Research On Fertigation System For Greenhouse Cucumber Based On Multi-information Fusion

The fertigation technology was a modern agriculture technology, combined irrigation and fertilization efficiently, this technology was controllable, the nutrient solution concentration, irrigation quantity and irrigation time could be precisely controlled. The utilization rate of fertilizer and water were improved, saved water and fertilizer, increased production and reduced environmental pollution and other advantages, the fertigation technology was vital significance for the development of facilities agriculture of China. At present, the fertigation technology was not mature enough in China, most of the irrigation equipment still used the control mode: fixed phase, pre-set time and pre-set irrigation quantity according to the experience, the environment information of greenhouse, the growth information of crop, and the information of nutrient solution have not been used in the control system of fertigation machinery, led to the concentration of nutrient solution and irrigation quantity could not be precisely variable controlled. So far, China has not yet developed the fertigation machinery with the intelligent decision-making system based on the environment information, the growth information of crop, and the nutrient solution information.In order to realize intelligent fertigation for facility cultivation of China, the greenhouse cucumber as the object in this paper, to research the fertigation system for greenhouse cucumber based on multi-information fusion, the fusion method of greenhouse environmental information and the growth information detection method for cucumber were explored, the multi-channel fertigation machinery was designed, and the intelligent control models of irrigation quantity and electrical conductivity of nutrient solution based on multi-information fusion were researched. The main research contents and conclusions were as follows:(1) The maldistribution of greenhouse environment was taken into accout, the greenhouse environment information detection system based on ECO-WATCH data logger was designed. To realize more homologous greenhouse environment information fusion by the adaptive weighted fusion algorithm, the fusion values of environmental information as the input for the fertigation decision-making system, so as to improve credibility,fault tolerance and reliability of the input for the system. The performance between the adaptive weighted fusion algorithm and the average fusion algorithm were compared, three evaluation parameters: the average absolute error, standard deviation and variation coefficient were choosed, the results showed that the adaptive weighted fusion algorithm was superior to the average fusion algorithm.(2) In order to detect the growth parameters of greenhouse cucumbers used on-line nondestructive test technique, to provide representative growth information of cucumbers for the fertigation system. The machine vision technology was used, to capture the cucumbers canopy image under natural lighting condition, and three segmentation algorithms: Excess Green minus Excess Red, Excess Green and normalized difference indices were used to extract the canopy area of the flora. Flora canopy characteristic parameters (covering ratio, canopy length and canopy width) were extracted. Flora canopy characteristic parameters were combined with the parameters of the flora obtained by artificial measurement: stem height, stem diameter, leaf number and fruit number, the inversion models for the growth parameters of greenhouse cucumbers were built. The new validation figures of cucumbers were built, to verify the performance of inversion models.The results showed that: the coefficients of determination (R2) between the inversion values and measured values reached significant level, when the cultivation modes: 4 lines×4 columns, 4 lines×3 columns and 4 lines×2 columns, different cultivation ways could be accurate inversioned, the performance was stable.(3) To realize multi-channel nutrient solution and water precisely mixing,achieved precision control of irrigation quantity and electrical conductivity. The multi-channel fertigation machinery was designed based on programmable logic controller and human machine interface. The cucumber as the object in this paper, to prepare the nutrient solution based on cucumber nutrient solution formula, the dilution model of nutrient solution was built, provided the basis for mixing. In order to solve the long time delay problem, the proportion-fuzzy control algorithm was used, to realize precisely mixing. The results showed that the mixing control algorithm was stable, the variation coefficient was less than 2.5 %, the control error of Ec value was less than 0.05 mS·cm-1 The irrigation uniformity index of each channel showed that the maximum variation coefficient of Ec and pH were 2.49 % and 0.98 %, respectively, evenness was greater than 98.16 %, the overall water distribution uniformity coefficient was greater than 97.98 %, showed that nutrient solution and water were mixed evenly.(4) To solve the problem of intelligent controlling irrigation quantity for greenhouse cucumber, the prediction models of transpiration rate for greenhouse cucumber, and evaporation model of coconut chaff were established, respectively, and the control model of irrigation quantity for greenhouse cucumber based on the multi-information fusion was built. The prediction model of nonlinear time series for greenhouse cucumber transpiration rate based on wavelet transformation (WT) and nonlinear autoregressive with external input(NARX) dynamic neural network was designed. The low-frequency and high-frequency time series by the method of wavelet decomposition-reconstruction were used to establish the prediction models of NARX dynamic neural sub networks respectively, the predicted value of the subnets were accumulated as the predicted value of transpiration rate of cucumber. The results showed that: the R2 between accumulated predictive values by two layers of WT-NARX and the measured values of transpiration rate was 0.974, the mean absolute error (MAE) was 4.42 g·h-1. The R2 between predicted value of the original time series by the NARX dynamic neural network and the measured value of transpiration rate was 0.856, the MAE was 10.09 g·h-1. The forecast performance of WT-NARX was better than NAR and BP neural network under the same structure of network. The simulated values by Penman-Monteith equation has a remarkable correlation to measure values, the R2 was 0.900, SE was 0.0083 g·m-2 s-1, relative average deviation (RAD) was 36.42 %.According to the accumulated thermal effectiveness and photosynthesis active radiation,average temperature and effective accumulated temperature, average humidity of each day,moisture evaporation equation of coconut chaff was established, the R2 was 0.956, SE was 207.73 g·m-2·d-1, RAD was 8.15 %. The irrigation quantity control model of greenhouse cucumber based on information fusion, RAD were 11.6 % and 3.41 %,respectively,when the time scale were 15 min and Id. The experimental result show that: RAD between the predicted values by the irrigation quantity control model based on multi-information fusion and measured values was 3.00 %, during the experiment (10 days).(5) To solve the problem of intelligent controlling electrical conductivity of nutrient solution, the information affected the growth of greenhouse cucumber: time, environmental and nutrient solution were taken into account, the evaluation indexs of greenhouse cucumber growth status by growth stage, accumulated electrical conductivity value,accumulated thermal effectiveness and photosynthesis active radiation, effective accumulated temperature, was established, respectively, the average percentage deviation from the standard value were 3.86 %, 6.15 %, 14.19 % and 6.77 %. Deviation from the standard value were 6.45 % and 5.83 %, when the evaluation index by three factors fusion index and four factors fusion index. To achieve online intelligent decision-making of electrical conductivity for fertigation machinery, the decision-making model of electrical conductivity based on multi-information fusion index was built.