Time-Space Domain Correlation Prediction Method of Air Pollutant Concentration

Abstract

The present invention relates to a method for predicting air pollutant concentration time-space correlation, including: step S1: taking PM2.5 as an example to predict target pollutants, and constructing a prediction model based on residual network and convolutional LSTM network; step S2: from Select appropriate training and test data from the environmental monitoring data to complete the initialization of the prediction model; Step S3: Train the prediction model step by step to obtain a neural network prediction model that can accurately predict PM2.5; Step S4: Use The verification set selects the hyperparameters (number of layers, number of nodes, learning rate) of the model until the model is optimal; Step S5: Use the verified prediction model to predict urban PM2.5. Compared with the existing technology, the present invention uses the convolutional LSTM network as the middle layer to realize the deep-level spatiotemporal correlation feature extraction of the spatial features extracted by the underlying ResNet network, thereby improving the prediction performance of the network model, and using the fully connected layer to receive volume The hidden state of the LSTM is accumulated to produce the final prediction result.

Description

technical field [0001] The invention relates to a method for predicting urban air pollutant concentration, in particular to a time-space correlation prediction method for air pollutant concentration. Background technique [0002] In recent years, the increasingly serious problem of air pollution has aroused widespread concern around the world. Pollutants such as PM2.5 and PM10 have a huge impact on people's life and health. The problem of air pollution is becoming more and more prominent. The analysis and prediction of air pollution are complex and dynamic, involving multiple departments, regions and fields. Accurate prediction of air pollution requires the processing of a large amount of related environmental data and environmental information. Various institutions attach importance to and focus on the improvement of air pollution response and processing capabilities, among which air pollution prediction technology is one of the current focus issues. At present, the new a...

AI Technical Summary

Problems solved by technology

However, in the face of the high dimensionality of air pollution data, the high diversity of influencing factors, and the massive pollution detection data, traditional numerical analysis models have encountered the following key problems: (1) The data sources used in the analysis models are too single, and most of them are only built on On a single set of pollution data, there is a lack of comprehensive consideration of other environmental factors, such as weather data; (2) In terms of space and time dimensions, traditional models lack the ability to mine the internal spatiotemporal correlation characteristics of pollution data, and cannot achieve deep internal data analysis. Contact extraction and response to the impact of abrupt weather environment; (3) The large-scale data application capacity of the model is limited, and it is difficult to mine the spatio-temporal correlation of pollutants from the perspective of big data

Images