Abstract: To explore a detection model with higher detection accuracy and suitable for capacitive soil moisture content sensors with low measurement frequency, 160 soil samples with moisture content ranging from 6.39% to 31.43% were configured with soil as the research object, the capacitance data of the soil in the range of measurement frequency ranging from 1 to 100 kHz were collected by using a digital bridge, and the multivariate linear regression (MLR), ridge regression (RR), support vector regression (SVR), decision tree regression algorithm (CART) and back propagation neural network (BPNN) algorithm models were established to predict soil moisture content, and the prediction performance of the model was compared and analyzed. The experimental results show that the prediction performance of the soil moisture content detection model constructed based on the SVR algorithm is better. By introducing sparrow search algorithm (SSA), particle swarm optimization algorithm (PSO) and whale optimization algorithm (WOA) to optimize and analyze the performance of the SVR algorithm model respectively, the detection model built by using PSO—SVR algorithm performs the best, with the R2, RMSE and RPD of the prediction set of 0.947, 0.012 and 4.363, respectively.  The model is used to predict 35 soil samples with moisture content ranging from 8.07%-26.43%, the regression coefficient R2 of the predicted versus measured values of moisture content was 0.953, and the root mean square error RMSE was 0.009, the absolute error range was -1.92%-1.68%. It is proved that the construction of soil water content prediction model based on PSO—SVR algorithm has high accuracy and good robustness.

Key words: soil, moisture content, prediction model, capacitance, measurement frequency, PSO—SVR

摘要： 为探究一种高检测精度且适用于低测量频率电容式土壤含水率传感器的检测模型，以土壤为研究对象，配置160份含水率为6.39%～31.43%的土壤样本，使用数字电桥采集土壤样本在测量频率为1～100kHz的电容数据，建立预测土壤含水率的多元线性回归（MLR）、岭回归（RR）、支持向量回归（SVR）、决策树回归（CART）和BP神经网络（BPNN）算法模型，对比分析模型的预测性能。结果表明，基于SVR算法构建的土壤含水率检测模型预测性能较佳。通过引入麻雀搜索（SSA）、粒子群优化算法（PSO）和鲸鱼优化算法（WOA）算法分别对SVR算法模型优化后的性能分析，采用PSO—SVR算法建立的检测模型性能最佳，其预测集的决定系数R2、均方根误差RMSE和相对百分比偏差RPD分别为0.947、0.012和4.363。利用该模型对35份含水率在8.07%～26.43%的土壤样本进行预测，含水率预测值与实测值的回归系数R2为0.953，RMSE为0.009，绝对误差范围为-1.92%～1.68%。由此证明，基于PSO—SVR算法构建土壤含水率预测模型具有较高的准确性和较好的鲁棒性。

关键词: 土壤, 含水率, 预测模型, 电容, 测量频率, PSO—SVR