Shared cloud integrated ecology air quality prediction and big data system

Abstract

The invention provides an air quality prediction and big data system of a shared cloud integrated ecology. The system comprises a data acquisition module, a network communication module, a microcontroller module and the shared cloud integrated ecology. The multi-source space-time air health state indexes and related meteorological environment indexes obtained by the multiple sites are transmitted to the shared cloud through a network communication module to form ecology; the shared cloud integrated ecology prediction module screens a plurality of data features from a data set, decomposes data of each feature into a plurality of modal components by using RPSEMD decomposition and divides the data set, and optimizes parameters of an MSSTNet model by using an improved alpha evolutionary algorithm to obtain optimal hyper-parameters; the data set of each modal component is tested and verified through an MSSTNet model; and a final prediction result is obtained after adaptive superposition. According to the method, through combination of multiple sites and multiple features, a shared cloud integrated ecology is integrated, the data complexity is reduced by using RPSEMD decomposition, meanwhile, the MSSTNet model is established to efficiently and accurately predict multi-scale multi-feature data, and the prediction precision is further improved for the model by using the improved alpha evolutionary algorithm. And a new thought and method are provided for air quality prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of air monitoring and data analysis, and in particular to an air quality prediction and big data system with a shared cloud integrated ecosystem. Background Art

[0002] In today's society, air quality has a profound impact on human health and quality of life. Accurately predicting air quality conditions is of great significance for environmental protection decisions, public health warnings, and people's living arrangements. Traditional air quality prediction methods, such as those based on statistical models, such as multivariate linear regression models, are simple and easy to use, but when dealing with complex and changeable air quality data, it is difficult to accurately capture the nonlinear relationship between various factors, resulting in limited prediction accuracy. Some early prediction methods based on machine learning, such as ordinary neural network models, are prone to falling into local optimal solutions, lack the ability to generalize data, and cannot effectively respond to the diverse air quality change characteristics in different regions and seasons.

[0003] With the development of artificial intelligence technology, evolutionary algorithms have gradually been applied to the field of air quality prediction. However, traditional evolutionary algorithms and prediction algorithms have slow convergence speeds during the optimization process, are prone to premature convergence, and have difficulty finding the global optimal solution, making it difficult for the air quality prediction models based on these algorithms to meet the actual needs of variable air quality measurements in terms of accuracy and stability. Therefore, how to improve the efficiency and accuracy of parameter optimization of prediction models in air quality prediction has become a key issue that needs to be urgently addressed in this field. Summary of the invention

[0004] Purpose of the invention: To solve the problems mentioned in the background technology, the present invention provides an air quality prediction and big data system with a shared cloud integrated ecosystem. Through multi-module integration and improvement of the alpha evolution algorithm, it outputs air quality prediction results more accurately and timely while displaying data intuitively, so that relevant departments can respond and implement countermeasures in a targeted manner.

[0005] Technical solution:

[0006] The present invention discloses an air quality prediction and big data system for a shared cloud integrated ecosystem, the system comprising:

[0007] Data collection module, used to monitor multi-source spatiotemporal air health indicators and related meteorological environment indicators belonging to multiple sites;

[0008] Network communication module, used for data transmission between the data acquisition module and the shared cloud integration ecosystem;

[0009] A microcontroller module for processing multi-source spatio-temporal air health status indicators and transmitting data to a shared cloud integration ecosystem through a network communication module;

[0010] The shared cloud integration ecosystem includes:

[0011] An air quality spatio-temporal prediction module that screens multi-source spatio-temporal feature sets of time, space, meteorology, and geography, uses the regenerative phase-shifted sine-assisted empirical mode decomposition (RPSEMD) to decompose the screened feature subset into multiple IMF modal components, divides the data set, constructs a multi-scale spatio-temporal prediction neural network (MSSTNet), uses the improved alpha evolution algorithm (IAE) to optimize the hyperparameters of the MSSTNet model, and obtains the prediction results of each IMF modal component and adaptively superimposes them to obtain the final prediction result;

[0012] An air health status visualization module that enables multi-terminal monitoring of the monitoring center, terminal server, and APP through the shared cloud integration ecosystem.

[0013] Furthermore, the specific steps for decomposing the feature subset into multiple IMFs using the regenerative phase-shifted sine-assisted empirical mode decomposition (RPSEMD) are as follows:

[0014] By applying the EMD algorithm to the original signal, a series of modal functions (IMFs) are obtained. The amplitude a k and frequency f k of the sine wave are determined according to the IMFs, and uniform sampling is performed on the phase δ ki to generate a sine wave s(t|a k , f k , δ ki ). The generated sine wave s(t|a k , f k , δ ki ) is added to the original signal, and the EMD algorithm is applied again to obtain the first IMF. The first IMF obtained by sampling all phases is averaged to obtain the final IMFc k (t). The IMFc k (t) is removed from the original signal, and the above steps are repeated until no new IMF can be obtained. The final signal is regarded as the residual r(t).

[0015] Furthermore, the improvement steps of the alpha evolution algorithm (IAE) are as follows:

[0016] S1 Initialize the matrix of MSSTNet hyperparameters, denoted as X, and calculate the parameter error index of MSSTNet;

[0017] S2 Introduce a reverse learning strategy to obtain the reverse solution of the MSSTNet parameters;

[0018] S3 Determine the update direction. If rand(0,1) < 0.5, calculate the probability matrix A, denoted as P a ; otherwise, calculate the rate matrix B, denoted as P b ;

[0019] S4 Update the current MSSTNet hyperparameter matrix E i , and obtain the current optimal MSSTNet hyperparameter x best ;

[0020] S5 Introduce Cauchy mutation to enhance the ability of the parameter optimization process to jump out of local optima, and impose boundary constraints on x best , and find the optimal MSSTNet hyperparameter x new b est ;

[0021] S6 Judge the termination condition. If it is satisfied, output the optimal hyperparameters; otherwise, repeat steps S3 - S5.

[0022] Furthermore, the improved alpha evolution algorithm IAE is used to optimize the MSSTNet parameters. A reverse learning strategy is introduced in this algorithm to increase the probability of searching for the optimal MSSTNet parameters. The MSSTNet hyperparameters ξ = [r, e] are encoded into the matrix form X = [x ij during initialization to increase the probability of searching for the optimal MSSTNet hyperparameters. The calculation formula of the reverse learning strategy is as follows:

[0023]

[0024] In the formula: x ij is the MSSTNet parameter of the i-th row and j-th dimension; x i ′ j is the reverse solution of x ij ; m is the elite reverse learning coefficient, randomly taking values in the range of [0, 1]; a ij and b ij respectively represent the maximum and minimum values in the j-th dimension;

[0025] Perform Cauchy mutation on the global optimal MSSTNet parameters to enhance the ability of the parameter optimization process to jump out of local optima and achieve the purpose of searching for better MSSTNet parameters. The Cauchy mutation calculation formula is as follows:

[0026] x newbest = x best + x best · Cauchy(0,1)

[0027] In the formula: Cauchy(0,1) represents the standard Cauchy function, x bestis the current global optimal MSSTNet parameter, x newbest is the global optimal MSSTNet parameter after Cauchy mutation;

[0028] The calculation formula of the update direction is:

[0029]

[0030] In the formula: c a and c b are the learning rates of P a and P b respectively; A is obtained by sampling with replacement the candidate MSSTNet parameters of the parameter matrix X; B is obtained by sampling without replacement the candidate MSSTNet parameters of the parameter matrix X; ω is the weight of the MSSTNet parameters in B;

[0031]

[0032] In the formula: E i is the i-th evolutionary MSSTNet parameter of the evolutionary matrix E; t is the current iteration number; P is the basis vector, which determines the starting position of the evolution; α is the attenuation factor; Δr i is the i-th random step size; θ is the control parameter; W i and L i are the sampled MSSTNet parameters of the matrix X, satisfying f(W i ) ≤ f(E i ) ≤ f(L i ).

[0033] Furthermore, the construction steps of the multi-scale spatio-temporal prediction neural network MSSTNet are as follows:

[0034] T1 Obtain multi-source spatio-temporal data of air quality;

[0035] T2 Construct an alternative multi-source spatio-temporal feature set, and remove features with weak correlation coefficients to obtain a feature subset closely related to air quality

[0036] T3 Use RPSEMD to decompose the feature subset to obtain N IMF modal components and the residual r(t);

[0037] T4 Divide all the decomposed sequences into a training set, a test set and a validation set;

[0038] T5 Construct an MSSTNet model, and use the improved alpha evolutionary algorithm IAE to optimize the hyperparameters of the MSSTNet model, and output the optimized MSSTNet model for prediction and verification;

[0039] The optimized T6 MSSTNet model outputs the prediction results of each decomposed sequence

[0040] The prediction results of each decomposed sequence of T7 pass through the adaptive operator and are superimposed to obtain the final air quality prediction result C of the p-th site p 。

[0041] Furthermore, the system also includes an air health status indicator sharing module, which is used for data sharing and model parameter sharing between newly built air quality monitoring stations and air quality monitoring stations established earlier. The air quality monitoring stations established earlier construct the MSSTNet model, and the IAE determines the hyperparameters of the model The newly built station uses the data sharing mechanism to migrate to its own site, denoted as The newly built station passes through the model for training. After the training is completed, the newly built station can use the KL divergence to evaluate whether the migrated hyperparameters are suitable for the data characteristics of the current site. If the KL divergence is less than the threshold, it can be retained for use. If the KL divergence is greater than the threshold, the air health status indicators recently monitored by the newly built station can be used as the model input to adjust the hyperparameters of the MSSTNet model on the basis of these hyperparameters, reducing the time for modeling and hyperparameter optimization of the model. The calculation formula of the KL divergence is:

[0042]

[0043] where: P(x) is the actual distribution of the monitored air quality indicators, and Q(x) is the corresponding distribution of the model-predicted air quality indicators

[0044] Furthermore, on the large screen interface of the monitoring center of the air health status visualization module, various visualization charts representing air health status indicators are displayed in real time, intuitively presenting the change trend and current status of the air health status; the status information of all terminal servers and site monitoring devices is centrally displayed, including the online / offline status of the devices, the running duration, the fault alarm situation, and the log information of each end is collected and stored, providing log query and analysis functions

[0045] The terminal server is configured with a fault self-checking mechanism. When the server detects a fault in itself, it tries to automatically restart or perform other preset recovery operations, and reports the fault information and recovery results to the monitoring center

[0046] The APP monitors the operating status of various devices in the site in real time. Users can monitor the operation of the devices through the APP to promptly detect device failures or abnormal situations. The APP displays real-time environmental data to ensure that the site is in a suitable operating environment and avoid device damage or unstable operation caused by environmental factors. Set data alarm rules. When the monitored data is abnormal or exceeds the preset threshold, the APP promptly sends an alarm notification to remind the user to pay attention and recommend effective countermeasures to prevent the problem from expanding.

[0047] Beneficial effects:

[0048] 1. The present invention constructs a cloud integration ecosystem to achieve efficient sharing of atmospheric environment monitoring data of multiple sites and intercommunication of prediction model parameters. By integrating multi-dimensional data such as meteorological environment indicators, site locations, and geographical information, the processing and analysis process of multi-feature data is optimized, and the data format is unified, which is more conducive to the stable transmission and accurate expression of data.

[0049] 2. The present invention first uses RPSEMD decomposition for multi-source spatio-temporal feature subsets, effectively solves the problem of mode mixing, reduces the complexity of prediction data, and uses the MSSTNet prediction model to decompose large kernel convolutions from a multi-scale perspective to capture spatial feature information at different scales. The combination of the two enables the model to accurately capture complex non-linear relationships in air quality data and effectively improve the prediction accuracy.

[0050] 3. The present invention designs the Alpha Evolution Algorithm AE to adopt a reverse learning strategy to generate a high-quality initial population, expand the optimization and avoid local extrema, strengthen the search ability. At the same time, the Cauchy mutation is introduced to enhance the exploration ability of the algorithm and accelerate convergence to improve the prediction accuracy. The Improved Alpha Evolution Algorithm IAE is used to optimize the hyperparameters of the MSSTNet model, which improves the global search ability and optimization efficiency of the model. Description of the Drawings

[0051] Figure 1 is the overall block diagram of the modules of the present invention;

[0052] Figure 2 is the air quality prediction flow chart of the present invention. Detailed Embodiments

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] Such as Figure 1 - Figure 2As shown in the figure, this embodiment discloses an air quality prediction and big data system for a shared cloud integration ecosystem, including: a data acquisition module: monitoring multi-source spatio-temporal air health status indicators and related meteorological environment indicators of multiple sites;

[0055] A network communication module: used for data transmission between the data acquisition module and the shared cloud integration ecosystem;

[0056] A microcontroller module: used to process multi-source spatio-temporal air health status indicators and transmit data to the shared cloud integration ecosystem through the network communication module;

[0057] The shared cloud integration ecosystem: includes an air quality spatio-temporal prediction module, an air health status indicator sharing module, and an air health status visualization module;

[0058] The air quality spatio-temporal prediction module screens a multi-source spatio-temporal feature set, including time features, space features, meteorological features, and geographical features, uses the regenerative phase-shifted sine-assisted empirical mode decomposition (RPSEMD) to decompose the screened feature subset into multiple IMF modal components, then divides the data set, constructs a multi-scale spatio-temporal prediction neural network (MSSTNet), uses the improved alpha evolution algorithm (IAE) to optimize the hyperparameters ξ of the MSSTNet model, outputs the optimal parameters to obtain the prediction results of each IMF modal component, and adaptively superimposes each IMF modal component to obtain the final prediction result; the air health status indicator sharing module can be used for data sharing and model parameter sharing between new sites and mature sites. The air health status data visualization module realizes multi-terminal monitoring of the monitoring center, terminal server, and APP through the shared cloud integration ecosystem.

[0059] Monitor multi-source spatio-temporal air health status indicators and related meteorological environment indicators of multiple sites; the air health status indicators are not limited to PM2.5, PM10, SO2, NO2, O3, CO; the meteorological environment indicators are not limited to: temperature and humidity, wind speed, rainfall; site location data; geographical data.

[0060] The multi-scale spatio-temporal prediction neural network MSSTNet, the specific steps are as follows:

[0061] Step 1: Obtain multi-source spatio-temporal data of air quality;

[0062] Step 2: Construct an alternative multi-source spatio-temporal feature set, and remove features with weak correlation coefficients to obtain a feature subset closely related to air quality

[0063] Step 3: Use RPSEMD to decompose the feature subset to obtain N IMF modal components and a residual r(t);

[0064] Step 4: All decomposed sequences Divide it into a training set, a test set and a validation set;

[0065] Step 5: Construct the MSSTNet model, and use the improved alpha evolution algorithm IAE to optimize the hyperparameters of the MSSTNet model, and output the optimized MSSTNet model for prediction and verification;

[0066] Step 6: The optimized MSSTNet model outputs the prediction results of each decomposed sequence

[0067] Step 7: The prediction results of each decomposed sequence pass through the adaptive operator and are superimposed to obtain the final air quality prediction result C of the p-th station p .

[0068] The calculation formula is as follows:

[0069]

[0070] In the formula, r and e are respectively the convolution kernel size and the convolution step length in the MSSTNet model, which is determined by the error size of the prediction results of each decomposed sequence;

[0071] The improved alpha evolution algorithm IAE has the following steps:

[0072] a. Initialize the matrix of MSSTNet hyperparameters, denoted as X, and calculate the MSSTNet parameter error index;

[0073] b. Introduce the reverse learning strategy to obtain the reverse solution of the MSSTNet parameters;

[0074] c. Judge the update direction. If rand(0,1) < 0.5, then calculate the probability matrix A, denoted as P a ; otherwise, calculate the probability matrix B, denoted as P b ;

[0075] d. Use the evolution operator to update the current MSSTNet hyperparameter matrix E i , and obtain the current optimal MSSTNet hyperparameter x best ;

[0076] e. Introduce Cauchy mutation to enhance the ability of the parameter optimization process to jump out of the local optimum, and impose boundary constraints on x best , and find the optimal MSSTNet hyperparameter x ne wb est ;

[0077] f. Judge the termination condition. If it is satisfied, output the optimal hyperparameters; otherwise, repeat steps 3-5;

[0078] The improved alpha evolution algorithm IAE is used to optimize the parameters of MSSTNet. The algorithm introduces a reverse learning strategy to improve the probability of searching for the optimal MSSTNet parameters and encodes the MSSTNet hyperparameters ξ=[r,e] into a matrix form X=[x ij ] During initialization, the probability of searching for the optimal MSSTNet hyperparameters is increased, and the reverse learning strategy calculation formula is as follows:

[0079]

[0080] Where: x ij is the MSSTNet parameter of the i-th row and j-th dimension; x′ ij For x ij The reverse solution; m is the elite reverse learning coefficient, which is randomly selected in the range of [0,1]; a ij and b ij They represent the maximum and minimum values ​​in the j dimension respectively.

[0081] The Cauchy mutation is performed on the global optimal MSSTNet parameters to enhance the ability of the parameter optimization process to jump out of the local optimum and achieve the purpose of searching for better MSSTNet parameters. The Cauchy mutation calculation formula is as follows:

[0082] x newbest =x best +x best Cauchy (0,1)

[0083] Where: Cauchy(0,1) represents the standard Cauchy function, x best is the current global optimal MSSTNet parameter, x newbest are the global optimal MSSTNet parameters after Cauchy mutation.

[0084] The update direction calculation formula is:

[0085]

[0086] Where: c a 、c b P a , P b The learning rate is ; A is the candidate MSSTNet parameters of the parameter matrix X obtained by sampling with replacement; B is the candidate MSSTNet parameters of the parameter matrix X obtained by sampling without replacement; ω is the weight of the MSSTNet parameters in B.

[0087]

[0088] Where: E iThe i-th evolutionary MSSTNet parameter of the evolutionary matrix E; t is the current iteration number; P is the basis vector, which determines the starting position of the evolution; α is the attenuation factor; Δr i is the i-th random step size; θ is the control parameter; W i and L i are the sampled MSSTNet parameters of matrix X, satisfying f(W i ) ≤ f(E i ) ≤ f(L i ).

[0089] Using the regenerative phase-shifted sine-assisted empirical mode decomposition RPSEMD to decompose the feature subset into multiple IMF modes, which is achieved by applying the EMD algorithm to the original signal to obtain a series of mode functions IMFs, and determining the amplitude a k and frequency f k of the sine wave according to the IMFs, uniformly sampling on the phase δ ki to generate the sine wave s(t|a k , f k , δ ki ); adding the generated sine wave s(t|a k , f k , δ ki ) to the original signal, applying the EMD algorithm again to obtain the first IMF, averaging the first IMF obtained by sampling all phases to obtain the final IMFc k (t), removing IMFc k (t) from the original signal, repeating the above steps until no new IMF can be obtained, and the final signal is regarded as the residual r(t); the final decomposition form is:

[0090]

[0091] where: c k (t) are the IMFs obtained by decomposing the feature subset, and r(t) is the residual.

[0092] This embodiment provides an air health status index sharing module, which can be used for data sharing and model parameter sharing between newly built air quality monitoring stations and air quality monitoring stations established earlier. In terms of data sharing, the air quality monitoring stations established earlier have accumulated a large amount of data, which can provide sufficient data support for the newly built stations.

[0093] In terms of parameter sharing, the air quality monitoring stations established earlier determine the hyperparameters of the model by constructing the MSSTNet model, IAE The newly built stations can use the data sharing mechanism to migrate these hyperparameters to their own stations, denoted as Subsequently, the newly established site is trained through the model After the training is completed, the newly established site can use the KL divergence to evaluate whether the transferred hyperparameters are suitable for the data characteristics of the current site. If the KL divergence is less than the threshold, it can be retained for use. If the KL divergence is greater than the threshold, based on this hyperparameter, the air health status indicators recently monitored by the newly established site can be used as the model input to adjust the hyperparameters of the MSSTNet model, reducing the modeling and model hyperparameter optimization time, thus avoiding a large amount of trial-and-error time and improving the efficiency of model optimization.

[0094] The formula for KL divergence is:

[0095]

[0096] In the formula: P(x) is the actual distribution of the monitored air quality indicators, and Q(x) is the corresponding distribution of the model-predicted air quality indicators.

[0097] The monitoring center can display various visual charts representing air health status indicators in real time on the large screen or interface of the monitoring center, intuitively presenting the change trend and current status of the air health status; centrally display the status information of all terminal servers and site monitoring devices, including device online / offline status, running duration, and fault alarm situations, facilitating a quick understanding of the overall device operation status; collect and store the log information of each end, providing log query and analysis functions to help technicians trace the cause of problems and optimize the system performance and stability;

[0098] The terminal server is configured with a fault self-checking mechanism. When the server detects a fault in itself, it attempts to automatically restart or perform other preset recovery operations, and reports the fault information and recovery results to the monitoring center;

[0099] The APP can monitor the running status of various devices in the site in real time. Users can intuitively understand whether each device is working properly through the APP and promptly discover device faults or abnormal situations; the APP will display real-time environmental data to ensure that the site is in a suitable operating environment and avoid device damage or unstable operation caused by environmental factors; set data alarm rules. When the monitored data is abnormal or exceeds the preset threshold, the APP will promptly send an alarm notification to remind users to pay attention and recommend effective countermeasures to prevent the problem from expanding.

[0100] The above description of the embodiments enables those skilled in the art to implement or use the present invention. Various modifications to the embodiments will be obvious to those skilled in the art. The general principles of the present invention can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention should not be limited to the embodiments shown herein, but should cover the broadest scope that conforms to the principles and novel features disclosed in the present invention.

Claims

1. A shared cloud integrated ecological air quality prediction and big data system, characterized in that: The system comprises: Data collection module, used to monitor multi-source spatiotemporal air health indicators and related meteorological environment indicators belonging to multiple sites; Network communication module, used for data transmission between the data acquisition module and the shared cloud integration ecosystem; A microcontroller module is used to process multi-source spatiotemporal air health status indicators and transmit data to a shared cloud integrated ecosystem through a network communication module; Shared cloud integration ecosystem, including: The air quality spatiotemporal prediction module selects the multi-source spatiotemporal feature sets of time, space, meteorology, and geography, decomposes the selected feature subsets into multiple IMF modal components using the regenerative phase-shifted sine-assisted empirical mode decomposition (RPSEMD), divides the data set, and constructs a multi-scale space-time prediction neural network MSSTNet. The improved alpha evolution algorithm (IAE) is used to optimize the hyperparameters of the MSSTNet model, outputs the optimal parameters, obtains the prediction results of each IMF modal component, and adaptively superimposes them to obtain the final prediction results; The air health status visualization module realizes multi-terminal monitoring of the monitoring center, terminal server, and APP through a shared cloud integrated ecosystem.

2. The air quality prediction and big data system of shared cloud integrated ecology according to claim 1, characterized in that: The regenerative phase-shifted sine-assisted empirical mode decomposition (RPSEMD) is used to decompose the feature subset into multiple IMF modes. The specific steps are as follows: By applying the EMD algorithm to the original signal, a series of modal functions IMFs are obtained, and the amplitude a of the sine wave is determined based on the IMFs. k and frequency f k , at phase δ ki uniform sampling to generate a sine wave s(t|a k , f k , δ ki ); the generated sine wave s(t|a k , f k , δ ki ) is added to the original signal, and the EMD algorithm is applied again to obtain the first IMF. The first IMF obtained by sampling all phases is averaged to obtain the final IMF c k (t), remove the IMF c from the original signal k (t), repeat the above steps until no new IMF can be obtained, and the final signal is regarded as the residual r(t).

3. The air quality prediction and big data system of shared cloud integrated ecology according to claim 1, characterized in that: The improvement steps of the Alpha Evolution Algorithm IAE are as follows: S1 initializes the MSSTNet hyperparameter matrix, denoted as X, and calculates the MSSTNet parameter error index; S2 introduces a reverse learning strategy to obtain the reverse solution of MSSTNet parameters; S3 determines the update direction. If rand(0,1) < 0.5, the probability matrix A is calculated, denoted as P a ; On the contrary, calculate the rate matrix B, denoted as P b ; S4 uses the evolution operator to update the current MSSTNet hyperparameter matrix E i , get the current optimal MSSTNet hyperparameter x best ; S5 introduces Cauchy mutation to enhance the ability of parameter optimization process to jump out of local optimum and best Apply boundary constraints to find the optimal MSSTNet hyperparameters x newbest ; S6 determines the termination condition. If it is met, the optimal hyperparameter is output; otherwise, steps S3-S5 are repeated.

4. The air quality prediction and big data system of the shared cloud integrated ecology according to claim 3 is characterized in that: The improved alpha evolution algorithm IAE is used to optimize the parameters of MSSTNet. The algorithm introduces a reverse learning strategy to improve the probability of searching for the optimal MSSTNet parameters. The MSSTNet hyperparameters ξ=[r,e] are encoded into a matrix form X=[x ij ] During initialization, the probability of searching for the optimal MSSTNet hyperparameters is increased, and the reverse learning strategy calculation formula is as follows: Where: x ij is the MSSTNet parameter of the i-th row and the j-th dimension; x i ' j For x ij The reverse solution; m is the elite reverse learning coefficient, which is randomly selected in the range of [0,1]; a ij and b ij Respectively represent the maximum and minimum values ​​in the j dimension; The Cauchy mutation is performed on the global optimal MSSTNet parameters to enhance the ability of the parameter optimization process to jump out of the local optimum and achieve the purpose of searching for better MSSTNet parameters. The Cauchy mutation calculation formula is as follows: x newbest =x best +x best ·Cauchy(0,1) Where: Cauchy(0,1) represents the standard Cauchy function, x best is the current global optimal MSSTNet parameter, x newbest is the global optimal MSSTNet parameter after Cauchy mutation; The update direction calculation formula is: Where: c a 、c b P a , P b The learning rate; A is obtained by sampling the candidate MSSTNet parameters of the parameter matrix X with replacement; B is obtained by sampling the candidate MSSTNet parameters of the parameter matrix X without replacement; ω is the weight of the MSSTNet parameters in B; Where: E i The i-th evolved MSSTNet parameter of the evolution matrix E; t is the current iteration number; P is the basis vector, which determines the starting position of the evolution; α is the decay factor; Δr i is the i-th random step length; θ is the control parameter; W i and L i is the sampled MSSTNet parameter of matrix X, satisfying f(W i )≤f(E i )≤f(L i ).

5. The air quality prediction and big data system of shared cloud integrated ecology according to claim 4 is characterized in that: The construction steps of the multi-scale space-time prediction neural network MSSTNet are as follows: T1 obtains multi-source spatiotemporal data of air quality; T2 constructs an alternative multi-source spatiotemporal feature set and removes features with weak correlation coefficients to obtain a feature subset that is closely related to air quality. T3 uses RPSEMD to decompose the feature subset and obtain N IMF modal components and residual r(t); T4 will decompose all the sequences Divide into training set, test set and validation set; T5 builds the MSSTNet model and uses the improved alpha evolution algorithm IAE to optimize the hyperparameters of the MSSTNet model, outputting the optimized MSSTNet model for prediction and verification; The T6 optimized MSSTNet model outputs the prediction results of each decomposed sequence The prediction results of each decomposed sequence of T7 are obtained through adaptive operators The final air quality forecast result C of the pth station is obtained by superposition. p .

6. The air quality prediction and big data system of shared cloud integrated ecology according to claim 1, characterized in that: The system also includes an air health status indicator sharing module, which is used to share data and model parameters between newly built air quality monitoring sites and older air quality monitoring sites. The older air quality monitoring sites build the MSSTNet model, and IAE determines the model's hyperparameters. New sites will use data sharing mechanisms to The site migrated to itself is recorded as New site through model After the training is completed, the newly built site can use KL divergence to evaluate whether the migrated hyperparameters are suitable for the data characteristics of the current site. If the KL divergence is less than the threshold, it can be retained for use. If the KL divergence is greater than the threshold, the air health status indicators recently monitored by the newly built site can be used as model input to adjust the MSSTNet model hyperparameters based on this hyperparameter, thereby reducing the modeling and model hyperparameter optimization time. The KL divergence calculation formula is: Where: P(x) is the actual distribution of monitored air quality indicators, and Q(x) is the corresponding distribution of air quality indicators predicted by the model.

7. The air quality prediction and big data system of shared cloud integrated ecology according to claim 1, characterized in that: The large screen interface of the monitoring center of the air health status visualization module displays various visualization charts representing air health status indicators in real time, intuitively presenting the changing trend and current status of the air health status; Centrally display the status information of all terminal servers and site monitoring devices, including the online and offline status of the equipment, the running time, the fault alarm situation, and collect and store the log information of each terminal, and provide log query and analysis functions; The terminal server is configured with a fault self-checking mechanism. When the server detects that it has a fault, it attempts to automatically restart or perform other preset recovery operations, and reports the fault information and recovery results to the monitoring center; The APP monitors the operating status of various equipment in the site in real time. Users can monitor the operation of the equipment through the APP and promptly discover equipment failures or abnormal conditions. The APP displays real-time environmental data to ensure that the site is in a suitable operating environment to avoid equipment damage or unstable operation due to environmental factors. Data alarm rules are set. When the monitoring data is abnormal or exceeds the preset threshold, the APP will promptly issue an alarm notification to remind users to pay attention and recommend effective response measures to prevent the problem from expanding.

Images