group set, a data set containing multiple different temperature and humidity feature groups. Each feature group represents the key features of a sample. For example, a feature group may include the temperature change range, humidity fluctuation amplitude, temperature and humidity change rate, and temperature and humidity periodic change characteristics, which can help the system identify anomaly patterns and trends, and improve the accuracy and robustness of anomaly detection.

[0058] Although this application makes various references to certain modules in the system according to the embodiments of this application, however, any number of different modules can be used and run on the user terminal and / or server. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved. In addition, the specific names of the functional units are only for easy distinction from each other and do not limit the protection scope of the present invention.

[0059] The above specific implementation manners do not constitute a limitation to the protection scope of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims

1. Temperature and humidity monitoring and early warning system for station equipment rooms, characterized in that, The system comprises: A monitoring background feature set acquisition module is used to collect geographical environment features of the target station equipment room and obtain a monitoring background feature set based on the collection results; The abnormal temperature and humidity monitoring information set acquisition module is used to perform frequent item mining on big data using the monitoring background feature set as a label to obtain the abnormal temperature and humidity monitoring information set of the equipment room of the sample station; A sample quality verification and screening module is used to use an NLI filter in combination with the monitoring background feature set to perform sample quality verification and screening on the abnormal temperature and humidity monitoring information set of the equipment room of the sample station to obtain a screening sample abnormal information set; The NLI filter is a nonlinear integration technology used for the fusion and screening of multiple data sources, to verify the quality and reduce noise of data, and to filter out unqualified, abnormal or irrelevant samples; A temperature and humidity feature extraction module is used to extract temperature and humidity features from the screening sample abnormal information set to obtain a screening sample temperature and humidity feature group set; A semantic difference analysis module, used for performing semantic difference analysis on the temperature and humidity feature group set of the screened samples, and combining and paraphrasing the screened sample abnormal information set according to the analysis result to obtain a newly added sample abnormal information set, and fusing the newly added sample abnormal information set with the screened sample abnormal information set to obtain a fused sample abnormal information set; A monitoring and early warning network layer acquisition module is used to perform generative adversarial training of the monitoring and early warning network layer based on the fusion sample abnormal information set to obtain the trained monitoring and early warning network layer; The temperature and humidity monitoring and early warning module is used to use the sensor array to monitor the temperature and humidity of the target station equipment room, obtain the target temperature and humidity monitoring data, and use the monitoring and early warning network layer to perform abnormal analysis on the target temperature and humidity monitoring data, and perform temperature and humidity monitoring and early warning according to the analysis results.

2. The temperature and humidity monitoring and early warning system for the station equipment room according to claim 1, characterized in that, The steps performed by the sample quality verification and screening module include: Generate hypothetical data and data labels based on the monitoring background feature set to obtain a hypothetical data set and a data label set, wherein the data labels include contradiction, implication and neutrality, and each hypothetical data corresponds to a data label; Using the hypothetical data set and the data label set to train the NLI recognizer built based on the BERT model until the training converges, thereby obtaining a trained NLI filter; Input the monitoring background feature set and the abnormal temperature and humidity monitoring information set of the sample station equipment room into the NLI filter to obtain a sample data label set; Based on the sample data label set, the sample station equipment room abnormal temperature and humidity monitoring information set is screened for sample quality verification to obtain a screened sample abnormal information set.

3. The temperature and humidity monitoring and early warning system for the station equipment room according to claim 2, characterized in that The steps performed by the sample quality verification and screening module include: When the sample data label in the sample data label set is implied, the corresponding sample station equipment room abnormal temperature and humidity monitoring information is added to the screening sample abnormal information set; When the sample data labels in the sample data label set are contradictory, the abnormal temperature and humidity monitoring information of the corresponding sample station equipment room will be eliminated; When the sample data label in the sample data label set is neutral, it is sent to the sample quality verification personnel for manual verification. When the manual verification passes, the corresponding abnormal temperature and humidity monitoring information of the sample station equipment room is added to the screened sample abnormal information set.

4. The temperature and humidity monitoring and early warning system for the station equipment room according to claim 1, characterized in that, The steps performed by the semantic difference analysis module include: Calculate the semantic difference degree between any one screened sample temperature and humidity feature group in the screened sample temperature and humidity feature group set and the remaining screened sample temperature and humidity feature groups, and obtain the screened sample temperature and humidity feature group semantic difference degree set; Taking the preset semantic difference degree set as a constraint, use the screened sample information corresponding to the screened sample temperature and humidity feature group semantic difference degree in the screened sample temperature and humidity feature group semantic difference degree set that is lower than the preset semantic difference degree set as a new template, and obtain the new template set; Combine and paraphrase the new template set to obtain the new sample abnormal information set.

5. The temperature and humidity monitoring and early warning system for the station equipment room according to claim 4, characterized in that, The steps performed by the semantic difference analysis module include: Randomly combine the new template set according to a preset combination method to obtain a new combined sample abnormal information set, where the preset combination method is to perform information crossover between any one new template and at least two new templates; Traverse and extract the keywords of the new template set to obtain the new template keyword group set, where each new template keyword group corresponds to a new template; Perform random synonym conversion on the new template keyword group set, and integrate the new template set according to the conversion result to obtain the new paraphrased sample abnormal information set; Use the new combined sample abnormal information set and the new paraphrased sample abnormal information set as the new sample abnormal information set.

6. The temperature and humidity monitoring and early warning system for the station equipment room according to claim 1, characterized in that, The steps performed by the monitoring and warning network layer obtaining module include: Divide the fused sample abnormal information set into a training set and a validation set according to a preset ratio; Construct an initial monitoring and warning network layer based on the training set; Verify the initial monitoring and warning network layer based on the validation set. If the verification passes, obtain the trained monitoring and warning network layer.

7. The temperature and humidity monitoring and early warning system for the station equipment room according to claim 6, characterized in that, The steps performed by the monitoring and warning network layer obtaining module include: Traverse the training set to extract feature vectors, and perform mean processing on the extraction results to obtain a concentrated feature vector; Introduce a random noise vector, combine it with the concentrated feature vector and input it into the generator to generate interference samples, and obtain an interference sample set; Use the training set and the interference sample set to train the discriminator to obtain a discrimination result, and obtain a discrimination loss coefficient based on the discrimination result; Use the discrimination loss coefficient to iteratively update the network parameters of the discriminator and the generator until the preset number of iterations is satisfied, and combine the generator and discriminator with updated network parameters to obtain the initially trained initial monitoring and warning network layer.

8. The temperature and humidity monitoring and early warning system for the station equipment room according to claim 1, characterized in that The steps performed by the temperature and humidity feature extraction module include: Pre-construct a feature extraction network layer based on a convolutional neural network; Use the feature extraction network layer to perform feature analysis on the screened sample abnormal information set to obtain the sample temperature and humidity feature group set.

Images