This paper proposes a prediction algorithm combining principal component analysis (PCA), grid search (GS) and K-nearest neighbor (KNN). First, to solve the multicollinearity problem in multiple regression, principal component analysis is applied to select the principal components of regression variables; then, the K-nearest neighbor regression prediction model is used for data training, and the grid search is used to obtain better prediction model parameters to solve the problem of parameter selection in the traditional prediction model of K-nearest neighbor regression; finally, the optimized prediction model is used to demonstrate the regional agricultural carbon emissions, taking Zhejiang Province of China as a case. The results show that the algorithm is superior to other prediction models in prediction accuracy, and can accurately predict regional agricultural carbon emissions.

Scientific analysis of regional agricultural carbon emission prediction models and empirical studies are of great practical significance to the realization of low-carbon agriculture, which can help revitalize and build up ecological and beautiful countryside in China. This paper takes agriculture in Guangdong Province, China, as the research object, and uses the extended STIPAT model to construct an indicator system for the factors influencing agricultural carbon emissions in Guangdong. Based on this system, a combined Isomap–ACO–ET prediction model combing the isometric mapping algorithm (Isomap), ant colony algorithm (ACO) and extreme random tree algorithm (ET) was used to predict agriculture carbon emissions in Guangdong Province under five scenarios. Effective predictions can be made for agricultural carbon emissions in Guangdong Province, which are expected to fluctuate between 11,142,200 tons and 11,386,000 tons in 2030. And compared with other machine learning and neural network models, the Isomap–ACO–ET model has a better prediction performance with an MSE of 0.00018 and an accuracy of 98.7%. To develop low-carbon agriculture in Guangdong Province, we should improve farming methods, reduce the intensity of agrochemical application, strengthen the development and promotion of agricultural energy-saving and emission reduction technologies and low-carbon energy sources, reduce the intensity of carbon emissions from agricultural energy consumption, optimize the agricultural planting structure, and develop green agricultural products and agro-ecological tourism according to local conditions. This will promote the development of agriculture in Guangdong Province in a green and sustainable direction.

The prediction of regional agricultural carbon emissions is of great significance to regional environmental protection and sustainable development of regional agriculture. This paper puts forward a combined prediction model integrating Partial Least Squares (PLS), Simulated Annealing (SA) and Adaptive Boosting (AdaBoost) to predict regional agricultural carbon emissions, which overcomes the shortcomings of insufficient accuracy of a single model prediction. This paper conducts a demonstrative study on the agricultural carbon emissions in Fujian Province, China to verify the feasibility and effectiveness of the PLS-SA-AdaBoost combined prediction model. The experimental results show that PLS-SA-AdaBoost combined prediction model has a higher precision than SA-AdaBoost model and PLS-SA-AdaBoost model; meanwhile PLS-SA-AdaBoost combined prediction model shows obvious advantages compared with other combined prediction models. In terms of five different scenarios, the paper adopts PLS-SA-AdaBoost combined prediction model to predict the future trend of agricultural carbon emissions in Fujian Province.

No abstract