基于MA-ARIMA-GASVR的食用菌温室温度预测

食用菌温室温度具有时变、非线性、多耦合特性,准确预测对稳定食用菌生产具有重要意义。本研究从挖掘温室历史温度数据时序信息角度出发,提出一种MA-ARIMA-GASVR组合方法建立温度预测模型,利用移动平均方法将历史温度序列分解成线性序列和残差序列,然后采用移动平均差分自回归模预测线性序列的趋势,再将移动平均差分自回归预测值、历史残差数据、历史温度数据作为支持向量回归模型的输入,并结合遗传算法优化支持向量回归模型参数改善其性能,从而获得更符合实际情况的温度预测值。最后选取实测温度数据作为训练集,对未来2d的温度进行预测验证。结果显示,MA-ARIMA-GASVR组合方法能更好地拟合原始温度数据,间隔1h的均方误差、平均绝对误差和平均绝对百分误差分别为0.18、0.36和1.34,均显示本研究方法预测精度优于支持向量回归、遗传算法优化的支持向量回归单一模型,也优于未经移动平均以及未经遗传算法优化的组合模型;此外,间隔6h的均方误差、平均绝对误差和平均绝对百分误差为0.29、0.52和1.95,说明本研究方法还能满足6 h以内的多步预测,为食用菌生产者预留更多调整时间。

Prediction of temperature in edible fungi greenhouse based on MA-ARIMA-GASVR

The temperature in edible fungi greenhouse has the characteristics of time-variant, nonlinear and multi-coupling, so accurate and effective temperature predictions can effectively help growers adjust the greenhouse environment and prevent edible fungus production and quality decline. Based on the perspective of mining the time-series information in historical temperature data. This paper described the specific steps to realize the MA-ARIMA-GASVR-based hybrid combination method to predict the temperature in the edible fungus greenhouse. Firstly, we assumed that the historical temperature series data of edible fungus greenhouse was a dynamic combination of linear and non-linear components, the historical temperature sequences were decomposed into linear sequences and residual sequences using the moving averages (MA) method. Then, time series analysis was conducted to established the model of the autoregressive integrated moving average (ARIMA) by using linear sequence after the decomposition of the moving averages, and the future trend of linear sequences was predicted by the established model. Afterward, to better fit the relationship between temperature trends and various noises in the environment, the autoregressive integrated moving average model prediction value, the historical residual data and the historical temperature data were employed as the input of the support vector regression (SVR) model, and the genetic algorithm (GA) was used to optimize the parameters of the support vector regression model to improve its performance, the parameters being optimized are penalty parameter and radial basis function kernel parameters in the support vector regression model. Finally, the hybrid model output was the temperature prediction value which was more in line with the actual situation. Moreover, the hybrid method was verified using the experimental data from the edible fungus greenhouse in Beijing. In this paper, a representative edible fungus greenhouse was selected as the experimental object according to the observation time requirements and the time-varying needs of edible fungus greenhouse temperature, which was located in the Daxing District of Beijing. A total of 2 208 measured edible fungus greenhouse temperature data were collected from July 1st, 2019 to September 30th, 2019 during the experiment. The experimental data acquisition device used the JXBS-7001 temperature monitoring sensor was used to automatically collect and record the experimental data. Three sets of sensors were deployed in the edible fungus greenhouse to record the experimental data set which included the average temperature data. We trained the proposed model by using data from July 3rd, 2019 to July 16th, 2019 and forecasted the temperature of the next two days and compared temperature prediction experiments with different models and different time intervals. The results indicated that the MA-ARIMA-GASVR-based hybrid model could better fit the original temperature data, the mean squared error, the mean absolute error and mean absolute percentage error of an hour interval temperature were 0.18, 0.36, 1.34, and three error evaluation indexes all showed that the prediction accuracy of the hybrid method in this paper was better than the single models of support vector regression and support vector regression optimized by genetic algorithm, and it was also superior to the hybrid methods which were not processed by moving averages method or optimized by genetic algorithm. Besides, the mean squared error, the mean absolute error and mean absolute percentage error of 6hours interval temperature were 0.29, 0.52, 1.95. the hybrid method in this paper can satisfy the multi-step prediction within 6 hours, which could provide more time for edible fungus producers to adjust the temperature in the greenhouse.