Regional atmospheric pollution contribution prediction method based on multi-source data fusion

Abstract

The invention discloses a regional atmospheric pollution contribution prediction method based on multi-source data fusion, and belongs to the technical field of pollution monitoring and data processing. Regional atmospheric pollution influence samples and regional atmospheric pollution contribution training labels are constructed through preprocessed multi-source pollution influence data and environment monitoring data; and then learning the regional atmospheric pollution contribution training sample and the regional atmospheric pollution contribution training label by using a machine learning model to obtain a regional atmospheric pollution contribution prediction model, and finally collecting atmospheric pollution influence data of a to-be-identified region corresponding to the target detection region. The regional atmospheric pollution contribution prediction model is adopted to identify the atmospheric pollution influence data of the to-be-identified region, so that multi-source data can be effectively fused, rapid and accurate pollution prediction is realized by utilizing machine learning, and the atmospheric pollution prediction precision can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the fields of pollution monitoring technology and data processing technology, and specifically relates to a method for predicting regional atmospheric pollution contribution based on multi-source data fusion. Background Art

[0002] Air pollution refers to the emission of various pollutants into the atmosphere due to human activities or natural processes, resulting in changes in the composition and properties of the atmosphere and negatively impacting humans, plants, animals, and the environment. Major pollutants include particulate matter (PM2.5, PM10), sulfur dioxide, nitrogen oxides, carbon monoxide, ozone, and volatile organic compounds (VOCs). These pollutants primarily originate from fossil fuel combustion, industrial emissions, vehicle exhaust, and agricultural activities. Air pollution poses a serious threat to human health, causing respiratory and cardiovascular diseases, and even cancer. It also contributes to environmental problems such as acid rain, the greenhouse effect, and ozone layer depletion, impacting ecological balance and climate change. Therefore, controlling air pollution and improving air quality are crucial for safeguarding human health and sustainable development. With the rapid development of industrialization and urbanization, air pollution is becoming increasingly serious, making accurate prediction of air pollution a critical component of environmental protection. Traditional air pollution prediction methods often rely on a single monitoring data source, which fails to fully capture the sources and transmission mechanisms of pollutants, resulting in limited prediction accuracy. Summary of the Invention

[0003] The present invention provides a method for predicting regional air pollution contribution based on multi-source data fusion, which is used to solve the problem of limited accuracy of air pollution prediction in the existing technology.

[0004] A method for predicting regional air pollution contribution based on multi-source data fusion, including:

[0005] Collecting multi-source pollution impact data and environmental monitoring data corresponding to the target monitoring area, and preprocessing the multi-source pollution impact data and environmental monitoring data to obtain preprocessed multi-source pollution impact data and environmental monitoring data;

[0006] Constructing regional air pollution contribution training data based on the pre-processed multi-source pollution impact data and environmental monitoring data; wherein the regional air pollution contribution training data includes regional air pollution impact samples and regional air pollution contribution training labels;

[0007] A machine learning model is used to learn regional air pollution contribution training samples and regional air pollution contribution training labels to obtain a regional air pollution contribution prediction model;

[0008] The atmospheric pollution impact data of the area to be identified corresponding to the target detection area are collected, and the atmospheric pollution impact data of the area to be identified are identified using a regional atmospheric pollution contribution prediction model to obtain a regional atmospheric pollution contribution prediction result.

[0009] In one possible implementation, collecting multi-source pollution impact data and environmental monitoring data corresponding to the target monitoring area includes:

[0010] Based on the preset data sampling frequency, meteorological data and pollution source data corresponding to the target monitoring area are collected at each time point to obtain multi-source pollution impact data;

[0011] Based on the preset data sampling frequency, the pollutant concentration data corresponding to the target monitoring area is collected at each time point to obtain environmental monitoring data.

[0012] In a possible implementation, preprocessing the multi-source pollution impact data and the environmental monitoring data to obtain the preprocessed multi-source pollution impact data and the environmental monitoring data includes:

[0013] Data cleaning, missing value filling and / or outlier processing are performed on the multi-source pollution impact data and environmental monitoring data to obtain pre-processed multi-source pollution impact data and environmental monitoring data.

[0014] In one possible implementation, constructing regional air pollution contribution training data based on the pre-processed multi-source pollution impact data and environmental monitoring data includes:

[0015] According to the pre-processed multi-source pollution impact data and environmental monitoring data, the pre-processed multi-source pollution impact data and environmental monitoring data corresponding to any N consecutive time points are used as regional atmospheric pollution impact samples;

[0016] According to the environmental monitoring data, the pollutant concentration increase rate corresponding to the Nth time point to the N+1th time point is used as the regional air pollution contribution training label.

[0017] In one possible implementation, a machine learning model is used to learn regional air pollution contribution training samples and regional air pollution contribution training labels to obtain a regional air pollution contribution prediction model, including:

[0018] Constructing a machine learning model and initializing hyperparameters of the machine learning model to obtain multiple hyperparameter particles;

[0019] Obtaining the fitness corresponding to each hyperparameter particle according to the regional atmospheric pollution contribution training sample and the regional atmospheric pollution contribution training label;

[0020] According to the fitness corresponding to each hyperparameter particle, the optimal hyperparameter particle is determined among all hyperparameter particles;

[0021] According to the optimal hyperparameter particle, a dual adaptive guidance strategy is adopted to guide and update the hyperparameter particle to obtain the hyperparameter particle after the guidance update;

[0022] A self-regulating information interaction strategy is used to interactively update the hyperparameter particles after the guided update to obtain the hyperparameter particles after the interactive update;

[0023] Adopting the adaptive differential evolution strategy to globally update the hyperparameter particles after interactive update, and obtain the hyperparameter particles after global update;

[0024] During each training process, the dual adaptive guidance strategy, self-adjusting information interaction strategy, and adaptive differential evolution strategy are repeatedly executed until the training end conditions are met. The optimal hyperparameter particle is re-determined based on the hyperparameter particle after the global update in the last training process.

[0025] Based on the re-determined optimal hyperparameter particles, the final hyperparameters of the machine learning model are obtained, and the regional atmospheric pollution contribution prediction model is obtained.

[0026] In a possible implementation, the method further includes: after executing any one of the dual adaptive guidance strategy, the self-adjusting information interaction strategy, and the adaptive differential evolution strategy, performing out-of-bounds processing on the hyperparameter particles.

[0027] In a possible implementation, obtaining the fitness corresponding to each hyperparameter particle according to the regional air pollution contribution training sample and the regional air pollution contribution training label includes:

[0028] For any hyperparameter particle, apply the hyperparameters contained in the hyperparameter particle to the machine learning model to obtain the machine learning model after applying the hyperparameters;

[0029] Using the regional air pollution contribution training sample as input data of the machine learning model after applying hyperparameters, using the regional air pollution contribution training label as the expected output of the machine learning model after applying the hyperparameters, and obtaining a loss function value;

[0030] After adding the loss function value to the preset constant term, a non-zero parameter value is obtained, and the reciprocal of the non-zero parameter value is taken to obtain the fitness corresponding to the hyperparameter particle;

[0031] Traverse each hyperparameter particle and obtain the fitness corresponding to each hyperparameter particle.

[0032] In a possible implementation, based on the optimal hyperparameter particle, a dual adaptive guidance strategy is used to guide and update the hyperparameter particle, and the hyperparameter particle after the guidance update is obtained, including:

[0033] For any hyperparameter particle, determine the Euclidean distance between the hyperparameter particle and other hyperparameter particles, and determine the neighboring particles corresponding to the hyperparameter particle based on the Euclidean distance;

[0034] Obtaining a neighborhood comprehensive position and a first influence factor corresponding to the neighborhood comprehensive position according to the neighborhood particle corresponding to the hyperparameter particle;

[0035] According to the optimal hyperparameter particle, obtaining a second influence factor corresponding to the optimal hyperparameter particle;

[0036] Obtaining a guided update speed corresponding to the hyperparameter particle according to the neighborhood comprehensive position, the first influence factor corresponding to the neighborhood comprehensive position, and the second influence factor corresponding to the optimal hyperparameter particle;

[0037] The hyperparameter particle is guided and updated according to the guidance update speed corresponding to the hyperparameter particle to obtain the hyperparameter particle after the guidance update.

[0038] In a possible implementation, a self-regulating information interaction strategy is used to interactively update the hyperparameter particles after the guided update, and the hyperparameter particles after the interactive update are obtained, including:

[0039] For any hyperparameter particle after the guide update, match the hyperparameter particle with two other hyperparameter particles to obtain a first information interaction particle and a second information interaction particle;

[0040] Obtaining a first information interaction factor according to the fitness corresponding to the first information interaction particle; obtaining a second information interaction factor according to the fitness corresponding to the second information interaction particle;

[0041] Get the adaptive inertia weight based on the current number of training times;

[0042] According to the first information interaction factor, the second information interaction factor, and the adaptive inertia weight, a sine and cosine function is used to interactively update the hyperparameter particles after the guided update to obtain the hyperparameter particles after the interactive update.

[0043] In a possible implementation, an adaptive differential evolution strategy is used to globally update the hyperparameter particles after interactive update, and the globally updated hyperparameter particles are obtained, including:

[0044] Based on the current number of training times, obtain the differential evolution factor;

[0045] For any hyperparameter particle after interactive update, randomly match the hyperparameter particle with two other hyperparameter particles to obtain the first differential evolution particle and the second differential evolution particle;

[0046] Performing differential evolution on the hyperparameter particles after interactive update according to the first differential evolution particles, the second differential evolution particles, and the differential evolution factor to obtain differential evolution particles;

[0047] According to the differential evolution particles, the hyperparameter particles after interactive update are globally updated to obtain the hyperparameter particles after global update.

[0048] The present invention provides a regional atmospheric pollution contribution prediction method based on multi-source data fusion. The method constructs regional atmospheric pollution impact samples and regional atmospheric pollution contribution training labels by pre-processing multi-source pollution impact data and environmental monitoring data. Then, a machine learning model is used to learn the regional atmospheric pollution contribution training samples and the regional atmospheric pollution contribution training labels to obtain a regional atmospheric pollution contribution prediction model. Finally, atmospheric pollution impact data of the to-be-identified area corresponding to the target detection area is collected, and the regional atmospheric pollution contribution prediction model is used to identify the atmospheric pollution impact data of the to-be-identified area. The method can effectively fuse multi-source data, and use machine learning to achieve fast and accurate pollution prediction, which can effectively improve the accuracy of atmospheric pollution prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0050] Figure 1 A flowchart of a method for predicting regional air pollution contribution based on multi-source data fusion provided by an embodiment of the present invention.

[0051] Figure 2 This is a flowchart of a regional air pollution contribution prediction model provided by an embodiment of the present invention.

[0052] The above drawings illustrate specific embodiments of the present invention, which will be described in more detail below. These drawings and the accompanying description are not intended to limit the scope of the present invention in any way, but rather to illustrate the concept of the present invention to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0053] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0054] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0055] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting regional air pollution contribution based on multi-source data fusion, comprising:

[0056] S101, collecting multi-source pollution impact data and environmental monitoring data corresponding to the target monitoring area, and preprocessing the multi-source pollution impact data and environmental monitoring data to obtain preprocessed multi-source pollution impact data and environmental monitoring data;

[0057] In one possible implementation, collecting multi-source pollution impact data and environmental monitoring data corresponding to the target monitoring area includes:

[0058] Based on the preset data sampling frequency, meteorological data and pollution source data corresponding to the target monitoring area are collected at each time point to obtain multi-source pollution impact data;

[0059] Based on the preset data sampling frequency, the pollutant concentration data corresponding to the target monitoring area is collected at each time point to obtain environmental monitoring data.

[0060] For example, data collection can be performed on a daily basis, or on a monthly or yearly basis, thereby achieving multi-source data collection.

[0061] Meteorological data can include weather-related data such as wind speed, temperature, humidity, and weather conditions. Pollution source data can include information on industrial and traffic emissions (e.g., the number of factories, factory size, and number of active vehicles). To improve forecast accuracy, emission reduction measures can also be incorporated. However, these measures are generally specific policies and require conversion to quantitative data for identification. For example, if a vehicle is subject to license plate restrictions, a parameter can be set to 1; if a vehicle is not subject to license plate restrictions, a parameter can be set to 0.

[0062] The pollutant concentration data may be PM2.5, PM10, SO2 (sulfur dioxide), NO2 (nitrogen dioxide) and the like. The embodiment of the present invention mainly predicts single data, and the staff may select the pollutant concentration data according to the actual situation.

[0063] S102: constructing regional air pollution contribution training data based on the pre-processed multi-source pollution impact data and the environmental monitoring data; wherein the regional air pollution contribution training data includes regional air pollution impact samples and regional air pollution contribution training labels;

[0064] By constructing input and output data pairs, multiple groups of data that can be learned can be formed, which facilitates the subsequent use of machine learning models for data learning.

[0065] S103, using a machine learning model to learn regional air pollution contribution training samples and regional air pollution contribution training labels to obtain a regional air pollution contribution prediction model;

[0066] Traditional algorithms such as particle swarm optimization and gradient descent optimization algorithms can be used to train machine learning models, and after training, a regional air pollution contribution prediction model can be obtained.

[0067] S104: Collect air pollution impact data of the area to be identified corresponding to the target detection area, and use a regional air pollution contribution prediction model to identify the air pollution impact data of the area to be identified to obtain a regional air pollution contribution prediction result.

[0068] The data structure of the atmospheric pollution impact data for the area to be identified should be the same as the data structure of the regional atmospheric pollution impact samples to ensure that the data can be identified normally.

[0069] It is worth noting that the multi-source pollution impact data are mainly some examples of the embodiments of the present invention. Some other data can also be used to construct the multi-source pollution impact data to achieve more accurate pollution prediction.

[0070] In a possible implementation, preprocessing the multi-source pollution impact data and the environmental monitoring data to obtain the preprocessed multi-source pollution impact data and the environmental monitoring data includes:

[0071] Data cleaning, missing value filling and / or outlier processing are performed on the multi-source pollution impact data and environmental monitoring data to obtain pre-processed multi-source pollution impact data and environmental monitoring data.

[0072] Data cleaning, missing value filling and / or outlier processing are relatively common technical means in this field and will not be described in detail in the embodiments of the present invention.

[0073] In one possible implementation, constructing regional air pollution contribution training data based on the pre-processed multi-source pollution impact data and environmental monitoring data includes:

[0074] According to the pre-processed multi-source pollution impact data and environmental monitoring data, the pre-processed multi-source pollution impact data and environmental monitoring data corresponding to any N consecutive time points are used as regional atmospheric pollution impact samples;

[0075] According to the environmental monitoring data, the pollutant concentration increase rate corresponding to the Nth time point to the N+1th time point is used as the regional air pollution contribution training label.

[0076] like Figure 2 As shown, a machine learning model is used to learn regional air pollution contribution training samples and regional air pollution contribution training labels to obtain a regional air pollution contribution prediction model, including:

[0077] S201: Build a machine learning model, initialize the hyperparameters of the machine learning model, and obtain multiple hyperparameter particles;

[0078] The machine learning model can be set as a convolutional neural network model. The hyperparameters of the machine learning model have corresponding upper and lower limits. Therefore, they can be randomly initialized between the upper and lower limits of the hyperparameters, and the initialized hyperparameters can be encoded into vectors to obtain hyperparameter particles. After repeated initialization, multiple hyperparameter particles can be obtained.

[0079] S202. Obtaining the fitness corresponding to each hyperparameter particle according to the regional air pollution contribution training sample and the regional air pollution contribution training label;

[0080] S203. Determine the optimal hyperparameter particle among all hyperparameter particles according to the fitness corresponding to each hyperparameter particle;

[0081] S204: Based on the optimal hyperparameter particle, a dual adaptive guidance strategy is used to guide and update the hyperparameter particle to obtain the hyperparameter particle after guidance and update;

[0082] S205, using a self-regulating information interaction strategy to interactively update the hyperparameter particles after the guided update, to obtain the hyperparameter particles after the interactive update;

[0083] S206, using an adaptive differential evolution strategy to globally update the hyperparameter particles after interactive update to obtain globally updated hyperparameter particles;

[0084] S207, repeatedly executing the dual adaptive guidance strategy, the self-adjusting information interaction strategy, and the adaptive differential evolution strategy during each training process until the training end condition is met (such as the current number of training times is greater than or equal to the preset maximum number of training times), and re-determining the optimal hyperparameter particle based on the hyperparameter particle after the global update in the last training process;

[0085] S208. Based on the re-determined optimal hyperparameter particle, the final hyperparameters of the machine learning model are obtained (i.e., the hyperparameters contained in the optimal hyperparameter particle are used as the final hyperparameters of the machine learning model), and a regional atmospheric pollution contribution prediction model is obtained.

[0086] Existing particle swarm algorithms are prone to falling into local optima and poor training results during the training of machine learning models, resulting in poor prediction of pollution contributions. Therefore, embodiments of the present invention provide an improved particle swarm algorithm that enhances global search capabilities, prevents the algorithm from falling into local optima during training, and ultimately improves the accuracy of pollution contribution prediction.

[0087] In a possible implementation, the method further includes: after executing any one of the dual adaptive guidance strategy, the self-adjusting information interaction strategy, and the adaptive differential evolution strategy, performing out-of-bounds processing on the hyperparameter particles.

[0088] In a possible implementation, obtaining the fitness corresponding to each hyperparameter particle according to the regional air pollution contribution training sample and the regional air pollution contribution training label includes:

[0089] For any hyperparameter particle, apply the hyperparameters contained in the hyperparameter particle to the machine learning model to obtain the machine learning model after applying the hyperparameters;

[0090] Using the regional air pollution contribution training sample as input data of the machine learning model after applying hyperparameters, using the regional air pollution contribution training label as the expected output of the machine learning model after applying the hyperparameters, and obtaining a loss function value;

[0091] After adding the loss function value to the preset constant term, a non-zero parameter value is obtained, and the reciprocal of the non-zero parameter value is taken to obtain the fitness corresponding to the hyperparameter particle;

[0092] Traverse each hyperparameter particle and obtain the fitness corresponding to each hyperparameter particle. It is worth noting that the larger the fitness value, the better the position of the hyperparameter particle in the solution space.

[0093] In a possible implementation, based on the optimal hyperparameter particle, a dual adaptive guidance strategy is used to guide and update the hyperparameter particle, and the hyperparameter particle after the guidance update is obtained, including:

[0094] S301. For any hyperparameter particle, determine the Euclidean distance between the hyperparameter particle and other hyperparameter particles, and determine the neighboring particles corresponding to the hyperparameter particle based on the Euclidean distance;

[0095] S301.1. Based on the Euclidean distance between a hyperparameter particle and other hyperparameter particles, determine the neighborhood radius:

[0096]

[0097] in, represents the i-th hyperparameter particle in the t-th training process, i = 1, 2, ..., NP, NP represents the total number of hyperparameter particles, represents the jth hyperparameter particle in the tth training process, R i represents the neighborhood radius corresponding to the i-th hyperparameter particle during the t-th training process; express and The Euclidean distance between

[0098] S301.2. Based on the neighborhood radius, determine other hyperparameter particles whose Euclidean distances are smaller than the neighborhood radius, and obtain neighboring particles corresponding to the hyperparameter particles.

[0099] The above-mentioned determination of neighborhood particles by dynamic neighborhood radius can effectively avoid the situation where there are no other hyperparameter particles in the neighborhood range corresponding to the hyperparameter particle, thereby ensuring that the training is effective.

[0100] S302. Obtaining a neighborhood comprehensive position and a first influence factor corresponding to the neighborhood comprehensive position based on the neighborhood particle corresponding to the hyperparameter particle;

[0101] S302.1. Based on the neighborhood particles corresponding to the hyperparameter particles, obtain the neighborhood comprehensive position:

[0102]

[0103] in, represents the neighborhood comprehensive position corresponding to the i-th hyperparameter particle during the t-th training process, represents the qth neighbor particle corresponding to the ith hyperparameter particle in the tth training process, q = 1, 2, ..., K, K represents the total number of neighbor particles, λ q Represents the neighborhood particles The corresponding weighting coefficient, f q Represents the neighborhood particles The corresponding fitness, f p Represents the neighborhood particles The corresponding fitness, represents the pth neighbor particle corresponding to the i-th hyperparameter particle in the t-th training process, p = 1, 2, ..., K;

[0104] S302.2. Obtain the first influence factor corresponding to the neighborhood comprehensive position:

[0105]

[0106] Among them, α1 represents the first influential factor, γ iq represents the positional superiority factor between the i-th hyperparameter particle and its corresponding q-th neighbor particle during the t-th training process, f i represents the fitness of the i-th hyperparameter particle during the t-th training process, f q represents the fitness of the qth neighbor particle, f best represents the fitness corresponding to the optimal hyperparameter particle, f worst Indicates the fitness corresponding to the particle with the worst hyperparameter;

[0107] S303. According to the optimal hyperparameter particle, the second influence factor corresponding to the optimal hyperparameter particle is obtained:

[0108]

[0109] Among them, α2 represents the second influential factor;

[0110] S304. According to the neighborhood comprehensive position, the first influence factor corresponding to the neighborhood comprehensive position, and the second influence factor corresponding to the optimal hyperparameter particle, the guided update speed corresponding to the hyperparameter particle is obtained:

[0111]

[0112] in, represents the guided update speed corresponding to the i-th hyperparameter particle during the t-th training process, represents the guided update speed corresponding to the i-th hyperparameter particle in the t+1-th training process, r1 represents the first random number between (0,1), and r2 represents the second random number between (0,1). represents the optimal hyperparameter particle;

[0113] S305. According to the guidance update speed corresponding to the hyperparameter particle, the hyperparameter particle is guided and updated, and the hyperparameter particle after the guidance update is obtained as follows:

[0114]

[0115] in, represents the hyperparameter particle after the i-th guided update.

[0116] The embodiment of the present invention adopts a dual adaptive guidance strategy to guide the update of hyperparameter particles, which can effectively learn the information of other hyperparameter particles in the neighborhood and the information of the optimal hyperparameter particle. While retaining the algorithm search accuracy, it improves the algorithm's ability to search the neighborhood range and improves the ability to find the global optimum.

[0117] In a possible implementation, a self-regulating information interaction strategy is used to interactively update the hyperparameter particles after the guided update, and the hyperparameter particles after the interactive update are obtained, including:

[0118] S401: For any hyperparameter particle after the guidance update, match the hyperparameter particle with two other hyperparameter particles to obtain a first information interaction particle and a second information interaction particle;

[0119] S402. Obtain a first information interaction factor according to the fitness corresponding to the first information interaction particle; obtain a second information interaction factor according to the fitness corresponding to the second information interaction particle;

[0120] The first information interaction factor is:

[0121]

[0122] The second information interaction factor is:

[0123]

[0124] Among them, χ m1 represents the first information interaction factor, χ m2 represents the second information interaction factor, e represents the natural constant, f m represents the fitness of the hyperparameter particle after the mth guided update, f mm1 represents the fitness corresponding to the first information interaction particle, f mm2 Indicates the fitness corresponding to the second information interaction particle;

[0125] S403. Based on the current number of training times, the adaptive inertia weight is obtained as:

[0126]

[0127] Where ω represents the adaptive inertia weight, ω max Represents the maximum value of the adaptive inertia weight, ω min Indicates the minimum value of the adaptive inertia weight, exp represents the exponential function with the natural constant e as the base, ε represents the constant term, which can be set to 500 or 1000, and T represents the preset maximum number of training times;

[0128] S404: Based on the first information interaction factor, the second information interaction factor, and the adaptive inertia weight, a sine-cosine function is used to interactively update the hyperparameter particle after the guided update, and the hyperparameter particle after the interactive update is obtained as follows:

[0129]

[0130] in, represents the hyperparameter particle after the mth guided update during the tth training process, represents the hyperparameter particle after the mth interaction update, m=1,2,…,NP, sin represents the sine function, r3 represents the third random number between (0,1), cos represents the cosine function, r4 represents the fourth random number between (0,1), π represents pi, represents the first information interaction particle, Represents the second information interaction particle.

[0131] The embodiment of the present invention adopts a self-adjusting information interaction strategy to interactively update the hyperparameter particles after the guidance update, which can effectively improve the algorithm's search ability for unfamiliar areas between multiple hyperparameter particles, and search in an irregular path, which can further improve the algorithm's global search ability while ensuring the algorithm's search precision.

[0132] In a possible implementation, an adaptive differential evolution strategy is used to globally update the hyperparameter particles after interactive update, and the globally updated hyperparameter particles are obtained, including:

[0133] S501. Based on the current number of training times, the differential evolution factor is obtained as:

[0134]

[0135] Among them, ξ represents the differential evolution factor, e represents the natural constant, r5 represents a random number between (0,1), and π represents the circumference of a circle;

[0136] S502: For any hyperparameter particle after interactive update, randomly match the hyperparameter particle with two other hyperparameter particles to obtain a first differential evolution particle and a second differential evolution particle;

[0137] S503: Perform differential evolution on the hyperparameter particles after interactive update according to the first differential evolution particles, the second differential evolution particles, and the differential evolution factor, and obtain the differential evolution particles:

[0138]

[0139] in, represents the hyperparameter particle after the nth interaction update during the tth training process, represents the differential evolution particle corresponding to the hyperparameter particle after the nth interaction update, n = 1, 2, ..., NP, represents the first differential evolution particle corresponding to the hyperparameter particle after the nth interaction update, represents the second differential evolution particle corresponding to the hyperparameter particle after the nth interaction update, r6 represents the sixth random number between (0, 1), r7 represents the seventh random number between (0, 1), r8 represents the eighth random number between (0, 1), and r9 represents the ninth random number between (0, 1);

[0140] S504 : performing a global update on the interactively updated hyperparameter particles according to the differential evolution particles to obtain the globally updated hyperparameter particles.

[0141] When the fitness of the differential evolution particle is greater than the fitness of the hyperparameter particle after interactive update, the differential evolution particle is used as the hyperparameter particle after global update; otherwise, the original hyperparameter particle after interactive update is directly used as the hyperparameter particle after global update.

[0142] The embodiment of the present invention adopts an adaptive differential evolution strategy to globally update the hyperparameter particles after interactive update, which can effectively improve the global search capability of the algorithm, and the differential evolution factor can be gradually reduced in the later stage of the algorithm, ensuring the convergence ability of the algorithm.

[0143] Through the mutual coordination of the above strategies, the global search capability, search speed and training effect of the algorithm can be effectively improved, and ultimately the prediction accuracy of regional atmospheric pollution contribution can be effectively improved.

[0144] The present invention provides a regional atmospheric pollution contribution prediction method based on multi-source data fusion. The method constructs regional atmospheric pollution impact samples and regional atmospheric pollution contribution training labels by pre-processing multi-source pollution impact data and environmental monitoring data. Then, a machine learning model is used to learn the regional atmospheric pollution contribution training samples and the regional atmospheric pollution contribution training labels to obtain a regional atmospheric pollution contribution prediction model. Finally, atmospheric pollution impact data of the to-be-identified area corresponding to the target detection area is collected, and the regional atmospheric pollution contribution prediction model is used to identify the atmospheric pollution impact data of the to-be-identified area. The method can effectively fuse multi-source data, and use machine learning to achieve fast and accurate pollution prediction, which can effectively improve the accuracy of atmospheric pollution prediction.

[0145] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0146] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0147] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0148] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0149] Those skilled in the art will understand that all or part of the steps in implementing the above facts and methods can be completed by instructing relevant hardware through a program, and the program involved or the program can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: the corresponding method steps are then brought out, and the storage medium can be ROM / RAM, a disk, an optical disk, etc.

[0150] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting regional air pollution contribution based on multi-source data fusion, characterized in that: include: Collecting multi-source pollution impact data and environmental monitoring data corresponding to the target monitoring area, and preprocessing the multi-source pollution impact data and environmental monitoring data to obtain preprocessed multi-source pollution impact data and environmental monitoring data; Constructing regional air pollution contribution training data based on the pre-processed multi-source pollution impact data and environmental monitoring data; wherein the regional air pollution contribution training data includes regional air pollution impact samples and regional air pollution contribution training labels; A machine learning model is used to learn regional air pollution contribution training samples and regional air pollution contribution training labels to obtain a regional air pollution contribution prediction model; The atmospheric pollution impact data of the area to be identified corresponding to the target detection area are collected, and the atmospheric pollution impact data of the area to be identified are identified using a regional atmospheric pollution contribution prediction model to obtain a regional atmospheric pollution contribution prediction result.

2. The method for predicting regional air pollution contribution based on multi-source data fusion according to claim 1 is characterized in that: Collect multi-source pollution impact data and environmental monitoring data corresponding to the target monitoring area, including: Based on the preset data sampling frequency, meteorological data and pollution source data corresponding to the target monitoring area are collected at each time point to obtain multi-source pollution impact data; Based on the preset data sampling frequency, the pollutant concentration data corresponding to the target monitoring area is collected at each time point to obtain environmental monitoring data.

3. The method for predicting regional air pollution contribution based on multi-source data fusion according to claim 2 is characterized in that: Preprocessing the multi-source pollution impact data and the environmental monitoring data to obtain the preprocessed multi-source pollution impact data and the environmental monitoring data includes: Data cleaning, missing value filling and / or outlier processing are performed on the multi-source pollution impact data and environmental monitoring data to obtain pre-processed multi-source pollution impact data and environmental monitoring data.

4. The method for predicting regional air pollution contribution based on multi-source data fusion according to claim 1, characterized in that: Based on the pre-processed multi-source pollution impact data and environmental monitoring data, regional air pollution contribution training data is constructed, including: According to the pre-processed multi-source pollution impact data and environmental monitoring data, the pre-processed multi-source pollution impact data and environmental monitoring data corresponding to any N consecutive time points are used as regional atmospheric pollution impact samples; According to the environmental monitoring data, the pollutant concentration increase rate corresponding to the Nth time point to the N+1th time point is used as the regional air pollution contribution training label.

5. The method for predicting regional air pollution contribution based on multi-source data fusion according to claim 1 is characterized in that: A machine learning model is used to learn regional air pollution contribution training samples and regional air pollution contribution training labels to obtain a regional air pollution contribution prediction model, including: Constructing a machine learning model and initializing hyperparameters of the machine learning model to obtain multiple hyperparameter particles; Obtaining the fitness corresponding to each hyperparameter particle according to the regional atmospheric pollution contribution training sample and the regional atmospheric pollution contribution training label; According to the fitness corresponding to each hyperparameter particle, the optimal hyperparameter particle is determined among all hyperparameter particles; According to the optimal hyperparameter particle, a dual adaptive guidance strategy is adopted to guide and update the hyperparameter particle to obtain the hyperparameter particle after the guidance update; A self-regulating information interaction strategy is used to interactively update the hyperparameter particles after the guided update to obtain the hyperparameter particles after the interactive update; Adopting the adaptive differential evolution strategy to globally update the hyperparameter particles after interactive update, and obtain the hyperparameter particles after global update; During each training process, the dual adaptive guidance strategy, self-adjusting information interaction strategy, and adaptive differential evolution strategy are repeatedly executed until the training end conditions are met. The optimal hyperparameter particle is re-determined based on the hyperparameter particle after the global update in the last training process. Based on the re-determined optimal hyperparameter particles, the final hyperparameters of the machine learning model are obtained, and the regional atmospheric pollution contribution prediction model is obtained.

6. The method for predicting regional air pollution contribution based on multi-source data fusion according to claim 5 is characterized in that: Also includes: After executing any one of the dual adaptive guidance strategy, the self-adjusting information interaction strategy and the adaptive differential evolution strategy, the hyperparameter particles are processed out of bounds.

7. The method for predicting regional air pollution contribution based on multi-source data fusion according to claim 5, characterized in that: According to the regional atmospheric pollution contribution training samples and regional atmospheric pollution contribution training labels, the fitness corresponding to each hyperparameter particle is obtained, including: For any hyperparameter particle, apply the hyperparameters contained in the hyperparameter particle to the machine learning model to obtain the machine learning model after applying the hyperparameters; Using the regional air pollution contribution training sample as input data of the machine learning model after applying hyperparameters, using the regional air pollution contribution training label as the expected output of the machine learning model after applying the hyperparameters, and obtaining a loss function value; After adding the loss function value to the preset constant term, a non-zero parameter value is obtained, and the reciprocal of the non-zero parameter value is taken to obtain the fitness corresponding to the hyperparameter particle; Traverse each hyperparameter particle and obtain the fitness corresponding to each hyperparameter particle.

8. The method for predicting regional air pollution contribution based on multi-source data fusion according to claim 7, characterized in that: According to the optimal hyperparameter particle, a dual adaptive guidance strategy is adopted to guide and update the hyperparameter particle to obtain the hyperparameter particle after the guidance update, including: For any hyperparameter particle, determine the Euclidean distance between the hyperparameter particle and other hyperparameter particles, and determine the neighboring particles corresponding to the hyperparameter particle based on the Euclidean distance; Obtaining a neighborhood comprehensive position and a first influence factor corresponding to the neighborhood comprehensive position according to the neighborhood particle corresponding to the hyperparameter particle; According to the optimal hyperparameter particle, obtaining a second influence factor corresponding to the optimal hyperparameter particle; Obtaining a guided update speed corresponding to the hyperparameter particle according to the neighborhood comprehensive position, the first influence factor corresponding to the neighborhood comprehensive position, and the second influence factor corresponding to the optimal hyperparameter particle; The hyperparameter particle is guided and updated according to the guidance update speed corresponding to the hyperparameter particle to obtain the hyperparameter particle after the guidance update.

9. The method for predicting regional air pollution contribution based on multi-source data fusion according to claim 8, characterized in that: The self-regulating information interaction strategy is used to interactively update the hyperparameter particles after the guided update, and the hyperparameter particles after the interactive update are obtained, including: For any hyperparameter particle after the guide update, match the hyperparameter particle with two other hyperparameter particles to obtain a first information interaction particle and a second information interaction particle; Obtaining a first information interaction factor according to the fitness corresponding to the first information interaction particle; obtaining a second information interaction factor according to the fitness corresponding to the second information interaction particle; Get the adaptive inertia weight based on the current number of training times; According to the first information interaction factor, the second information interaction factor, and the adaptive inertia weight, a sine and cosine function is used to interactively update the hyperparameter particles after the guided update to obtain the hyperparameter particles after the interactive update.

10. The method for predicting regional air pollution contribution based on multi-source data fusion according to claim 9, characterized in that: Adopting the adaptive differential evolution strategy to globally update the hyperparameter particles after interactive update, the hyperparameter particles after global update are obtained, including: Based on the current number of training times, obtain the differential evolution factor; For any hyperparameter particle after interactive update, randomly match the hyperparameter particle with two other hyperparameter particles to obtain the first differential evolution particle and the second differential evolution particle; Performing differential evolution on the hyperparameter particles after interactive update according to the first differential evolution particles, the second differential evolution particles, and the differential evolution factor to obtain differential evolution particles; According to the differential evolution particles, the hyperparameter particles after interactive update are globally updated to obtain the hyperparameter particles after global update.

Images