Matrix strawberry evapotranspiration single-step prediction model and method based on NARX neural network

Abstract

The invention relates to a matrix strawberry evapotranspiration single-step prediction model and a matrix strawberry evapotranspiration single-step prediction method based on an NARX neural network. S1, data acquisition and processing; s1.1, acquiring corresponding weight data based on an electronic scale, measuring irrigation volume and displacement based on a weighing method, and measuring evapotranspiration based on a water balance method; s1.2, microclimate factors are collected; a sensor is arranged above the planting pot; the sensors comprise a temperature sensor, a relative humidity sensor and an effective optical radiation sensor, and the sensors are electrically connected with a background server, record greenhouse microclimate factor changes in real time and upload the greenhouse microclimate factor changes to the background server; s1.4, data processing is carried out; s2, constructing an evapotranspiration prediction model based on an NARX neural network; s2.1, NARX neural network model training is carried out; and S2.2, verifying the NARX neural network model.

Description

Technical Field

[0001] The invention relates to a single-step prediction model and method for substrate strawberry evapotranspiration based on a NARX neural network, and belongs to the technical field of artificial intelligence control and prediction in precision agriculture. Background Art

[0002] Evapotranspiration is the main link of crop water consumption and the main basis for making irrigation decisions. Crop evapotranspiration is highly coupled with environmental factors such as temperature and humidity, light intensity, and soil moisture content, so it is very difficult to establish an accurate calculation model. Artificial intelligence algorithm neural network models have a strong learning ability for various nonlinear coupling relationships and can map various complex input-output relationships. However, there are differences in the learning ability of different neural network models, so it is necessary to scientifically select neural network models to establish a prediction model to accurately estimate future evapotranspiration and facilitate the implementation of irrigation decisions.

[0003] The Penman formula is a widely used traditional evapotranspiration calculation model, which calculates the evapotranspiration of a reference crop (generally open-air alfalfa) on the same day based on the statistics of multiple meteorological factors on the day. The parameters of this formula are complex to select and are not suitable for greenhouse crops with closed microclimate characteristics.

[0004] With the development of artificial intelligence technology, BP neural network models have been applied to evapotranspiration forecasting. Most studies are limited to establishing a static mapping relationship between the current environmental influencing factors and the current evapotranspiration, and cannot achieve advance prediction. Moreover, the results calculated by the Penman formula are mostly used as standard training values ​​for BP neural networks, which cannot be applied to greenhouse irrigation decisions. Precision irrigation is an important part of achieving "water and fertilizer integration" and improving strawberry quality, so it is necessary to develop a neural network model that can predict evapotranspiration in advance.

[0005] The NARX (non-linear auto-regressive with exogenous inputs) network is a recurrent network model. The model is not only driven by its own historical state, but also receives external historical variable inputs to jointly predict the output at the current moment. Since the network has a time delay unit, it is good at mining the underlying historical development logic of time-dependent sequences and is a type of network model with dynamic memory function. The NARX neural network model can predict the evaporation volume of the next time step based on a small amount of historical data.

[0006] The present invention creatively selects the NARX model, which can accurately realize the advance prediction of the evapotranspiration of matrix strawberries. From the establishment to the practice of this model, it is necessary to collect environmental data in real time and continuously calculate the evapotranspiration of each period. The technical difficulty of evapotranspiration measurement is also creatively solved by the present invention. Summary of the invention

[0007] The technical problem to be solved by the present invention is generally to provide a single-step prediction model and method for the evapotranspiration of matrix strawberry based on NARX neural network. Strawberry is a widely planted matrix crop. The present invention uses matrix strawberry as research material, selects NARX network model for evapotranspiration prediction, provides a complete implementation plan, and is also convenient for implanting automatic irrigation decision-making systems for other matrix crops.

[0008] To solve the above problems, the technical solution adopted by the present invention is:

[0009] A single-step prediction method for substrate strawberry evaporation based on NARX neural network, the prediction method comprising the following steps:

[0010] S1, data acquisition and processing;

[0011] S1.1, obtain the corresponding weight data based on the electronic scale, measure the irrigation volume and drainage volume based on the weighing method, and measure the evaporation volume based on the water balance method;

[0012] S1.2, microclimate factor collection: a sensor is arranged above the planting pot; the sensor includes a temperature sensor, a relative humidity sensor and an effective light radiation sensor, the sensor is electrically connected to the background server, and records the changes of the greenhouse microclimate factors in real time and uploads them to the background server;

[0013] S1.3, leaf index measurement; the leaf index is the number of leaf layers per unit area, that is,

[0014] Leaf area index = total leaf area planted area (1);

[0015] S1.4, data processing;

[0016] First, set the statistical period; then, calculate the average temperature, average relative humidity and total effective light radiation in the greenhouse by period; secondly,

[0017] The evaporation is calculated using the water balance method using the formula:

[0018] ET = Δw3 - Δw2 - Δw1 (2);

[0019] Where: Δw1, Δw2, Δw3 are the mass changes of the planting pot, drainage bucket and flow bucket in each period, unit: g;

[0020] Δw3 takes the average value of the statistics of traffic bucket a and traffic bucket b in each time period;

[0021] S2, construction of evapotranspiration prediction model based on NARX neural network;

[0022] S2.1, NARX neural network model training;

[0023] The NARX neural network consists of an input layer, a time delay layer, a hidden layer, and an output layer. The transfer function of the hidden layer uses a Sigmoid function, the output layer uses a linear function, and the root mean square error RMSE and the determination coefficient R 2 Measures the training accuracy of the model; among them,

[0024] First, determine the root mean square error;

[0025]

[0026] Then, determine the coefficient;

[0027]

[0028] Where: N is the total sample size; y(t), are the predicted value, measured value and the average of the measured value respectively; several basic candidate models are obtained through NARX open-loop network training;

[0029] S2.2, NARX neural network model validation;

[0030] First, the basic candidate model of the NARX neural network is transferred to the de-delay model and verified with the validation set. The temperature and humidity, light radiation, leaf index, evapotranspiration and historical delay data of each quantity at time t are input to predict the evapotranspiration at time t+1. Then, the minimum root mean square error (RMSE) and the coefficient of determination (R) are selected. 2 The maximum de-delay model is used as the local strawberry evaporation forecast model;

[0031] Further, in step S1.1, first, a strawberry planting pot, a flow barrel and a drainage barrel are respectively provided, and the planting pot, the flow barrel and the drainage barrel are respectively equipped with corresponding electronic scales, and the mass change of the planting pot is monitored by the planting pot and the corresponding electronic scale; the strawberry irrigation amount is monitored by the flow barrel and the corresponding electronic scale; the strawberry drainage amount is monitored by the drainage barrel and the corresponding electronic scale, and the evaporation amount is calculated by the water balance method;

[0032] In step S1.3, first, the leaf area is measured by a leaf area meter, and the strawberry leaves are classified into three types, namely, large leaves, medium leaves and small leaves; then, the average of the three types of leaf areas is taken as the average leaf area A of the same period. aver ; Then count the total number of leaves N, with NA aver Measure the leaf index as total leaf area; set the measurement frequency.

[0033] Furthermore, the planting pots include planting pots a and planting pots b, the planting pots a are equipped with electronic scales a, and the planting pots b are equipped with electronic scales c, and the mass changes of the planting pots a and the planting pots b are monitored respectively according to the electronic scales a and c;

[0034] Planting pot a and planting pot b are equipped with drainage bucket a and drainage bucket c respectively; drainage bucket a and drainage bucket c are equipped with electronic scale b and electronic scale d respectively, and the drainage volume of strawberries in planting pot a and planting pot b is monitored by electronic scale b and electronic scale d respectively;

[0035] Flow bucket a and flow bucket b are equipped with electronic scale f and electronic scale e respectively. Electronic scale f and electronic scale e respectively measure the irrigation volume of drop arrows equal to the planting pots. The average value of the measurement results of the two flow buckets is used as the irrigation volume of a single planting pot during the statistical period.

[0036] Based on the mass changes of planting pots, drainage buckets and flow buckets, the evaporation volume is calculated using the principle of water balance method;

[0037] Electronic scale a, electronic scale c, electronic scale b, electronic scale d, electronic scale f, and electronic scale e transmit weight to the background server in real time.

[0038] Furthermore, equal number of seedlings are arranged in planting pot a and planting pot b; each seedling is equipped with a dripping arrow for irrigation;

[0039] Flow barrel a and flow barrel b are respectively equipped with the same number of drop arrows as planting pot a and planting pot b;

[0040] The lower part of the planting pot a is connected to the drainage pipe a; the lower outlet of the drainage pipe a is located above the drainage bucket a;

[0041] The lower part of the planting pot b is connected to the drainage pipe b; the lower outlet of the drainage pipe b is located above the drainage bucket c;

[0042] Drainage barrel a and drainage barrel c are closed barrels;

[0043] The lower outlet of the drain pipe a is sealed with the drain bucket a;

[0044] The lower outlet of the drainage pipe b is sealed with the drainage barrel c.

[0045] Furthermore, microclimate factors include temperature, humidity, and total effective light radiation.

[0046] Further, in step S1.3, Δw1, Δw2, and Δw3 are obtained by an electronic scale;

[0047] The data of the corresponding statistical time periods are stored according to the different growth stages of strawberries.

[0048] Further, in step S2.1, the following steps are included:

[0049] S2.1.1, map the same sequence data to the interval [-1,1], the formula is as follows:

[0050]

[0051] Where: xs is the normalized value; x is the measured value of a factor in a sample sequence; x min is the minimum value of a sequence; x max is the maximum value of a sample sequence.

[0052] The data were normalized during the flower bud differentiation period, flowering period, and flowering and fruiting period; the flowering and fruiting period included winter and spring and summer;

[0053] S2.1.2, in the open-loop model of the NARX neural network model, set the parameters, where the variables of the input layer include temperature and humidity, light radiation, leaf index and evapotranspiration, the variables of the output layer are evapotranspiration, the delay layer dn, and the number of hidden layer units N;

[0054] S2.1.3, according to the growth stage, select and set the training set data of a certain period and input them into the NARX open-loop neural network model for training, and use the LM algorithm for iterative calculation;

[0055] S2.1.4, select the root mean square error RMSE is less than the set value and the determination coefficient R 2 Multiple open-loop network models greater than the set value are used as NARX basic candidate models.

[0056] A single-step prediction model for substrate strawberry evapotranspiration based on a NARX neural network is used to execute the above prediction method; the model includes a NARX neural network and a plurality of planting pots.

[0057] The NARX network is a recurrent network model that is not only driven by its own historical state, but also receives external historical variable inputs to jointly predict the output at the current moment. It is a type of network model with dynamic memory function.

[0058] The present invention can use a small amount of historical data to perform single-step advance prediction. The present invention collects and processes various data through a background server and selects the NARX network model to predict evapotranspiration. The NARX neural network model can predict the evapotranspiration of the next time step based on a small amount of historical data. From the establishment to the practice of this model, it is necessary to collect environmental data in real time and continuously calculate the evapotranspiration of each time period. The present invention solves the technical difficulty of evapotranspiration measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the automatic weighing method of the present invention.

[0060] Figure 2 Schematic diagram of the NARX neural network of the present invention.

[0061] Figure 3 It is a schematic diagram of the NARX open-loop network structure of the present invention.

[0062] Figure 4 It is a diagram of the NARX delay removal network structure of the present invention.

[0063] Figure 5 It is a schematic diagram of the NARX evapotranspiration advance prediction model of the present invention.

[0064] Figure 6 It is a schematic diagram comparing the predicted values ​​and the measured values ​​of the NARX model of the present invention.

[0065] Figure 7 It is a schematic diagram of the BP neural network model results of the present invention.

[0066] Among them, 1. planting pot a; 2. electronic scale a; 3. drainage bucket a; 4. electronic scale b; 5. planting pot b; 6. electronic scale c; 7. drainage bucket c; 8. electronic scale d; 9. flow bucket a; 10. flow bucket b; 11. electronic scale e; 12. electronic scale f; 13. electric valve a; 14. electric valve b; 15. electric valve c; 16. electric valve d; 17. dropping arrow. DETAILED DESCRIPTION

[0067] like Figure 1 ～ 7. This embodiment provides a real-time data acquisition scheme, a evaporation measurement method, and specific steps for establishing a NARX neural network model, which specifically include the following steps:

[0068] S1, data acquisition and processing;

[0069] S1.1, evaporation measurement;

[0070] First, two planting pots a1 and b5 can be set up for experimental control. Electronic scales a2 and c6 are respectively used to monitor changes in the quality of the planting pots. Based on the electronic scales to obtain the corresponding weight data, drainage barrels a3 and c7 are respectively equipped with electronic scales b4 and d8 to monitor the drainage of strawberries. Flow barrels a9 and b10 are equipped with electronic scales f12 and e11 to measure the equal amount of drip irrigation in the planting pots, and the average value is used as the statistical irrigation amount of a single planting pot during the period. Based on the water balance method, the evaporation ET of each planting pot is calculated separately, and the sum of the evaporation of the two planting pots is used as the training verification data. There can be several planting pots.

[0071] As a specific preference, the specifications of the planting pot a1 and the planting pot b5 are: length × width × height = 50cm × 20cm × 20cm, with a total of 6 seedlings in each pot. A drip arrow 17 is inserted into each plant for sufficient irrigation, and the drip arrow 17 is preferably a Netafim drip arrow. The planting pot a1 and the planting pot b5 are respectively equipped with a drainage barrel a3 and a drainage barrel c7; the drainage barrel a3 and the drainage barrel c7 are respectively sealed square barrels for collecting the drainage of the corresponding planting pot a1 and the planting pot b5. In addition, the flow barrel a9 and the flow barrel b10 are another type of sealed square barrels, which are used to measure the irrigation volume in each time period, and each square barrel is provided with 6 groups of drip arrows equal to the planting pot. Electronic scales that can transmit weight in real time are provided under the planting pots, drainage barrels and flow barrels, see Figure 1 .

[0072] The drainage barrel a3, the drainage barrel c7, the flow barrel a9, and the flow barrel b10 are respectively installed with electronic scales b4, electronic scales d8, electronic scales f12, and electronic scales e11, and the fertilizer liquid is regularly emptied at 24:00 every day.

[0073] The electronic scale can transmit the weight to the background server in real time; due to the weak water holding capacity of the matrix, the matrix strawberry irrigation is characterized by small amounts and multiple times, which continuously meets the crop's daily water demand. The fertilization system is frequently started, and the irrigation time is very short, generally about 10 minutes, and the non-constant flow is the main irrigation process. The existing commonly used Hall flow sensors only have reliable accuracy for constant flow, so the method of setting a flow bucket to directly count the irrigation amount in the present invention can effectively circumvent this problem, which is simple, economical, easy to implement, and innovative.

[0074] S1.2, microclimate factor collection;

[0075] Temperature, relative humidity and effective light radiation sensors are installed about 1.5m above the planting pots to record the changes of greenhouse microclimate factors in real time 24 hours a day, with a collection frequency of 10 seconds. The sensors are electrically connected to the backend server to record the changes of greenhouse microclimate factors in real time and upload them to the backend server;

[0076] S1.3, leaf index measurement;

[0077] As strawberries grow from seedlings to maturity, their canopy layers gradually grow and transpiration becomes stronger. The leaf index is an important parameter for measuring the characteristics of this crop. The leaf index is the number of leaf layers per unit area, i.e.

[0078] Leaf area index = total leaf area planted area (1);

[0079] First, the leaf area is measured with a leaf area meter, and the strawberry leaves can be classified into three types, namely large leaves, medium leaves and small leaves; then, the average of the three types of leaf areas is taken as the average leaf area A of the same period. aver ; Then count the total number of leaves N, with NAaver Use the leaf index as a measure of the total leaf area; set the measurement frequency to 7 to 10 days.

[0080] S1.4, data processing;

[0081] First, set the statistical period; then, count the average temperature, average relative humidity and total effective light radiation in the greenhouse by time period, such as 0.5 hour and 1 hour.

[0082] By default, the leaf index between two measurement days does not change and is recorded according to the last statistical result. The leaf index is inserted into the corresponding statistical period on a daily basis.

[0083] The evaporation is calculated using the water balance method using the formula:

[0084] ET = Δw3 - Δw2 - Δw1 (2);

[0085] Where: Δw1, Δw2, Δw3 are the mass changes of the planting pot, drainage bucket and flow bucket in each period, unit: g;

[0086] Δw3 takes the average of the statistics of each period of the two flow buckets d9 and d10, which can avoid the statistical error of irrigation volume caused by the flow difference of the drip arrow 17. If the number of plants is too small, the evapotranspiration is difficult to measure and the accidental error increases. Therefore, the final measured value of evapotranspiration takes the total amount of all experimental planting pots, and the NARX neural network model is established by training with this data.

[0087] Since strawberries have different water consumption capacities at different growth stages, it is necessary to store various data in stages. The present invention preferably establishes a model according to the flower bud differentiation stage, flowering stage, flowering and fruiting stage (winter), and flowering and fruiting stage (spring and summer), or several of the stages may be selected.

[0088] The data include: average temperature during the period, average humidity during the period, total effective light radiation during the period, leaf index and total evapotranspiration during the period.

[0089] Secondly, store the corresponding data according to different growth stages: flower bud differentiation period, flowering period, flowering and fruiting period (winter), flowering and fruiting period (spring and summer);

[0090] S2, construction of evapotranspiration prediction model based on NARX neural network;

[0091] S2.1, NARX neural network model training;

[0092] The NARX neural network consists of an input layer, a time delay layer, a hidden layer, and an output layer. The transfer function of the hidden layer uses a Sigmoid function, and the output layer uses a linear function. Figure 2The input layer and output layer are determined by the problem being solved, and the delay layer and hidden layer structure can be adjusted freely. The model structure determines the data fitting ability of the NARX neural network, which can be expressed by the root mean square error RMSE and the coefficient of determination R 2 Measures the training accuracy of the model.

[0093] (1) Root mean square error

[0094]

[0095] (2) Determination coefficient

[0096]

[0097] Where: N is the total sample size; y(t), are the predicted value, measured value and the average value of the measured value respectively; several basic candidate models are obtained through NARX open-loop network training. The specific steps are as follows:

[0098] S2.1.1, different data sets may have large differences in magnitude due to different dimensions. In order to avoid difficulties in convergence of neural network model training, the training data needs to be normalized. Map the same sequence data to the interval [-1,1], the formula is as follows:

[0099]

[0100] Where: x s is the normalized value; x is the measured value of a factor in a sample sequence; x min is the minimum value of one of the sequences; x max is the maximum value of a sample sequence.

[0101] During the flower bud differentiation period, flowering period, flowering and fruiting period (winter), and flowering and fruiting period (spring and summer), each data was normalized.

[0102] S2.1.2, in the NARX open-loop model (e.g. Figure 3 As shown in the figure, multiple model structures are determined for training. For example, the input layer can be the following variables: temperature and humidity, light radiation and evaporation, the output layer can be one variable, namely evaporation, the delay layer dn=6, and the hidden layer unit is 12. The delay layer dn can also be adjusted to 4, and the hidden layer unit can be adjusted to 16.

[0103] S2.1.3, according to the growth stage, select a certain period of training set data and input them into the NARX neural network model for training. For example, if the step length is 1 hour, a prediction model for the forecast period of 1 hour can be established. The number of iterations is 200, and the algorithm using LM (Levenberg-Marquardt) can generally achieve a faster convergence process.

[0104] S2.1.4, minimum root mean square error RMSE and coefficient of determination R 2 The largest model is not necessarily the one with the strongest generalization ability, so the model with a smaller root mean square error RMSE and a smaller coefficient of determination R is selected. 2 Larger multiple open-loop network models are used as NARX basic candidate models.

[0105] S2.2, NARX neural network model validation;

[0106] In different growth stages, the NARX basic candidate model was transferred to the de-delay model, and the validation set was used to verify it respectively. The model with the smallest root mean square error (RMSE) and the coefficient of determination (R) was selected. 2 The largest de-delayed model is used as the local strawberry evapotranspiration forecast model. By inputting the temperature and humidity, light radiation, leaf index, evapotranspiration and historical delay data of each quantity at time t, the evapotranspiration at time t+1 can be predicted, such as Figure 4 .

[0107] The present invention can use a small amount of historical data for single-step advance prediction. From September to May of the following year, the XX plantation in XX County, XX City, China, respectively carried out irrigation experiments on matrix strawberries. The greenhouse temperature, humidity, effective light radiation and evapotranspiration were automatically measured, and the leaf index was regularly measured manually, and the data was stored in the terminal. The 1-hour data of each growth stage was taken to establish the NARX prediction model. In the flowering and fruiting period (spring and summer), the leaf index is stable, and the model ignores this influencing factor.

[0108] The one-hour data of greenhouse temperature, humidity, effective light radiation and evapotranspiration in March and April were used as the training set to train the open-loop model, and the data from May 1 to May 14 were used as the validation set to input the delayed model to verify the advance prediction ability of the neural network. Some experimental data are shown in Table 1.

[0109]

[0110] The NARX advance prediction model inputs the x variable (including average temperature and humidity, effective light radiation) in period t, the y variable (evapotranspiration) in period t, and the corresponding 5-step historical delay data. The model outputs the evapotranspiration in period t+1. The structure of the advance prediction model is shown in Figure 5 .

[0111] Combination Figure 6 The verification results in May showed that the coefficient of determination R 2 =0.9978, RMSE = 2.55g. Prediction curve Figure 6 As shown, the predicted value is in good agreement with the measured value, and the error can be controlled within a reasonable range, which can fully meet the needs of irrigation decision-making.

[0112] When the BP neural network is trained with the data from March and April, the input factors are the average temperature, average humidity, and effective light radiation in 1 hour t period, and the model output is the evaporation of strawberries in 1 hour t period. The number of hidden layer units is the same as that of the NARX model, and the training algorithm is still Levenberg-Marquardt. After training, the model root mean square error RMSE = 18.9g. The prediction results of the BP neural network for May are as follows Figure 7 As shown in Figure 2, the results show that the predicted data curve is far from the measured curve. This is sufficient to prove that the dynamic learning mechanism of the NARX neural network model gives it a more powerful nonlinear approximation and prediction ability.

[0113] The present invention is fully described for a clearer disclosure, and the prior art is not listed one by one.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein with equivalents; it is obvious for those skilled in the art to combine multiple technical solutions of the present invention. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention. The technical contents not described in detail in the present invention are all known technologies.

Claims

1. A single-step prediction method for matrix strawberry evapotranspiration based on NARX neural network, characterized in that: The prediction method includes the following steps; S1, data acquisition and processing; S1.1, obtain the corresponding weight data based on the electronic scale, measure the irrigation volume and drainage volume based on the weighing method, and measure the evaporation volume based on the water balance method; S1.2, microclimate factor collection: a sensor is arranged above the planting pot; the sensor includes a temperature sensor, a relative humidity sensor and an effective light radiation sensor, the sensor is electrically connected to the background server, and records the changes of the greenhouse microclimate factors in real time and uploads them to the background server; S1.3, leaf index measurement; the leaf index is the number of leaf layers per unit area, that is, Leaf area index = total leaf area planted area (1); S1.4, data processing; First, set the statistical period; then, calculate the average temperature, average relative humidity and total effective light radiation in the greenhouse by period; Secondly, The evaporation is calculated by the water balance method, the formula is: ET = Δw3 - Δw2 - Δw1 (2); Where: Δw1, Δw2, Δw3 are the mass changes of the planting pot, drainage bucket and flow bucket in each period, unit: g; Δw3 takes the average value of the statistics of traffic bucket a and traffic bucket b in each time period; S2, construction of evapotranspiration prediction model based on NARX neural network; S2.1, NARX neural network model training; The NARX neural network consists of an input layer, a time delay layer, a hidden layer, and an output layer. The transfer function of the hidden layer uses the Sigmoid function, the output layer uses a linear function, and the root mean square error RMSE and the determination coefficient R 2 Measures the training accuracy of the model; among them, First, determine the root mean square error; Then, determine the coefficient; Where: N is the total sample size; y(t), are the predicted value, measured value and the average of the measured value respectively; several basic candidate models are obtained through NARX open-loop network training; S2.2, NARX neural network model validation; First, the basic candidate model of the NARX neural network is transferred to the de-delay model and verified with the validation set. The temperature and humidity, light radiation, leaf index, evapotranspiration and historical delay data of each quantity at time t are input to predict the evapotranspiration at time t+1. Then, the minimum root mean square error (RMSE) and the coefficient of determination (R) are selected. 2 The maximum de-delay model is used as the local strawberry evaporation forecast model.

2. The method for predicting the evaporation of strawberries based on the NARX neural network according to claim 1, wherein: In step S1.1, first, a strawberry planting pot, a flow barrel and a drainage barrel are respectively provided, and the planting pot, the flow barrel and the drainage barrel are respectively equipped with corresponding electronic scales, and the quality change of the planting pot is monitored by the planting pot and the corresponding electronic scale; the strawberry irrigation amount is monitored by the flow barrel and the corresponding electronic scale; and the strawberry drainage amount is monitored by the drainage barrel and the corresponding electronic scale; Calculate evaporation using the water balance method; In step S1.3, first, the leaf area is measured by a leaf area meter, and the strawberry leaves are classified into three types, namely, large leaves, medium leaves and small leaves; then, the average of the three types of leaf areas is taken as the average leaf area A of the same period. aver ; Then count the total number of leaves N, with NA aver Measure the leaf index as total leaf area; set the measurement frequency.

3. The method for predicting the evaporation of strawberries based on the NARX neural network according to claim 2, characterized in that: The planting pots include planting pot a (1) and planting pot b (5), the planting pot a (1) is equipped with an electronic scale a (2), and the planting pot b (5) is equipped with an electronic scale c (6), and the mass changes of the planting pot a (1) and the planting pot b (5) are monitored respectively according to the electronic scale a (2) and the electronic scale c (6); The planting pot a (1) and the planting pot b (5) are respectively equipped with a drainage bucket a (3) and a drainage bucket c (7); the drainage bucket a (3) and the drainage bucket c (7) are respectively equipped with an electronic scale b (4) and an electronic scale d (8), and the drainage volume of the strawberries in the planting pot a (1) and the planting pot b (5) is respectively monitored by the electronic scale b (4) and the electronic scale d (8); Flow bucket a (9) and flow bucket b (10) are respectively equipped with electronic scale f (12) and electronic scale e (11), and electronic scale f (12) and electronic scale e (11) respectively measure the irrigation amount of the drop arrows equal to the planting pot, and the average value of the measurement results of the two flow buckets is used as the irrigation amount of a single planting pot in the statistical period; Based on the mass changes of planting pots, drainage buckets and flow buckets, the evaporation volume is calculated using the principle of water balance method; Electronic scale a (2), electronic scale c (6), electronic scale b (4), electronic scale d (8), electronic scale f (12), and electronic scale e (11) transmit weights to the backend server in real time.

4. The single-step prediction method for matrix strawberry evaporation based on NARX neural network according to claim 3 is characterized in that: An equal number of seedlings are arranged in the planting pot a (1) and the planting pot b (5); each seedling is equipped with a dripping arrow (17) for irrigation; The flow barrel a (9) and the flow barrel b (10) are respectively equipped with the same number of drop arrows (17) as the planting pot a (1) and the planting pot b (5); The lower part of the planting pot a (1) is connected to the drainage pipe a; the lower outlet of the drainage pipe a is located above the drainage bucket a (3); The lower part of the planting pot b (5) is connected to the drainage pipe b; the lower outlet of the drainage pipe b is located above the drainage bucket c (7); The drainage barrel a (3) and the drainage barrel c (7) are closed barrels; The lower outlet of the drain pipe a is sealed with the drain bucket a (3); The lower outlet of the drainage pipe b is sealed with the drainage barrel c (7).

5. The single-step prediction method for matrix strawberry evapotranspiration based on NARX neural network according to claim 4 is characterized in that: The dropping arrow (17) is Netafim's dropping arrow.

6. The single-step prediction method for matrix strawberry evapotranspiration based on NARX neural network according to claim 1, characterized in that: Microclimate factors include temperature, humidity, and total effective light radiation.

7. The single-step prediction method for matrix strawberry evapotranspiration based on NARX neural network according to claim 1, characterized in that: In step S1.3, Δw1, Δw2, and Δw3 are obtained by an electronic scale; Store the data of the corresponding statistical time periods according to the different growth stages of strawberries.

8. The single-step prediction method for matrix strawberry evapotranspiration based on NARX neural network according to claim 1, characterized in that: In step S2.1, the following steps are included: S2.1.1, map the same sequence data to the interval [-1,1], the formula is as follows: Where: x s is the normalized value; x is the measured value of a factor in a sample sequence; x min is the minimum value of a sequence; x max is the maximum value of a sample sequence; The data were normalized during the flower bud differentiation period, flowering period, and flowering and fruiting period. The flowering and fruiting period included winter, spring, and summer. S2.1.2, in the open-loop model of the NARX neural network model, set the parameters, where the variables of the input layer include temperature and humidity, light radiation, leaf index and evapotranspiration, the variables of the output layer are evapotranspiration, the delay layer dn, and the number of hidden layer units N; S2.1.3, according to the growth stage, select and set the training set data of a certain period and input them into the NARX open-loop neural network model for training, and use the LM algorithm for iterative calculation; S2.1.4, select the root mean square error RMSE is less than the set value and the determination coefficient R 2 Multiple open-loop network models greater than the set value are used as NARX basic candidate models.

9. A single-step prediction model for substrate strawberry evapotranspiration based on NARX neural network, characterized in that: Used to execute the prediction method described in claim 1; the model includes a NARX neural network and a plurality of planting pots.

Images