Urban green land soil fertility index evaluation system and method based on instant data

Abstract

The invention discloses an urban green land soil fertility index evaluation system and method based on instant data, and relates to the technical field of soil fertility evaluation. Extracting input features of the prediction model of the indoor detection indexes; training a prediction model of the indoor detection indexes by using the extracted input features; obtaining real-time data of the indoor detection indexes, and comprehensively evaluating the soil fertility according to the real-time data of the on-site rapid detection indexes and the real-time data of the indoor detection indexes; on-site rapid detection indexes related to soil fertility are monitored in real time, indoor detection indexes which can only be obtained through experimental analysis are obtained according to the change condition of the on-site rapid detection indexes, soil fertility indexes can be generated in time, the soil fertility can be evaluated, treatment and adjustment can be conducted in time according to the change condition of the soil fertility, and the purposes of maintaining the soil fertility and improving the soil fertility quality are achieved. The yield and the quality of crops are improved.

Description

Technical Field

[0001] The invention relates to the technical field of soil fertility assessment, and in particular to an urban green space soil fertility index assessment system and method based on real-time data. Background Art

[0002] Soil fertility refers to the ability of soil to provide the conditions such as water, nutrients, air and heat required for plant growth. It is a comprehensive reflection of the physical, chemical and biological properties of soil. The level of soil fertility directly affects the growth and yield of crops. Soil fertility assessment is an important task in agricultural scientific research and agricultural production management. It involves a comprehensive evaluation of multiple aspects such as nutrient content, physical properties, chemical properties and biological activity in the soil. The existing soil improvement methods usually improve the soil in a targeted manner according to the results of soil fertility assessment after obtaining the indicators of soil fertility. However, most of the soil fertility indicators need to be obtained by sampling the soil and conducting experimental analysis. Experimental analysis takes a lot of time and it is difficult to obtain real-time data of soil fertility indicators, so it is impossible to evaluate soil fertility in real time. Real-time evaluation can timely discover problems in the soil and take corresponding measures to improve and repair it, so as to maintain the health and productivity of the soil. Therefore, how to estimate the analysis results of soil fertility indicators in the laboratory in real time based on the instant acquisition of sensors has become an urgent problem to be solved. Summary of the invention

[0003] The purpose of the present invention is to provide a system and method to solve the problems raised in the prior art.

[0004] To achieve the above object, the present invention provides the following technical solution: a method for evaluating soil fertility index of urban green space based on real-time data, comprising: S11, obtaining historical data of soil fertility indicators; the soil fertility indicators include on-site rapid measurement indicators and indoor detection indicators, the on-site rapid measurement indicators are measured instantly by detection equipment, and the indoor detection indicators are measured by chemical analysis methods in the laboratory; S12, preprocessing the soil fertility index to extract input features of the prediction model of the indoor detection index; S13, using the extracted input features to train a prediction model for indoor detection indicators; S14, obtaining the real-time data of the on-site rapid test indicators and the historical data of the indoor test indicators, obtaining the real-time data of the indoor test indicators through the prediction model of the indoor test indicators, and making a comprehensive assessment of the soil fertility based on the real-time data of the on-site rapid test indicators and the indoor test indicators.

[0005] In step S12, the step of extracting input features of the prediction model of indoor detection indicators further includes the following steps: S21, obtain the historical data of soil fertility indoor test indicators, let t1 and t2 represent the time when the soil fertility indoor test indicators are measured twice by the laboratory chemical analysis method, obtain the change of soil fertility on-site rapid test indicators within the time [t1, t2], let x i (t) represents the change of the i-th soil fertility rapid test index in the time [t1, t2]; S22, analyzing the correlation between the indoor soil fertility detection index and the on-site rapid measurement index, obtaining the actual values ​​of the indoor soil fertility detection index and the on-site rapid measurement index at the same time from the historical data, taking one indoor soil fertility detection index as the fitting target, and using the remaining on-site rapid measurement indexes as input to regress the fitting target, and obtaining the goodness of fit according to the regression result; excluding one on-site rapid measurement index G from the input, and regressing the fitting target again, if the change amplitude of the goodness of fit is less than the threshold, it is judged that the on-site rapid measurement index G of soil fertility is not related to the fitting target, and if the change amplitude of the goodness of fit is not less than the threshold, it is judged that the on-site rapid measurement index G of soil fertility is related to the fitting target; excluding all on-site rapid measurement indicators of soil fertility one by one, and obtaining the on-site rapid measurement indicators of soil fertility related to the fitting target; changing the fitting target, repeating the steps to obtain the correlation between the on-site rapid measurement index of soil fertility and the indoor detection index; adding the on-site rapid measurement indicators whose number of associated indoor detection indicators is not less than 1 to the set U; S23, obtaining the actual values ​​of the soil fertility indoor detection index and the on-site rapid test index at the same time from the historical data, taking one soil fertility indoor detection index as the fitting target, and regressing the fitting target with the other remaining soil fertility indoor detection indexes as input, and obtaining the goodness of fit according to the regression result; excluding one soil fertility indoor detection index F from the input, and regressing the fitting target again, if the change amplitude of the goodness of fit is less than the threshold, it is judged that the soil fertility indoor detection index F is not related to the fitting target, and if the change amplitude of the goodness of fit is not less than the threshold, it is judged that the soil fertility indoor detection index F is related to the fitting target; excluding all soil fertility indoor detection indicators one by one, and obtaining the soil fertility indoor detection indicators related to the fitting target; changing the fitting target, repeating the steps to obtain the correlation between the soil fertility indoor detection index and the internal indoor detection indicators, and obtaining the number of other indoor detection indicators associated with each soil fertility indoor detection indicator; S24, calculate the average value ax of the i-th soil fertility on-site rapid test index in the set U within the time [t1, t2] i (t2), ; Obtain the average values ​​of all soil fertility on-site rapid test indicators in set U within the time [t1, t2], the measured values ​​of soil fertility indoor test indicators at t1 and the time difference t2-t1 as the initial input features of the prediction model of indoor test indicators.

[0006] Some soil indicators that can be measured immediately, such as soil moisture, pH value and conductivity, are related to indicators that need to be determined after sampling experiments, such as soil organic matter and nutrient content. Through these indicators that can be measured immediately, we can indirectly infer some information about soil fertility; for example, soil moisture changes can reflect the rainfall in the monitored area. When the average soil moisture value increases abnormally, it means that the rainfall in the monitored area has increased, which will lead to soil fertility loss. Combined with the historical experimental analysis results of soil fertility indoor detection indicators, the soil fertility indoor detection indicators can be predicted; there is no need to continuously conduct experimental analysis on the soil, and the real-time situation of soil fertility can be obtained. Only the on-site rapid measurement indicators associated with the indoor detection indicators need to be used for prediction, which not only saves computing resources, but also avoids the interference of redundant data on the prediction effect.

[0007] In step S12, extracting the input features of the prediction model of the indoor detection index also includes the following steps: S31, let y1, y2, ..., yn represent the initial input features of the prediction model of the indoor detection index, and n is the number of the initial input features; normalize the initial input features, and perform nonlinear feature extraction on the initial input features through a stacked autoencoder; S32, k1, k2, ..., km represent the dimension of the output feature of the self-encoding layer, k1, k2, ..., km decreases step by step, and the self-encoding layer includes an encoder and a decoder; the n-dimensional initial input feature is reduced to a k1-dimensional input feature by the encoder of the first self-encoding layer, and the k1-dimensional input feature is restored to the n-dimensional initial input feature by the decoder of the first self-encoding layer, and the reconstruction error between the restored n-dimensional initial input feature and the initial input features y1, y2, ..., yn is calculated, and the reconstruction error is minimized by using the error back propagation algorithm for training, and the training of the first self-encoding layer is completed; S33, let a=1, for the a+1th autoencoder layer, use the autoencoder to reduce the ka-dimensional input features to ka+1-dimensional input features, and then use the decoder to restore the ka+1-dimensional input features to ka-dimensional input features, calculate the reconstruction error between the restored ka-dimensional initial input features and the ka-dimensional input features at the beginning, use the error back propagation algorithm to train to minimize the reconstruction error, and complete the training of the a+1th autoencoder layer; S34, add 1 to the value of a, and return to step S33 until a=m, completing the training of all autoencoder layers; S35, connect the decoders of m autoencoding layers one by one in descending order of the dimensions k1, k2, ..., km of the output features to complete the connection of the encoders; connect the encoder of the mth autoencoding layer to its own decoder, connect the decoders of the remaining m-1 autoencoding layers one by one in ascending order of the dimensions km-1, ..., k1 of the output features, connect all encoders and decoders to complete the construction of the autoencoding neural network; input the initial input features into the autoencoding neural network to obtain km-dimensional input features.

[0008] There are many parameters for soil fertility indicators. For example, micro-nutrients and macro-nutrients contain many detection indicators, which have an adverse effect on the construction of deep neural network models. Therefore, it is necessary to extract features of the parameters. However, linear feature extraction methods are prone to lose the nonlinear relationship between parameters. Therefore, a nonlinear autoencoder neural network is used for feature extraction.

[0009] In step S13, the method of using the extracted input features to train the prediction model of the indoor detection index further includes the following steps: According to the number of other indoor detection indicators associated with each soil fertility indoor detection indicator, the indoor detection indicators are sorted in descending order of quantity, and a prediction model is trained separately for each indoor detection indicator according to the sorting result; For the vth indoor detection index in the sorting, the structure of the deep neural network model is built and the hyperparameters are set. The km-dimensional input features after nonlinear feature extraction are divided into a training set and a test set. The km-dimensional input features in the training set and the predicted values ​​of the indoor detection indicators associated with v before v in the sorting are used as input, and the actual value of the soil fertility indoor detection index at time t2 is used as output to train the parameters of the deep neural network model. After the parameter training of the deep neural network model is completed, the km-dimensional input features in the test set are input into the deep neural network model to obtain the predicted value of the soil fertility indoor detection index at time t2, and the determination coefficient between the predicted value and the actual value of the soil fertility indoor detection index at time t2 is calculated as the evaluation index of the deep neural network model. The hyperparameters of the deep neural network model are adjusted. When the evaluation index reaches the maximum value, the obtained deep neural network model is the prediction model of the vth indoor detection index in the sorting.

[0010] The indoor detection indicators are predicted in sequence, and the acquired indoor detection indicators can serve the subsequent indoor detection indicators, thereby improving the prediction effect.

[0011] In step S14, the step of obtaining the real-time data of the indoor detection index through the prediction model of the indoor detection index further includes the following steps: For the vth indoor test index in the sorting, obtain the change x of the soil fertility field rapid test index in the time interval [R, T]i (t), get the actual value of the soil fertility indoor test index at time R, T is the current time, R is the time before time T when the soil fertility indoor test index was measured by the laboratory chemical analysis method, and calculate the average value ax of the soil fertility field rapid test index in the set U within the time [R, T] i (T), , the average value of the soil fertility on-site rapid test index within the time [R, T], the actual value of the soil fertility indoor detection index at time R and the time difference TR are input into the autoencoder neural network for feature extraction, and the result of feature extraction and the predicted value of the indoor detection index associated with v before v in the sorting are input into the prediction model of the vth indoor detection index to obtain the predicted value of the vth indoor detection index at time T.

[0012] In step S14, the comprehensive assessment of soil fertility based on the real-time data of the on-site rapid test indicators and the indoor test indicators also includes the following steps: Get the actual value of the on-site rapid test index of soil fertility and the predicted value of the indoor test index at the current time T, and calculate the single fertility coefficient Pj, Pj=Cj / Bj, where Cj is the single fertility index of index j. When index j is the on-site rapid test index, the single fertility index of index j is the actual value. When index j is the indoor test index, the single fertility index of index j is the predicted value. Bj is the evaluation standard value of index j. Calculate the comprehensive soil fertility index P, , where N is the number of soil fertility indicators, APj is the average value of all individual fertility coefficients, and mPj is the minimum value of all individual fertility coefficients; the soil fertility evaluation is given according to the soil fertility comprehensive index.

[0013] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a soil fertility index assessment system for urban green space based on real-time data, comprising a soil data acquisition module, a soil fertility analysis module, a soil data storage module and a soil treatment module; the output end of the soil data acquisition module is connected to the input end of the soil data storage module and the soil fertility analysis module, and is used to obtain soil fertility indicators, and the soil fertility indicators include on-site rapid measurement indicators and indoor detection indicators; the output end of the soil fertility analysis module is connected to the input end of the soil treatment module and the soil data storage module, and the indoor detection indicators of soil fertility are obtained according to the on-site rapid measurement indicators of soil fertility, and the soil fertility is evaluated according to the soil fertility indicators; the soil data storage module is connected to the soil fertility analysis module, and is used to store historical data of soil fertility indicators; the soil fertility analysis module generates soil improvement suggestions according to the soil fertility evaluation results.

[0014] The soil data acquisition module also includes an instant data acquisition unit, an experimental analysis unit and a data transmission unit. The instant data acquisition unit performs instant measurement of soil fertility on-site rapid test indicators through detection equipment; the experimental analysis unit performs chemical analysis on the collected soil samples to obtain soil fertility indoor detection indicators; the data transmission unit is used to transmit the soil fertility indicators of the soil data acquisition module to the soil fertility analysis module. The soil fertility analysis module also includes a correlation analysis unit, a nonlinear feature extraction unit, an indoor detection indicator prediction unit and an on-site rapid test indicator analysis unit; the correlation analysis unit is used to analyze the correlation between soil fertility indicators, the nonlinear feature extraction unit is used to perform nonlinear feature extraction on the input of the indoor detection indicator prediction model; the indoor detection indicator prediction unit is used to predict the indoor detection indicators of soil fertility; the on-site rapid test indicator analysis unit generates the input of the indoor detection indicator prediction model according to the on-site rapid test indicators of soil fertility. The soil fertility management module also includes a soil structural improvement unit and a soil nutrient improvement unit; the soil structural improvement unit generates soil maintenance measure recommendations based on the soil fertility evaluation results; the soil nutrient improvement unit generates soil fertilization measure recommendations based on the soil fertility evaluation results.

[0015] Compared with the prior art, the beneficial effects of the present invention are: real-time monitoring of on-site rapid test indicators related to soil fertility, obtaining indoor detection indicators that require experimental analysis based on changes in on-site rapid test indicators, being able to instantly generate soil fertility indicators and evaluate soil fertility, and being able to process and adjust soil fertility changes in a timely manner, thereby maintaining soil fertility and improving the yield and quality of crops. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a structural schematic diagram of an urban green space soil fertility index assessment system based on real-time data. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Example: Figure 1As shown, the present invention provides a technical solution, a soil fertility index assessment system for urban green space based on real-time data, comprising a soil data acquisition module, a soil fertility analysis module, a soil data storage module and a soil treatment module; the output end of the soil data acquisition module is connected to the input end of the soil data storage module and the soil fertility analysis module, and is used to obtain soil fertility indicators, and the soil fertility indicators include on-site rapid measurement indicators and indoor detection indicators; the output end of the soil fertility analysis module is connected to the input end of the soil treatment module and the soil data storage module, and the indoor detection indicators of soil fertility are obtained according to the on-site rapid measurement indicators of soil fertility, and the soil fertility is evaluated according to the soil fertility indicators; the soil data storage module is connected to the soil fertility analysis module, and is used to store historical data of soil fertility indicators; the soil fertility analysis module generates soil improvement suggestions according to the soil fertility evaluation results.

[0019] The soil data acquisition module also includes an instant data acquisition unit, an experimental analysis unit and a data transmission unit. The instant data acquisition unit performs instant measurement of soil fertility on-site rapid test indicators through detection equipment; the experimental analysis unit performs chemical analysis on the collected soil samples to obtain soil fertility indoor detection indicators; the data transmission unit is used to transmit the soil fertility indicators of the soil data acquisition module to the soil fertility analysis module. The soil fertility analysis module also includes a correlation analysis unit, a nonlinear feature extraction unit, an indoor detection indicator prediction unit and an on-site rapid test indicator analysis unit; the correlation analysis unit is used to analyze the correlation between soil fertility indicators, the nonlinear feature extraction unit is used to perform nonlinear feature extraction on the input of the indoor detection indicator prediction model; the indoor detection indicator prediction unit is used to predict the indoor detection indicators of soil fertility; the on-site rapid test indicator analysis unit generates the input of the indoor detection indicator prediction model according to the on-site rapid test indicators of soil fertility. The soil fertility management module also includes a soil structural improvement unit and a soil nutrient improvement unit; the soil structural improvement unit generates soil maintenance measure recommendations based on the soil fertility evaluation results; the soil nutrient improvement unit generates soil fertilization measure recommendations based on the soil fertility evaluation results.

[0020] Embodiment: The present invention provides a technical solution, a method for evaluating soil fertility index of urban green space based on real-time data, comprising: S11, obtaining historical data of soil fertility indicators; the soil fertility indicators include on-site rapid test indicators and indoor test indicators, the on-site rapid test indicators are measured instantly by testing equipment, and the indoor test indicators are measured by laboratory chemical analysis methods; for the collection method of surface soil samples, a 20 m×20 m sample plot is laid out in the plot near the monitoring point, and 5 soil sampling points are arranged in a plum blossom shape in the sample plot; ring knife samples, analysis samples, and microbial samples are taken; for profile sample collection: after the profile is dug, ring knife samples, analysis samples, and microbial samples are collected at 0-20 cm, 20-40, and 40-60 cm; for water-stable large aggregate sample collection: the original soil samples collected from multiple mixed sampling points are placed in a container that is not easy to deform (such as a hard plastic box, a wide-mouth plastic bottle, etc.), and merged into one sample; the indoor test indicators are determined by sample analysis, and at the same time, the on-site rapid test indicators can also be determined by sample analysis. When there is a sample analysis result, the sample analysis result shall prevail, and when there is no sample analysis result, the test equipment shall prevail; Specifically, the indoor detection indicators are organic matter, total nitrogen and available potassium; the on-site rapid test indicators are soil moisture, pH value and conductivity; organic matter sampling is measured by potassium dichromate oxidation-external heating method, total nitrogen sampling is measured by Kjeldahl nitrogen determination method, and available potassium sampling is measured by AB-DTPA extraction / ICPOES method; soil moisture and conductivity are measured by soil moisture sensor, and pH value is measured by soil in-situ pH meter; S12, pre-processing the soil fertility index to extract the input features of the prediction model of the indoor detection index, including steps S21 to S24: S21, obtain the historical data of soil fertility indoor test indicators, let t1 and t2 represent the time when the soil fertility indoor test indicators are measured twice by the laboratory chemical analysis method, obtain the change of soil fertility on-site rapid test indicators within the time [t1, t2], let x i (t) represents the change of the i-th soil fertility rapid test index in the time [t1, t2]; S22, analyzing the correlation between the indoor soil fertility detection index and the on-site rapid measurement index, obtaining the actual values ​​of the indoor soil fertility detection index and the on-site rapid measurement index at the same time from the historical data, taking one indoor soil fertility detection index as the fitting target, and using the remaining on-site rapid measurement indexes as input to regress the fitting target, and obtaining the goodness of fit according to the regression result; excluding one on-site rapid measurement index G from the input, and regressing the fitting target again, if the change amplitude of the goodness of fit is less than the threshold, it is judged that the on-site rapid measurement index G of soil fertility is not related to the fitting target, and if the change amplitude of the goodness of fit is not less than the threshold, it is judged that the on-site rapid measurement index G of soil fertility is related to the fitting target; excluding all on-site rapid measurement indicators of soil fertility one by one, and obtaining the on-site rapid measurement indicators of soil fertility related to the fitting target; changing the fitting target, repeating the steps to obtain the correlation between the on-site rapid measurement index of soil fertility and the indoor detection index; adding the on-site rapid measurement indicators whose number of associated indoor detection indicators is not less than 1 to the set U; S23, obtaining the actual values ​​of the soil fertility indoor detection index and the on-site rapid test index at the same time from the historical data, taking one soil fertility indoor detection index as the fitting target, and regressing the fitting target with the other remaining soil fertility indoor detection indexes as input, and obtaining the goodness of fit according to the regression result; excluding one soil fertility indoor detection index F from the input, and regressing the fitting target again, if the change amplitude of the goodness of fit is less than the threshold, it is judged that the soil fertility indoor detection index F is not related to the fitting target, and if the change amplitude of the goodness of fit is not less than the threshold, it is judged that the soil fertility indoor detection index F is related to the fitting target; excluding all soil fertility indoor detection indicators one by one, and obtaining the soil fertility indoor detection indicators related to the fitting target; changing the fitting target, repeating the steps to obtain the correlation between the soil fertility indoor detection index and the internal indoor detection indicators, and obtaining the number of other indoor detection indicators associated with each soil fertility indoor detection indicator; S24, calculate the average value ax of the i-th soil fertility on-site rapid test index in the set U within the time [t1, t2] i (t2), ; Obtain the average values ​​of all soil fertility on-site rapid test indicators in set U within the time [t1, t2], the measured values ​​of soil fertility indoor test indicators at t1 and the time difference t2-t1 as the initial input features of the prediction model of indoor test indicators.

[0021] Nonlinear feature extraction is performed on the initial input features through steps S31 to S35: S31, let y1, y2, ..., yn represent the initial input features of the prediction model of the indoor detection index, and n is the number of the initial input features; normalize the initial input features, and perform nonlinear feature extraction on the initial input features through a stacked autoencoder; S32, k1, k2, ..., km represent the dimension of the output feature of the self-encoding layer, k1, k2, ..., km decreases step by step, and the self-encoding layer includes an encoder and a decoder; the n-dimensional initial input feature is reduced to a k1-dimensional input feature by the encoder of the first self-encoding layer, and the k1-dimensional input feature is restored to the n-dimensional initial input feature by the decoder of the first self-encoding layer, and the reconstruction error between the restored n-dimensional initial input feature and the initial input features y1, y2, ..., yn is calculated, and the reconstruction error is minimized by using the error back propagation algorithm to complete the training of the first self-encoding layer; k1, k2, ..., km decrease step by step, such as when the number of indoor detection indicators and on-site rapid measurement indicators is 30, they are reduced in the order of 20, 14, 10, and 5; S33, let a=1, for the a+1th autoencoder layer, use the autoencoder to reduce the ka-dimensional input features to ka+1-dimensional input features, and then use the decoder to restore the ka+1-dimensional input features to ka-dimensional input features, calculate the reconstruction error between the restored ka-dimensional initial input features and the ka-dimensional input features at the beginning, use the error back propagation algorithm to train to minimize the reconstruction error, and complete the training of the a+1th autoencoder layer; S34, add 1 to the value of a, and return to step S33 until a=m, completing the training of all autoencoder layers; S35, connect the decoders of m autoencoding layers one by one in descending order of the dimensions k1, k2, ..., km of the output features to complete the connection of the encoders; connect the encoder of the mth autoencoding layer to its own decoder, connect the decoders of the remaining m-1 autoencoding layers one by one in ascending order of the dimensions km-1, ..., k1 of the output features, connect all encoders and decoders to complete the construction of the autoencoding neural network; input the initial input features into the autoencoding neural network to obtain km-dimensional input features.

[0022] S13, using the extracted input features to train the prediction model of indoor detection indicators: According to the number of other indoor detection indicators associated with each soil fertility indoor detection indicator, the indoor detection indicators are sorted in descending order of quantity, and a prediction model is trained separately for each indoor detection indicator according to the sorting result; For the vth indoor detection index in the sorting, the structure of the deep neural network model is built and the hyperparameters are set. The km-dimensional input features after nonlinear feature extraction are divided into a training set and a test set. The km-dimensional input features in the training set and the predicted values ​​of the indoor detection indicators associated with v before v in the sorting are used as input, and the actual value of the soil fertility indoor detection index at time t2 is used as output to train the parameters of the deep neural network model. After the parameter training of the deep neural network model is completed, the km-dimensional input features in the test set are input into the deep neural network model to obtain the predicted value of the soil fertility indoor detection index at time t2, and the determination coefficient between the predicted value and the actual value of the soil fertility indoor detection index at time t2 is calculated as the evaluation index of the deep neural network model. The hyperparameters of the deep neural network model are adjusted. When the evaluation index reaches the maximum value, the obtained deep neural network model is the prediction model of the vth indoor detection index in the sorting.

[0023] The deep neural network model includes: input layer, hidden layer and output layer in the network structure; the number of neurons in the input layer and hidden layer is adjusted according to the determination coefficient, and the number of neurons in the output layer is 1; similarly, the number of hidden layers can also try different values ​​and be determined according to the determination coefficient; the activation functions of the input layer, hidden layer and output layer can select relu, sigmoid and other functions; hyperparameters such as learning rate and number of training times are adjusted accordingly according to actual data.

[0024] S14, obtaining the real-time data of the on-site rapid test indicators and the historical data of the indoor test indicators, obtaining the real-time data of the indoor test indicators through the prediction model of the indoor test indicators, and performing a comprehensive assessment of soil fertility based on the real-time data of the on-site rapid test indicators and the indoor test indicators; For the vth indoor test index in the sorting, obtain the change x of the soil fertility field rapid test index in the time interval [R, T] i (t), get the actual value of the soil fertility indoor test index at time R, T is the current time, R is the time before time T when the soil fertility indoor test index was measured by the laboratory chemical analysis method, and calculate the average value ax of the soil fertility field rapid test index in the set U within the time [R, T] i (T), , the average value of the soil fertility on-site rapid test index within the time [R, T], the actual value of the soil fertility indoor detection index at time R and the time difference TR are input into the autoencoder neural network for feature extraction, and the result of feature extraction and the predicted value of the indoor detection index associated with v before v in the sorting are input into the prediction model of the vth indoor detection index to obtain the predicted value of the vth indoor detection index at time T; the first indoor detection index in the sorting is predicted only by the data after nonlinear feature extraction to obtain the predicted value; for the next indoor detection index in the sorting, if it is associated with the first indoor detection index, it is predicted by the data after nonlinear feature extraction and the predicted value of the first indoor detection index; the indoor detection index with more associated indoor detection indicators is predicted first, in order to improve the prediction effect of subsequent indicators.

[0025] Specifically, the indoor detection indicators are organic matter, total nitrogen and available potassium; the on-site rapid detection indicators are soil moisture, pH value and electrical conductivity; among them, there is an obvious correlation between organic matter and total nitrogen and available potassium, and there is a certain correlation between total nitrogen and available potassium. The number of other indoor detection indicators associated with organic matter, total nitrogen and available potassium is 2, and they are at the same priority in the sorting. Considering that organic matter has a significant impact on total nitrogen and available potassium, as long as the data of organic matter is obtained, the data of total nitrogen and available potassium can be determined, and the prediction model of organic matter is trained first, and then the prediction models are trained in the order of total nitrogen, available potassium and available potassium; when the indoor indicator increases the soil bulk density, the correlation with organic matter, total nitrogen and available potassium decreases, and is placed at the end of the sequence; In the training process of the organic matter prediction model, t1 and t2 are the times when the indoor soil fertility detection indicators are measured by laboratory chemical analysis methods, and accurate training data exists; the data at t2 is predicted based on the data at t1, and the initial input data includes the average values ​​of soil moisture, pH value and conductivity between t2-t1 and t1 to t2, and the organic matter content at t1. The on-site rapid measurement indicators include but are not limited to soil moisture, pH value and conductivity, which can be increased, so feature extraction is required; in particular, if the indoor detection indicators can be roughly estimated on-site, they can also be used as the input of the prediction model, that is, t is added to the initial input data. The on-site rapid measurement value of organic matter at t2 is used, and the initial input data is subjected to nonlinear feature extraction by the stacked autoencoder to obtain the input t2 of the prediction model; the organic matter prediction model is obtained after the hyperparameters of the prediction model are trained; after the experimental measurement is completed at t2, at t4 after t2, only the on-site rapid measurement indicators can be obtained because no experimental measurement is performed, and when it is necessary to obtain the organic matter data at t4, the organic matter content at t2, the average values ​​of soil moisture, pH value and conductivity between t4-t2 and t2 to t4, and the on-site rapid measurement value of organic matter at t4 are subjected to nonlinear feature extraction, and then input into the organic matter prediction model to obtain the organic matter content at t4.

[0026] For the prediction model of total nitrogen, if it is difficult to make a rough estimate on site and it is impossible to conduct rapid on-site measurement, the initial input data are the average values ​​of soil moisture, pH value and conductivity between t2-t1 and t1 to t2, the organic matter content at t1, the predicted value of organic matter content at t2 and the total nitrogen content at t1. Except for the organic matter content at t2, all of them are accurate values. The initial input data are nonlinearly extracted to train the prediction model of total nitrogen. When predicting total nitrogen at t4, the organic matter content at t2, the average values ​​of soil moisture, pH value and conductivity between t4-t2 and t2 to t4, and the predicted value of organic matter content at t4 are subjected to nonlinear feature extraction and input into the prediction model of total nitrogen to obtain the total nitrogen content at t4.

[0027] Get the actual value of the on-site rapid test index of soil fertility and the predicted value of the indoor test index at the current time T, and calculate the single fertility coefficient Pj, Pj=Cj / Bj, where Cj is the single fertility index of index j. When index j is the on-site rapid test index, the single fertility index of index j is the actual value. When index j is the indoor test index, the single fertility index of index j is the predicted value. Bj is the evaluation standard value of index j. Calculate the comprehensive soil fertility index P, , where N is the number of soil fertility indicators, APj is the average value of all single fertility coefficients, and mPj is the minimum value of all single fertility coefficients; the soil fertility evaluation is given according to the soil fertility comprehensive index. The evaluation standard value is obtained by referring to standards LY / T1225-1999, NY / T 1121.6-2006, LY / T 1228-2015, DB31 / T 661 Appendix F, etc.

[0028] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for evaluating soil fertility index of urban green space based on real-time data, characterized in that: The following steps are involved: S11, obtaining historical data of soil fertility indicators; the soil fertility indicators include on-site rapid measurement indicators and indoor detection indicators, the on-site rapid measurement indicators are measured instantly by detection equipment, and the indoor detection indicators are measured by chemical analysis methods in the laboratory; S12, preprocessing the soil fertility index to extract input features of the prediction model of the indoor detection index; S13, using the extracted input features to train a prediction model for indoor detection indicators; S14, obtaining the real-time data of the on-site rapid test indicators and the historical data of the indoor test indicators, obtaining the real-time data of the indoor test indicators through the prediction model of the indoor test indicators, and making a comprehensive assessment of the soil fertility based on the real-time data of the on-site rapid test indicators and the indoor test indicators.

2. The method for evaluating soil fertility index of urban green space based on real-time data according to claim 1 is characterized in that: In step S12, the step of extracting input features of the prediction model of indoor detection indicators further includes the following steps: S21, obtain the historical data of soil fertility indoor test indicators, let t1 and t2 represent the time when the soil fertility indoor test indicators are measured twice by the laboratory chemical analysis method, obtain the change of soil fertility on-site rapid test indicators within the time [t1, t2], let x i (t) represents the change of the i-th soil fertility rapid test index in the time [t1, t2]; S22, analyzing the correlation between the indoor soil fertility detection index and the on-site rapid measurement index, obtaining the actual values ​​of the indoor soil fertility detection index and the on-site rapid measurement index at the same time from the historical data, taking one indoor soil fertility detection index as the fitting target, and using the remaining on-site rapid measurement indexes as input to regress the fitting target, and obtaining the goodness of fit according to the regression result; excluding one on-site rapid measurement index G from the input, and regressing the fitting target again, if the change amplitude of the goodness of fit is less than the threshold, it is judged that the on-site rapid measurement index G of soil fertility is not related to the fitting target, and if the change amplitude of the goodness of fit is not less than the threshold, it is judged that the on-site rapid measurement index G of soil fertility is related to the fitting target; excluding all on-site rapid measurement indicators of soil fertility one by one, and obtaining the on-site rapid measurement indicators of soil fertility related to the fitting target; changing the fitting target, repeating the steps to obtain the correlation between the on-site rapid measurement index of soil fertility and the indoor detection index; adding the on-site rapid measurement indicators whose number of associated indoor detection indicators is not less than 1 to the set U; S23, obtaining the actual values ​​of the soil fertility indoor detection index and the on-site rapid test index at the same time from the historical data, taking one soil fertility indoor detection index as the fitting target, and regressing the fitting target with the other remaining soil fertility indoor detection indexes as input, and obtaining the goodness of fit according to the regression result; excluding one soil fertility indoor detection index F from the input, and regressing the fitting target again, if the change amplitude of the goodness of fit is less than the threshold, it is judged that the soil fertility indoor detection index F is not related to the fitting target, and if the change amplitude of the goodness of fit is not less than the threshold, it is judged that the soil fertility indoor detection index F is related to the fitting target; excluding all soil fertility indoor detection indicators one by one, and obtaining the soil fertility indoor detection indicators related to the fitting target; changing the fitting target, repeating the steps to obtain the correlation between the soil fertility indoor detection index and the internal indoor detection indicators, and obtaining the number of other indoor detection indicators associated with each soil fertility indoor detection indicator; S24, calculate the average value ax of the i-th soil fertility on-site rapid test index in the set U within the time [t1, t2] i (t2), ; Obtain the average values ​​of all soil fertility on-site rapid test indicators in set U within the time [t1, t2], the measured values ​​of soil fertility indoor test indicators at t1 and the time difference t2-t1 as the initial input features of the prediction model of indoor test indicators.

3. The method for evaluating soil fertility index of urban green space based on real-time data according to claim 2 is characterized in that: In step S12, extracting the input features of the prediction model of the indoor detection index also includes the following steps: S31, let y1, y2, ..., yn represent the initial input features of the prediction model of the indoor detection index, and n is the number of the initial input features; normalize the initial input features, and perform nonlinear feature extraction on the initial input features through a stacked autoencoder; S32, k1, k2, ..., km represent the dimension of the output feature of the self-encoding layer, k1, k2, ..., km decreases step by step, and the self-encoding layer includes an encoder and a decoder; the n-dimensional initial input feature is reduced to a k1-dimensional input feature by the encoder of the first self-encoding layer, and the k1-dimensional input feature is restored to the n-dimensional initial input feature by the decoder of the first self-encoding layer, and the reconstruction error between the restored n-dimensional initial input feature and the initial input features y1, y2, ..., yn is calculated, and the reconstruction error is minimized by using the error back propagation algorithm for training, and the training of the first self-encoding layer is completed; S33, let a=1, for the a+1th autoencoder layer, use the autoencoder to reduce the ka-dimensional input features to ka+1-dimensional input features, and then use the decoder to restore the ka+1-dimensional input features to ka-dimensional input features, calculate the reconstruction error between the restored ka-dimensional initial input features and the ka-dimensional input features at the beginning, use the error back propagation algorithm to train to minimize the reconstruction error, and complete the training of the a+1th autoencoder layer; S34, add 1 to the value of a, and return to step S33 until a=m, completing the training of all autoencoder layers; S35, connect the decoders of m autoencoding layers one by one in descending order of the dimensions k1, k2, ..., km of the output features to complete the connection of the encoders; connect the encoder of the mth autoencoding layer to its own decoder, connect the decoders of the remaining m-1 autoencoding layers one by one in ascending order of the dimensions km-1, ..., k1 of the output features, connect all encoders and decoders to complete the construction of the autoencoding neural network; input the initial input features into the autoencoding neural network to obtain km-dimensional input features.

4. The method for evaluating soil fertility index of urban green space based on real-time data according to claim 3 is characterized in that: In step S13, the method of using the extracted input features to train the prediction model of the indoor detection index further includes the following steps: According to the number of other indoor detection indicators associated with each soil fertility indoor detection indicator, the indoor detection indicators are sorted in descending order of quantity, and a prediction model is trained separately for each indoor detection indicator according to the sorting result; For the vth indoor detection index in the sorting, the structure of the deep neural network model is built and the hyperparameters are set. The km-dimensional input features after nonlinear feature extraction are divided into a training set and a test set. The km-dimensional input features in the training set and the predicted values ​​of the indoor detection indicators associated with v before v in the sorting are used as input, and the actual value of the soil fertility indoor detection index at time t2 is used as output to train the parameters of the deep neural network model. After the parameter training of the deep neural network model is completed, the km-dimensional input features in the test set are input into the deep neural network model to obtain the predicted value of the soil fertility indoor detection index at time t2, and the determination coefficient between the predicted value and the actual value of the soil fertility indoor detection index at time t2 is calculated as the evaluation index of the deep neural network model. The hyperparameters of the deep neural network model are adjusted. When the evaluation index reaches the maximum value, the obtained deep neural network model is the prediction model of the vth indoor detection index in the sorting.

5. The method for evaluating soil fertility index of urban green space based on real-time data according to claim 4 is characterized in that: In step S14, the step of obtaining the real-time data of the indoor detection index through the prediction model of the indoor detection index further includes the following steps: For the vth indoor test index in the sorting, obtain the change x of the soil fertility field rapid test index in the time interval [R, T] i (t), get the actual value of the soil fertility indoor test index at time R, T is the current time, R is the time before time T when the soil fertility indoor test index was measured by the laboratory chemical analysis method, and calculate the average value ax of the soil fertility field rapid test index in the set U within the time [R, T] i (T), , the average value of the soil fertility on-site rapid test index within the time [R, T], the actual value of the soil fertility indoor detection index at time R and the time difference TR are input into the autoencoder neural network for feature extraction, and the result of feature extraction and the predicted value of the indoor detection index associated with v before v in the sorting are input into the prediction model of the vth indoor detection index to obtain the predicted value of the vth indoor detection index at time T.

6. The method for evaluating soil fertility index of urban green space based on real-time data according to claim 5 is characterized in that: In step S14, the comprehensive assessment of soil fertility based on the real-time data of the on-site rapid test indicators and the indoor test indicators also includes the following steps: Get the actual value of the on-site rapid test index of soil fertility and the predicted value of the indoor test index at the current time T, and calculate the single fertility coefficient Pj, Pj=Cj / Bj, where Cj is the single fertility index of index j. When index j is the on-site rapid test index, the single fertility index of index j is the actual value. When index j is the indoor test index, the single fertility index of index j is the predicted value. Bj is the evaluation standard value of index j. Calculate the comprehensive soil fertility index P, , where N is the number of soil fertility indicators, APj is the average value of all individual fertility coefficients, and mPj is the minimum value of all individual fertility coefficients; the soil fertility evaluation is given according to the soil fertility comprehensive index.

7. A soil fertility index assessment system for urban green space based on real-time data, characterized in that: It includes a soil data acquisition module, a soil fertility analysis module, a soil data storage module and a soil management module; the output end of the soil data acquisition module is connected to the input end of the soil data storage module and the soil fertility analysis module, and is used to obtain soil fertility indicators, and the soil fertility indicators include on-site rapid measurement indicators and indoor detection indicators; the output end of the soil fertility analysis module is connected to the input end of the soil management module and the soil data storage module, and the indoor detection indicators of soil fertility are obtained according to the on-site rapid measurement indicators of soil fertility, and the soil fertility is evaluated according to the soil fertility indicators; the soil data storage module is connected to the soil fertility analysis module, and is used to store historical data of soil fertility indicators; the soil fertility analysis module generates soil improvement suggestions according to the soil fertility evaluation results.

8. The urban green space soil fertility index evaluation system based on real-time data according to claim 7 is characterized in that: The soil data acquisition module also includes an instant data acquisition unit, an experimental analysis unit and a data transmission unit. The instant data acquisition unit instantly measures the on-site rapid test index of soil fertility through detection equipment; the experimental analysis unit performs chemical analysis on the collected soil samples to obtain indoor detection index of soil fertility; the data transmission unit is used to transmit the soil fertility index of the soil data acquisition module to the soil fertility analysis module.

9. The urban green space soil fertility index evaluation system based on real-time data according to claim 8 is characterized in that: The soil fertility analysis module also includes a correlation analysis unit, a nonlinear feature extraction unit, an indoor detection index prediction unit and an on-site rapid measurement index analysis unit; the correlation analysis unit is used to analyze the correlation between soil fertility indicators, and the nonlinear feature extraction unit is used to perform nonlinear feature extraction on the input of the indoor detection index prediction model; the indoor detection index prediction unit is used to predict the indoor detection index of soil fertility; the on-site rapid measurement index analysis unit generates the input of the indoor detection index prediction model according to the on-site rapid measurement index of soil fertility.

10. The urban green space soil fertility index evaluation system based on real-time data according to claim 9 is characterized in that: The soil fertility management module also includes a soil structural improvement unit and a soil nutrient improvement unit; the soil structural improvement unit generates soil maintenance measure recommendations based on the soil fertility evaluation results; the soil nutrient improvement unit generates soil fertilization measure recommendations based on the soil fertility evaluation results.

Images