Greenhouse microclimate mid-and-long term prediction method based on rolling BP neural network

Abstract

The invention proposes a greenhouse microclimate mid-and-long term prediction method based on a rolling BP neural network, and the method comprises the steps: constructing one BP neural network at each moment according to prediction time, and finally forming a rolling BP neural network group. Specifically, the method comprises two operation stages: employing an automatic coder to carry out the non-supervised learning, obtaining good initial network parameters, optimizing the network parameters through an improved local PSO (Particle Swarm Optimization) method, and building an initial BP neural network; taking the output of a former network as a part of input of a latter network on the basis of the initial BP neural network, and carrying out the rolling training and prediction. The method can achieve the more accurate prediction of the mid-and-long term environment change trend of greenhouses in different seasons in different regions, and effectively improves the greenhouse microclimate prediction precision.

Description

technical field [0001] The invention belongs to the field of protected agricultural environment prediction, in particular to a method for medium and long-term prediction of greenhouse microclimate based on rolling BP neural network. Background technique [0002] The efficient production of greenhouses depends on a suitable greenhouse microclimate environment, and the establishment of a high-precision greenhouse microclimate medium- and long-term prediction model is of great significance for the realization of optimal control of the greenhouse environment. Although the threshold control method commonly used in greenhouses is simple and easy to implement, it has high energy consumption and poor system stability. Based on proportional-integral-derivative (Proportion-Integral-Derivative, PID) controllers and model predictive control (Model Predictive Control, MPC) and other automatic control methods, it has high reliability and low energy consumption, but it needs to predict mul...

AI Technical Summary

Problems solved by technology

[0003] At present, scholars at home and abroad have established microclimate simulation models based on BP neural network for different greenhouse environments, and achieved good results. Studies have shown that artificial neural networks are feasible in the prediction of greenhouse microclimate environments, but most of these prediction models can only perform single-step predictions. , that is, short-term forecasting, which cannot realize medium and long-term forecasting, and cannot meet the requirements of optimal regulation

In addition, the use of BP neural network modeling has certain advantages, but it also has some defects and shortcomings, such as easy to fall into local minimum, over-reliance on the selection of initial weights, and poor generalization ability. Therefore, the accuracy of BP neural network prediction There is still a lot of room for improvement

Many researchers in the past did not propose improved methods for the defects of the BP neural network, and only selected the best results to show. In fact, these results are not convincing to a certain extent.

Images