Field corn fertilization system based on ipso-lstm algorithm

Abstract

The application relates to a field corn fertilization system based on an IPSO-LSTM algorithm, and relates to a field corn fertilization system based on an IPSO-LSTM algorithm. The application aims to solve the problem that no existing technology can determine accurate fertilizer application amount according to different soils. The system comprises a collection module, a relay module, a main controller module and a remote monitoring module. The collection module acquires soil parameters of a fertilization area, performs first-stage processing, and sends the processed data to the relay module. The relay module performs second-stage processing on the data to obtain optimal fusion results. The main controller module takes the optimal fusion result data as input, predicts the fertilizer application amount through a trained IPSO-LSTM fertilizer application amount prediction model, controls the fertilizer ratio, and thus completes the function of autonomous fertilization. The remote monitoring module interacts with the main controller module to realize the function of remote monitoring. The application is used in the technical field of agricultural water fertilization.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of agricultural water fertilization, in particular to a field corn fertilization system based on an IPSO-LSTM algorithm. BACKGROUND

[0002] With the continuous development of the country and the improvement of people's living standards, higher requirements are put forward for the yield and quality of grain. Therefore, reasonable use of limited arable land, improvement of unit area grain yield and realization of grain self-sufficiency are the basic principles for solving the problem of grain. As an important raw material for agricultural production, the use rate of chemical fertilizer directly affects the yield and quality of grain.

[0003] Deep learning is a powerful machine learning technology. By collecting relevant data in agricultural production, the data is used to establish a fertilization amount prediction model for crop growth, which can help agricultural workers make more accurate fertilization decisions, thereby ensuring the rationality of the fertilization amount. Therefore, in modern agricultural production, the organic combination of Internet of Things technology, deep learning technology and water and fertilizer integration technology can effectively improve the efficiency of agricultural production, and has extremely important significance for the long-term development of modern agriculture in China. SUMMARY

[0004] The purpose of the application is to solve the problem that there is no accurate chemical fertilizer application amount determined according to different soils, and a field corn fertilization system based on an IPSO-LSTM algorithm is provided.

[0005] A field corn fertilization system based on an IPSO-LSTM algorithm comprises a collection module, a relay module, a main controller module and a remote monitoring module.

[0006] The collection module comprises a plurality of collection ends. The collection ends obtain soil parameters of the fertilization area through sensors, perform first-level processing on the soil parameters of the fertilization area, and send the processed data to the relay module.

[0007] The relay module is responsible for receiving the data sent by the collection module and performing second-level processing on the data to obtain an optimal fusion result, and sending the optimal fusion result to the main controller module.

[0008] The main controller module takes the optimal fusion result data sent by the relay as input, predicts the fertilization amount through the trained IPSO-LSTM fertilization amount prediction model, controls the fertilizer ratio, and thus completes the function of autonomous fertilization.

[0009] The IPSO is a particle swarm optimization algorithm, and the LSTM is a long short-term memory neural network.

[0010] The remote monitoring module is used for data interaction with the main controller module, thereby realizing the function of remote monitoring.

[0011] The beneficial effects of the present application are:

[0012] The IPSO-LSTM algorithm integrates the improved particle swarm optimization algorithm and the long short-term memory neural network, can adjust in real time according to soil characteristics, environmental conditions and crop demand and other factors, can optimize a large number of fertilization parameters, and accurately predict and adjust through the neural network. Compared with the traditional trial and error method, the system can find the optimal solution faster, effectively eliminates the uncertainty of relying on experience and experimental data, improves the accuracy and efficiency of fertilization decision-making. Practice has proved that the system can more accurately determine the fertilizer application amount, reduce waste and environmental pollution, and improve the yield and quality of corn. Compared with the traditional fertilization method, the system has better performance, effectively improves the agricultural production benefit. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 It is a kind of field corn fertilization system overall architecture block diagram based on IPSO-LSTM algorithm provided by the application;

[0014] Figure 2 It is a kind of field corn fertilization system system flow chart based on IPSO-LSTM algorithm provided by the application;

[0015] Figure 3 The present application provides a kind of field corn fertilization system based on IPSO-LSTM algorithm two kinds of particle swarm optimization result chart, a) is learning rate optimization result, b) is neuron number optimization result, c) is learning rate optimization result, d) neuron number optimization result;

[0016] Figure 4 The present application provides a kind of field corn fertilization system based on IPSO-LSTM algorithm prediction nitrogen application amount contrast chart;

[0017] Figure 5 The present application provides a kind of field corn fertilization system based on IPSO-LSTM algorithm prediction phosphorus application amount contrast chart;

[0018] Figure 6 The present application provides a kind of field corn fertilization system based on IPSO-LSTM algorithm prediction potassium application amount contrast chart. DETAILED DESCRIPTION

[0019] Specific embodiment one: the present embodiment a kind of field corn fertilization system based on IPSO-LSTM algorithm includes acquisition module, relay module, main controller module and remote monitoring module;

[0020] The collection module includes a plurality of collection ends, the collection end acquires the soil parameters of the fertilization area through a sensor, and the soil parameters of the fertilization area are subjected to first-level processing and then sent to a relay module;

[0021] The relay module is responsible for receiving the data sent by the collection module and performing second-level processing on the data to obtain an optimal fusion result, and the optimal fusion result is sent to a main controller module;

[0022] The main controller module takes the optimal fusion result data sent by the relay as input, predicts the fertilizer amount through an internally deployed IPSO-LSTM fertilizer amount prediction model, controls the fertilizer amount ratio in combination with components such as electromagnetic valves and flow rate sensors, and thus completes the function of autonomous fertilization;

[0023] The IPSO is a particle swarm optimization algorithm, and the LSTM is a long short-term memory neural network;

[0024] The remote monitoring module is used for data interaction with the main controller module, thereby realizing the function of remote monitoring.

[0025] Specific implementation method two: different from the specific implementation method one, the soil parameters are soil nitrogen, phosphorus, and potassium element content, soil pH value, soil temperature and humidity, soil light intensity, and soil moisture;

[0026] The sensor is a nitrogen, phosphorus, and potassium element sensor, a pH value sensor, a temperature and humidity sensor, a light intensity sensor, and a soil moisture sensor;

[0027] The collection module is powered by a storage battery and solar power.

[0028] The nitrogen, phosphorus, and potassium sensor measures the content of nitrogen (N), phosphorus (P), and potassium (K) in the soil. The measurement result is the content value of these nutrients. For example, the nitrogen, phosphorus, and potassium measurement result may show that the content of nitrogen in the soil is 100 mg / L, the content of phosphorus is 20 mg / L, and the content of potassium is 150 mg / L.

[0029] The other steps and parameters are the same as those of the specific implementation method one.

[0030] Specific implementation method three: different from the specific implementation method one or two, the collection end acquires the soil parameters of the fertilization area through a sensor, and the soil parameters of the fertilization area are subjected to first-level processing and then sent to a relay module; the specific process is as follows:

[0031] The soil nitrogen, phosphorus, and potassium sensor is connected to the collection end through an RS485 interface, the collection end performs median average filtering on the collected soil nitrogen, phosphorus, and potassium content data to obtain n data X1, X2,..., X n, the variance of the median average filtered data is calculated The average and variance of the soil nitrogen, phosphorus and potassium content data are sent to the relay module through the LoRa wireless communication network.

[0032] The first stage uses a median average filtering algorithm to perform nonlinear filtering on data with large errors. The principle of median filtering is to sort and count the data within the filtering window, eliminate a certain number of maximum and minimum values in the data, and calculate the average of the remaining values, which is the output value of the median filtering. The filtered result is as close to the true value as possible, thereby eliminating the influence of abnormal data on the whole.

[0033] The other steps and parameters are the same as those in embodiment one or two.

[0034] Embodiment four: The difference between this embodiment and one of embodiments one to three is that the relay module is responsible for receiving the data sent by the collection module and performing second-level processing on the data to obtain the optimal fusion result The specific process is as follows:

[0035] The optimal weighting factor is obtained by using the variance, and the n data X1, X2,...X n The optimal fusion result is calculated

[0036] The optimal weighting factor is calculated as shown in formula (4):

[0037]

[0038]

[0039] Where W i * is the optimal weighting factor;

[0040] The optimal mean square error is calculated as shown in formula (5):

[0041]

[0042] Where σ min is the optimal mean square error;

[0043] The optimal fusion result is shown in formula (6):

[0044]

[0045] The other steps and parameters are the same as those in one of embodiments one to three.

[0046] Embodiment five: different from one of embodiments one to four is that the relay module needs to be externally connected with an SD memory card to save and backup data.

[0047] The relay and the collection end adopt a common design. Compared with the collection end, the relay does not need to be connected with sensors, but needs to store a large amount of data. Therefore, the relay module needs to be externally connected with an SD memory card to save and backup data.

[0048] The other steps and parameters are the same as one of embodiments one to four.

[0049] Embodiment six: different from one of embodiments one to five is that the main controller module adopts i.MX6ULL as a processing core.

[0050] The main controller module is connected with a screen through a network interface to complete the functions of remote and local monitoring.

[0051] The main controller module includes an i.MX6ULL chip, and the hardware circuit mainly consists of a minimum system circuit, a power supply circuit, a communication circuit, a functional interface circuit, a relay control circuit and an alarm circuit.

[0052] The other steps and parameters are the same as one of embodiments one to five.

[0053] Embodiment seven: different from one of embodiments one to six is that the process of obtaining the trained IPSO-LSTM fertilization amount prediction model is as follows:

[0054] Step 1: obtaining a training set;

[0055] The training set has a fertilization amount label corresponding to the nitrogen, phosphorus and potassium content data in the soil;

[0056] The training set has a fertilization amount label, and each sample in the training set contains a set of feature data of nitrogen, phosphorus and potassium. In the training process, the model will be adjusted and optimized according to the input features and the corresponding fertilization amount label to minimize the difference between the predicted value and the actual fertilization amount.

[0057] Step 2: inputting the training set into the IPSO-LSTM model for training until the maximum iteration number t is reached max to obtain the trained IPSO-LSTM model;

[0058] The input of the IPSO-LSTM model is the soil nitrogen, phosphorus and potassium data;

[0059] The output of the IPSO-LSTM model is the prediction result of the fertilization amount in the soil nitrogen, phosphorus and potassium content data;

[0060] When the training iteration number In this method, the inertia weight w is adjusted by decreasing the inertia weight w1 using a concave function. For example, in 100 training iterations, the inertia weight w is set to w1 for iterations 1-50 (inclusive). In the early stages of model training, it is hoped that the particle swarm can explore the search space more extensively in order to find a possible global optimum in the possible solution space. The method of adjusting the inertia weight by decreasing the concave function uses a large inertia weight in the initial iteration, prompting the particles to move in the search space with larger steps. Then, the inertia weight is gradually decreased, causing the particles to gradually converge to a possible solution.

[0061] When the number of training iterations In this method, the inertia weight w decreases linearly by w2; for example, in 100 training iterations, the inertia weight w is w2 from iteration 51 to 1000. In the later stages of model training, it is hoped that the particle swarm will gradually focus on the possible solution space to find the global optimum more accurately. The linearly decreasing inertia weight method uses a large inertia weight in the initial iteration, and then linearly decreases the inertia weight according to a certain rule in each iteration to slow down the movement speed of the particles, thereby locating the solution more accurately.

[0062] In the formula, t is the current iteration number; t max This represents the maximum number of iterations.

[0063] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0064] Specific Implementation Method Eight: This implementation method differs from one of Specific Implementation Methods One to Seven in that the calculation formulas for the concave function decreasing adjustment inertia weight w1 and the linear decreasing inertia weight w2 are as follows;

[0065]

[0066]

[0067] In the formula, w max The maximum value of the inertial weight; w min The minimum inertia weight is t; t is the current iteration number; t max This represents the maximum number of iterations. This is a control factor.

[0068] This patented Particle Swarm Optimization (IPSO) algorithm proposes a strategy of using a concave function to decrease the inertia weight, thereby preventing premature convergence. By adjusting the inertia weight using a concave function, the accuracy and precision of the PSO algorithm's optimization can be improved.

[0069] A fertilizer application prediction model was established and pre-trained based on the IPSO-LSTM neural network;

[0070] The IPSO-LSTM neural network based fertilization amount prediction model refers to a method for predicting the fertilization amount required by crops in a future period of time by using an improved particle swarm optimization algorithm (IPSO) and a long short-term memory neural network (LSTM). The model is based on historical fertilization data, uses the IPSO algorithm to determine the optimal neural network structure and hyperparameters, and then uses the LSTM neural network to train and optimize the data to predict the growth status and fertilization demand of crops in a future period of time.

[0071] The other steps and parameters are the same as one of embodiments one to seven.

[0072] Embodiment nine: This embodiment is different from one of embodiments one to eight in that the maximum inertia weight w max is 0.9, and the minimum inertia weight w min is 0.4.

[0073] The other steps and parameters are the same as one of embodiments one to eight.

[0074] Embodiment ten: This embodiment is different from one of embodiments one to nine in that the main controller module takes the optimal fusion result data sent by the relay as input, and predicts the fertilization amount by using the internally deployed trained IPSO-LSTM fertilization amount prediction model; the specific process is as follows:

[0075] The optimal fusion result data sent by the relay to be tested is input into the trained IPSO-LSTM fertilization amount prediction model, and the fertilization amount prediction value is output.

[0076] The other steps and parameters are the same as one of embodiments one to nine.

[0077] The following examples are used to verify the beneficial effects of the present application:

[0078] Example one:

[0079] IPSO-LSTM fertilization amount prediction model simulation and analysis are performed;

[0080] Experimental data collection, this time the corn variety is Fuer116, the collection site is the water saving company test field in the east of Suihua City, Suihua City is located in Songnen Plain, the black land can be cultivated to 40 centimeters, the annual sunshine hours is 2790.4 hours, the annual average precipitation is 460.3 millimeters, and the terrain is flat, which provides an excellent environment for the vigorous growth of corn and other crops. According to the operating characteristics of the fertilizer machine, the data collection period is from 2019 to 2021, and the collection unit is per hectare. 50 hectares of corn soil nitrogen (N) phosphorus (P) potassium (K) content, fertilizer nitrogen (N) phosphorus (P2O5) potassium (K2O) content and yield data are collected every year, invalid data is removed, and finally 140 groups of valid samples with 7 attributes are obtained. The sample data is divided into training set and test set according to the ratio of 5:2, and the data is as follows:

[0081] Part of the experimental field data table

[0082]

[0083]

[0084] In order to reasonably select the number of neurons and learning rate of LSTM neural network, IPSO-LSTM and PSO-LSTM models are trained. First, input the collected 100 training samples into IPSO-LSTM prediction model and PSO-LSTM prediction model. The initial parameters of the two models are set as follows: the particle population size is 20, the iteration number is 100, the learning factors c1 and c2 are both set to 2, the optimizer uses Adma algorithm, in order to speed up the training speed and reduce the calculation amount of subsequent model deployment, the number of hidden layers of LSTM neural network is set to 1. The search range of hidden layer neurons and learning rate is [1, 100] and [0.0001, 0.01] respectively, and the corresponding search speed interval is [-5, 5] and [-0.0005, 0.0005] respectively. The optimization process of the prediction model is as shown in Figure 3 .

[0085] It can be known from Figure 3 that the optimization results of IPSO for the number of hidden layer neurons and learning rate of LSTM neural network are 92 and 0.9702x10 -2 respectively. The optimal number of hidden layer neurons and learning rate of LSTM neural network obtained by PSO algorithm are 89 and 0.995x10 -2 respectively. Compared with the traditional particle swarm algorithm, the IPSO algorithm has higher optimization accuracy for the learning rate of LSTM neural network.

[0086] In order to compare the performance of the model more widely and reliably, the IPSO-LSTM model and the PSO-LSTM model are established based on the above parameters, and the classical machine learning BP neural network and the LSTM neural network without optimization algorithm are selected to construct the fertilizer application amount prediction model. In order to evaluate the prediction accuracy of each model, the absolute error of the prediction data is calculated by subtracting the true value from the predicted value, and the calculation formula is:

[0087] E AE = x i - y i

[0088] In the formula, x i is the predicted value; y i is the measured value; n is the sample number;

[0089] At the same time, four error analysis methods are used to evaluate the prediction accuracy of the model, and the mean absolute percentage error (MAPE) is used as a statistical indicator to measure the prediction accuracy of each group of models, and the calculation formula is:

[0090]

[0091] In the formula, x i is the predicted value; y i is the measured value; n is the sample number;

[0092] The mean square error (MSE) and the root mean square error (RMSE) show the dispersion of the data, and the calculation formula is:

[0093]

[0094]

[0095] In the formula, x i is the predicted value; y i is the measured value; n is the sample number;

[0096] The mean absolute error (MAE) is used to avoid mutual offset between the prediction errors of each model, so as to accurately evaluate each model, and the calculation formula is:

[0097]

[0098] In the formula, x i is the predicted value; y i is the measured value; n is the sample number;

[0099] The 4 groups of models are trained, the same 40 groups of test sets are used for N, P and K fertilizer application prediction, and compared with the true value. The N, P and K fertilizer application prediction results are as follows Figure 4 、 5 、6. It is concluded from the following Figure 4 、 5 、6: the prediction error of the fertilizer application prediction model based on BP neural network is relatively large, which verifies the inference that traditional machine learning is not easy to capture the relationship between data when processing nonlinear data, and is prone to overfitting. The prediction values of LSTM model and PSO-LSTM model are relatively balanced, but there is a large error in the prediction of individual samples. Compared with the other three groups of models, in the 40 test samples, the IPSO-LSTM model predicts the nitrogen, phosphorus and potassium fertilizer application, and the fitting degree of the actual nitrogen, phosphorus and potassium fertilizer application is the highest.

[0100] Among them, the accuracy of each model to predict the fertilizer application is compared;

[0101]

[0102] As can be seen from the table: the MAPE, MSE, RMSE and MAE of the ISPO-LSTM prediction model are 1.2, 1.445, 1.202 and 0.968, respectively, compared with the BP prediction model, each index is increased by 57.1%, 68.3%, 43.7% and 46.9%, respectively; compared with the PSO-LSTM prediction model, each index is increased by 36.8%, 43.8%, 25.1% and 35.1%, respectively; compared with the LSTM prediction model, each index is increased by 47.8%, 50.7%, 19.8% and 45.1%, respectively. The error analysis of the four groups of models shows that the fertilizer application prediction model based on IPSO-LSTM can effectively improve the prediction accuracy of fertilizer application, and the maximum error with the true measurement value is 1.2 kg / hm 2 , which meets the actual agricultural fertilizer error requirement.

[0103] The present application also has other various embodiments, and those skilled in the art can make various corresponding changes and modifications according to the present application without departing from the spirit and essence of the present application, but these corresponding changes and modifications should all belong to the protection scope of the claims attached to the present application.

Claims

1. An open field corn fertilization system based on IPSO-LSTM algorithm, characterized in that: The system comprises a collection module, a relay module, a main controller module and a remote monitoring module; The collection module comprises a plurality of collection ends, which acquire soil parameters of the fertilization area through sensors, perform first-level processing on the soil parameters of the fertilization area, and send the processed data to the relay module; The relay module is responsible for receiving the data sent by the collection module and performing second-level processing on the data to obtain an optimal fusion result, which is sent to the main controller module; The main controller module takes the optimal fusion result data sent by the relay module as input, predicts the fertilizer amount through the trained IPSO-LSTM fertilizer amount prediction model, controls the fertilizer ratio, and thus completes the function of autonomous fertilization; The IPSO is a particle swarm optimization algorithm, and the LSTM is a long short-term memory neural network; The remote monitoring module is used for data interaction with the main controller module, thereby realizing the function of remote monitoring; The soil parameters are soil nitrogen, phosphorus and potassium element content, soil pH value, soil temperature and humidity, soil light intensity and soil moisture; The sensors are nitrogen, phosphorus and potassium element sensors, pH value sensors, temperature and humidity sensors, light intensity sensors and soil moisture sensors; The collection module is powered by a storage battery and solar power; The collection end acquires soil parameters of the fertilization area through sensors, performs first-level processing on the soil parameters of the fertilization area, and sends the processed data to the relay module; the specific process is as follows: The soil nitrogen, phosphorus and potassium sensor is connected with the collection end through an RS485 interface, the collection end performs median average filtering on the collected soil nitrogen, phosphorus and potassium content data to obtain one data , calculates the variance of the median average filtered data , and sends the average value and the variance of the soil nitrogen, phosphorus and potassium content data to the relay module through a LoRa wireless communication network. The relay module is responsible for receiving the data sent by the collection module and performing second-level processing on the data to obtain an optimal fusion result ; The specific process is as follows: The optimal weighting factor is obtained by using variance, and the optimal fusion result is calculated based on the optimal weighting factor and the data obtained by median average filtering The optimal fusion result is calculated ;​ The optimal weighting factor is calculated according to formula (4): （4） wherein is the optimal weighting factor; The optimal mean square error is calculated according to formula (5): （5） wherein is the optimal mean square error; the optimal fusion result see equation (6): （6） The relay module needs to be connected with an SD memory card to save and backup data; The main controller module adopts i.MX6ULL as the processing core; The main controller module is connected with a screen through a network interface to complete the functions of remote and local monitoring; The process of obtaining the trained IPSO-LSTM fertilizer amount prediction model is as follows: Step 1: Obtain a training set; The training set has fertilizer amount labels corresponding to soil nitrogen, phosphorus and potassium content data; Step 2: input the training set into the IPSO-LSTM model for training until the maximum number of iterations is reached to obtain a trained IPSO-LSTM model; The input of the IPSO-LSTM model is soil nitrogen, phosphorus and potassium data; The output of the IPSO-LSTM model is the prediction result of the fertilizer amount in the soil nitrogen, phosphorus and potassium content data; When the number of training iterations , the inertia weight is adjusted by taking the concave function decreasing inertia weight ; When the number of training iterations , the inertia weight is taken as a linearly decreasing inertia weight ; In the formula, is the current iteration number; is the maximum iteration number; The concave function decreases the adjustment inertia weight and the linear decreasing inertia weight The calculation formula is: wherein is the maximum value of the inertia weight; is the minimum value of the inertia weight; is the current iteration number; is the maximum iteration number; is the control factor; the maximum value of the inertia weight is 0.9, the minimum value of the inertia weight is 0.4; The main controller module takes the optimal fusion result data sent by the relay module as input, and predicts the fertilizer amount through the trained IPSO-LSTM fertilizer amount prediction model; the specific process is as follows: The optimal fusion result data sent by the relay module is input into the trained IPSO-LSTM fertilizer amount prediction model, and the fertilizer amount prediction value is output.

Images