Dust concentration early warning method, computer storage medium and terminal equipment

Abstract

The present invention relates to a dust concentration early warning method, a computer storage medium and a terminal device. The early warning model includes an input feature extraction module, a prediction module, a time series feature extraction module and an early warning output module. First, historical features are extracted through historical concentration data, and on this basis, the predicted concentration data for a period of time in the future is predicted through the prediction module. Then, the two are combined and input into the time series feature extraction module to extract the time series features of the combination of historical concentration data and predicted concentration data. Finally, according to the time series features, the early warning result is output. It can not only improve the accuracy of dust concentration early warning, but also monitor in real time and warn in advance. By identifying and warning potential dust hazards in advance, it provides strong support for enterprises and supervision, greatly improves its real-time and response speed, and achieves early warning and prevention.

Description

Technical Field

[0001] The present invention relates to the field of neural network machine learning, and in particular to a dust concentration early warning method, a computer storage medium and a terminal device. Background Art

[0002] In the context of the current rapid industrial development, the number of workplaces in dusty environments has increased significantly, and the resulting health problems and accident risks have become increasingly prominent. Working in a high-dust environment for a long time, workers face the threat of occupational diseases such as pneumoconiosis and respiratory diseases, which seriously damages the health of workers. In addition, explosions and fire accidents caused by dust accumulation pose a major threat to production safety. At the same time, dust emissions pollute the surrounding environment, destroy the ecological balance, and affect sustainable development. In view of the above background, the dust environment early warning system came into being. Its core goal is to monitor dust concentration in real time and promptly discover potential health and safety risks, so as to take effective measures to prevent accidents, protect the health of workers, ensure production safety, and reduce pollution to the environment. Through early warning, the early warning system prompts enterprises to take measures to reduce dust exposure, effectively reduce the incidence of occupational diseases, enhance production safety, reduce production accident rates, and ensure production continuity and efficiency. At the same time, it reduces dust emissions, improves the quality of the surrounding environment, promotes sustainable development, reduces direct and indirect economic losses caused by accidents and health problems, and improves enterprise production efficiency and competitiveness.

[0003] Although dust environment early warning technology has played an important role in protecting workers' health and improving production safety, there are still some challenges and problems that need to be solved, such as real-time and response speed: in some cases, the real-time and response speed of the early warning system may not be sufficient to avoid sudden accidents in time, especially in an environment where dust concentration changes rapidly. Early warning and prevention are particularly important.

[0004] Therefore, how to improve the real-time and response speed of dust warnings so as to achieve early identification and prevention is a technical problem that needs to be solved urgently in this field. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a dust concentration early warning method, comprising:

[0006] Construct and train an early warning model that takes historical concentration data at several past time points as input and issues an early warning result as output, including an input feature extraction module, a prediction module, a time series feature extraction module, and an early warning output module;

[0007] An input feature extraction module is used to input historical concentration data at several past time points to extract historical features;

[0008] The prediction module is connected to the output of the input feature extraction module, and is used to input historical features and make predictions based on them to obtain predicted concentration data at several future time points;

[0009] A time series feature extraction module, connected to the input of the input feature extraction module and the output of the prediction module, for inputting historical concentration data at a number of past time points and predicted concentration data at a number of future time points to extract time series features;

[0010] The warning output module is connected to the output of the time series feature extraction module, and is used to input the time series features and output the warning results accordingly;

[0011] Obtain historical concentration data for several past time points in the current scene and input it into the trained warning model to determine whether to issue a warning.

[0012] Furthermore, during the model construction process, a feature extraction module is input, including: a first convolutional layer, a first pooling layer, and a hidden layer;

[0013] The first convolutional layer is used to input historical concentration data at several past time points and extract local features of the historical concentration data;

[0014] The first pooling layer, whose input is connected to the output of the first convolutional layer, is used to reduce the local feature size to extract the key features of the local features;

[0015] The hidden layer, whose input is connected to the output of the first pooling layer, is used to learn the complex relationship between features to extract historical features.

[0016] Further, the prediction module includes: a second convolutional layer, a second pooling layer and a first fully connected layer;

[0017] The second convolutional layer, whose input is connected to the output of the hidden layer, is used to extract higher-level features of the historical features;

[0018] The second pooling layer, whose input is connected to the output of the second convolutional layer, is used to reduce the size of higher-level features to extract key features of higher-level features;

[0019] The first fully connected layer, whose input is connected to the output of the second convolutional layer, is used to integrate the features extracted by the previous layers for the final regression task to obtain the predicted concentration data at several future time points.

[0020] Furthermore, the temporal feature extraction module includes a third convolutional layer, a random dropout layer, a residual unit, a fourth convolutional layer and a third pooling layer;

[0021] The third convolutional layer, whose input is connected to the input of the first convolutional layer and the output of the first fully connected layer, is used to input the historical concentration data at several past time points and the predicted concentration data at several future time points to generate comprehensive features;

[0022] A residual module, whose input is connected to the output of the input feature extraction module, and whose output is connected to the input of the random activation layer to pass the historical features to the random activation layer;

[0023] The random dropout layer, whose input is connected to the output of the third convolutional layer and the output of the residual module, is used to input historical features and comprehensive features, and perform preprocessing to obtain fused features;

[0024] The fourth convolutional layer, whose input is connected to the output of the random dropout layer, is used to input the processed fusion features and further extract deeper features of the fusion features;

[0025] The third pooling layer, whose input is connected to the output of the fourth convolutional layer, is used to reduce the spatial size of deeper features to obtain the final temporal features.

[0026] Furthermore, the warning output module includes a second fully connected layer, which is used to complete the final classification task according to the time series characteristics and output 0 or 1; where 0 indicates no warning and 1 indicates warning.

[0027] Furthermore, the first convolution layer includes N channels; the third convolution layer includes N+M channels, which are respectively used to input historical concentration data at a time point or predicted concentration data at a time point; each channel configures a higher weight coefficient for the historical concentration data or / and predicted concentration data close to the current time according to the order of the time points; N is the number of time points of the historical concentration data; M is the number of time points of the predicted concentration data.

[0028] Furthermore, during the model training process, the regression loss function of the prediction module uses the mean absolute error MAE, and the classification loss function of the warning output module uses the logarithmic hyperbolic cosine Log-Cosh.

[0029] Furthermore, the loss function of the entire model is:

[0030] L=2×MAE+Log-Cosh.

[0031] On the other hand, the present invention further provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-mentioned dust concentration warning methods.

[0032] On the other hand, the present invention also provides a terminal device, including a memory and a processor; the memory stores a program code that can be executed by the processor; the program code is used to execute any of the above-mentioned dust concentration warning methods.

[0033] The dust concentration early warning method, computer storage medium and terminal device provided by the present invention first extract historical features through historical concentration data, and then predict the predicted concentration data for a period of time in the future through a prediction module on this basis; then combine the two and input the time series feature extraction module to extract the time series features of the combination of historical concentration data and predicted concentration data; finally, output the early warning result according to the time series features. It can not only improve the accuracy of dust concentration early warning, but also enable real-time monitoring and early warning. By identifying and warning potential dust hazards in advance, it provides strong support for enterprises and supervision, greatly improves its real-time performance and response speed, and achieves early warning and prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a structural schematic diagram of an embodiment of an early warning model of the dust concentration early warning method of the present invention;

[0035] Figure 2 It is a structural schematic diagram of another embodiment of the early warning model of the dust concentration early warning method of the present invention;

[0036] Figure 3 It is a structural schematic diagram of an embodiment of a residual module of the warning model of the dust concentration warning method of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] It should be noted that if the embodiments of the present invention involve directional indications, such as up, down, left, right, front, back, etc., then the directional indication is only used to explain the relative position relationship, movement status, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly. In addition, if the embodiments of the present invention involve descriptions of "first, second", "S1, S2", "step one, step two", etc., then such descriptions are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of indicated technical features or indicating the execution order of the method, etc. Those skilled in the art can understand that anything that does not violate the gist of the invention under the technical concept of the invention should be included in the protection scope of the present invention.

[0039] The present invention provides a dust concentration early warning method, comprising:

[0040] S1: Construct and train an early warning model that takes historical concentration data at several past time points as input and issues an early warning result as output, such as Figure 1 As shown, it includes an input feature extraction module A, a prediction module B, a time series feature extraction module C and a warning output module D;

[0041] An input feature extraction module is used to input historical concentration data at several time points in the past to extract historical features;

[0042] The prediction module is connected to the output of the input feature extraction module, and is used to input historical features and make predictions based on them to obtain predicted concentration data at several future time points;

[0043] A time series feature extraction module, connected to the input of the input feature extraction module and the output of the prediction module, for inputting historical concentration data at a number of past time points and predicted concentration data at a number of future time points to extract time series features;

[0044] The warning output module is connected to the output of the time series feature extraction module, and is used to input the time series features and output the warning results accordingly;

[0045] S2: Obtain historical concentration data for the current scene at several past time points and input it into the trained warning model to determine whether to issue a warning.

[0046] In this embodiment, a dust concentration warning method of the present invention is given. The warning model constructed by the method first extracts historical features through historical concentration data, and then predicts the predicted concentration data for a period of time in the future through a prediction module; then the two are combined and input into a time series feature extraction module to extract the time series features of the combination of historical concentration data and predicted concentration data; finally, the warning result is output according to the time series features. For example, the monitoring equipment collects dust estimation data for the past 12 consecutive days in real time as input data {a1, a2, ..., a12}, firstly outputs the predicted concentration values ​​y1, y2, y3 for the next 3 days through the input feature extraction module and the prediction module; then the two {y1, ..., y3, a1, ......., a12} are input into the time series feature extraction module, and the time series feature extraction module and the warning output module output (0, 1), 0 represents normal, 1 represents warning; it monitors the dust concentration in real time through data collection and analysis, once the model outputs the warning result, the system will alarm immediately, and can give a warning 3 days in advance, that is: three days before the dust concentration exceeds the standard, the air condition is rectified, which can not only improve the accuracy of the dust concentration warning, but also can monitor in real time and give early warning, and provide strong support for enterprises and supervision by identifying and warning potential dust hazards in advance, greatly improving its timeliness and response speed, and achieving early warning and prevention, and at least achieving the following significant effects:

[0047] Protecting workers' health: By real-time monitoring and control of dust concentration and early warning, the risk of workers' exposure to harmful dust can be effectively reduced, and the occurrence of occupational diseases, especially pneumoconiosis and other respiratory diseases, can be reduced, thereby improving the overall health and quality of life of workers.

[0048] Improve production safety: Dust accumulation may lead to explosion and fire accidents. The early warning system can issue alarms in time, enabling enterprises to take prompt action to prevent accidents, protect the lives of employees, and reduce production stagnation and economic losses caused by accidents.

[0049] Reduce environmental pollution: Effectively controlling dust emissions helps reduce air pollution, improves the air quality of the working environment and surrounding communities, complies with environmental regulations, and promotes green and sustainable development.

[0050] Economic Benefits: Reduced medical expenses and compensation: By reducing occupational diseases through preventive measures, companies can reduce medical expenses and related compensation costs caused by employee illnesses.

[0051] Improve production efficiency: reduce downtime caused by accidents and occupational diseases, ensure the continuity and stability of production, and improve production efficiency and product quality.

[0052] Enhance corporate image: Implementing effective dust management measures can enhance the company's external image and attract more employees and customers who pay attention to the working environment.

[0053] Compliance and legal risk management: Comply with relevant occupational health and environmental regulations to avoid fines and legal risks due to illegal operations.

[0054] Promote technological innovation and industrial upgrading: Promote the research and development and application of dust prevention technology and equipment, and promote technological progress and industrial upgrading of the entire industry.

[0055] Social benefits: By reducing occupational diseases and environmental pollution, improving public health, enhancing the overall welfare of society, and promoting the harmonious and stable development of society.

[0056] In summary, the dust environment early warning system not only brings direct benefits at the economic level, but also has a profound positive impact on the social and environmental levels. It is an important measure to achieve sustainable safe production and protect the health of workers.

[0057] S1:

[0058] Model building, such as Figure 2 As shown:

[0059] A: Input feature extraction module

[0060] The input feature extraction module may optionally adopt any form of convolutional neural network (CNN). Preferably, it includes: a first convolutional layer, a first pooling layer, and a hidden layer;

[0061] The first convolution layer is used to input the historical concentration data of several time points in the past and extract the local features of the historical concentration data; specifically, the first convolution layer is the input layer of the model, and its main function is to extract the local features of the input data. In the convolution layer, the convolution kernel slides on the input data, and generates feature maps through convolution operations. These feature maps capture the local patterns of the input data, such as edges, textures and other information, to obtain local features. More specifically, the first convolution layer, preferably, has a convolution kernel = 3, a step size = 3, a padding = 2, and an activation function of rule. Preferably, the first convolution layer includes N channels, each of which is used to input the historical concentration data of a time point; each channel configures a higher weight coefficient for the historical concentration data close to the current time according to the order of the time points. For a specific example, the historical concentration data of the past 12 days are collected and input through 12 channels respectively. For the historical concentration data of the first day, the second day to the twelfth day, higher weight coefficients are configured in turn to increase the weight of the most recent time point and improve its role and guidance for subsequent predictions.

[0062] The first pooling layer, whose input is connected to the output of the first convolutional layer, is used to reduce the local feature size to extract the key features of the local features; specifically, the first pooling layer is used to reduce the spatial size of the feature map, thereby reducing the number of parameters and the amount of calculation in the subsequent layers, while improving the generalization ability of the model to extract the first key features, because it extracts the most critical features through maximum pooling or average pooling.

[0063] The hidden layer, whose input is connected to the output of the first pooling layer, is used to learn the complex relationship between features to extract historical features.

[0064] B. Prediction Module

[0065] A prediction module, optionally including: a second convolutional layer, a second pooling layer, and a first fully connected layer;

[0066] The second convolutional layer, whose input is connected to the output of the hidden layer, is used to extract higher-level features of the historical features. Specifically, this is another convolutional layer in the network, which further extracts higher-level features from the feature map output by the first convolutional layer. More specifically, the second convolutional layer has a preferred convolution kernel of 3, stride of 7, padding of 3, and an activation function of rule.

[0067] The second pooling layer, whose input is connected to the output of the second convolutional layer, is used to reduce the size of higher-level features to extract key features of higher-level features; specifically, the second pooling layer is used to reduce the size of higher-level features and improve the generalization ability of the model to extract the second key features; the second pooling layer, similar to the first pooling layer, continues to reduce the dimension of the feature map and further extracts key features.

[0068] The first fully connected layer, whose input is connected to the output of the second convolutional layer, is used to integrate the features extracted by the previous layer for the final regression task to obtain the predicted concentration data at several future time points. Specifically, in the fully connected layer, each neuron is connected to all neurons in the previous layer, which enables the network to learn the complex relationship between features and make the final prediction.

[0069] C: Time series feature extraction module

[0070] A temporal feature extraction module, optionally including a third convolution layer, a random dropout layer, a residual unit, a fourth convolution layer, and a third pooling layer;

[0071] The third convolutional layer, whose input is connected to the input of the first convolutional layer and the output of the first fully connected layer, is used to input historical concentration data at several past time points and predicted concentration data at several future time points to generate comprehensive features; specifically, this layer inputs the initial historical concentration data and the predicted concentration data at future time points output by the prediction module, continues to extract the high-level features of the two input data, extracts the spatial features of the input data through a sliding window (convolution kernel), and generates comprehensive features. More specifically, the third convolutional layer preferably has a convolution kernel = 3, a step size = 7, a padding = 2, and an activation function of rule. More preferably, similar to the first convolutional layer, the third convolutional layer optionally includes N+M channels, which are used to input historical concentration data and predicted concentration data at a time point respectively; each channel configures a higher weight coefficient for the historical concentration data or / and predicted concentration data close to the current time according to the order of the time points; N is the number of time points for the historical concentration data; M is the number of time points for the predicted concentration data. In the same example, historical concentration data of the past 12 days are collected and predicted concentration data of the next 3 days are predicted at the same time. They are input through 12+3 channels respectively. For the historical concentration data of the first day, second day to the twelfth day in the past, higher weight coefficients are configured in turn, and for the predicted concentration data of the first day, second day and third day in the future, lower weight coefficients are configured in turn to increase the weight of the most recent time point and improve its accuracy and guiding significance for subsequent predictions. For example, today is August 15, then the weight coefficient of the historical concentration data of August 14 is higher than the weight coefficient of the historical concentration data of August 13, and the weight coefficient of the predicted concentration data of August 16 is higher than the weight coefficient of the predicted concentration data of August 17.

[0072] The residual module has an input connected to the output of the input feature extraction module, and an output connected to the input of the random dropout layer to transfer the historical features to the random dropout layer. Specifically, the residual module can be selected as follows: Figure 3 As shown in the figure, by introducing cross-layer connections, the degradation problem in deep network training can be solved. This is because in traditional deep neural networks, as the number of network layers increases, the gradient of back propagation will gradually decrease or increase, making it difficult for the network to converge or become unstable. When the network becomes very deep, a small change in the low-level parameters will cause a drastic change in the high-level parameters, making it difficult for the optimization algorithm to find the optimal solution. By introducing the residual module, the residual network allows the gradient to propagate by skipping a certain number of layers, making it easier to train even very deep networks; this design allows the network to learn the residual function, that is, the difference between the input and the output, making it easier to train the network with deeper layers. Specifically, the input of the residual module is connected to the output of the hidden layer, and the output is connected to the input of the random inactivation layer, which can create a shortcut in the network to directly pass the input features that have not undergone complex transformations.

[0073] The random dropout layer, whose input is connected to the output of the third convolutional layer and the output of the residual module, is used to input historical features and comprehensive features, and perform preprocessing to obtain fused features. Specifically, the random dropout layer is a regularization technique used to prevent overfitting of the neural network. Since the historical features and comprehensive features are intricate, it is optional to randomly set the output of some neurons to zero during the training process, thereby reducing the complex co-adaptation relationship between neurons, and obtaining the preprocessed fused features as the input of the subsequent convolutional layer, which can further improve the generalization ability of the model, and by using a single network to approximate the average prediction effect of all sparse networks during testing.

[0074] The fourth convolutional layer, whose input is connected to the output of the random dropout layer, is used to input the processed fusion features and further extract deeper features of the fusion features; specifically, the fourth convolutional layer further processes the data to extract deeper features. More specifically, the fourth convolutional layer preferably has a convolution kernel of 5, a step size of 3, a padding of 2, and an activation function of rule.

[0075] The third pooling layer, whose input is connected to the output of the fourth convolutional layer, is used to reduce the spatial size of deeper features to obtain the final temporal features. Specifically, the third pooling layer further reduces the spatial size of the deeper feature map, thereby reducing the number of parameters and the amount of calculation, while extracting key features to obtain temporal features of associated comprehensive features and historical features.

[0076] D: Warning output module

[0077] The warning output module includes a second fully connected layer, which is used to complete the final classification task and output the warning result according to the time series characteristics. Specifically, the fully connected layer integrates the features extracted by the previous layer for the final classification and outputs 0 / 1, where 0 means no warning and 1 means issuing a warning.

[0078] In summary, several preferred embodiments of constructing an early warning model in the dust concentration early warning method of the present invention are given. It extracts historical features by extracting local features of input data - key features of local features; then extracts higher-level features of historical features - key features of higher-level features - and integrates them to obtain predicted concentration data at several future time points, thereby completing the prediction of future concentration data by historical concentration data; then the historical concentration data and the predicted concentration data are input into a time series feature extraction module together, and a comprehensive feature is first generated. Then, a cross-layer connection is introduced through a residual module, and the preceding historical features are passed to a random inactivation layer. The historical features are fused with the comprehensive features to obtain fused features - deeper features of the fused features - and the final time series features are obtained after dimensionality reduction - and finally, the final classification is completed according to the time series features, and a warning result of 0 / 1 is output. This is a dust concentration early warning model with accurate identification and early warning.

[0079] Model training:

[0080] Specifically, you can choose to collect concentration data for several days, with the previous several days as historical concentration data and the next several days as predicted concentration data, build training sets and validation sets, train the above-built early warning model, and continuously adjust the model parameters to improve the accuracy and generalization ability of the model. Specifically, evaluate the trained model, and you can choose to use evaluation indicators such as mean square error (MSE) and mean absolute error (MAE). If the model performance does not meet the requirements, you can further optimize the model parameters.

[0081] Preferably, in the prediction module, the regression loss function uses the mean absolute error MAE, and in the warning output module, the classification loss function uses the logarithmic hyperbolic cosine Log-Cosh loss function; the loss function of the entire model is:

[0082] L = 2 × MAE + Log-Cosh

[0083] In this embodiment, several preferred embodiments of the training warning model in the dust concentration warning method of the present invention are given. The regression loss function uses the mean absolute error because it is insensitive to abnormal values ​​and is suitable for data with more noise. For example, the concentration increase at a certain moment does not have a significant impact on data prediction. The classification loss function uses the logarithmic hyperbolic preselection. The hyperbolic cosine is a relatively smooth loss function. The hyperbolic cosine is used to calculate the prediction error. Its advantage is that for very small errors, log(cosh(x)) and (x 2)×12, and for large errors it is very close to abs(x)-log2. The logcosh loss function can have the advantages of MSE while not being affected too much by outliers. More importantly, the entire model loss function = 2×MAE+Log-Cosh. Multiplying the regression loss by 2 and superimposing it with the classification loss can increase the impact of future predicted concentration data on early warning judgments, further identify and warn of concentration exceeding the standard, and further improve the overall real-time and response speed of the model.

[0084] S2: Get the historical concentration data of the current scene at several time points in the past. You can install portable or fixed dust concentration monitors in key areas such as factories to be monitored. You can install them near production lines or worker operation areas to ensure continuous monitoring of dust levels in the air. For example:

[0085] ‌Belt transport location‌: The dust concentration monitor should be installed above the belt transport transfer point, and the recommended installation height is between 0.5 meters and 2 meters. This position can effectively monitor the dust concentration and help take necessary measures to reduce dust emissions in a timely manner‌.

[0086] Production workshop: The best location is to install it on a pillar or wall in the workshop, generally at a height of about 2.5 meters. Such a location can cover the entire workshop and ensure that changes in dust concentration are monitored in a timely manner, thereby protecting the health of employees and the safety of the production environment.

[0087] ‌Grinding machine area‌: The dust concentration monitor should be installed above the grinding machine, with a recommended height between 0.5m and 2m. This position helps monitor the dust concentration generated during the grinding process, and take timely control measures to reduce the harm of dust to workers‌.

[0088] ‌Finished product stacking area‌: When it is necessary to monitor the dust concentration of materials such as coal powder and flour, the installation location of the monitor should be considered above the material stacking point. For example, if the coal powder is stacked at a height of 15 meters, it is recommended to install the monitor about 18 meters away from the stacking point to ensure the accuracy of concentration monitoring.

[0089] Before installation, you need to fully understand the performance, specifications and usage of the equipment, select a suitable installation location, and avoid places that affect dust concentration, such as vents, heat sources, or exhaust ports of production equipment. Prepare the required installation tools, such as screwdrivers, wrenches, cables, etc., to ensure a smooth installation process. The installation steps include installing the bracket, installing the monitor, connecting the power and data transmission lines, and performing a power-on test to ensure that the equipment is working properly.

[0090] More specifically, the dust concentration measurement method preferably adopts the light absorption method. When light waves pass through linear substances, they interact with the substances. Part of the light wave is absorbed by the medium and converted into heat energy: part of the light wave is scattered by the medium and deviates from the original propagation direction, and the rest still passes through the medium in the original propagation direction. The relationship between the transmitted light intensity and the incident light intensity conforms to the Lambert-Beer law. Based on the Lambert-Beer law, the light absorption dust concentration sensor can measure the incident light intensity and the outgoing light intensity and calculate the dust concentration. This method can accurately measure in high-concentration gas-solid two-phase flow.

[0091] Subsequently, the monitored historical concentration data is input into the early warning model of this application to obtain whether the early warning result is obtained. If yes, an early warning signal is issued, otherwise, continuous monitoring is carried out.

[0092] On the other hand, the present invention further provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-mentioned dust concentration warning methods.

[0093] On the other hand, the present invention also provides a terminal device, including a memory and a processor; the memory stores a program code that can be executed by the processor; the program code is used to execute any of the above-mentioned dust concentration warning methods.

[0094] Exemplarily, the program code may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the program code in the terminal device.

[0095] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the terminal device may also include an input / output device, a network access device, a bus, etc.

[0096] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0097] The memory may be an internal storage unit of the terminal device, such as a hard disk or a memory. The memory may also be an external storage device of the terminal device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the terminal device. Further, the memory may also include both an internal storage unit of the terminal device and an external storage device. The memory is used to store the program code and other programs and data required by the terminal device. The memory may also be used to temporarily store data that has been output or is to be output.

[0098] The above-mentioned computer storage medium and terminal device are created based on the above-mentioned dust concentration warning method, and their technical functions and beneficial effects are no longer repeated here. The various technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the various technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0099] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A dust concentration early warning method, characterized in that: include: Construct and train an early warning model that takes historical concentration data at several past time points as input and issues an early warning result as output, including an input feature extraction module, a prediction module, a time series feature extraction module, and an early warning output module; An input feature extraction module is used to input historical concentration data at several time points in the past to extract historical features; The prediction module is connected to the output of the input feature extraction module, and is used to input historical features and make predictions based on them to obtain predicted concentration data at several future time points; A time series feature extraction module, connected to the input of the input feature extraction module and the output of the prediction module, for inputting historical concentration data at a number of past time points and predicted concentration data at a number of future time points to extract time series features; The warning output module is connected to the output of the time series feature extraction module, and is used to input the time series features and output the warning results accordingly; Obtain historical concentration data for several past time points in the current scene and input it into the trained warning model to determine whether to issue a warning; An input feature extraction module includes: a first convolutional layer, a first pooling layer, and a hidden layer; The first convolutional layer is used to input the historical concentration data at several time points in the past and extract the local feature; The first pooling layer, whose input is connected to the output of the first convolutional layer, is used to reduce the local feature size to extract the key features of the local features; The hidden layer, whose input is connected to the output of the first pooling layer, is used to learn the complex relationship between features to extract historical features; The prediction module includes: a second convolutional layer, a second pooling layer, and a first fully connected layer; The second convolutional layer, whose input is connected to the output of the hidden layer, is used to extract higher-level features of the historical features; The second pooling layer, whose input is connected to the output of the second convolutional layer, is used to reduce the size of higher-level features to extract key features of higher-level features; The first fully connected layer, whose input is connected to the output of the second convolutional layer, is used to integrate the features extracted by the previous layers for the final regression task to obtain the predicted concentration data at several future time points; A temporal feature extraction module, including a third convolutional layer, a random dropout layer, a residual unit, a fourth convolutional layer, and a third pooling layer; The third convolutional layer, whose input is connected to the input of the first convolutional layer and the output of the first fully connected layer, is used to input the historical concentration data at several past time points and the predicted concentration data at several future time points to generate comprehensive features; A residual module, whose input is connected to the output of the input feature extraction module, and whose output is connected to the input of the random activation layer to pass the historical features to the random activation layer; The random dropout layer, whose input is connected to the output of the third convolutional layer and the output of the residual module, is used to input historical features and comprehensive features, and perform preprocessing to obtain fused features; The fourth convolutional layer, whose input is connected to the output of the random dropout layer, is used to input the processed fusion features and further extract deeper features of the fusion features; The third pooling layer, whose input is connected to the output of the fourth convolutional layer, is used to reduce the spatial size of deeper features to obtain the final temporal features.

2. The dust concentration early warning method according to claim 1, characterized in that: The warning output module includes a second fully connected layer, which is used to complete the final classification task according to the time series characteristics and output 0 or 1; 0 means no warning and 1 means warning.

3. The dust concentration early warning method according to claim 2, characterized in that: The first convolution layer includes N channels; the third convolution layer includes N+M channels, which are used to input the historical concentration data of a time point or the predicted concentration data of a time point respectively; each channel configures a higher weight coefficient for the historical concentration data or / and predicted concentration data close to the current time according to the order of the time points; N is the number of time points of the historical concentration data; M is the number of time points of the predicted concentration data.

4. The dust concentration early warning method according to any one of claims 1 to 3, characterized in that: During the model training process, the regression loss function of the prediction module uses the mean absolute error (MAE), and the classification loss function of the warning output module uses the logarithmic hyperbolic cosine (Log Cosh).

5. The dust concentration early warning method according to claim 4, characterized in that: The loss function of the entire model is: L=2×MAE+Log Cosh.

6. A computer storage medium, characterized in that: An executable program code is stored; the executable program code is used to execute the dust concentration early warning method described in any one of claims 1-5.

7. A terminal device, characterized in that: It comprises a memory and a processor; the memory stores a program code executable by the processor; the program code is used to execute the dust concentration early warning method described in any one of claims 1-5.

Images