A Deep Learning-Based Air Quality Monitoring Method

Abstract

The application discloses an air quality monitoring method based on deep learning, and relates to the technical field of air quality monitoring.The historical sample data and corresponding air quality labels are constructed through historical monitoring data of air quality; an air quality prediction model is constructed by using a deep learning model; the air quality prediction model is trained according to the historical sample data and corresponding air quality labels and by using a multi-strategy guided optimization algorithm to obtain the trained air quality prediction model; real-time sample data with the same data structure as the historical sample data are collected, and the trained air quality prediction model is used to identify the real-time sample data to obtain air quality prediction information; an air quality monitoring report is generated according to the air quality prediction information, and the air quality monitoring report is fed back to the equipment designated by the staff, the prediction of air quality is realized through the deep learning technology, and the timeliness of air quality monitoring is improved.

Description

Technical Field

[0001] This application relates to the field of air quality monitoring technology, and more specifically, to an air quality monitoring method based on deep learning. Background Technology

[0002] With the continuous advancement of urbanization and the rapid development of urban construction and transportation, factors such as industrial emissions, vehicle exhaust, and heating emissions have brought significant impacts to the urban ecological environment, making air pollution an increasingly pressing concern. Constructing effective air quality prediction models is crucial for pollution early warning and prevention, holding significant practical importance. However, because air quality changes are influenced by multiple factors including meteorology, pollutant emissions, and geographical location, traditional air quality monitoring typically utilizes specialized instruments (such as spectroscopic and electrochemical methods) to precisely measure the concentrations of pollutants like PM2.5, PM10, SO2, NO2, CO, and O3. While this method offers the advantage of accurate monitoring data, it often suffers from a delay of approximately one hour, making it difficult to meet the needs of real-time early warning. Summary of the Invention

[0003] This application aims to provide a deep learning-based air quality monitoring method, which addresses the problem that existing technologies cannot meet the requirement of real-time early warning.

[0004] This application provides a deep learning-based air quality monitoring method, including:

[0005] Collect historical air quality monitoring data, and construct historical sample data and corresponding air quality labels based on the historical air quality monitoring data;

[0006] An air quality prediction model is constructed using a deep learning model. Based on the historical sample data and the corresponding air quality labels, the air quality prediction model is trained using a multi-strategy guided optimization algorithm to obtain the trained air quality prediction model.

[0007] Real-time sample data with the same data structure as the historical sample data is collected, and the trained air quality prediction model is used to identify the real-time sample data to obtain air quality prediction information.

[0008] An air quality monitoring report is generated based on the air quality forecast information, and the air quality monitoring report is fed back to the equipment designated by the staff to complete the air quality monitoring based on deep learning.

[0009] In one possible implementation, historical air quality monitoring data is collected, including:

[0010] Based on a preset data sampling frequency, historical air quality monitoring data for multiple sampling time points are obtained by collecting PM2.5 concentration, PM10 concentration, SO2 concentration, NO2 concentration, O3 concentration, CO concentration, air temperature, air pressure, wind speed, relative humidity, wind direction, and rainfall.

[0011] In one possible implementation, historical sample data and corresponding air quality labels are constructed based on the historical air quality monitoring data, including:

[0012] Obtain the air quality monitoring indicators specified by the staff; wherein the air quality monitoring indicators are one of PM2.5 concentration, PM10 concentration, SO2 concentration, NO2 concentration, O3 concentration or CO concentration;

[0013] Historical sample data is constructed using historical monitoring data at any N consecutive sampling time points, and the air quality monitoring index at the N+1th sampling time point is used as the air quality label corresponding to the historical sample data.

[0014] Repeatedly obtain multiple different historical sample data and the corresponding air quality labels for the historical sample data.

[0015] In one possible implementation, an air quality prediction model is constructed using a deep learning model, including: constructing an air quality prediction model using a CNN-LSTM model.

[0016] In one possible implementation, the air quality prediction model is trained using a multi-strategy guided optimization algorithm based on the historical sample data and the corresponding air quality labels, resulting in a trained air quality prediction model, including:

[0017] Based on the hyperparameters of the air quality prediction model, a training population is generated; wherein, the training population includes multiple different individuals; each individual includes all the hyperparameters of the air quality prediction model to be trained.

[0018] For any given individual, the individual's fitness is obtained based on the historical sample data and the corresponding air quality labels.

[0019] The individual with the highest fitness is identified as the current best individual, and the current worst individual is identified as the current worst individual.

[0020] Based on the fitness of the current best individual and the fitness of the current worst individual, a multi-point guided search strategy is used to perform multi-information search on the individual to obtain the individual after multi-information search;

[0021] A chain-like memory decay-guided search strategy is used to perform information fusion search on the individuals after the multi-information search to obtain the individuals after information fusion search.

[0022] An adaptive global search is performed on the individuals after the information fusion search using a quantum-guided search strategy to obtain the individuals after the adaptive global search.

[0023] Determine the current number of training iterations;

[0024] Based on the current number of training iterations and the preset maximum number of training iterations, a training phase is obtained; wherein, the training phase includes a continue training phase or a complete training phase;

[0025] If the training phase is a continuation training phase, then the individual after the adaptive global search is used for the next training.

[0026] If the training phase is completed, the individual after the adaptive global search will redetermine the current optimal individual to obtain the target individual;

[0027] Based on the target individual, obtain the trained air quality prediction model.

[0028] In one possible implementation, for any given individual, the fitness of the individual is obtained based on the historical sample data and the corresponding air quality labels, including:

[0029] For any individual, the hyperparameters contained in the individual are applied to the air quality prediction model to obtain the air quality prediction model after applying the hyperparameters.

[0030] The input of the air quality prediction model after applying hyperparameters is constructed using the historical sample data, and the expected output of the air quality prediction model after applying hyperparameters is constructed using the air quality labels corresponding to the historical sample data, and the loss function value corresponding to each individual is obtained.

[0031] Add the loss function value corresponding to the individual to a preset constant value to obtain the non-zero value term corresponding to the individual;

[0032] The fitness of an individual is obtained by taking the reciprocal of the non-zero numerical value corresponding to that individual.

[0033] In one possible implementation, based on the fitness of the current best individual and the fitness of the current worst individual, a multi-point guided search strategy is used to perform a multi-information search on the individuals to obtain individuals after the multi-information search, including:

[0034] Based on the fitness of the current best individual and the fitness of the current worst individual, determine the guidance strength parameter of each individual in the solution space;

[0035] Based on the guidance strength parameter of each individual in the solution space, determine the guidance factor between any individual and other individuals;

[0036] For any given individual, obtain multi-point guided search volume based on the guiding factors between that individual and other individuals;

[0037] Based on the multi-point guided search volume and the guidance intensity parameter, a multi-information search is performed on the individual to obtain the individual after the multi-information search.

[0038] In one possible implementation, a chain-like memory decay-guided search strategy is used to perform an information fusion search on the individuals after the multi-information search, resulting in individuals after the information fusion search, including:

[0039] The individuals obtained after the multi-information search are arranged in descending order of fitness to obtain the arranged individuals and their corresponding serial numbers;

[0040] Based on the current number of training iterations, and based on the current number of training iterations and the individual's corresponding sequence number, determine the update dimension corresponding to the sorted individuals;

[0041] Based on the fitness of the current best individual and the fitness of the current worst individual, determine the corresponding memory decay factor for the individuals after the ranking.

[0042] Based on the updated dimension corresponding to the individuals after the arrangement and the memory decay factor, an information fusion search is performed on the individuals after the arrangement using a chain structure to obtain the individuals after information fusion search.

[0043] In one possible implementation, an adaptive global search is performed on the individuals obtained after the information fusion search using a quantum-guided search strategy, resulting in individuals after the adaptive global search, including:

[0044] Obtain the best neighboring individual of the individual after information fusion search;

[0045] Determine the first Euclidean distance between the current best individual and the individuals after information fusion search, and confirm the second Euclidean distance between the individual after information fusion search and its corresponding neighborhood best individual;

[0046] Based on the current best individual, the best individual in the neighborhood, the first Euclidean distance, and the second Euclidean distance, the quantum-guided search parameters are obtained using an exponential function;

[0047] Based on the quantum-guided search parameters, an adaptive global search is performed on the individuals obtained after the information fusion search to obtain the individuals obtained after the adaptive global search.

[0048] In one possible implementation, generating an air quality monitoring report based on the air quality forecast information and feeding the air quality monitoring report back to a device designated by the staff includes:

[0049] An air quality monitoring report is generated using a preset fixed template based on the air quality forecast information and the corresponding real-time sample data, and the air quality monitoring report is fed back to the device designated by the staff.

[0050] Beneficial effects:

[0051] This application provides a deep learning-based air quality monitoring method. It constructs historical sample data and corresponding air quality labels from historical air quality monitoring data; it then uses a deep learning model to build an air quality prediction model, training the model using a multi-strategy guided optimization algorithm based on the historical sample data and its corresponding air quality labels to obtain a trained air quality prediction model; it collects real-time sample data with the same data structure as the historical sample data, and uses the trained air quality prediction model to identify the real-time sample data, obtaining air quality prediction information; it generates an air quality monitoring report based on the air quality prediction information and feeds the report back to a device designated by the staff. By using deep learning technology to predict air quality, the timeliness of air quality monitoring is improved. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of an air quality monitoring method based on deep learning proposed in one embodiment of this application.

[0054] Figure 2 This is a flowchart illustrating the acquisition of a trained air quality prediction model according to an embodiment of this application. Detailed Implementation

[0055] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0056] like Figure 1 As shown in the figure, this application provides a deep learning-based air quality monitoring method, including:

[0057] S101. Collect historical air quality monitoring data, and construct historical sample data and corresponding air quality labels based on the historical air quality monitoring data.

[0058] In one possible implementation, historical air quality monitoring data is collected, including:

[0059] Based on a preset data sampling frequency, PM2.5 concentration, PM10 concentration, SO2 concentration, NO2 concentration, O3 concentration, CO concentration, air temperature, air pressure, wind speed, relative humidity, wind direction, and rainfall are collected to obtain historical air quality monitoring data for multiple sampling time points. It is worth noting that all the above data are expressed numerically and are only preferred examples in this application embodiment; other air quality-related data may also be included.

[0060] In one possible implementation, historical sample data and corresponding air quality labels are constructed based on the historical air quality monitoring data, including:

[0061] Obtain the air quality monitoring indicators specified by the staff; wherein the air quality monitoring indicators are one of PM2.5 concentration, PM10 concentration, SO2 concentration, NO2 concentration, O3 concentration or CO concentration;

[0062] Optionally, PM2.5 concentration is usually a major indicator that has a significant impact on cities. Therefore, in this embodiment, the air quality monitoring indicator can be directly set to PM2.5 concentration, and staff are not allowed to specify it. This can also achieve the prediction of air quality.

[0063] Historical sample data is constructed using historical monitoring data at any N consecutive sampling time points, and the air quality monitoring index at the N+1th sampling time point is used as the air quality label corresponding to the historical sample data.

[0064] Repeatedly obtain multiple different historical sample data and the corresponding air quality labels for the historical sample data.

[0065] S102. An air quality prediction model is constructed using a deep learning model. Based on the historical sample data and the air quality labels corresponding to the historical sample data, the air quality prediction model is trained using a multi-strategy guided optimization algorithm to obtain the trained air quality prediction model.

[0066] In one possible implementation, an air quality prediction model is constructed using a deep learning model, including: constructing an air quality prediction model using a CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory) model.

[0067] In existing technologies, deep learning models are prone to getting trapped in local optima during training, which often leads to the model failing to accurately learn the correlation between sample data and labels, ultimately resulting in the trained model being unable to accurately complete the data recognition task. Therefore, this application employs a multi-strategy guided optimization algorithm to train the air quality prediction model to improve the training effect and thus enhance the prediction accuracy of the air quality prediction model.

[0068] S103. Collect real-time sample data with the same data structure as the historical sample data, and use the trained air quality prediction model to identify the real-time sample data to obtain air quality prediction information.

[0069] The trained air quality prediction model has the ability to identify data. In order to ensure the recognition effect, it is necessary to collect real-time sample data with the same data structure as the historical sample data. Then, the real-time sample data is used as input to obtain air quality prediction information.

[0070] Optionally, both historical and real-time sample data can be normalized before use, which can not only reduce data complexity but also improve data recognition accuracy.

[0071] S104. Generate an air quality monitoring report based on the air quality prediction information, and feed the air quality monitoring report back to the device designated by the staff to complete the deep learning-based air quality monitoring.

[0072] Air quality forecast information is essentially a prediction of air quality monitoring indicators. A warning threshold can be set in advance, and an early warning can be issued if the air quality forecast information exceeds the warning threshold.

[0073] like Figure 2 As shown, based on the historical sample data and the corresponding air quality labels, the air quality prediction model is trained using a multi-strategy guided optimization algorithm to obtain the trained air quality prediction model, including:

[0074] S201. Based on the hyperparameters of the air quality prediction model, a training population is generated; wherein, the training population includes multiple different individuals; each individual includes all the hyperparameters to be trained in the air quality prediction model;

[0075] For example, for a CNN-LSTM model, the hyperparameters to be trained can be the connection weights between its network layers. These hyperparameters to be trained have upper and lower bounds. They can be randomly initialized between the upper and lower bounds of the hyperparameters to be trained, and the randomly initialized hyperparameters to be trained can be used to form a vector to obtain individuals. Multiple individuals can be repeatedly obtained to form a training population.

[0076] S202. For any individual, obtain the individual's fitness based on the historical sample data and the air quality label corresponding to the historical sample data;

[0077] In one possible implementation, for any given individual, the fitness of the individual is obtained based on the historical sample data and the corresponding air quality labels, including:

[0078] For any individual, the hyperparameters contained in the individual are applied to the air quality prediction model to obtain the air quality prediction model after applying the hyperparameters.

[0079] The input to the air quality prediction model after applying hyperparameters is constructed using the historical sample data, and the expected output of the air quality prediction model after applying hyperparameters is constructed using the air quality labels corresponding to the historical sample data, and the loss function value corresponding to each individual is obtained; such as the cross-entropy loss function value or the root mean square loss function value.

[0080] Add the loss function value corresponding to the individual to a preset constant value to obtain the non-zero value term corresponding to the individual;

[0081] The fitness of an individual is obtained by taking the reciprocal of the non-zero numerical value corresponding to that individual.

[0082] Optionally, the default constant value can be set to 0.001.

[0083] S203. Determine the individual with the highest fitness as the current best individual and the current worst individual;

[0084] S204. Based on the fitness of the current best individual and the fitness of the current worst individual, a multi-point guided search strategy is used to perform multi-information search on the individual to obtain the individual after multi-information search.

[0085] S205. A chain-like memory decay-guided search strategy is used to perform information fusion search on the individuals after the multi-information search to obtain the individuals after information fusion search.

[0086] S206. Adaptive global search is performed on the individuals after the information fusion search using a quantum-guided search strategy to obtain the individuals after the adaptive global search.

[0087] S207. Determine the current number of training iterations;

[0088] S208. Based on the current number of training iterations and the preset maximum number of training iterations, obtain the training stage; wherein, the training stage includes a continue training stage or a complete training stage;

[0089] If the current number of training iterations is less than the maximum number of training iterations, then the training phase is determined as the "continue training phase"; otherwise, the training phase is determined as the "complete training phase".

[0090] S209. If the training phase is a continued training phase, then the individual after the adaptive global search is used for the next training.

[0091] S210. If the training phase is not completed, the individual after the adaptive global search will redetermine the current optimal individual to obtain the target individual.

[0092] S211. Based on the target individual, obtain the trained air quality prediction model.

[0093] The hyperparameters contained in the target individual can be used as the final hyperparameters of the air quality prediction model to obtain the trained air quality prediction model.

[0094] Optionally, after each individual update, boundary crossing can be performed to ensure that the individual lies between its upper and lower limits. For example, it can be randomly generated between the upper and lower limits.

[0095] In one possible implementation, based on the fitness of the current best individual and the fitness of the current worst individual, a multi-point guided search strategy is used to perform a multi-information search on the individuals to obtain individuals after the multi-information search, including:

[0096] Based on the fitness of the current best individual and the fitness of the current worst individual, the guidance strength parameter for each individual in the solution space is determined as follows:

[0097]

[0098]

[0099] in, Indicates the first i The fitness of each individual i =1,2,...,NP, where NP represents the total number of individuals. This indicates the fitness of the worst-performing individual. This represents the fitness of the current best individual. Indicates the first i The degree of fitness corresponding to each individual Indicates the first j The degree of fitness corresponding to each individual Indicates the first i Guidance strength parameters corresponding to each individual;

[0100] Based on the guidance strength parameter of each individual in the solution space, the guidance factor between any individual and other individuals is determined as follows:

[0101]

[0102] in, Indicates the first t During the training process, the first i Individual and the first j The individual in the first d Guiding factors on the hyperparameters, d =1,2,...,D, where D represents the total dimension of hyperparameters in an individual. This represents the base quantity for bootstrapping and is set to 100. Represents the natural constant. This represents the first preset constant term, which is set to 20; T represents the preset maximum number of training iterations. Indicates the first j The guidance strength parameters corresponding to each individual. Indicates the first i Individual and the first j Euclidean distance between individuals Indicates the first t During the training process, the first i The first individual d dimensional hyperparameters, Indicates the first t During the training process, the first j The first individual d dimensional hyperparameters;

[0103] For any given individual, based on the guiding factors between that individual and other individuals, the multi-point guided search volume is obtained as follows:

[0104]

[0105] in, Indicates the first t During the training process, the first j The first individual d The multi-point guided search volume corresponding to the dimensional hyperparameter. Represented as the first j A random number between (0, 1) generated by each individual;

[0106] Based on the multi-point guided search volume and guidance strength parameters, a multi-information search is performed on the individual to obtain the individual after the multi-information search:

[0107]

[0108]

[0109] in, Indicates the first i The individual's first after multiple information searches d dimensional hyperparameters, Indicates the first t During the training process, the first i The search speed of each individual d dimensional hyperparameters, Indicates the first t+ During the first training session i The search speed of each individual d dimensional hyperparameters, This represents the first random number between (0, 1).

[0110] By employing a multi-point guided search strategy to perform multi-information search on the individuals, the gravitational pull of multiple individuals can be effectively utilized to achieve multi-point guided search. This allows for searching more solution space in the early stages of the algorithm. Then, as all individuals gradually gather near the optimal position, the search automatically transforms into a local search, balancing the global and local search aspects of the algorithm.

[0111] In one possible implementation, a chain-like memory decay-guided search strategy is used to perform an information fusion search on the individuals after the multi-information search, resulting in individuals after the information fusion search, including:

[0112] The individuals obtained after the multi-information search are arranged in descending order of fitness to obtain the arranged individuals and their corresponding serial numbers;

[0113] Based on the current number of training iterations, and according to the individual's corresponding index, the update dimension for the sorted individuals is determined as follows:

[0114]

[0115] in, Indicates the first m The update dimension corresponding to each individual after the permutation. m =1,2,...,NP Indicates rounding up. This represents an exponential function with base e. This represents the disturbance factor, and is set to 0.1; Represents a second random number between (0,1);

[0116] Based on the fitness of the current best individual and the fitness of the current worst individual, the memory decay factor corresponding to the individuals after the permutation is determined as follows:

[0117]

[0118] in, Indicates the first m The memory decay factor corresponding to each individual after the permutation, Indicates the first m The fitness of the individuals after each permutation;

[0119] Based on the update dimension corresponding to the arranged individuals and the memory decay factor, an information fusion search is performed on the arranged individuals using a chain structure to obtain the individuals after information fusion search, including:

[0120] Based on the update dimension corresponding to the individuals after the arrangement, randomly select from the individuals after the arrangement. There are several dimensions to be searched, and based on the memory decay factor, an information fusion search is performed on the dimensions to be searched for the individuals after the chaining structure is adopted, to obtain the individuals after information fusion search:

[0121]

[0122] in, Indicates the first t During the training process, the first m The first individual after the permutation s One hyperparameter for the dimension to be searched. Indicates the first m The first individual after information fusion search s There are 10 hyperparameters for the search dimension. The non-search dimensions in the individuals after sorting remain unchanged before and after the search. This represents a third random number between (0,1). This represents the fourth random number between (0,1). Indicates the current optimal individual and Hyperparameters of the corresponding dimension Indicates the first t During the training process, the first m The -1th individual in the permutation s One hyperparameter for the dimension to be searched.

[0123] Employing a chain-like memory decay-guided search strategy to perform information fusion search on individuals after the multi-information search further enhances the solution space exploration capability. Individuals in better positions learn more and better information, effectively improving the training speed of the algorithm. Conversely, individuals in worse positions only remember part of their own information and learn less, effectively ensuring population diversity and improving the algorithm's ability to find the global optimum. As the population gradually converges in the later stages of the algorithm, this chain-like memory decay-guided search strategy also ensures training accuracy.

[0124] In one possible implementation, an adaptive global search is performed on the individuals obtained after the information fusion search using a quantum-guided search strategy, resulting in individuals after the adaptive global search, including:

[0125] Obtain the best neighboring individual of the individual after information fusion search;

[0126] A neighborhood radius can be preset, and then the Euclidean distance between any individual after information fusion search and other individuals after information fusion search can be obtained. If the Euclidean distance is less than the neighborhood radius, then other individuals can be considered as neighborhood individuals of any individual. Then, the individual with the highest fitness among the neighborhood individuals is found to obtain the optimal neighborhood individual corresponding to the individual after information fusion search.

[0127] Determine the first Euclidean distance between the current best individual and the individuals after information fusion search, and confirm the second Euclidean distance between the individual after information fusion search and its corresponding neighborhood best individual;

[0128] Based on the current best individual, the best neighboring individual, the first Euclidean distance, and the second Euclidean distance, the quantum-guided search parameters are obtained using an exponential function:

[0129]

[0130] in, Indicates the first t +1 training session n The quantum-guided search parameters corresponding to the d-th hyperparameter of an individual after information fusion search. Indicates the first t During the training process, the first n The d-th dimension hyperparameter of an individual after information fusion search. n =1,2,...,NP This represents the global boot control parameter value, which can be set to 1.5; This indicates the value of the neighborhood guidance control parameter, which can be set to 1; This represents the visibility coefficient, and is set to 2. Indicates the first Euclidean distance. Indicates the second Euclidean distance. The d-th dimension hyperparameter represents the optimal individual in the neighborhood;

[0131] It is worth noting that in this embodiment of the application, the process of generating individuals is taken as the first training process, and the process of starting training with the formula is taken as the second training process. The training is repeated continuously until the training ends.

[0132] Based on the quantum-guided search parameters, an adaptive global search is performed on the individuals obtained after the information fusion search, resulting in the following individuals after the adaptive global search:

[0133]

[0134]

[0135] in, Indicates the first n The d-th dimension hyperparameter of an individual after an adaptive global search. Indicates the first t During the training process, the first n The search speed of the individual after information fusion search is the first. d dimensional hyperparameters, Indicates the first t +1 training session n The search speed of the individual after information fusion search is the first. d dimensional hyperparameters, Indicates the absolute value sign. Represents the cosine function. sin This represents the sine function.

[0136] Optionally, global search may slow down algorithm training; therefore, a greedy strategy or simulated annealing can be used to control the global search process. For example, simulated annealing can be used to probabilistically accept inferior solutions.

[0137] The quantum-guided search strategy is used to perform an adaptive global search on the individuals after the information fusion search. This can effectively help the algorithm escape local optima. As the population gathers in the later stages of the algorithm, the search speed decreases, which can ensure the convergence of the algorithm.

[0138] In one possible implementation, generating an air quality monitoring report based on the air quality forecast information and feeding the air quality monitoring report back to a device designated by the staff includes:

[0139] An air quality monitoring report is generated using a preset fixed template based on the air quality forecast information and the corresponding real-time sample data, and the air quality monitoring report is fed back to the device designated by the staff.

[0140] For example, a fixed template can be preset, which includes a field for filling in air quality forecast information and a field for filling in real-time sample data corresponding to the air quality forecast information. By filling in the data in the corresponding fields, an air quality monitoring report can be generated, enabling staff to process it quickly.

[0141] This application provides a deep learning-based air quality monitoring method. It constructs historical sample data and corresponding air quality labels from historical air quality monitoring data; it then uses a deep learning model to build an air quality prediction model, training the model using a multi-strategy guided optimization algorithm based on the historical sample data and its corresponding air quality labels to obtain a trained air quality prediction model; it collects real-time sample data with the same data structure as the historical sample data, and uses the trained air quality prediction model to identify the real-time sample data, obtaining air quality prediction information; it generates an air quality monitoring report based on the air quality prediction information and feeds the report back to a device designated by the staff. By using deep learning technology to predict air quality, the timeliness of air quality monitoring is improved.

[0142] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0143] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0144] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0145] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0146] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0147] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0148] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A deep learning-based air quality monitoring method, characterized in that, The method comprises the following steps: Collect historical monitoring data of air quality, and construct historical sample data and air quality labels corresponding to the historical sample data according to the historical monitoring data of air quality; A deep learning model is used to construct an air quality prediction model, and the air quality prediction model is trained according to the historical sample data and the air quality labels corresponding to the historical sample data by using a multi-strategy guided optimization algorithm to obtain the trained air quality prediction model, which comprises the following steps: Based on the hyperparameters of the air quality prediction model, a training population containing multiple individuals is generated; wherein the individuals include all the to-be-trained hyperparameters of the air quality prediction model; The fitness of each individual is obtained according to the historical sample data and the corresponding air quality labels, and the current optimal individual and the current worst individual are determined according to the fitness; According to the fitness of the current optimal individual and the fitness of the current worst individual, a multi-point guided search strategy is used to perform multi-information search on the individuals to obtain the individuals after multi-information search; The chain memory decay guided search strategy is used to perform information fusion search on the individuals after multi-information search, which comprises the following steps: the individuals after multi-information search are arranged in descending order of fitness to obtain the arranged individuals and their corresponding serial numbers; the update dimension of the arranged individuals is determined according to the current training number and the serial numbers of the individuals corresponding to the current training number; the memory decay factor of the arranged individuals is determined according to the fitness of the current optimal individual and the fitness of the current worst individual; the arranged individuals are subjected to information fusion search by using a chain structure according to the update dimension of the arranged individuals and the memory decay factor, to obtain the individuals after information fusion search; The quantum guided search strategy is used to perform adaptive global search on the individuals after information fusion search, which comprises the following steps: the neighborhood optimal individual corresponding to the individual after information fusion search is obtained; the first Euclidean distance between the current optimal individual and the individual after information fusion search is determined, and the second Euclidean distance between the individual after information fusion search and its corresponding neighborhood optimal individual is determined; the quantum guided search parameter is obtained by using an exponential function according to the current optimal individual, the neighborhood optimal individual, the first Euclidean distance and the second Euclidean distance; the adaptive global search is performed on the individual after information fusion search according to the quantum guided search parameter to obtain the individual after adaptive global search; It is judged whether the training number is greater than the maximum training number, if yes, the individual after adaptive global search is used for next training, otherwise, the current optimal individual is re-determined according to the individual after adaptive global search to obtain a target individual, and the trained air quality prediction model is obtained according to the target individual; Real-time sample data with the same data structure as the historical sample data are collected, and the trained air quality prediction model is used to identify the real-time sample data to obtain air quality prediction information; An air quality monitoring report is generated according to the air quality prediction information, and the air quality monitoring report is fed back to the equipment designated by the staff to complete the air quality monitoring based on deep learning. 2.The deep learning-based air quality monitoring method of claim 1, wherein, The historical monitoring data of air quality is collected, including: On the basis of a preset data sampling frequency, the PM2.5 concentration, the PM10 concentration, the SO2 concentration, the NO2 concentration, the O3 concentration, the CO concentration, the air temperature, the air pressure, the wind speed, the relative humidity, the wind direction and the rainfall are collected to obtain the historical monitoring data of air quality corresponding to a plurality of sampling time points. 3.The deep learning-based air quality monitoring method of claim 2, wherein, The historical sample data and the air quality labels corresponding to the historical sample data are constructed according to the historical monitoring data of air quality, including: An air quality monitoring index specified by a worker is obtained; wherein the air quality monitoring index is one of the PM2.5 concentration, the PM10 concentration, the SO2 concentration, the NO2 concentration, the O3 concentration or the CO concentration; The historical sample data is constructed with the historical monitoring data at any N consecutive sampling time points, and the air quality monitoring index at the N+1 sampling time point is taken as the air quality label corresponding to the historical sample data; The plurality of different historical sample data and the air quality labels corresponding to the historical sample data are repeatedly obtained.

4. The deep learning-based air quality monitoring method of claim 3, wherein, The air quality prediction model is constructed by using a deep learning model, including: the CNN-LSTM model is used to construct the air quality prediction model. 5.The deep learning-based air quality monitoring method of claim 1, wherein, For any individual, the fitness of the individual is obtained according to the historical sample data and the air quality labels corresponding to the historical sample data, including: For any individual, the hyperparameters contained in the individual are applied to the air quality prediction model to obtain the air quality prediction model after applying the hyperparameters; The input of the air quality prediction model after applying the hyperparameters is constructed with the historical sample data, and the expected output of the air quality prediction model after applying the hyperparameters is constructed with the air quality labels corresponding to the historical sample data, to obtain the loss function value corresponding to the individual; The loss function value corresponding to the individual is added to a preset constant value to obtain a non-zero numerical term corresponding to the individual; The fitness of the individual is obtained by taking the reciprocal of the non-zero numerical term corresponding to the individual.

6. The deep learning-based air quality monitoring method of claim 5, wherein, According to the fitness of the current optimal individual and the fitness of the current worst individual, a multi-point guided search strategy is used to search the individual with multiple information to obtain the individual after the multiple information search, including: According to the fitness of the current optimal individual and the fitness of the current worst individual, the guide intensity parameter of each individual in the solution space is determined; According to the guide intensity parameter of each individual in the solution space, the guide factor between any individual and other individuals is determined; For any individual, the multi-point guided search amount is obtained according to the guide factor between the individual and other individuals; According to the multi-point guided search amount and the guide intensity parameter, the individual is searched with multiple information to obtain the individual after the multiple information search. 7.The deep learning-based air quality monitoring method of claim 1, wherein, The air quality monitoring report is generated according to the air quality prediction information, and the air quality monitoring report is fed back to the device specified by the worker, including: The air quality monitoring report is generated by using a preset fixed template on the basis of the air quality prediction information and the real-time sample data corresponding to the air quality prediction information, and the air quality monitoring report is fed back to the device specified by the worker.

Images