在温室智能控制中，为准确判断温室内的环境状况，需要使用多传感器信息融合技术对温室环境信息进行综合，从而向温室管理者提供准确的温室环境信息，辅助温室管理者推断出农作物生长状态和下一步的温室调控措施，进而为温室农作物提供一个较为理想的生长环境，达到提高温室农作物产量和质量的目的。因此，本文针对传感器采集温室环境参数受外界因素干扰，导致采集数据非均匀分布的问题，以温室智能控制中多传感器信息融合算法为侧重点主要做了如下工作：

        (1) 针对传感器在工作区域内，由于出现设备故障、外界环境干扰或人为因素干预等，导致采集数据出现异常的问题，研究设计了一种基于格拉布斯准则的改进神经网络预测算法。首先，通过格拉布斯准则将传感器所获得的检测值分为正常和异常值两类。然后，运用改进神经网络预测模型对正常值进行训练，并用训练好的网络模型来预测测量中的异常值并替换之。最后，通过算术平均值法计算出温室内的温室环境信息。实验结果表明：改进神经网络预测模型的MAE值、MSE值和RMSE值均低于同类预测算法，更接近真实值；本文算法求得的标准差也低于未修正异常数据时求得的标准差。故本文算法对异常数据进行预处理后的融合结果比未修正异常数据的融合结果更加准确。

(2) 针对异质多传感器缺少有效信息融合模型的问题，研究设计了一种改进D-S证据理论算法对异质多传感器信息进行融合。该算法首先引入相容系数来表征证据之间的相容性，得到每个命题的权值矩阵，接着重新分配每个焦元的BPA，得到新的证据源。然后，引入可信度概念，使用证据可信度与证据焦元的平均支持度对合成规则进行改进，从而得到融合结果。实验结果表明：与其他算法相比，本文算法能在一定程度上解决D-S证据理论在处理高度冲突证据时存在的问题，融合出的结果更加合理，收敛速度更快。

(3) 设计多传感器信息融合算法在温室智能控制中的信息融合模型，将基于格拉布斯准则的改进神经网络预测算法和改进D-S证据理论算法分别作为局部、全局融合中心的核心算法。同时监控温室内栽培黄瓜的环境因子，并使用本文算法对其进行信息融合，从而准确判断温室环境状况，实施相应的调控措施，使温室黄瓜能在适宜的环境中生长。

        In the intelligent control of greenhouse, in order to accurately judge the environmental conditions in the greenhouse, it is necessary to use multi-sensor information fusion technology to synthesize the greenhouse environmental information, In order to provide the greenhouse managers with accurate greenhouse environment information, assist the greenhouse managers to infer the crop growth state and the next greenhouse control measures, and then provide a more ideal growth environment for greenhouse crops, so as to improve the yield and quality of greenhouse crops. Therefore, in view of the problem that the environmental parameters collected by sensors are interfered by external factors, which leads to the non-uniform distribution of the collected data, the following work is mainly done with the multi-sensor information fusion algorithm in the greenhouse intelligent control as the focus:
        (1) To solve the problem of abnormal data acquisition caused by equipment failure, external environment interference or human factors, an improved neural network prediction algorithm based on the Grubbs criterion was designed. Firstly, the detection value obtained by the sensor is divided into normal value and abnormal value by the Grubbs criterion. Then the improved neural network prediction model is used to train the normal value, and the trained network model is used to predict and replace the abnormal value in the measurement. Finally, the greenhouse environment information in the temperature room is calculated by the arithmetic average method. The experimental results show that the MAE value, MSE value and RMSE value of the improved neural network prediction model are lower than those of the same kind of prediction algorithm, and more close to the real value; the standard deviation of the algorithm is also lower than that of the uncorrected abnormal data, and the fusion result after preprocessing the abnormal data is more accurate than that of the uncorrected abnormal data.
        (2) To solve the problem that heterogeneous multisensor lacks effective information fusion model, an improved D-S evidence theory algorithm is designed to fuse heterogeneous multisensor information. Firstly, the consistency coefficient is introduced to represent the consistency between evidences, and the weight matrix of each proposition is obtained. Then, BPA of each focal element is redistributed, and a new evidence source is obtained. Then, the concept of credibility is introduced, and the synthesis rules are improved by using the credibility of evidence and the average support of the focal element of evidence, and the fusion results are obtained. The experimental results show that: compared with other methods, this method can solve the problems of D-S evidence theory in dealing with highly conflicting evidence to a certain extent, and the fusion result is more reasonable and the convergence speed is faster.
        (3) The information fusion model of multi-sensor information fusion algorithm in greenhouse intelligent control is designed. The improved neural network prediction algorithm based on Grubbs criterion and the improved D-S evidence theory algorithm are regarded as the core algorithms of local and global fusion centers respectively. At the same time, monitoring the environmental factors of Cucumber in greenhouse, and using this algorithm to carry out information fusion, so as to accurately judge the environmental conditions of greenhouse, and implement the corresponding control measures, so that the greenhouse cucumber can grow in the appropriate environment.