温室环境温度预测自适应机理模型参数在线识别方法

温室小气候模型具有非线性、强干扰、时变等特性,常用的线性化模型或离线辨识模型无法有效预测温室环境动态变化规律,模型精度不高直接影响温室环境控制性能。针对模型的这种复杂特性,该文以温度模型为例,提出利用连续-离散递推预测误差算法,对非线性模型参数和状态进行在线估计,以提高模型精度。首先,考虑加热和通风2种控制输入,建立了温室温度模型。然后,论述了该算法的原理和优势,算法通过引入并调整增益矩阵,对系统噪声进行在线更新,减少了系统噪声初始化偏差对参数估计收敛性的影响。最后,基于试验温室实测数据,对温室温度模型进行了在线辨识仿真,结果表明,利用连续-离散递推预测误差算法辨识模型拟合度为93.7%,扩展卡尔曼滤波器拟合度为89.5%,该文提出的算法在预测温室温度方面更为有效。

Online identification method of parameters for greenhouse temperature prediction self-adapting mechanism model

Abstract: The greenhouse climate model has the characteristics of nonlinearity, strong disturbance and time variance. The commonly used methods to deal with the complex model are to linearize the original nonlinear models or to do some offline identification research. Without considering the complex characteristics completely, the usual modeling methods cannot predict the greenhouse climate dynamic behaviour effectively. In this paper, on account of the complex characteristics of greenhouse climate system, taking the temperature model as the example, the continuous-discrete recursive error algorithm was used to identify combined parameter and state online. Firstly, describe the greenhouse temperature model. The greenhouse temperature is affected by the heat load imposed on the greenhouse by the sun, the energy lost to the external air because of transmission through the greenhouse cover, the heat transfer between the internal air and soil, the heat loss by crop transpiration, the heat lost through natural ventilation of the roof windows and the energy supply from the heating system. In all of the model parameters, due to the fact that external solar radiation has a great effect on greenhouse temperature, radiation conversion factor changes over time and the parameters related to heating and ventilation are fundamental but difficult to obtain, this paper attempted to estimate and update 5 key parameters online. Secondly, continuous-discrete recursive prediction error algorithm to estimate combined parameters and states online was developed. This algorithm is appropriate for a continuous-discrete system, which is defined as a dynamic system with a continuous state function, and the observation function is discrete. The algorithm estimates the parameters and states by minimizing the error sum of squares between predicted values and measured values, which is usual in this technology. Compared with other traditional estimation algorithm such as the extended Kalman filter, the big difference of the algorithm is that there is no need to set the initial value of system noise precisely. It defines the system noise in real time by introducing an extra parameter as gain matrix and estimating it online. The algorithm adjusts the system noise to match the model predicted values and actual values by regulating the gain matrix. As estimating the system noise for the greenhouse temperature system beforehand is extremely difficult, the advantage of the proposed algorithm enhances the feasibility of its application in practice. At last, in order to test the developed algorithm, the model identification results were compared between the continuous-discrete recursive prediction error algorithm and the extended Kalman filter in MATLAB. The simulation was based on the measured data containing outside temperature, outside solar radiation, control inputs and temperature of an experiment greenhouse. The results showed that the proposed algorithm could lead to a higher model fit value of 93.7% compared with 89.5% of the extended Kalman filter. The gain matrix varied from zero to non-zero, and it nearly maintained stable at a non-zero constant in the end of each test day. The changing process indicated that there were errors between the model predicted value and the measured value in the initial, and the errors could be compensated by regulating the gain matrix. The different values of gain matrix in 2 days showed that system noise may vary largely in different condition and it should not be set as a constant. From the simulation results and the data analysis, it can be known that the proposed continuous-discrete recursive prediction error algorithm can estimate the temperature well and improve the model accuracy and validity.