Spatial pollutant concentration fitting method and system based on recurrent convolutional neural network

Abstract

The present invention relates to a spatial pollutant concentration fitting method and system based on a recurrent convolutional neural network. The method comprises the following steps: S1, collecting real-time data from sensors within each purifier and monitor and calculating a single-time slice concentration fitting map of pollutants in the sensor neighborhood; S2, combining the results of step S1 with the sensor point map and calculating a preprocessing feature vector for the single-time slice concentration of spatial pollutants; S3, calculating a pre-fitting map of the single-time slice concentration of spatial pollutants based on the results of step S2; S4, combining the results of step S3 with a three-dimensional map of the indoor space and calculating a feature vector for the single-time slice concentration of spatial pollutants; S5, calculating a multi-layer residual input vector based on the results of step S2; S6, calculating the feature vector of the current final fitting map based on the results of steps S4 and S5; and S7, calculating a three-dimensional numerical map of spatial pollutant concentration based on the results of step S6. The present invention has high fitting accuracy for spatial pollutants and can effectively track pollutants.

Description

Technical Field

[0001] The present invention relates to the technical field of public space air purification, and in particular to a spatial pollutant concentration fitting method and system based on a recurrent convolutional neural network. Background Art

[0002] In public space air purification solutions, the traditional method is to read fixed point values ​​and control the purifier to purify when any of them exceed the standard value; a more advanced method is to use traditional methods to fit spatial pollutants, but the fitting accuracy of traditional methods is low, pollutants cannot be tracked, the spatial pollutant fitting effect is not ideal, and the working status and efficiency of the purifier cannot be adjusted according to actual conditions. Summary of the Invention

[0003] In response to the problems existing in the prior art, the purpose of the present invention is to provide a spatial pollutant concentration fitting method and system based on a recurrent convolutional neural network, which has high fitting accuracy for spatial pollutants and can effectively track pollutants.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] A spatial pollutant concentration fitting method based on a recurrent convolutional neural network comprises the following steps:

[0006] S1: Collect real-time data from sensors in each indoor space purifier and monitor, input the data into a pre-trained indoor space sensor monitoring effect pre-training module, and output a single time slice concentration fitting graph of the neighborhood pollutants of the indoor space sensor;

[0007] S2: Input the single-time-slice concentration fitting map of the indoor space sensor's neighborhood pollutants and their corresponding placement points in the indoor space purifier point map and the monitor point map into the trained convolutional neural network spatial pollutant single-time-slice concentration feature vector pre-calculation module, extract the sensor's historical performance feature vector, and output the spatial pollutant single-time-slice concentration pre-processing feature vector;

[0008] S3: Input the spatial pollutant single time slice concentration preprocessing feature vector into the trained deconvolution neural network pollutant single time slice concentration prefitting module, and output the spatial pollutant single time slice concentration prefitting map;

[0009] S4: Input the spatial pollutant single time slice concentration pre-fitting map and the indoor space three-dimensional map into the trained convolutional neural network spatial pollutant single time slice concentration fitting module, extract the spatial pollutant concentration feature vector, and output the spatial pollutant single time slice concentration feature vector;

[0010] S5: Input the preprocessed feature vector of the single time slice concentration of the spatial pollutant into the trained recurrent neural network residual layer vector input module, match the input structure required by the recurrent residual neural network, and output a multi-layer residual input vector;

[0011] S6: Input the spatial pollutant single time slice concentration feature vector and the multi-layer residual input vector into the trained cyclic residual neural network spatial pollutant concentration feature vector calculation module; input the combined vector of the multi-layer residual input vector and the input of the previous residual layer into each residual layer, and use the cyclic neural network to fit the current spatial pollutant with the spatial pollutant single time slice concentration feature vector and the multi-layer residual input vector at the previous time point to generate the feature vector of the current final fitting graph;

[0012] S7: Input the feature vector of the current final fitting graph into the trained deconvolution neural network pollutant concentration fitting module, and output a three-dimensional numerical graph of spatial pollutant concentration.

[0013] Preferably, step S5 is performed before step S3 or synchronously with step S3 or before step S4 or synchronously with step S4.

[0014] Preferably, step S8 is also included, which is to input the characteristic vector of the current final fitting graph and the multi-layer residual input vector into the trained cyclic residual neural network purification efficiency fitting module; input the merged vector of the multi-layer residual input vector and the input of the previous residual layer into each residual layer, and use the cyclic neural network to combine the characteristic vector of the final fitting graph at the previous time point and the multi-layer residual input vector to evaluate the current purification efficiency, and output multi-angle evaluation results.

[0015] Preferably, step S8 is performed before step S7 or is performed synchronously with step S7.

[0016] In addition, the present invention also discloses a spatial pollutant concentration fitting system based on a recurrent convolutional neural network, which includes

[0017] The indoor space sensor monitoring effect pre-training module is used to collect real-time data from the sensors in each purifier and monitor in the indoor space, and analyze and process it to obtain a single time slice concentration fitting diagram of the neighborhood pollutants of the indoor space sensor;

[0018] The convolutional neural network spatial pollutant single time slice concentration feature vector pre-calculation module is used to process the neighborhood pollutant single time slice concentration fitting map of the indoor space sensor and combine it with the placement points in the indoor space purifier point map and the monitor point map to obtain the spatial pollutant single time slice concentration pre-processing feature vector;

[0019] The deconvolution neural network pollutant single time slice concentration pre-fitting module is used to process the spatial pollutant single time slice concentration pre-processing feature vector to obtain the spatial pollutant single time slice concentration pre-fitting map;

[0020] The convolutional neural network spatial pollutant single time slice concentration fitting module is used to process the pre-fitted map of spatial pollutant single time slice concentration and obtain the spatial pollutant single time slice concentration feature vector by combining it with the three-dimensional map of the indoor space;

[0021] The recurrent neural network residual layer vector input module is used to process the preprocessed feature vector of the single time slice concentration of spatial pollutants to obtain a multi-layer residual input vector;

[0022] The cyclic residual neural network spatial pollutant concentration feature vector calculation module is used to process the spatial pollutant single time slice concentration feature vector and the multi-layer residual input vector to obtain the feature vector of the current final fitting graph;

[0023] The deconvolution neural network pollutant concentration fitting module is used to process the eigenvector of the current final fitting graph to obtain a three-dimensional numerical graph of spatial pollutant concentration.

[0024] Preferably, it also includes a cyclic residual neural network purification efficiency fitting module, which is used to process the characteristic vector of the current final fitting graph and the multi-layer residual input vector to obtain a spatial pollutant purification efficiency evaluation result.

[0025] After adopting the above scheme, the present invention collects and processes the real-time data of indoor space purifiers and monitors through a cyclic convolutional neural network to obtain a single-time slice concentration fitting map of pollutants in the sensor field. The single-time slice concentration preprocessing feature vector of spatial pollutants is obtained by combining the distribution point map of each sensor. The pre-fitting map of the single-time slice concentration of spatial pollutants and the multi-layer residual input vector can be further derived through the trained neural network. The single-time slice concentration pre-fitting map of spatial pollutants is combined with the spatial three-dimensional map to obtain the single-time slice concentration feature vector of spatial pollutants. The feature vector of the current final fitting map can be fitted and generated by combining the multi-layer residual input vector. Finally, the trained neural network is used to obtain a three-dimensional numerical map of spatial pollutant concentration with high fitting accuracy and tracking effectiveness for spatial pollutants. According to the three-dimensional numerical map of spatial pollutant concentration, the purification system can reasonably plan the operating status and efficiency of each purifier in real time, improve the purification efficiency of the entire purification system and enhance the energy management and control capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0027] like Figure 1 As shown, the present invention discloses a spatial pollutant concentration fitting method based on a recurrent convolutional neural network, which includes the following steps:

[0028] S1: Collect real-time data from sensors in each purifier and monitor in the indoor space, input the data into the trained indoor space sensor monitoring effect pre-training module, and output the single time slice concentration fitting graph of the neighborhood pollutants of the indoor space sensor.

[0029] S2: Input the single-time-slice concentration fitting map of the neighborhood pollutants of the indoor space sensor and its corresponding placement points in the indoor space purifier point map and the monitor point map into the trained convolutional neural network spatial pollutant single-time-slice concentration feature vector pre-calculation module, extract the sensor historical performance feature vector, and output the spatial pollutant single-time-slice concentration pre-processing feature vector.

[0030] S3: Input the spatial pollutant single time slice concentration preprocessing feature vector into the trained deconvolution neural network pollutant single time slice concentration prefitting module, and output the spatial pollutant single time slice concentration prefitting map.

[0031] S4: Input the pre-fitted map of spatial pollutant single time slice concentration and the three-dimensional map of indoor space into the trained convolutional neural network spatial pollutant single time slice concentration fitting module, extract the spatial pollutant concentration feature vector, and output the spatial pollutant single time slice concentration feature vector.

[0032] S5: Input the preprocessed feature vector of the single time slice concentration of spatial pollutants into the trained recurrent neural network residual layer vector input module, match the input structure required by the recurrent residual neural network, and output a multi-layer residual input vector.

[0033] S6: Input the spatial pollutant single time slice concentration feature vector and the multi-layer residual input vector into the trained cyclic residual neural network spatial pollutant concentration feature vector calculation module; input the merged vector of the multi-layer residual input vector and the input of the previous residual layer into each residual layer, and use the cyclic neural network to fit the current spatial pollutant in combination with the spatial pollutant single time slice concentration feature vector and the multi-layer residual input vector at the previous time point to generate the feature vector of the current final fitting graph.

[0034] S7: Input the feature vector of the current final fitting graph into the trained deconvolution neural network pollutant concentration fitting module, and output a three-dimensional numerical graph of spatial pollutant concentration.

[0035] S8: Input the feature vector of the current final fitting graph and the multi-layer residual input vector into the trained recurrent residual neural network purification efficiency fitting module; input the merged vector of the multi-layer residual input vector and the input of the previous residual layer into each residual layer, and use the recurrent neural network to combine the feature vector of the final fitting graph at the previous time point and the multi-layer residual input vector to evaluate the current purification efficiency, and output the multi-angle evaluation results.

[0036] It should be noted that, in the above steps, step S5 may be executed before step S3 or simultaneously with step S3, or before step S4 or simultaneously with step S4, and it is sufficient that step S5 is completed before step S6. Step S8 may be executed before step S7 or simultaneously with step S7.

[0037] Based on the same inventive concept, the present invention also discloses a spatial pollutant concentration fitting system based on a recurrent convolutional neural network, which includes

[0038] The indoor space sensor monitoring effect pre-training module is used to collect real-time data from the sensors in each purifier and monitor in the indoor space, and analyze and process it to obtain a single time slice concentration fitting diagram of the neighborhood pollutants of the indoor space sensor;

[0039] The convolutional neural network spatial pollutant single time slice concentration feature vector pre-calculation module is used to process the neighborhood pollutant single time slice concentration fitting map of the indoor space sensor and combine it with the placement points in the indoor space purifier point map and the monitor point map to obtain the spatial pollutant single time slice concentration pre-processing feature vector;

[0040] The deconvolution neural network pollutant single time slice concentration pre-fitting module is used to process the spatial pollutant single time slice concentration pre-processing feature vector to obtain the spatial pollutant single time slice concentration pre-fitting map;

[0041] The convolutional neural network spatial pollutant single time slice concentration fitting module is used to process the pre-fitted map of spatial pollutant single time slice concentration and obtain the spatial pollutant single time slice concentration feature vector by combining it with the three-dimensional map of the indoor space;

[0042] The recurrent neural network residual layer vector input module is used to process the preprocessed feature vector of the single time slice concentration of spatial pollutants to obtain a multi-layer residual input vector;

[0043] The cyclic residual neural network spatial pollutant concentration feature vector calculation module is used to process the spatial pollutant single time slice concentration feature vector and the multi-layer residual input vector to obtain the feature vector of the current final fitting graph;

[0044] The deconvolution neural network pollutant concentration fitting module is used to process the eigenvectors of the current final fitting graph to obtain a three-dimensional numerical graph of spatial pollutant concentration;

[0045] The cyclic residual neural network purification efficiency fitting module is used to process the eigenvector of the current final fitting graph and the multi-layer residual input vector to obtain the spatial pollutant purification efficiency evaluation results.

[0046] The key to this invention lies in collecting and processing real-time data from indoor space purifiers and monitors using a recurrent convolutional neural network (RCN) to generate a sensor-domain pollutant single-time-slice concentration fitting map. This is then combined with the distribution point maps of each sensor to generate a pre-processed feature vector for the spatial pollutant single-time-slice concentration. The trained neural network can then be used to further derive the pre-processed spatial pollutant single-time-slice concentration pre-fit map and a multi-layer residual input vector. The pre-processed spatial pollutant single-time-slice concentration pre-fit map is combined with the three-dimensional spatial map to generate a feature vector for the spatial pollutant single-time-slice concentration. This feature vector is then combined with the multi-layer residual input vector to generate the feature vector of the final fitting map. Finally, the trained neural network is used to generate a three-dimensional numerical map of spatial pollutant concentration that has high fitting accuracy and is effective for tracking spatial pollutants. Furthermore, the feature vector of the final fitting map and the multi-layer residual input vector are calculated using the RCNN purification efficiency fitting module to generate a spatial pollutant purification efficiency assessment. Based on the three-dimensional numerical map of spatial pollutant concentration, the purification system can rationally plan the operating status and efficiency of each purifier in real time, improving the purification efficiency of the entire purification system and enhancing energy management capabilities. Combined with the spatial pollutant purification efficiency assessment, a feedback mechanism can be further established to promote intelligent and efficient air purification operations.

[0047] The above description is merely an embodiment of the present invention and does not limit the technical scope of the present invention. Therefore, any minor modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A spatial pollutant concentration fitting method based on a recurrent convolutional neural network is characterized by: The following steps are involved: S1: Collect real-time data from sensors in each indoor space purifier and monitor, input the data into a pre-trained indoor space sensor monitoring effect pre-training module, and output a single time slice concentration fitting graph of the neighborhood pollutants of the indoor space sensor; S2: Input the single-time-slice concentration fitting map of the indoor space sensor's neighborhood pollutants and their corresponding placement points in the indoor space purifier point map and the monitor point map into the trained convolutional neural network spatial pollutant single-time-slice concentration feature vector pre-calculation module, extract the sensor's historical performance feature vector, and output the spatial pollutant single-time-slice concentration pre-processing feature vector; S3: Input the spatial pollutant single time slice concentration preprocessing feature vector into the trained deconvolution neural network pollutant single time slice concentration prefitting module, and output the spatial pollutant single time slice concentration prefitting map; S4: Input the spatial pollutant single time slice concentration pre-fitting map and the indoor space three-dimensional map into the trained convolutional neural network spatial pollutant single time slice concentration fitting module, extract the spatial pollutant concentration feature vector, and output the spatial pollutant single time slice concentration feature vector; S5: Input the preprocessed feature vector of the single time slice concentration of the spatial pollutant into the trained recurrent neural network residual layer vector input module, match the input structure required by the recurrent residual neural network, and output a multi-layer residual input vector; S6: Input the spatial pollutant single time slice concentration feature vector and the multi-layer residual input vector into the trained cyclic residual neural network spatial pollutant concentration feature vector calculation module; input the combined vector of the multi-layer residual input vector and the input of the previous residual layer into each residual layer, and use the cyclic neural network to fit the current spatial pollutant with the spatial pollutant single time slice concentration feature vector and the multi-layer residual input vector at the previous time point to generate the feature vector of the current final fitting graph; S7: Input the feature vector of the current final fitting graph into the trained deconvolution neural network pollutant concentration fitting module, and output a three-dimensional numerical graph of spatial pollutant concentration.

2. The spatial pollutant concentration fitting method based on recurrent convolutional neural network according to claim 1 is characterized in that: Step S5 is executed before step S3 or synchronously with step S3 or before step S4 or synchronously with step S4.

3. The spatial pollutant concentration fitting method based on recurrent convolutional neural network according to claim 1 or 2, characterized in that: It also includes step S8, which is to input the characteristic vector of the current final fitting graph and the multi-layer residual input vector into the trained cyclic residual neural network purification efficiency fitting module; input the merged vector of the multi-layer residual input vector and the input of the previous residual layer into each residual layer, and use the cyclic neural network to combine the characteristic vector of the final fitting graph and the multi-layer residual input vector at the previous time point to evaluate the current purification efficiency, and output multi-angle evaluation results.

4. The spatial pollutant concentration fitting method based on recurrent convolutional neural network according to claim 3 is characterized in that: Step S8 is executed before step S7 or synchronously with step S7.

5. Spatial pollutant concentration fitting system based on recurrent convolutional neural network, characterized by: include The indoor space sensor monitoring effect pre-training module is used to collect real-time data from the sensors in each purifier and monitor in the indoor space, and analyze and process it to obtain a single time slice concentration fitting diagram of the neighborhood pollutants of the indoor space sensor; The convolutional neural network spatial pollutant single time slice concentration feature vector pre-calculation module is used to process the neighborhood pollutant single time slice concentration fitting map of the indoor space sensor and combine it with the placement points in the indoor space purifier point map and the monitor point map to obtain the spatial pollutant single time slice concentration pre-processing feature vector; The deconvolution neural network pollutant single time slice concentration pre-fitting module is used to process the spatial pollutant single time slice concentration pre-processing feature vector to obtain the spatial pollutant single time slice concentration pre-fitting map; The convolutional neural network spatial pollutant single time slice concentration fitting module is used to process the pre-fitted map of spatial pollutant single time slice concentration and obtain the spatial pollutant single time slice concentration feature vector by combining it with the three-dimensional map of the indoor space; The recurrent neural network residual layer vector input module is used to process the preprocessed feature vector of the single time slice concentration of spatial pollutants to obtain a multi-layer residual input vector; The cyclic residual neural network spatial pollutant concentration feature vector calculation module is used to process the spatial pollutant single time slice concentration feature vector and the multi-layer residual input vector to obtain the feature vector of the current final fitting graph; The deconvolution neural network pollutant concentration fitting module is used to process the eigenvector of the current final fitting graph to obtain a three-dimensional numerical graph of spatial pollutant concentration.

6. The spatial pollutant concentration fitting system based on recurrent convolutional neural network according to claim 5, characterized in that: It also includes a cyclic residual neural network purification efficiency fitting module, which is used to process the characteristic vector of the current final fitting graph and the multi-layer residual input vector to obtain a spatial pollutant purification efficiency evaluation result.

Images