Environmental risk monitoring platform based on multi-modal information fusion

Abstract

The invention discloses an environmental risk monitoring platform based on multi-modal information fusion, and relates to the technical field of environmental risk monitoring, and the platform comprises an air parameter sequence obtaining module which carries out the continuous monitoring, and obtains a plurality of air parameter sequences; the circulation influence coefficient array acquisition module is used for performing air circulation influence analysis to obtain a circulation influence coefficient array; the air quality risk parameter acquisition module is used for carrying out air quality change risk prediction and carrying out influence correction on a plurality of air quality risk parameters; and the judgment and early warning module is used for carrying out judgment and early warning according to the multiple adjusted air quality risk parameters. The technical problems that in the prior art, environment risk monitoring is not comprehensive and inaccurate, and early warning lags are solved, and the technical effects that by means of multi-module cooperation, distributed sensors are utilized, the personnel density is analyzed, and all-directional dynamic accurate monitoring and risk early warning of the air quality of the target space are achieved, and environmental safety is guaranteed are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental risk monitoring, and particularly to an environmental risk monitoring platform based on multi-modal information fusion. Background Art

[0002] With the rapid development of industrialization and urbanization, environmental risk monitoring has fallen into numerous dilemmas. Factories discharge pollutants wantonly, vehicle exhaust fills the air, and with the high concentration of urban population and dense buildings with poor ventilation, the air quality has deteriorated sharply, bringing many obstacles to the monitoring work. The traditional monitoring means rely on few and scattered fixed stations for air quality monitoring, and can only measure a small number of indicators at fixed intervals, unable to accurately capture the air quality everywhere in the target area, and monitoring gaps can be seen everywhere. Once a sudden pollution source or meteorological change occurs, due to the fixed detection interval, it is difficult to detect the air quality fluctuation in time, and the early warning is seriously lagged. Particularly crucial is that when considering the influencing factors of air quality comprehensively, in the past, the assessment of air quality risk only looked at the pollutant concentration. For factors such as population density, in places like shopping malls and office buildings, the activities of people change the air composition, hinder the circulation, and exacerbate the pollution, but the traditional methods cannot quantitatively process these changes, resulting in a huge deviation between the assessment result and the actual situation, and it is difficult to meet the urgent need for high-precision environmental risk monitoring at present.

[0003] The prior art has technical problems such as incomplete, inaccurate environmental risk monitoring and lagged early warning. Summary of the Invention

[0004] The present application provides an environmental risk monitoring platform based on multi-modal information fusion, which is used to solve the technical problems of incomplete, inaccurate environmental risk monitoring and lagged early warning in the prior art.

[0005] In view of the above problems, the present application provides an environmental risk monitoring platform based on multi-modal information fusion.

[0006] The present application provides an environmental risk monitoring platform based on multi-modal information fusion, and the platform includes:

[0007] An air parameter sequence acquisition module, configured to continuously monitor and acquire air parameters at multiple spatial positions through distributed sensors deployed in a target space, and obtain multiple air parameter sequences; a circulation influence coefficient array acquisition module, configured to acquire a personnel density array at the multiple spatial positions, perform air circulation influence analysis, and obtain a circulation influence coefficient array; an air quality risk parameter acquisition module, configured to perform air quality change risk prediction based on the multiple air parameter sequences, obtain multiple air quality risk parameters, and perform influence correction on the multiple air quality risk parameters by using multiple circulation influence coefficients in the circulation influence coefficient array to obtain multiple adjusted air quality risk parameters; a discrimination and early warning module, configured to generate comprehensive environmental risk parameters based on the multiple adjusted air quality risk parameters and in combination with the personnel density array, and perform discrimination and early warning.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] The air parameter sequence acquisition module continuously monitors and acquires air parameters at multiple spatial positions to obtain multiple air parameter sequences; the circulation influence coefficient array acquisition module performs air circulation influence analysis to obtain a circulation influence coefficient array; the air quality risk parameter acquisition module performs air quality change risk prediction to obtain multiple air quality risk parameters, and performs influence correction on the multiple air quality risk parameters to obtain multiple adjusted air quality risk parameters; the discrimination and early warning module is configured to generate comprehensive environmental risk parameters based on the multiple adjusted air quality risk parameters and perform discrimination and early warning. It achieves the technical effect of realizing all-round dynamic and accurate monitoring and risk warning of the air quality in the target space by means of multi-module collaboration, using distributed sensors and analyzing the personnel density, so as to ensure environmental safety. Description of the Drawings

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0011] Figure 1 It is a schematic structural diagram of an environmental risk monitoring platform based on multi-modal information fusion provided by an embodiment of this application;

[0012] Figure 2 It is a schematic flow diagram of an air parameter sequence acquisition module in an environmental risk monitoring platform based on multi-modal information fusion provided by an embodiment of this application.

[0013] Description of reference numerals: Air parameter sequence acquisition module 10, flow influence coefficient array acquisition module 20, air quality risk parameter acquisition module 30, discrimination and early warning module 40. Detailed implementation manner

[0014] This application provides an environmental risk monitoring platform based on multi-modal information fusion to solve the technical problems of incomplete, inaccurate environmental risk monitoring and lagging early warning in the prior art.

[0015] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0016] Embodiment, as Figure 1 shown, this application provides an environmental risk monitoring platform based on multi-modal information fusion, and the platform includes:

[0017] An air parameter sequence acquisition module 10, configured to continuously monitor and acquire air parameters at multiple spatial positions through distributed sensors arranged in a target space, and obtain multiple air parameter sequences.

[0018] Specifically, the air parameter sequence acquisition module conducts comprehensive and continuous monitoring by means of a distributed sensor network deployed in the target space. When deploying a distributed sensor network in the target space, multiple factors need to be considered comprehensively. First, it is planned according to the physical structure of the target space. For indoor spaces such as factories and office buildings, sensor points are arranged based on the room layout, ventilation system distribution, and areas with frequent personnel activities. In large factories, sensors are densely arranged along production lines, personnel operation areas, and locations close to possible pollution sources (such as chemical raw material storage areas and equipment exhaust outlets) to ensure that air quality changes can be captured in a timely manner. For different spatial functional areas such as production workshops, warehouses, and office areas, the distribution density of sensors is set differently according to their sensitivity to air quality and potential risk levels. For example, since there are more air pollutants emitted related to production processes in the production workshop, the sensors are more densely distributed to accurately monitor air quality fluctuations; while in the office area, the sensors are relatively evenly distributed, focusing on air parameters related to personnel comfort (such as temperature, humidity, carbon dioxide concentration, etc.). At the same time, the ventilation characteristics of the space are fully considered, and sensors are reasonably placed at key positions such as ventilation openings, return air vents, and corners where air circulation is relatively poor. The sensors near the ventilation openings are used to monitor the impact of outside air entering on indoor air quality, and the sensors at the return air vents help to evaluate the quality change after indoor air circulation. The sensors at the corner positions can promptly detect the accumulation of pollutants that may occur due to poor air circulation. In addition, according to the spatial scale of the target space, the coverage range of the sensors is reasonably determined to ensure that all areas can be effectively monitored and avoid monitoring blind spots. For larger spaces, a hierarchical and zonal layout method needs to be adopted to improve the accuracy and comprehensiveness of monitoring. Through such meticulous layout planning, the distributed sensor network can comprehensively and accurately obtain the air parameter information at each position in the target space. The sensors are distributed at multiple key spatial positions and can capture various parameters in the air environment in real time. During the long-term operation process, the air parameter information of each spatial position is accurately obtained at multiple different time points. These parameters cover important indicators such as harmful gas concentration. Over time, rich parameter data is accumulated for each spatial position. By integrating the parameter data of the same spatial position at different times, multiple air parameter sets are formed. Each air parameter set completely records the air parameter status of a specific spatial position at different time nodes. Further, these air parameter sets are arranged in order of time sequence, and finally multiple air parameter sequences are successfully obtained, providing basic data support for subsequent in-depth analysis of the dynamic evolution of air quality.

[0019] The air circulation influence coefficient array acquisition module 20 is used to obtain the personnel density array of the multiple spatial positions, conduct air circulation influence analysis, and obtain the air circulation influence coefficient array.

[0020] Specifically, the main function of the circulation influence coefficient array acquisition module 20 is to acquire the personnel density array and conduct air circulation influence analysis, thereby obtaining the circulation influence coefficient array. This module first constructs the personnel density array, including a variety of data collection methods, and the specific approach is as follows: The variety of data collection methods include a personnel counting system based on video image analysis. The high-definition cameras therein cover various areas. By using deep learning object detection algorithms, it can identify individual personnel in the picture in real time and accurately, and continuously count the change in the number of personnel at each spatial position through tracking algorithms. At the same time, infrared induction counters are carefully installed at each entrance and exit of the space. Relying on the induction characteristics of infrared rays on the human body, whenever a person enters or exits, the counter can quickly and accurately record the personnel flow information, providing reliable supplementary data for calculating the real-time number of personnel in each area. In addition, Wi-Fi probe technology is also fully utilized. By monitoring the signal strength fluctuation of devices connecting to Wi-Fi in the space and the change in the number of connected devices, and with the help of complex signal processing algorithms, it can effectively calculate the personnel distribution in different areas, further enriching and improving the collection of personnel quantity information, obtaining the personnel quantity information at multiple spatial positions in the target space, and at the same time combining the spatial area data of each spatial position. The personnel quantity and the spatial area are comprehensively calculated. For each spatial position, such as spatial position A, when the personnel quantity obtained is n A , and the corresponding spatial area is S A At this time, strictly in accordance with the personnel density calculation formula ρ A , the personnel density value is calculated

[0021] The personnel density array presents the density of personnel distribution in the target space in an intuitive form. For example, in areas with a high density of people, the personnel density value is relatively high; while in areas with a low density of people, the personnel density value is relatively low. After the construction of the personnel density array is completed, the module enters the stage of analyzing the impact of air circulation, which involves pre-training the air circulation impact classification channel. At the beginning of the training, a set of sample personnel densities is collected from the rich air circulation test data accumulated over historical time. For scenarios with different sample personnel densities, with the help of high-precision air velocity measuring instruments, such as advanced hot-wire anemometers and laser Doppler anemometers, the proportion of the reduction in air circulation velocity under different personnel density conditions is measured, and these measurement results are accurately labeled as a set of sample circulation impact coefficients. Subsequently, using the set of sample personnel densities and the set of sample circulation impact coefficients as supervised training data, the multi-layer perceptron (MLP) neural network algorithm is used to construct the air circulation impact classification channel. During the training process, the algorithm continuously adjusts the parameters inside the channel. Through multiple iterations of optimization, the channel gradually learns the complex internal relationship between the personnel density and the proportion of the reduction in air circulation velocity. This process continues until the model reaches a convergence state, that is, the error between the predicted result of the model and the actual measurement result reaches the minimum. At this time, a high-precision air circulation impact classification channel that has completed training is obtained. Finally, multiple personnel density data in the constructed personnel density array are sequentially input into the trained air circulation impact classification channel. Based on the complex relationships and rules it has learned in advance, the channel outputs the corresponding circulation impact coefficients for each input personnel density data. These output circulation impact coefficients are integrated to successfully construct a circulation impact coefficient array. This array accurately reflects the degree of influence of the personnel distribution on the air circulation in the target space. For example, in areas with a high density of people, the circulation impact coefficient may be relatively large, indicating that people have a strong obstructive effect on air circulation; while in areas with a low density of people, the circulation impact coefficient may be relatively small, and the air circulation is relatively smooth. This circulation impact coefficient array will provide a key basis for the correction of subsequent air quality risk parameters, further improving the accuracy and reliability of the entire environmental risk monitoring platform.

[0022] The air quality risk parameter acquisition module 30 is used to predict the risk of air quality change based on the multiple air parameter sequences to obtain multiple air quality risk parameters, and use the multiple circulation impact coefficients in the circulation impact coefficient array to correct the influence on the multiple air quality risk parameters to obtain multiple adjusted air quality risk parameters.

[0023] Specifically, the air quality risk parameter acquisition module 30 predicts the air quality change risk based on multiple air parameter sequences and corrects the prediction results through the circulation influence coefficient array to obtain more accurate adjusted air quality risk parameters. The module utilizes multiple air parameter sequences continuously monitored and processed by distributed sensors in the early stage. These sequences contain air parameter information at different positions in the target space at multiple moments, such as the concentration of harmful gases. The air parameter sequences are successively input into the pre-trained air quality risk prediction channel. The air quality risk prediction channel is based on the long short-term memory network (LSTM) algorithm model, which is obtained through machine learning training with a large amount of historical air quality detection data. It can deeply explore the hidden change rules and trends in the air parameter sequences and predict the changes in air quality after a preset time length in the future for each air parameter sequence, thereby obtaining multiple air quality risk parameters. These parameters intuitively reflect the degree of air quality risk that may be faced in different spatial positions in the future. For example, a higher value may indicate a greater risk of air quality deterioration. After obtaining the air quality risk parameters, the module immediately proceeds with the impact correction work. It reads the circulation influence coefficient array generated by the circulation influence coefficient array acquisition module. This array contains multiple circulation influence coefficients corresponding to spatial positions, and these coefficients reflect the degree of influence of personnel density on air circulation. For each air quality risk parameter, a specific correction calculation method is adopted to multiply the air quality risk parameter by (1 + the corresponding circulation influence coefficient). The principle behind this is that when the circulation influence coefficient is large, it means that the personnel density has a greater obstruction to air circulation, and the air quality risk parameter needs to be adjusted accordingly to increase, so as to more accurately reflect the actual air quality risk situation. In this way, each air quality risk parameter is corrected and calculated one by one, and finally multiple adjusted air quality risk parameters are obtained. These adjusted parameters fully consider the influence of personnel density on air quality. Compared with the uncorrected air quality risk parameters, they can more accurately evaluate the air quality risk in the target space, provide more reliable data support for the subsequent discrimination and warning module, and help improve the accuracy of the air quality risk assessment of the entire environmental risk monitoring platform.

[0024] The discrimination and warning module 40 is used to generate comprehensive environmental risk parameters based on the multiple adjusted air quality risk parameters and in combination with the personnel density array, and conduct discrimination and warning.

[0025] Specifically, the discrimination and early warning module 40 first performs the weight allocation work. Based on the personnel density array obtained from the circulation influence coefficient array acquisition module, which contains the personnel density information of multiple spatial positions. Using a proportion-based weight calculation method, for each spatial position, accurately calculate the ratio of the personnel density at this position to the total personnel density of all spatial positions, and determine this ratio as the weight value of this position. For example, assume there are spatial positions A, B, and C, with their personnel densities being ρA = 0.3 people per square meter, ρB = 0.5 people per square meter, and ρC = 0.2 people per square meter respectively. After calculation, the sum of all personnel densities is 1 person per square meter. Thus, the weight of spatial position A is determined to be 0.3 ÷ 1 = 0.3; the weight of spatial position B is 0.5 ÷ 1 = 0.5; the weight of spatial position C is 0.2 ÷ 1 = 0.2. This weight calculation method can achieve reasonable weight allocation according to the actual distribution ratio of personnel in each spatial position, thereby accurately reflecting the relative contribution degree of the personnel density in each area to the overall environmental risk. In this way, corresponding weights are assigned to each adjusted air quality risk parameter to ensure that the influence of the personnel density factor can be reasonably reflected in the subsequent calculation of the comprehensive environmental risk parameter. Next, calculate the comprehensive environmental risk parameter. Using the allocated weights, perform weighted calculation on the multiple adjusted air quality risk parameters obtained from the air quality risk parameter acquisition module, multiply each adjusted air quality risk parameter by its corresponding weight, and then add these products to obtain the comprehensive environmental risk parameter. This parameter comprehensively considers the air quality risk and the personnel density factor, and comprehensively reflects the environmental risk situation in the target space. For example, in an area with a high density of people and a high adjusted air quality risk parameter, the comprehensive environmental risk parameter will increase significantly, indicating that the environmental risk in this area is at a high level. Finally, perform the discrimination and early warning operation, and strictly compare the calculated comprehensive environmental risk parameter with the pre-set environmental risk threshold. If the comprehensive environmental risk parameter is greater than the environmental risk threshold, it means that the environmental risk in the target space has exceeded the acceptable range. At this time, the module immediately activates the early warning mechanism. The early warning methods are diverse, and may include emitting strong sound and light alarm signals to attract the attention of on-site personnel; at the same time, sending early warning information to relevant management personnel via text messages, push notifications, etc., informing them of the specific situation and location of the environmental risk, so as to take corresponding measures in a timely manner, such as strengthening ventilation, evacuating personnel, checking pollution sources, etc., to reduce the environmental risk and ensure the health and safety of personnel. If the comprehensive environmental risk parameter is less than or equal to the environmental risk threshold, it is determined that the current environmental risk is in a controllable state and no early warning is carried out temporarily. However, the module will continuously monitor the target space and be ready to respond to possible risk changes at any time to ensure that the environmental safety is always within the controllable range. Through such a rigorous and efficient work process, the discrimination and early warning module realizes the real-time and accurate monitoring and early warning of environmental risks, providing reliable environmental risk protection for the target space.

[0026] In one embodiment, as Figure 2 shown, the air parameter sequence acquisition module 10 includes:

[0027] An air parameter set acquisition unit, configured to continuously monitor and acquire air parameters at multiple spatial positions at multiple moments through distributed sensors deployed in a target space, to obtain multiple air parameter sets, where each air parameter set includes the air parameters at the multiple moments, and each air parameter includes a harmful gas concentration; an air parameter sequence acquisition unit, configured to sort the multiple air parameter sets according to the multiple moments to obtain multiple air parameter sequences.

[0028] Specifically, the air parameter set acquisition unit relies on a carefully deployed distributed sensor network to build an all-round air parameter monitoring system in the target space. These sensors are distributed at multiple key spatial positions, and these sensors are reasonably distributed at multiple key spatial positions according to factors such as the physical structure, functional partition, and air flow characteristics of the target space. For example, in an indoor environment, sensors will be arranged in areas with frequent human activities (such as office areas, rest areas, etc.), near potential pollution sources (such as kitchens, bathrooms, equipment machine rooms, etc.), and at key air exchange nodes such as ventilation openings and return air openings. In an industrial plant environment, in addition to the above areas, sensors will also be mainly deployed at key positions such as workstations where harmful gas emissions may occur during the production process and raw material storage areas to ensure that air parameter information in all regions of the target space can be comprehensively and accurately captured. During continuous operation, at multiple different time points, the sensors accurately capture the air parameter information of each spatial position. Among them, each air parameter covers harmful gas concentration indicators that have important impacts on environmental quality and human health, such as the concentrations of common harmful gases such as carbon monoxide, sulfur dioxide, and nitrogen dioxide. Over time, for each spatial position, rich and diverse air parameter data are gradually accumulated. By systematically integrating the air parameters obtained at different moments at the same spatial position, multiple air parameter sets are successfully formed. Each air parameter set completely and detailedly records the air parameter status of this position at each time node, providing a solid data basis for subsequent in-depth analysis of the dynamic changes in air quality.

[0029] After obtaining multiple air parameter sets, the air parameter sequence acquisition unit undertakes the important task of constructing an air parameter sequence. It strictly sorts these air parameter sets according to the chronological order. Arranged in an orderly manner along the time context, the air parameter sets at different times in the same spatial position are successively concatenated. Through such a sorting process, multiple air parameter sequences are successfully obtained. It provides key input data for subsequent work such as further exploring the law of air quality change and predicting air quality risks, and is an important link for the entire environmental risk monitoring platform to accurately grasp the dynamic of air quality.

[0030] In one implementation, the circulation influence coefficient array acquisition module 20 includes:

[0031] A personnel density array construction unit, which is used to obtain the number of people in the multiple spatial positions, combine with the spatial areas of the multiple spatial positions, calculate multiple personnel densities, and construct a personnel density array; a classification channel training unit, which is used to pre-train an air circulation influence classification channel; a circulation influence coefficient output unit, which is used to input the multiple personnel densities in the personnel density array into the air circulation influence classification channel respectively, output multiple circulation influence coefficients, and construct the circulation influence coefficient array.

[0032] Specifically, the personnel density array construction unit is used to obtain the number of people information in multiple spatial positions within the target space. This process is achieved through various data collection means, such as personnel counting devices installed at key positions, personnel statistics systems based on image recognition technology, etc., to ensure that the real-time number of people in each spatial position can be accurately obtained. At the same time, accurate spatial area data of each spatial position is obtained, and these data come from architectural design drawings, on-site measurements, or spatial modeling based on advanced surveying and mapping technologies. Then, the number of people in each spatial position is comprehensively calculated with the corresponding spatial area, that is, personnel density = number of people / spatial area. Through such calculations, the personnel density values of each spatial position are obtained. Finally, these calculated personnel density values are orderly integrated according to the order of spatial positions, and a personnel density array is successfully constructed. This array presents the density of personnel distribution in the target space in an intuitive form. For example, in areas with a high density of people, the personnel density value is high, while in areas with a low density of people, the personnel density value is low, providing basic data for subsequent analysis of the impact of personnel distribution on air circulation.

[0033] When the classification channel training unit pre-trains the air circulation impact classification channel, it first uses a multi-layer perceptron (MLP) neural network. The MLP consists of an input layer, multiple hidden layers, and an output layer. When constructing the network structure, the number of hidden layers and the number of neurons in each layer are reasonably determined according to the complexity of the input sample personnel density set features and the expected output sample circulation impact coefficient set. For example, if the sample data dimension is high and the relationship is complex, more hidden layers and neurons are set to enhance the learning ability and expression ability of the network. Each sample in the sample personnel density set is used as an input vector and input into the input layer of the MLP. Each dimension of these input vectors corresponds to the personnel density value at a spatial position. After passing through the input layer, the input vectors are successively subjected to complex non-linear transformations through the hidden layers. The neurons in the hidden layer are connected by weights. Each neuron performs a weighted sum of the input data and is processed by an activation function (such as the commonly used ReLU function) and then passed to the next layer. During the training process, the backpropagation algorithm is used to adjust the connection weights between neurons. The backpropagation algorithm calculates the gradient of the error with respect to the weight of each neuron from the output layer in reverse according to the error between the predicted circulation impact coefficient output by the output layer and the actual sample circulation impact coefficient, and then updates the weights according to the gradient descent method, so that the error predicted by the network gradually decreases. The training process continues, continuously inputting sample data into the network for forward propagation and backpropagation to update the weights until the model converges. The judgment basis for convergence can be to set a small error threshold. When the prediction error continuously falls below this threshold, it is considered that the model has learned the stable relationship between personnel density and air circulation impact coefficient. At this time, the training is stopped, and a high-precision air circulation impact classification channel after training is obtained. This channel can accurately predict the corresponding circulation impact coefficient according to the newly input personnel density data, providing strong support for accurately evaluating the impact of personnel density on air circulation subsequently, and thus improving the performance of the entire environmental risk monitoring platform.

[0034] The air circulation influence coefficient output unit first obtains the constructed personnel density array from the personnel density array acquisition module. This array contains accurate personnel density information at multiple spatial positions within the target space. Then, it sequentially extracts each personnel density value in the personnel density array. These personnel density values are individually input into the pre-trained air circulation influence classification channel. Based on the multi-layer perceptron (MLP) neural network model and the relationship pattern between personnel density and air circulation influence learned during the training process, the air circulation influence classification channel deeply analyzes and calculates each input personnel density value, and outputs the air circulation influence coefficient corresponding to each input personnel density. For example, when a relatively high personnel density value is input, the channel outputs a relatively large air circulation influence coefficient according to the pattern it has learned, indicating that the air circulation is more obstructed under this personnel density; conversely, when a relatively low personnel density value is input, the output air circulation influence coefficient is relatively small, indicating that the air circulation is relatively smooth. As each personnel density value in the personnel density array is sequentially input and processed, a series of air circulation influence coefficients are continuously obtained. Finally, these output air circulation influence coefficients are integrated in the order corresponding to the personnel density values to successfully construct the air circulation influence coefficient array. This array comprehensively and accurately reflects the differences in the degree of influence of personnel density at different spatial positions within the target space on air circulation, provides a key basis for the calibration of subsequent air quality risk parameters, and further improves the accuracy and reliability of the air quality risk assessment of the entire environmental risk monitoring platform.

[0035] In one implementation, the classification channel training unit includes:

[0036] The sample air circulation influence coefficient set annotation unit is used to collect the sample personnel density set according to the air circulation test data within the historical time, and test the reduction ratio of the air circulation speed under different sample personnel densities, and label it as the sample air circulation influence coefficient set; the air circulation influence classification channel training unit is used to use the sample personnel density set and the sample air circulation influence coefficient set as supervised training data to construct and train the air circulation influence classification channel until convergence, and obtain the trained air circulation influence classification channel.

[0037] Specifically, the sample circulation impact coefficient set annotation unit first deeply mines a large amount of air circulation test data accumulated over historical time. These data are the monitoring and records of air circulation conditions under different environmental conditions, spatial layouts, and personnel activities in the past. Using data processing techniques and screening strategies, a representative sample personnel density set is carefully selected from the massive data. This process requires comprehensive consideration of various factors, including personnel density in different space types (such as offices, workshops, public places, etc.), different time periods (weekdays, holidays, peak hours, etc.), and different personnel aggregation patterns, to ensure that the collected samples can comprehensively cover the range of possible personnel density changes in the target space. After obtaining the sample personnel density set, for each sample personnel density, a test of the air circulation speed reduction ratio is carried out using high-precision air circulation test instruments. The test process is carried out under simulated actual environments or conditions based on historical scenario reproduction to ensure the authenticity and reliability of the test results. For each specific sample personnel density, by accurately measuring the air circulation speed in the presence of personnel and comparing it with the baseline air circulation speed without personnel interference, the air circulation speed reduction ratio is calculated. Finally, each sample personnel density is accurately matched and annotated with its corresponding air circulation speed reduction ratio to form a sample circulation impact coefficient set. Each element in this set clearly records the degree of influence of a specific personnel density on the air circulation speed, providing rich, accurate, and targeted data support for the training of subsequent air circulation impact classification channels, and is an important cornerstone for the entire environmental risk monitoring platform to accurately evaluate the impact of personnel density on air circulation.

[0038] When the air circulation impact classification channel training unit constructs a model for accurately predicting the impact of personnel density on air circulation, it uses an MLP (Multi-Layer Perceptron) neural network and employs the sample personnel density set and the sample circulation impact coefficient set as supervised training data for training until the model converges to obtain a trained channel. First, construct the MLP neural network structure. Determine the number of input layer neurons according to the feature dimensions of the sample personnel density set to ensure that complete personnel density information can be received. Design an appropriate number of hidden layers, and the number of hidden layers and the number of neurons in each layer will be balanced according to the complexity of the data and the learning ability of the model. The output layer is determined according to the target output form of the sample circulation impact coefficient set. For example, if predicting continuous circulation impact coefficient values, the output layer is a single neuron; if performing a classification task (such as grading according to the degree of circulation impact), the number of output layer neurons is the same as the number of categories. Next, sequentially input each sample data in the sample personnel density set into the input layer of the MLP neural network. After the data is received by the neurons in the input layer, it is passed to the hidden layer through the connection weights. The neurons in the hidden layer use activation functions (such as the commonly used ReLU function) to perform non-linear transformation on the input data, enhancing the model's ability to express complex relationships. After being processed by multiple hidden layers, the data finally reaches the output layer, and the predicted circulation impact coefficient or classification result is output. Calculate the error between the predicted result and the actual value of the corresponding sample in the sample circulation impact coefficient set. Usually, loss functions such as the mean square error (MSE) are used to quantify the error size. According to the error value, use the backpropagation algorithm to calculate the gradient of the error with respect to the connection weights of each neuron backward from the output layer. Then, according to the gradient descent method, adjust the connection weights between neurons according to the calculated gradient so that the model can reduce the prediction error in the next iteration. Repeat the above processes of data input, prediction, error calculation, and weight adjustment, and continuously iterate to train the model. As the training progresses, the model's fitting ability to the sample data gradually increases, and the prediction error gradually decreases. Continue training until the model converges. The convergence criterion can be that when the value of the loss function no longer significantly decreases in consecutive multiple iteration cycles, or reaches a pre-set minimum error threshold. At this time, the fully trained MLP neural network becomes a trained air circulation impact classification channel, which can accurately predict the corresponding circulation impact coefficient according to the input personnel density data, providing a reliable tool for the assessment of the impact of personnel density on air circulation in subsequent actual environmental monitoring, and effectively improving the performance of the entire environmental risk monitoring platform.

[0039] In one implementation, the air quality risk parameter acquisition module 30 includes:

[0040] A quality risk prediction channel training unit is used to pre-train an air quality risk prediction channel; an air quality risk parameter acquisition unit inputs the multiple air parameter sequences into the air quality risk prediction channel respectively to perform air quality change risk prediction and obtains multiple air quality risk parameters after a preset future time length.

[0041] Specifically, when the quality risk prediction channel training unit pre-trains the air quality risk prediction channel, it uses the Long Short-Term Memory (LSTM) algorithm to achieve accurate prediction of the air quality change risk. First, an LSTM network structure is constructed. According to the characteristics of the input sample air parameter sequence set and the form of the expected output sample air quality risk parameter set, the input layer, hidden layer, and output layer of the network are determined. The number of neurons in the input layer matches the feature dimension of the air parameter sequence to receive complete air parameter information. The hidden layer consists of multiple LSTM units. The LSTM units control the transmission and forgetting of information through their unique gating structures (forget gate, input gate, output gate), and can effectively capture the long-term dependencies in the air parameter sequence, which is crucial for understanding the trend of air quality changes. The output layer is designed according to the prediction target of the air quality risk parameters. For example, if predicting a single air quality risk indicator, the output layer is a single neuron; if predicting multiple related indicators, the number of neurons in the output layer increases accordingly. Each sample sequence in the sample air parameter sequence set is sequentially input into the input layer of the LSTM network. After the data at each time step in the sequence is received by the neurons in the input layer, it is processed through the gating mechanism of the LSTM units in the hidden layer. The forget gate determines which information in the previous hidden state needs to be forgotten, the input gate controls the input of new information, the cell state is responsible for storing long-term information, and the output gate determines which information in the hidden state is used as the output at the current time step. After processing through multiple time steps, the LSTM network conducts in-depth analysis on the entire sample air parameter sequence and finally outputs the predicted air quality risk parameters at the output layer. Calculate the error between the prediction result and the actual value of the corresponding sample in the sample air quality risk parameter set, and use the Mean Squared Error (MSE) to quantify the size of the error. Then, use the Backpropagation Through Time (BPTT) algorithm to calculate the gradient of the error with respect to each parameter (including weights and biases) in the LSTM network starting from the output layer in reverse. According to the calculated gradient, use the Stochastic Gradient Descent method to adjust the network parameters so that the model can reduce the prediction error in the next iteration. Repeat the above processes of data input, prediction, error calculation, and parameter adjustment to continuously iterate and train the LSTM network. As the training progresses, the network's understanding of the relationship between the sample air parameter sequence and the air quality risk parameters deepens, and the prediction error gradually decreases. Continue training until the model converges. The convergence criterion can be that when the value of the loss function no longer decreases significantly in consecutive multiple iteration cycles, or reaches a pre-set minimum error threshold. At this time, the fully trained LSTM network becomes the trained air quality risk prediction channel, which can accurately predict the air quality risk parameters after a preset time length in the future based on the input air parameter sequence, providing strong technical support for subsequent air quality risk assessment and environmental monitoring.

[0042] The air quality risk parameter acquisition unit realizes the accurate assessment of the air quality change risk through collaborative work with the air quality risk prediction channel. This unit first obtains multiple air parameter sequences obtained in the early stage. These air parameter sequences detail the air parameter information at different spatial positions at multiple moments, including important indicators such as harmful gas concentration and particulate matter concentration, providing a rich data basis for subsequent predictions. Then, these multiple air parameter sequences are sequentially input into the pre-trained air quality risk prediction channel. The air quality risk prediction channel conducts in-depth analysis and calculation on each input air parameter sequence based on the long short-term memory network (LSTM) algorithm model and the rules learned during the training process. Inside the channel, the algorithm effectively captures the dynamic trends and potential patterns of air quality changes according to the characteristics and temporal relationships of the air parameter sequences, and finally outputs multiple air quality risk parameters after a preset time length in the future. These parameters intuitively reflect the degree of air quality risk that different positions in the target space may face in a future period. For example, if the value of the air quality risk parameter at a certain spatial position is relatively high, it indicates that the risk of air quality deterioration in this area in the future preset time is relatively large, and situations such as excessive harmful gas concentration and decline in air quality grade may occur. This provides key input data for the subsequent discrimination and warning module, helps to timely detect potential air quality problems, provides an important basis for taking corresponding measures to ensure the air quality safety in the target space, and thus improves the effectiveness and practicality of the entire environmental risk monitoring platform.

[0043] In one implementation manner, the quality risk prediction channel training unit includes:

[0044] The sample air quality risk parameter set annotation unit is used to collect a set of sample air parameter sequences according to the air quality detection data within the historical time, and collect the air quality risk parameters after a preset time length following different sample air parameter sequences, and label them as the sample air quality risk parameter set; the air quality risk prediction channel construction unit is used to use the set of sample air parameter sequences and the sample air quality risk parameter set as supervised training data, and use machine learning to construct and train the air quality risk prediction channel until convergence, and obtain the trained air quality risk prediction channel.

[0045] Specifically, the sample air quality risk parameter set annotation unit first comprehensively sorts out the vast amount of air quality detection data accumulated over historical time, including detailed information on the air quality in the target space under different time periods, different meteorological conditions, different geographical regions, and different pollution source emissions. Using data mining and screening techniques, a representative set of sample air parameter sequences is selected. During the collection process, various factors that may affect air quality are fully considered to ensure that the selected sample sequences can comprehensively reflect the diversity and complexity of air quality changes. For example, it includes air parameter sequences from areas with severe industrial pollution during peak production periods, as well as sample sequences from nature reserves in an ecological balance state; it collects air parameters in high-temperature and high-humidity environments in summer and also involves data during cold and dry winters. For each collected sample air parameter sequence, the unit further tracks the actual changes in air quality over a preset time length in the future. Through subsequent continuous air quality monitoring and professional data analysis and evaluation, the corresponding air quality risk parameters are obtained. The determination of these risk parameters comprehensively considers the change trends of various air quality indicators, such as the rising or falling rate of harmful gas concentrations, the fluctuation range of particulate matter concentrations, and the grade changes of the Air Quality Index (AQI). For example, if within the preset time after a certain sample air parameter sequence, the harmful gas concentration continuously rises and exceeds a certain threshold, and at the same time the AQI grade changes from good to slightly polluted, then according to the pre-set evaluation criteria, the air quality risk parameter corresponding to this sample sequence is determined to be a high-risk level. Finally, each sample air parameter sequence is accurately matched and annotated with its corresponding air quality risk parameter to form a sample air quality risk parameter set. These carefully annotated sample data provide a data basis for the construction and training of subsequent air quality risk prediction channels.

[0046] The air quality risk prediction channel construction unit uses LSTM (Long Short-Term Memory network), with the sample air parameter sequence set and the sample air quality risk parameter set as the supervised training data, to construct and train a high-precision air quality risk prediction channel. First, determine the number of neurons in the input layer of the LSTM network according to the feature dimension of the sample air parameter sequence to ensure that the air parameter information can be completely received. Design a hidden layer containing multiple LSTM units. The unique gating structure (forget gate, input gate, output gate) of the LSTM unit enables it to effectively capture the long-term dependencies in the air parameter sequence, which is crucial for understanding the complex dynamic process of air quality changes. Set the output layer according to the prediction target of the air quality risk parameter. For example, if predicting a single air quality risk value, the output layer is a single neuron; if multiple risk indicators are involved in the prediction, multiple output neurons are set accordingly. Input each sequence in the sample air parameter sequence set into the input layer of the LSTM network in turn. At each time step, the data enters the LSTM units in the hidden layer through the input layer. The forget gate decides which information needs to be forgotten based on the previous hidden state and the current input. The input gate controls the input of new information. The cell state is responsible for storing long-term information. The output gate determines which information in the hidden state is passed as the output at the current time step to the next layer. Through the processing of multiple time steps, the LSTM network deeply analyzes the entire air parameter sequence and finally outputs the predicted air quality risk parameter at the output layer. Calculate the error between the prediction result and the actual value of the corresponding sample in the sample air quality risk parameter set, and use the mean square error (MSE) loss function to quantify the error size. Then use the backpropagation through time (BPTT) algorithm to calculate the gradient of the error with respect to each parameter (including weights and biases) of the LSTM network from the output layer in reverse. According to the calculated gradient, adjust the network parameters by means of the stochastic gradient descent method to make the model continuously reduce the prediction error in subsequent iterations. Continuously repeat the above process of data input, prediction, error calculation, and parameter adjustment for iterative training. As the training progresses, the LSTM network becomes more and more accurate in grasping the relationship between the air parameter sequence and the air quality risk parameter in the sample data, and the prediction error gradually decreases. The training continues until the model converges. The convergence criterion is that when the value of the loss function no longer significantly decreases in consecutive multiple iteration cycles, or reaches the pre-set minimum error threshold. At this time, the fully trained LSTM network becomes the trained air quality risk prediction channel, which can accurately predict the air quality risk parameter after a preset time length in the future according to the input air parameter sequence, providing reliable technical support for air quality risk assessment and environmental monitoring, and effectively improving the performance of the entire environmental risk monitoring platform.

[0047] In one implementation, the air quality risk parameter acquisition module 30 includes:

[0048] A correction coefficient calculation unit is configured to calculate a plurality of correction coefficients based on the plurality of circulation influence coefficients; an influence correction calculation unit is configured to perform influence correction calculation on the plurality of circulation influence coefficients by using the plurality of correction coefficients to obtain a plurality of adjusted air quality risk parameters.

[0049] Specifically, the correction coefficient calculation unit calculates corresponding correction coefficients based on the plurality of circulation influence coefficients. This unit first obtains the plurality of circulation influence coefficients, which are quantitative representations of how the personnel density at different spatial positions in the target space affects the air circulation condition. They reflect how the personnel distribution changes the flow characteristics of air in the space, thereby affecting the diffusion and distribution of air quality. When calculating the correction coefficients, first, the obtained circulation influence coefficients are normalized, and their values are mapped to a specific interval, such as the interval [0, 1] or [-1, 1], for subsequent calculation and comparison. Then, according to the formula used, the correction coefficient = 0.5 + 0.5 × circulation influence coefficient. When the circulation influence coefficient is 0.6, the calculated correction coefficient is 0.8. Through such a calculation process, corresponding correction coefficients are calculated for each circulation influence coefficient. These correction coefficients will serve as important adjustment factors and play a role in the subsequent influence correction calculation unit, used to accurately correct the air quality risk parameters, so that the air quality risk assessment can more comprehensively and accurately consider the impact of personnel density on air quality, thereby improving the reliability of the entire environmental risk monitoring platform.

[0050] The impact correction calculation unit uses correction coefficients to accurately correct air quality risk parameters, thereby obtaining adjusted air quality risk parameters that can better reflect the actual situation. This unit first receives multiple correction coefficients from the correction coefficient calculation unit and multiple air quality risk parameters obtained from the air quality risk parameter acquisition module. These air quality risk parameters are the risk situations of air quality changes predicted based on air parameter sequences, but the key factor of the impact of personnel density on air circulation has not been considered. For each air quality risk parameter, the unit performs impact correction calculations in combination with its corresponding correction coefficient. Specifically, the air quality risk parameter is multiplied by (1 + correction coefficient). The principle behind this is that the correction coefficient reflects the degree of influence of personnel density on air circulation. When the correction coefficient is positive, it indicates that personnel density has an obstructive effect on air circulation, and the air quality risk parameter needs to be increased accordingly to more accurately reflect the actual risk. When the correction coefficient is negative (if this is the case, it may indicate that personnel density has a promoting effect on air circulation in some special cases, but this is relatively rare), the air quality risk parameter will be decreased accordingly. By performing such correction calculations on each air quality risk parameter in turn, multiple adjusted air quality risk parameters are finally obtained. These adjusted parameters fully consider the impact of personnel density on air circulation and can more accurately reflect the air quality risk status at different positions in the target space. They provide a more reliable data basis for the subsequent discrimination and warning module, helping to improve the accuracy of the air quality risk assessment of the entire environmental risk monitoring platform, and thus more effectively ensuring the environmental safety and personnel health in the target space. For example, in a densely populated area, if the correction coefficient is large, the corrected air quality risk parameter will be significantly increased, more accurately indicating the relatively high air quality risk that may exist in this area, so as to take corresponding measures for prevention and treatment in a timely manner.

[0051] In one implementation manner, the discrimination and warning module 40 includes:

[0052] A weight allocation unit is used to allocate a plurality of weights according to a plurality of personnel densities within the personnel density array; a comprehensive environmental risk parameter acquisition unit is used to perform weighted calculation on the plurality of adjusted air quality risk parameters by using the plurality of weights to obtain a comprehensive environmental risk parameter; a comprehensive environmental risk parameter judgment unit is used to judge whether the comprehensive environmental risk parameter is greater than an environmental risk threshold. If so, a warning is issued; if not, no warning is issued. Specifically, the weight allocation unit accurately allocates corresponding multiple weights based on a plurality of personnel density data within the personnel density array. When receiving the personnel density array, a weight calculation method based on proportion is adopted. For each spatial position, the ratio of the personnel density at this position to the total personnel density of all spatial positions is accurately calculated, and this ratio is determined as the weight value of this position. For example, there are spatial positions A, B, and C, and their personnel densities are ρA = 0.3 persons per square meter, ρB = 0.5 persons per square meter, and ρC = 0.2 persons per square meter respectively. After calculation, the sum of all personnel densities is 1 person per square meter. Thus, the weight of spatial position A is determined to be 0.3÷1 = 0.3; the weight of spatial position B is 0.5÷1 = 0.5; the weight of spatial position C is 0.2÷1 = 0.2. This weight calculation method can realize the reasonable allocation of weights according to the actual distribution ratio of personnel in each spatial position, so as to accurately reflect the relative contribution degree of the personnel density in each region to the overall environmental risk. Through such a rigorous weight allocation process, the dynamic changes of environmental risks can be reflected more accurately.

[0053] The comprehensive environmental risk parameter acquisition unit first receives two sets of key data: one set is a plurality of weights calculated by the weight allocation unit based on the personnel density array, and these weights include risk proportion information of different personnel density regions; the other set is a plurality of adjusted air quality risk parameters, which truly reflect the potential risk status of air quality in each region. Then the weighted calculation link is started, and each adjusted air quality risk parameter is matched with the corresponding weight one by one for multiplication operation. In areas with high personnel density and high weights, the air quality risk parameter accounts for a larger proportion in the final result, and vice versa. After all the multiplication operations are completed, these products are then accumulated and summed to finally successfully obtain a comprehensive environmental risk parameter.

[0054] After the comprehensive environmental risk parameter judgment unit obtains the comprehensive environmental risk parameters output by the comprehensive environmental risk parameter acquisition unit, it immediately calls the built-in comparison algorithm to compare these parameters with the environmental risk thresholds set in advance through multi-dimensional data analysis and model construction. This unit has pre-stored environmental risk thresholds set according to the nature, use, and relevant environmental standards of the target space. For example, for places with extremely high air quality requirements such as hospital operating rooms and electronic chip manufacturing workshops, the environmental risk thresholds are set very strictly, between 0.3 and 0.5; while for general offices, shopping malls and other places, the environmental risk thresholds are relatively higher, between 0.8 and 1.0. When the comparison result shows that the comprehensive environmental risk parameter is greater than the environmental risk threshold, it indicates that the environmental condition in the target space has deviated from the safe range, and the risk of air quality deterioration shows a sharp upward trend, which is very likely to cause serious negative impacts on key areas such as personnel life and health, and production operation processes. At this time, the judgment unit quickly activates the preset multi-modal early warning system, which realizes the coordinated operation of multiple early warning means based on intelligent linkage technology. On the one hand, at the scene of the target space, through the sound and light alarm module integrated with Internet of Things devices, high-intensity sound and light alarms are immediately triggered to prompt the on-site personnel of the emergency risk state with strong sound and light signals, prompting them to quickly follow the preset emergency procedures to take protective actions; on the other hand, with the help of communication interface technology, it is linked with the SMS gateway, system message push platform, etc., and the structured detailed early warning information is accurately pushed to the management personnel terminal, and the information covers professional content such as the geographical coding information of the risk area, the risk quantification assessment level, the estimated impact range and diffusion trend, etc., so that the management personnel can quickly start the emergency plan according to the early warning information.

[0055] On the contrary, if the comprehensive environmental risk parameter is less than or equal to the environmental risk threshold, it is determined that the current environmental risk is in a controllable state and no early warning is issued for the time being. But this does not mean that the unit stops working. Instead, it will continuously monitor the comprehensive environmental risk parameter, regularly (such as every 5 minutes or 10 minutes) re-obtain the latest parameter value and compare it with the threshold to ensure that the environmental risk is always under monitoring. Once it is found that the comprehensive environmental risk parameter has an upward trend and is approaching the threshold, the judgment unit will issue an early warning prompt in advance so that relevant personnel can take preventive measures in time to prevent the further deterioration of the environmental risk and ensure that the environmental safety in the target space is always within a controllable range.

[0056] It should be noted that the above order of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above specific embodiments of the present specification have been described. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-tasking and parallel processing are also possible or may be beneficial.

[0057] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

[0058] This specification and the drawings are merely illustrative of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. The environmental risk monitoring platform based on multimodal information fusion is characterized by: The platform includes: An air parameter sequence acquisition module is used to continuously monitor and acquire air parameters at multiple spatial locations through distributed sensors deployed in the target space to obtain multiple air parameter sequences; A circulation influence coefficient array acquisition module is used to acquire the personnel density array of the plurality of spatial positions, perform air circulation influence analysis, and obtain a circulation influence coefficient array; An air quality risk parameter acquisition module, configured to perform air quality change risk prediction according to the plurality of air parameter sequences to obtain a plurality of air quality risk parameters, and to perform impact correction on the plurality of air quality risk parameters using a plurality of circulation influence coefficients in the circulation influence coefficient array to obtain a plurality of adjusted air quality risk parameters; The discrimination and early warning module is used to generate comprehensive environmental risk parameters according to the multiple adjusted air quality risk parameters in combination with the personnel density array, and to perform discrimination and early warning.

2. The environmental risk monitoring platform based on multimodal information fusion according to claim 1 is characterized in that: The air parameter sequence acquisition module includes: An air parameter set acquisition unit, configured to continuously monitor and acquire air parameters at multiple spatial locations at multiple times through distributed sensors arranged in the target space, and obtain multiple air parameter sets, wherein each air parameter set includes air parameters at the multiple times, and each air parameter includes a concentration of harmful gases; The air parameter sequence acquisition unit is used to sort the multiple air parameter sets according to the multiple time points to obtain multiple air parameter sequences.

3. The environmental risk monitoring platform based on multimodal information fusion according to claim 1 is characterized in that: The circulation influence coefficient array acquisition module includes: A personnel density array construction unit is used to obtain the number of personnel in the multiple spatial positions, calculate multiple personnel densities based on the spatial areas of the multiple spatial positions, and construct a personnel density array; Classification channel training unit, used to pre-train the classification channels affected by air circulation; The circulation influence coefficient output unit is used to input the multiple personnel densities in the personnel density array into the air circulation influence classification channel respectively, output multiple circulation influence coefficients, and construct the circulation influence coefficient array.

4. The environmental risk monitoring platform based on multimodal information fusion according to claim 3 is characterized in that: The classification channel training unit comprises: The sample circulation influence coefficient set annotation unit is used to collect the sample personnel density set according to the air circulation test data in the historical time, and test the reduction ratio of the air circulation speed under different sample personnel densities, and annotate it as the sample circulation influence coefficient set; The air circulation impact classification channel training unit is used to use the sample personnel density set and the sample circulation impact coefficient set as supervised training data to construct and train the air circulation impact classification channel until convergence, thereby obtaining a trained air circulation impact classification channel.

5. The environmental risk monitoring platform based on multimodal information fusion according to claim 1 is characterized in that: The air quality risk parameter acquisition module includes: Quality risk prediction channel training unit, used to pre-train air quality risk prediction channels; The air quality risk parameter acquisition unit inputs the multiple air parameter sequences into the air quality risk prediction channel respectively, performs air quality change risk prediction, and obtains multiple air quality risk parameters after a preset time length in the future.

6. The environmental risk monitoring platform based on multimodal information fusion according to claim 5 is characterized in that: The quality risk prediction channel training unit includes: The sample air quality risk parameter set labeling unit is used to collect a sample air parameter sequence set based on the air quality detection data in the historical time, and collect the air quality risk parameters of the preset time length after different sample air parameter sequences, and label them as a sample air quality risk parameter set; The air quality risk prediction channel construction unit is used to use the sample air parameter sequence set and the sample air quality risk parameter set as supervised training data, use machine learning, construct and train the air quality risk prediction channel until convergence, and obtain a trained air quality risk prediction channel.

7. The environmental risk monitoring platform based on multimodal information fusion according to claim 1 is characterized in that: The air quality risk parameter acquisition module includes: A correction coefficient calculation unit, used for calculating and obtaining a plurality of correction coefficients according to the plurality of circulation influence coefficients; The impact correction calculation unit is used to use the multiple correction coefficients to perform impact correction calculations on the multiple circulation impact coefficients to obtain multiple adjusted air quality risk parameters.

8. The environmental risk monitoring platform based on multimodal information fusion according to claim 1 is characterized in that: The discrimination and warning module comprises: A weight allocation unit, used for allocating and obtaining a plurality of weights according to a plurality of personnel densities in the personnel density array; A comprehensive environmental risk parameter acquisition unit, configured to adopt the multiple weights to perform weighted calculation on the multiple adjusted air quality risk parameters to obtain a comprehensive environmental risk parameter; The comprehensive environmental risk parameter judgment unit is used to judge whether the comprehensive environmental risk parameter is greater than the environmental risk threshold. If so, an early warning is issued; if not, no early warning is issued.

Images