Multi-mode sensing assisted toxic gas dynamic monitoring system

Abstract

The invention provides a multi-mode perception-assisted toxic gas dynamic monitoring system, and relates to the technical field of toxic gas monitoring, and the system comprises the following steps: communicating with a multi-mode sensor and receiving environment perception data; the environment temperature information and the environment humidity information are processed through a toxic gas concentration deviation prediction model, and toxic gas concentration prediction deviation is obtained; when the poisonous gas concentration prediction deviation is larger than or equal to a concentration deviation threshold value, the first poisonous gas concentration information is corrected and then stored; when the poisonous gas concentration prediction deviation is smaller than a concentration deviation threshold value, the first poisonous gas concentration information is stored. According to the invention, the technical problem of inaccurate monitoring data caused by insufficient sensitivity when a sensor monitors low-concentration poisonous gas in the prior art can be solved, the influence of environment temperature and humidity on the concentration of the poisonous gas is considered through the multi-mode sensor, the concentration measurement precision is improved, and the concentration of the poisonous gas is accurately measured.

Description

Technical Field

[0001] The present application relates to the technical field of toxic gas monitoring, and in particular to a toxic gas dynamic monitoring system assisted by multimodal sensing. Background Art

[0002] In the context of rapid development of modern industry, the application scope of toxic and harmful gases is expanding. These gases may leak during production, storage and transportation, posing a serious threat to the safety of relevant personnel. For example, many organic chemicals (VOCs) and by-products in the production process (such as ammonia, carbon monoxide, hydrogen sulfide, etc.) contain toxic gases, which may cause immediate harm (such as physical discomfort, morbidity, death) and long-term harm (such as disability and cancer) to workers. However, there are some technical problems with existing toxic gas monitoring technology. Sensors often have insufficient sensitivity when monitoring low-concentration gases, making it difficult to accurately identify and quantify the presence of gases. Secondly, environmental conditions such as temperature and humidity can also affect the concentration measurement of toxic gases, resulting in possible deviations in the monitoring results.

[0003] In summary, the prior art has a technical problem in that the monitoring data is inaccurate due to the insufficient sensitivity of the sensor when monitoring low concentrations of toxic gases. Summary of the invention

[0004] The purpose of this application is to provide a dynamic monitoring system for toxic gases assisted by multimodal sensing, so as to solve the technical problem in the prior art that the monitoring data is inaccurate due to the insufficient sensitivity of the sensor when monitoring low concentrations of toxic gases.

[0005] In view of the above problems, the present application provides a toxic gas dynamic monitoring system assisted by multimodal perception, wherein the toxic gas dynamic monitoring system assisted by multimodal perception includes: an environmental data acquisition module for communicating with a multimodal sensor and receiving environmental perception data, wherein the environmental perception data includes environmental temperature information, environmental humidity information and first toxic gas concentration information; a deviation prediction module for processing the environmental temperature information and the environmental humidity information through a toxic gas concentration deviation prediction model to obtain a toxic gas concentration prediction deviation; a concentration correction module for correcting the first toxic gas concentration information when the toxic gas concentration prediction deviation is greater than or equal to a concentration deviation threshold, obtaining second toxic gas concentration information, and storing the second toxic gas concentration information; a data storage module for storing the first toxic gas concentration information when the toxic gas concentration prediction deviation is less than the concentration deviation threshold.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The environmental data acquisition module is used to communicate with the multimodal sensor and receive environmental perception data, wherein the environmental perception data includes environmental temperature information, environmental humidity information and first toxic gas concentration information; the deviation prediction module is used to process the environmental temperature information and the environmental humidity information through the toxic gas concentration deviation prediction model to obtain the toxic gas concentration prediction deviation; the concentration correction module is used to correct the first toxic gas concentration information when the toxic gas concentration prediction deviation is greater than or equal to the concentration deviation threshold, obtain the second toxic gas concentration information, and store the second toxic gas concentration information; the data storage module is used to store the first toxic gas concentration information when the toxic gas concentration prediction deviation is less than the concentration deviation threshold. That is to say, by using a multimodal sensor for collection, introducing a toxic gas concentration deviation prediction model, considering the influence of environmental temperature and humidity on the toxic gas concentration, predicting the concentration deviation, and correcting the deviation of the gas sensor under different environmental conditions according to the comparison result of the predicted deviation and the threshold, so as to improve the accuracy of concentration measurement and perform accurate measurement of toxic gas concentration.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0010] Figure 1 This is a schematic diagram of the structure of the multi-modal sensing-assisted toxic gas dynamic monitoring system of this application;

[0011] Figure 2 A schematic diagram of the process of receiving environmental perception data in the multimodal perception-assisted toxic gas dynamic monitoring system of this application.

[0012] Explanation of reference numerals: environmental data acquisition module 11 , deviation prediction module 12 , concentration correction module 13 , data storage module 14 . DETAILED DESCRIPTION

[0013] This application solves the technical problem in the prior art that the monitoring data is inaccurate due to the lack of sensitivity of the sensor when monitoring low-concentration toxic gases by providing a dynamic monitoring system for toxic gases assisted by multimodal sensing. By using multimodal sensors for data collection, introducing a toxic gas concentration deviation prediction model, considering the influence of ambient temperature and humidity on the concentration of toxic gases, predicting the concentration deviation, and correcting the deviation of the gas sensor under different environmental conditions based on the comparison results of the predicted deviation and the threshold, the accuracy of the concentration measurement is improved, and accurate measurement of the concentration of toxic gases is performed.

[0014] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0015] For examples, please see the attached Figure 1 The present application provides a multi-modal sensing-assisted toxic gas dynamic monitoring system, wherein the multi-modal sensing-assisted toxic gas dynamic monitoring system includes:

[0016] The environmental data acquisition module 11 is used to communicate with the multimodal sensor and receive environmental perception data, wherein the environmental perception data includes environmental temperature information, environmental humidity information and first toxic gas concentration information.

[0017] Specifically, a multimodal sensor refers to a sensor that can measure multiple environmental parameters at the same time, such as temperature, humidity, light, toxic gas concentration, etc. Multimodal sensing can be achieved through a combination of multiple sensors, such as a thermistor temperature sensor, a humidity capacitance sensor, and a gas sensor, or using a single composite sensor module. The multimodal sensor establishes a communication connection with the monitoring system through a wireless method (such as Wi-Fi, Bluetooth, ZigBee, etc.) or a wired method (such as Ethernet, serial communication, etc.) to receive data in real time. After the connection is established, the monitoring system receives a series of environmental sensing data from the multimodal sensor, including ambient temperature information, ambient humidity information, and first toxic gas concentration information (such as NO2, CO, C2H4, CH4, etc.). Multimodal sensors can also monitor toxic gases in the environment through physical or chemical methods. For example, some sensors may use electrochemical principles. When a specific gas reacts with an electrode in the sensor, a current signal is generated, thereby detecting the presence and concentration of the gas. By communicating with multimodal sensors, multi-dimensional data of the regional environment can be obtained, the gas conditions in the environment can be monitored in real time or regularly, and abnormal conditions such as toxic gas leakage or excessive concentration can be discovered in time, so that corresponding measures can be taken to protect people's health and safety.

[0018] The deviation prediction module 12 is used to process the ambient temperature information and the ambient humidity information through a toxic gas concentration deviation prediction model to obtain a toxic gas concentration prediction deviation.

[0019] Specifically, according to the specific toxic gas type and the specific multimodal sensor model, the historical monitoring data of toxic gas concentration in the past period of time is collected, including the historical record value of ambient temperature, the historical record value of ambient humidity and the monitoring deviation identification value of toxic gas concentration. The historical record value of ambient temperature and the historical record value of ambient humidity are used as input variables, and the monitoring deviation identification value of toxic gas concentration is used as the output supervision variable to determine the target variable that the model needs to learn, and train to obtain the prediction model of toxic gas concentration deviation. The collected ambient temperature and humidity data are sorted into a suitable format, including data cleaning, data standardization or normalization, so as to facilitate input into the prediction model. The processed ambient temperature information and ambient humidity information are input into the trained toxic gas concentration deviation prediction model to generate the corresponding prediction results, and the deviation of toxic gas concentration is obtained, which represents the difference between the actual monitored toxic gas concentration and the expected concentration. The deviation of toxic gas concentration is predicted by the toxic gas concentration deviation prediction model, so as to better monitor and control the toxic gas concentration in the environment.

[0020] The concentration correction module 13 is used to correct the first toxic gas concentration information when the toxic gas concentration prediction deviation is greater than or equal to a concentration deviation threshold, obtain second toxic gas concentration information, and store the second toxic gas concentration information.

[0021] Specifically, a concentration deviation threshold is set in advance to determine whether the predicted deviation is within an acceptable range. In practical applications, the concentration deviation threshold needs to be determined according to specific circumstances, such as the type and toxicity of toxic gases; specific requirements of the monitoring environment, such as industrial safety standards or environmental regulations; and the accuracy and reliability of the sensor. For example, the concentration deviation threshold is set to 5ppm, that is, when the predicted deviation exceeds 5ppm, correction is performed. When the predicted deviation of the toxic gas concentration is greater than or equal to the concentration deviation threshold, it means that the prediction result may be inaccurate or unstable and needs to be corrected. The correction method may include using a prediction model to adjust the first toxic gas concentration information according to the predicted deviation, or using other more complex correction algorithms, such as using a specific mathematical formula (for example, subtracting the predicted deviation) to correct the concentration value. After the correction is completed, the new concentration information is recorded, which is called the second toxic gas concentration information, which more accurately reflects the actual toxic gas concentration. The second toxic gas concentration information is stored in a database, data log file, or cloud storage system for subsequent calls and historical record tracking. By correcting the toxic gas concentration information, more accurate toxic gas concentration data can be obtained, so as to better monitor and control the toxic gas concentration in the environment, which helps to protect public health and safety and prevent abnormal situations such as toxic gas leakage or excessive concentration.

[0022] The data storage module 14 is used to store the first toxic gas concentration information when the toxic gas concentration prediction deviation is less than the concentration deviation threshold.

[0023] Specifically, when the predicted deviation of the toxic gas concentration is less than the concentration deviation threshold, the prediction result is considered reliable, and the current first toxic gas concentration information is stored. The stored content includes: the first toxic gas concentration information at the current moment, the relevant ambient temperature and humidity information, the predicted deviation value and the relevant timestamp, etc. The stored data is written to a database, data log file or cloud storage system to ensure the persistence and accessibility of the data. The data can be stored in a structured form for subsequent retrieval and analysis. For example, using JSON, CSV or database table format. By directly storing data with small prediction deviations, unnecessary correction steps are avoided, saving computing resources and time.

[0024] Further, as attached Figure 2 As shown, the environmental data acquisition module 11 in the multimodal sensing-assisted toxic gas dynamic monitoring system is also used for:

[0025] Communicate with a multimodal sensor and receive several groups of preset time zone environmental perception data of several locations, wherein the preset time zone is a distribution time zone of a preset time step forward from the data collection moment; the several groups of preset time zone environmental perception data include several ambient temperature time series information, several ambient humidity time series information and several toxic gas concentration time series information; perform spatiotemporal fusion on the several ambient temperature time series information to obtain the ambient temperature information; perform spatiotemporal fusion on the several ambient humidity time series information to obtain the ambient humidity information; perform spatiotemporal fusion on the several toxic gas concentration time series information to obtain the first toxic gas concentration information.

[0026] Specifically, multimodal sensors not only monitor temperature, humidity and the concentration of toxic gases, but also need to collect data from multiple locations to form cross-spatial data coverage. Collect environmental perception data for a specific time period at multiple locations, that is, collect environmental perception data for a specific time step before the data collection time. For example, if the preset time step is 1 hour, then the preset time zone is a time period of 1 hour from the data collection time. The time step refers to the time interval for data collection within the preset time zone. The preset time step is a pre-set time interval, which can be set according to actual conditions or according to monitoring requirements. The collected sets of preset time zone environmental perception data include several sets of ambient temperature time series information, several sets of ambient humidity time series information and several sets of toxic gas concentration time series information, that is, the ambient temperature time series, ambient humidity time series and toxic gas concentration time series of multiple locations. The time series information refers to the environmental parameter data sequence that changes over time.

[0027] Spatiotemporal fusion is the core step of data processing. By jointly analyzing data from multiple locations and times, a spatiotemporal feature set is obtained. Based on the data fusion algorithm at different locations and times, the overall temperature, humidity and gas concentration information is generated. Taking several ambient temperature time series information as an example, outlier detection is performed for the ambient temperature data collected at each location in the region within the preset time zone, and possible abnormal data, that is, those data points that obviously deviate from the normal distribution, are eliminated. After eliminating the outliers, the central trend value (such as the mean) of the temperature data at each location is calculated to obtain the ambient temperature time domain fusion value at each location. The time domain fusion values ​​of all locations are aggregated together to form a data set, which represents the temperature distribution of the entire region in the current time period. The time domain fusion values ​​of these locations are also analyzed for outliers to eliminate possible extreme location temperatures, such as abnormally high or low temperatures at local special locations. The mean of the remaining time domain fusion values ​​is calculated to obtain the ambient temperature information of the entire region. The final calculation result is the representative temperature level of the current region. After spatiotemporal fusion, the abnormal data in local time periods and spaces are excluded, so it is more representative.

[0028] The execution process of environmental humidity information and first toxic gas concentration information is similar to that of environmental temperature information, so we will not elaborate on it here. By performing spatiotemporal fusion processing on each type of environmental data (temperature, humidity, concentration) one by one, the stability and representativeness of each data is ensured. After spatiotemporal fusion, not only the influence of extreme values ​​in individual locations or time periods is eliminated, but also the reliability of the overall data is enhanced, which is convenient for the subsequent comprehensive assessment and monitoring of the regional environment.

[0029] Furthermore, the environmental data acquisition module 11 in the multimodal sensing-assisted toxic gas dynamic monitoring system is also used for:

[0030] Performing a central tendency analysis on the ambient temperature time series information at the first position of the plurality of ambient temperature time series information to obtain a time-domain fusion value of the ambient temperature at the first position; performing a central tendency analysis on the ambient temperature time series information at the Nth position of the plurality of ambient temperature time series information to obtain a time-domain fusion value of the ambient temperature at the Nth position; performing a central tendency analysis on the ambient temperature time-domain fusion value at the first position until the ambient temperature time-domain fusion value at the Nth position to obtain the ambient temperature information.

[0031] Specifically, the ambient temperature time series information of the first position is arbitrarily extracted from a number of ambient temperature time series information, that is, the ambient temperature data collected at the first position in the preset time zone is subjected to central trend analysis. Central trend analysis is used to extract representative values ​​from time series data to eliminate the impact of short-term fluctuations on the overall results. Common methods include: calculating the average temperature of a certain location in a specific time period to obtain a representative value; reflecting the central trend by calculating the median, especially when there are extreme values, it can better represent the overall temperature level; if the time series data has an obvious central value, the mode can also be used to reflect the temperature level of the location.

[0032] For the time series information of the ambient temperature at the first location, a central trend analysis is performed. First, the abnormal temperature data is eliminated, that is, the values ​​that are significantly deviated from the majority of the data are filtered out. Then, the mean of the non-outlier temperature data is calculated to obtain the time domain fusion value of the ambient temperature at the first location. In data analysis, outliers are abnormal temperature data caused by sensor failure, extreme environmental events or other accidental factors, which may have an adverse effect on the results, so they usually need to be identified and removed.

[0033] Similarly, the above calculation is performed for the Nth position ambient temperature time series information of several ambient temperature time series information, that is, the central tendency analysis. After removing the abnormal temperature data, the mean of the non-outlier temperature data is counted to obtain the time-domain fusion value of the ambient temperature at multiple positions. When the time-domain fusion value of each position has been obtained, these values ​​are subjected to central tendency analysis, and the overall aggregation of the temperature levels at different positions is performed to generate a regional-level ambient temperature information. The method of removing outliers can be used again to detect outliers on the time-domain fusion values ​​from the first position to the Nth position to remove possible abnormal position data. These outliers may be caused by special environmental conditions or sensor abnormalities at certain locations and do not represent the overall temperature trend. After removing the outliers, the remaining data is averaged or the median is used as the overall ambient temperature information. Through the two-level central tendency analysis, the temperature data of each position is firstly removed from abnormalities and the mean is obtained, and then the temperature values ​​of extreme positions are removed at the regional level and the average is calculated to ensure that the obtained ambient temperature information can reflect the actual temperature conditions of most positions and avoid the interference of local abnormal data.

[0034] Furthermore, the environmental data acquisition module 11 in the multimodal sensing-assisted toxic gas dynamic monitoring system is also used for:

[0035] An outlier analysis is performed on the time series information of the ambient temperature at the first position to obtain the ambient temperature at a non-outlier moment of the first position; and the mean value of the ambient temperature at the non-outlier moment of the first position is calculated to obtain a time domain fusion value of the ambient temperature at the first position.

[0036] Specifically, outliers refer to values ​​that deviate significantly from most data, which often lead to a decrease in the representativeness of the overall temperature data. In the ambient temperature time series data, outliers may be abnormal temperature data caused by sensor failure, extreme environmental events, or other accidental factors. Outlier analysis is a statistical method used to identify abnormal values ​​or outliers in a data set. Common statistical methods include z-score analysis and IQR (interquartile range) method, which can identify data points that are statistically significantly deviated. The main method used here is to identify outliers through cluster analysis. Specifically, the ambient temperature data is clustered according to the temperature deviation threshold, and data points with similar temperatures are divided into the same cluster. After clustering, the number of data points in each cluster is checked, and clusters with too little data are eliminated. If the number of ambient temperature data points in a cluster is less than or equal to the ambient temperature number threshold, then this cluster will be considered unrepresentative and needs to be deleted. After completing the outlier detection, the outlier data is removed from the time series, and the remaining first position non-outlier moment ambient temperature is retained.

[0037] For all the remaining non-outlier temperature data, the statistical mean is the result of adding up the number of data of the ambient temperature at all non-outlier moments of the first position. It is a common statistic to measure the trend of the data set. For example, assuming that the ambient temperature data at the non-outlier moment of the first position are as follows: 20℃, 22℃, 21℃, 23℃, 21℃, the mean is 21.4℃. After calculating the mean, the time-domain fusion value of the ambient temperature at the first position is obtained, which represents the representative temperature level of the ambient temperature at the first position in the preset time zone. The mean after eliminating outlier data can reduce the impact of abnormal fluctuations, making the time-domain fusion value more accurate and stable. Through outlier analysis and non-outlier data mean calculation, the abnormal fluctuations in the temperature data are effectively filtered out, and a more representative temperature time-domain fusion value for each position is obtained, which improves the accuracy and reliability of the data.

[0038] Furthermore, the environmental data acquisition module 11 in the multimodal sensing-assisted toxic gas dynamic monitoring system is also used for:

[0039] According to the temperature deviation threshold, the ambient temperatures at multiple moments of the ambient temperature time series information of the first position are clustered to obtain multiple clusters of ambient temperatures; clusters in which the number of ambient temperatures is less than or equal to the ambient temperature number threshold are deleted to obtain the ambient temperature at the non-outlier moment of the first position.

[0040] Specifically, the temperature deviation threshold refers to the allowable temperature fluctuation range, which is used to judge the similarity between different temperature data. It is usually a preset standard used to judge whether the difference between temperature data is large enough that it should be divided into different categories. For example, if the deviation threshold is set to 1°C, then any two temperature readings with a difference of more than 1°C will be considered large enough to be divided into different clusters, and temperature data with a difference of no more than 1°C will be considered to belong to the same cluster. According to the temperature deviation threshold, the temperature data at multiple moments of the first position ambient temperature time series information are clustered to form multiple clusters. Clustering is an unsupervised learning method that groups data points into multiple clusters so that the similarity of data points in the same cluster is high, while the similarity of data points in different clusters is low.

[0041] According to the characteristics of the data and the purpose of clustering, choose a suitable clustering algorithm, such as K-means, hierarchical clustering, DBSCAN, etc. If it is known that the data set should be divided into a specific number of clusters, use the K-means algorithm, pre-specify the number of clusters, and assign the data point to the cluster with the closest cluster center; hierarchical clustering gradually merges temperature values ​​with high similarity into clusters until the temperature difference between all clusters exceeds the deviation threshold; DBSCAN can cluster close temperature data into clusters according to density, which works well when detecting outliers because outliers are often distributed in areas with low density. In cluster analysis, the algorithm groups data based on the characteristics and similarity of the data itself, rather than relying on known labels or results. Assign each data point in the data set to a cluster. The number of clusters can be pre-specified or automatically determined by the algorithm. Each cluster contains a group of similar data points that have a higher similarity than the data points in other clusters under a certain metric.

[0042] After clustering, the first position ambient temperature time series information is divided into several clusters, each of which contains multiple temperature values ​​that are close to each other in temperature. The ambient temperature quantity threshold refers to the minimum amount of data required in each cluster, which is used to determine whether the number of data points in a cluster is large enough so that the cluster is considered representative. Clusters with less than this threshold number will be considered unrepresentative or abnormal and will be eliminated. The significance of setting the quantity threshold is that the natural law of temperature change usually forms clusters with larger data volumes, while outliers often form small clusters or individual points. Check the number of ambient temperature data points in each cluster, and delete those clusters whose number is less than or equal to the ambient temperature quantity threshold, and remove those small clusters that may be caused by noise or outliers, so as to obtain a cleaner and more reliable data set.

[0043] After removing small clusters, the ambient temperature data of the remaining clusters are considered to be the ambient temperature at non-outlier moments, representing the actual temperature fluctuations at the location, and excluding outliers caused by equipment failures or transient changes in the environment. They usually come from larger clusters and represent the main temperature patterns or trends in the environment. Outlier detection based on cluster analysis is more flexible than traditional methods (such as simple upper and lower limit settings) because it can adaptively identify multiple concentrated areas instead of simply removing extreme values. It is suitable for variable environmental conditions, can identify multiple temperature concentration trends, and can be screened according to the amount of data, so that the retained temperature data can better reflect the actual situation. Through temperature deviation threshold clustering and small cluster removal methods, the temperature data is screened and optimized, and the processed ambient temperature at non-outlier moments is more stable and reliable, and can more accurately reflect the actual ambient temperature situation.

[0044] Furthermore, the deviation prediction module 12 in the multimodal sensing-assisted toxic gas dynamic monitoring system is also used for:

[0045] Taking the toxic gas type and the multimodal sensor model as constraints, historical monitoring data of toxic gas concentration is collected, wherein the historical monitoring data of toxic gas concentration includes historical record values ​​of ambient temperature, historical record values ​​of ambient humidity and monitoring deviation identification values ​​of toxic gas concentration; the historical record values ​​of ambient temperature and the historical record values ​​of ambient humidity are used as input variables, and the toxic gas concentration monitoring deviation identification value is used as an output supervision variable to train the toxic gas concentration deviation prediction model.

[0046] Specifically, a specific type of multimodal sensor is used to monitor a specific type of toxic gas. Different sensor models may have different monitoring sensitivities and accuracies for different types of toxic gases. For example, semiconductor gas sensors are suitable for gas monitoring such as CH4, CO, H2S and other complex gases. The objects to be detected run fast and have high sensitivity, but generally all kinds of gas monitoring need to be monitored at a high temperature of 200 degrees to operate normally; electrochemical gas sensors are suitable for NO2, NO, SO2 and other gases. They have good temperature selectivity and high sensitivity, but the service life of gas sensors is short.

[0047] When collecting historical monitoring data, only specific types of toxic gases and specific models of sensors are considered to ensure the pertinence of the data and the accuracy of the model. Collect historical monitoring data on toxic gas concentrations over a period of time, including historical record values ​​of ambient temperature, historical record values ​​of ambient humidity, and monitoring deviation identification values ​​of toxic gas concentrations. The historical record value of ambient temperature refers to the historical temperature data during the monitoring period. The historical record value of ambient humidity refers to the historical humidity data during the monitoring period. The monitoring deviation identification value of toxic gas concentration refers to the deviation of toxic gas concentration caused by changes in temperature or humidity in historical monitoring, that is, the deviation between the toxic gas concentration and the actual concentration, which indicates the difference between the monitored value and the true value.

[0048] The historical record values ​​of ambient temperature and ambient humidity are used as input variables, and the toxic gas concentration monitoring deviation identification value is used as the output supervision variable, that is, the target variable that the model needs to predict, indicating the deviation between the actual monitored toxic gas concentration and the expected concentration. The model is trained using the input variables and the output supervision variables to obtain the first toxic gas concentration deviation prediction model. After that, the data set whose output accuracy is lower than the threshold in the model is trained again, that is, the part of the model prediction that is inaccurate is refined, and the third, fourth, and Mth toxic gas concentration deviation prediction models are constructed in sequence, until the number of samples in the M+1th model construction data set is less than the data volume threshold. The first toxic gas concentration deviation prediction model, the second toxic gas concentration deviation prediction model, and the Mth toxic gas concentration deviation prediction model are fused to obtain a toxic gas concentration deviation prediction model. Using historical temperature and humidity data and concentration deviation identification data, a concentration deviation prediction model is trained to estimate the error of toxic gas concentration monitoring according to the changes in temperature and humidity in real-time monitoring, so as to monitor the concentration more accurately.

[0049] Furthermore, the deviation prediction module 12 in the multimodal sensing-assisted toxic gas dynamic monitoring system is also used for:

[0050] A first model construction data set is constructed with the ambient temperature historical record value and the ambient humidity historical record value as input variables and the toxic gas concentration monitoring deviation identification value as output supervision variable; a first toxic gas concentration deviation prediction model is trained according to the first model construction data set; a second model construction data set whose output accuracy is less than or equal to an output accuracy threshold is extracted from the first toxic gas concentration deviation prediction model; when the second model construction data set is greater than or equal to a data volume threshold, a second toxic gas concentration deviation prediction model is trained according to the second model construction data set; until the M+1th model construction data set is less than the data volume threshold, the first toxic gas concentration deviation prediction model, the second toxic gas concentration deviation prediction model, and the Mth toxic gas concentration deviation prediction model are merged to obtain the toxic gas concentration deviation prediction model.

[0051] Specifically, the ambient temperature history record value and the ambient humidity history record value are input variables, which are generally considered to be factors that affect the toxic gas concentration monitoring results. The toxic gas concentration monitoring deviation identification value is the output supervision variable of the model, which is the target that the model is expected to predict or identify. By comparing the actual monitored toxic gas concentration with the standard value or expected value, the monitoring deviation identification value can be obtained, which is used to evaluate the accuracy and reliability of the sensor. The input variables and the output supervision variables are combined to form a data set for training the first model, which should contain enough samples to ensure that the model can learn the relationship between the input variables and the output supervision variables.

[0052] Data cleaning is performed on the first model building data set to remove missing values ​​and outliers to ensure the accuracy of model training. According to the size and complexity of the data set, a suitable machine learning algorithm is selected. Linear regression is selected for simple and easy implementation, decision tree regression is selected for more complex nonlinear relationships, and neural network is selected for large-scale data and complex nonlinear relationships. According to the selected algorithm, the initial parameter settings of the model are determined, such as the learning rate, the number of layers and the number of nodes (for neural networks), or the depth of the tree (for decision trees). For example, a neural network model is selected for training. According to the number of input features, the number of nodes in the input layer is set (such as 2 input nodes corresponding to temperature and humidity respectively); one or more hidden layers are designed to determine the number of nodes and activation functions; an output layer is set to output the predicted value of the toxic gas concentration deviation. The data set is divided into a training set and a validation set, such as 80% for training and 20% for validation. Use the training set to train the model so that the model can output the corresponding deviation value when the environmental temperature and humidity information is input. According to the loss function value of the model during the training process (such as the mean square error), the model parameters are adjusted to better fit the data. Test the output of the model on the validation set, calculate the difference between the predicted value and the actual deviation value, and use the mean square error (MSE) or mean absolute error (MAE) to evaluate the accuracy of the model. The mean square error refers to the average of the squares of the differences between the predicted value and the actual value. It has a higher penalty effect on large errors (deviations) and is suitable for use when the deviation has a greater impact on the prediction results. The mean absolute error refers to the average of the absolute values ​​of the differences between the predicted value and the actual value. It treats all deviations linearly and is suitable for treating all prediction errors equally. Set an accuracy threshold, such as 95%. After training is completed, evaluate the output accuracy of the model, that is, the accuracy of the predicted deviation value. If the accuracy is lower than the accuracy threshold, it means that the model may not be able to accurately predict the deviation under certain conditions.

[0053] Extract samples with output accuracy less than or equal to the output accuracy threshold from the first toxic gas concentration deviation prediction model to construct a new data set representing environmental conditions that are difficult for the first model to accurately predict, i.e., the second model construction data set. Check whether the sample size of the second model construction data set reaches the specified data volume threshold. If the data volume is insufficient, stop further model construction and enter the model fusion stage. If the second model construction data set meets the data volume requirement, use the data set to train a new deviation prediction model, i.e., the second toxic gas concentration deviation prediction model, focusing on the data that the first model performs poorly in the prediction, and improving the prediction ability under these environmental conditions. The above process is repeated, and each time the inaccurate part of the previous model prediction is refined, and the third, fourth, and up to the Mth toxic gas concentration deviation prediction models are constructed in turn, until the number of samples in the M+1th model construction data set is less than the data volume threshold. Through this iterative process, each new model specifically handles samples that cannot be accurately predicted in the previous model, and continuously improves the accuracy of the overall prediction system under complex environmental conditions.

[0054] The first toxic gas concentration deviation prediction model, the second toxic gas concentration deviation prediction model, and the Mth toxic gas concentration deviation prediction model are fused, and different weights are assigned according to the performance of each model. The prediction results of each model are multiplied by the corresponding weights and averaged to obtain the toxic gas concentration deviation prediction model. Model fusion is a technology that combines the prediction results of multiple models to generate a more accurate prediction result. The prediction performance is improved by dynamic adjustment and multi-model fusion, ensuring the adaptability and reliability of the model under different conditions. The final fusion model can combine the advantages of each model to improve the prediction accuracy of toxic gas concentration deviation.

[0055] Furthermore, the deviation prediction module 12 in the multimodal sensing-assisted toxic gas dynamic monitoring system is also used for:

[0056] Add the first model construction data set, the second model construction data set, up to the Mth model construction data set to obtain a first data volume; calculate the data volume ratio of the first model construction data set, the second model construction data set, up to the Mth model construction data set, and the first data volume respectively to obtain a first weight, a second weight, up to the Mth weight; according to the first weight, the second weight, up to the Mth weight, perform weighted mean full connection on the first toxic gas concentration deviation prediction model, the second toxic gas concentration deviation prediction model, up to the Mth toxic gas concentration deviation prediction model, to obtain the toxic gas concentration deviation prediction model.

[0057] Specifically, the first model construction data set, the second model construction data set, and the M-th model construction data set are added together to obtain a total data volume, namely, the first data volume. For each model construction data set, the ratio of its data volume to the total data volume is calculated as the weight of the model in the fusion process, and the first weight, the second weight, and the M-th weight are obtained. The output of each model is combined using the weighted average method, namely, the output result of the first toxic gas concentration deviation prediction model, the second toxic gas concentration deviation prediction model, and the M-th toxic gas concentration deviation prediction model. The final prediction result is obtained by multiplying the prediction result of each model by its corresponding weight and then taking the weighted average. Make full use of the prediction ability of each model while taking into account the importance or contribution of each model. The toxic gas concentration deviation prediction model is obtained by weighted mean full connection. Weighted mean full connection is a commonly used operation in machine learning and deep learning, especially in neural networks, where multiple inputs are combined into one output by weighted summation. Multiply the output value of each model by its corresponding weight, sum all the weighted model output values, get a sum, and normalize the summed result to ensure the rationality and interpretability of the output result. A comprehensive prediction model is obtained by weighted averaging the outputs of multiple models. Use the data set size of each model as a weight to ensure that the model with a larger data volume has a larger proportion in the final prediction.

[0058] In summary, the multimodal sensing-assisted toxic gas dynamic monitoring system provided by this application has the following technical effects:

[0059] The environmental data acquisition module is used to communicate with the multimodal sensor and receive environmental perception data, wherein the environmental perception data includes environmental temperature information, environmental humidity information and first toxic gas concentration information; the deviation prediction module is used to process the environmental temperature information and the environmental humidity information through the toxic gas concentration deviation prediction model to obtain the toxic gas concentration prediction deviation; the concentration correction module is used to correct the first toxic gas concentration information when the toxic gas concentration prediction deviation is greater than or equal to the concentration deviation threshold, obtain the second toxic gas concentration information, and store the second toxic gas concentration information; the data storage module is used to store the first toxic gas concentration information when the toxic gas concentration prediction deviation is less than the concentration deviation threshold. That is to say, by using a multimodal sensor for collection, introducing a toxic gas concentration deviation prediction model, considering the influence of environmental temperature and humidity on the toxic gas concentration, predicting the concentration deviation, and correcting the deviation of the gas sensor under different environmental conditions according to the comparison result of the predicted deviation and the threshold, so as to improve the accuracy of concentration measurement and perform accurate measurement of toxic gas concentration.

[0060] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0061] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A multi-modal sensing-assisted toxic gas dynamic monitoring system, characterized in that: include: The environmental data acquisition module is used to communicate with the multimodal sensor and receive environmental perception data, wherein the environmental perception data includes environmental temperature information, environmental humidity information and first toxic gas concentration information; The deviation prediction module is used to process the ambient temperature information and the ambient humidity information through a toxic gas concentration deviation prediction model to obtain a toxic gas concentration prediction deviation; The concentration correction module is used for correcting the first toxic gas concentration information when the toxic gas concentration prediction deviation is greater than or equal to the concentration deviation threshold, obtaining second toxic gas concentration information, and storing the second toxic gas concentration information; The data storage module is used to store the first toxic gas concentration information when the toxic gas concentration prediction deviation is less than the concentration deviation threshold.

2. The multi-modal sensing-assisted toxic gas dynamic monitoring system according to claim 1, characterized in that: Communicate with multimodal sensors and receive environmental perception data, including: Communicate with the multimodal sensor and receive several groups of preset time zone environmental perception data of several locations, wherein the preset time zone is a distribution time zone of a preset time step forward from the data collection moment; The plurality of groups of preset time zone environment sensing data include a plurality of environmental temperature time series information, a plurality of environmental humidity time series information and a plurality of toxic gas concentration time series information; Performing spatiotemporal fusion on the plurality of ambient temperature time series information to obtain the ambient temperature information; Performing spatiotemporal fusion on the plurality of environmental humidity time series information to obtain the environmental humidity information; The plurality of toxic gas concentration time series information are temporally and spatially fused to obtain the first toxic gas concentration information.

3. The multi-modal sensing-assisted toxic gas dynamic monitoring system according to claim 2, characterized in that: Performing spatiotemporal fusion on the plurality of ambient temperature time series information to obtain the ambient temperature information includes: Performing a central trend analysis on the first position ambient temperature time series information of the plurality of ambient temperature time series information to obtain a time domain fusion value of the first position ambient temperature; Until a central trend analysis is performed on the Nth position ambient temperature time series information of the plurality of ambient temperature time series information to obtain a time domain fusion value of the Nth position ambient temperature; Central trend analysis is performed on the time-domain fusion value of the ambient temperature at the first position to the time-domain fusion value of the ambient temperature at the Nth position to obtain the ambient temperature information.

4. The multi-modal sensing-assisted toxic gas dynamic monitoring system according to claim 1, characterized in that: The toxic gas concentration deviation prediction model construction step includes: Taking the toxic gas type and the multimodal sensor model as constraints, collect the historical monitoring data of the toxic gas concentration, wherein the historical monitoring data of the toxic gas concentration includes the historical record value of the ambient temperature, the historical record value of the ambient humidity and the toxic gas concentration monitoring deviation identification value; The environmental temperature historical record value and the environmental humidity historical record value are used as input variables, and the toxic gas concentration monitoring deviation identification value is used as an output supervision variable to train the toxic gas concentration deviation prediction model.

5. The multi-modal sensing-assisted toxic gas dynamic monitoring system according to claim 4, characterized in that: The environmental temperature historical record value and the environmental humidity historical record value are used as input variables, and the toxic gas concentration monitoring deviation identification value is used as an output supervisory variable to train the toxic gas concentration deviation prediction model, including: The environmental temperature historical record value and the environmental humidity historical record value are used as input variables, and the toxic gas concentration monitoring deviation identification value is used as an output supervision variable to construct a first model construction data set; Building a data set based on the first model to train a first toxic gas concentration deviation prediction model; Extracting a second model construction data set whose output accuracy is less than or equal to an output accuracy threshold value in the first toxic gas concentration deviation prediction model; When the second model construction data set is greater than or equal to the data amount threshold, training a second toxic gas concentration deviation prediction model according to the second model construction data set; Until the M+1th model construction data set is less than the data volume threshold, the first toxic gas concentration deviation prediction model, the second toxic gas concentration deviation prediction model, and the Mth toxic gas concentration deviation prediction model are integrated to obtain the toxic gas concentration deviation prediction model.

6. The multi-modal sensing-assisted toxic gas dynamic monitoring system according to claim 5, characterized in that: The first toxic gas concentration deviation prediction model, the second toxic gas concentration deviation prediction model, and the Mth toxic gas concentration deviation prediction model are integrated to obtain the toxic gas concentration deviation prediction model, including: Adding the first model building data set, the second model building data set, and up to the Mth model building data set to obtain a first data volume; Respectively calculate the data volume ratios of the first model construction data set, the second model construction data set, up to the M-th model construction data set, and the first data volume to obtain a first weight, a second weight, up to the M-th weight; According to the first weight, the second weight up to the Mth weight, weighted mean full connection is performed on the first toxic gas concentration deviation prediction model, the second toxic gas concentration deviation prediction model up to the Mth toxic gas concentration deviation prediction model to obtain the toxic gas concentration deviation prediction model.

7. The multi-modal sensing-assisted toxic gas dynamic monitoring system according to claim 3, characterized in that: Performing a central trend analysis on the first position ambient temperature time series information of the plurality of ambient temperature time series information to obtain a time domain fusion value of the first position ambient temperature includes: Performing outlier analysis on the time series information of the ambient temperature at the first position to obtain the ambient temperature at a non-outlier moment of the first position; The average value of the ambient temperature at the non-outlier time of the first position is calculated to obtain a time-domain fusion value of the ambient temperature at the first position.

8. The multi-modal sensing-assisted toxic gas dynamic monitoring system according to claim 7, characterized in that: Performing outlier analysis on the time series information of the ambient temperature at the first position to obtain the ambient temperature at a non-outlier time of the first position includes: Clustering the ambient temperatures at multiple moments of the ambient temperature time series information of the first position according to the temperature deviation threshold to obtain multiple clusters of ambient temperatures; The clusters whose number of ambient temperatures is less than or equal to the ambient temperature number threshold are deleted to obtain the ambient temperature of the first position at the non-outlier time.

Images