Training method and prediction method of tremella aurantialba greenhouse growth environment model based on LSTM neural network

Abstract

The invention discloses a training method and a prediction method of a tremella aurantialba greenhouse growth environment model based on an LSTM neural network. The training method comprises the following steps: acquiring original growth environment variable data of tremella aurantialba in a greenhouse; after the original growth environment variable data of the tremella aurantialba are preprocessed, data sets are divided according to the growth stages of the tremella aurantialba; training a tremella aurantialba greenhouse growth environment model based on the LSTM neural network by adopting a modulation Adam optimizer; and performing hyper-parameter optimization on the tremella aurantialba greenhouse growth environment model based on a Bayesian optimization algorithm, and determining an optimal tremella aurantialba greenhouse growth environment model of each growth stage of the tremella aurantialba. According to the invention, a greenhouse control system can be helped to provide a more suitable growth environment for the growth of the tremella aurantialba, and the yield of the tremella aurantialba is improved.

Description

Technical Field

[0001] The present invention relates to the field of agricultural technology, and in particular to a training method and a prediction method of a golden ear greenhouse growth environment model based on an LSTM neural network. Background Art

[0002] In the field of agricultural planting technology, greenhouse control systems are the mainstream technical means for current crop cultivation. Greenhouse control systems collect environmental variables in greenhouses through various environmental sensors and upload the collected environmental variables to the Internet through agricultural Internet of Things technology. Therefore, users can remotely monitor the growth environment of crops in greenhouses in real time, and can remotely control the environment in greenhouses through Internet of Things technology to adapt to the growth needs of crops. However, this strategy method relies on the experience of crop planting experts, and there is still a certain lag in controlling the greenhouse environment, which cannot fully meet the growth needs of crops. Therefore, in order to increase crop yields and reduce manpower pressure, it is necessary to promote the development of agricultural planting technology through modern deep learning technology. Deep learning technology can make reasonable predictions of environmental variables in the future by learning a large amount of historical data, and a large amount of greenhouse environmental historical data can be easily obtained through the current agricultural Internet of Things technology, which further improves the prediction accuracy and reliability of deep learning technology.

[0003] In recent years, with the continuous deepening of deep learning technology research, more and more deep learning models have been proposed by scholars. Among them, the recurrent neural network (RNN) model is a neural network model that can analyze and predict time series data. Compared with traditional neural network models, the RNN model shows stronger adaptability in time series data analysis. However, the RNN model also has problems such as gradient disappearance, gradient explosion, and insufficient long-term memory capacity. Based on these problems, a long short-term memory (LSTM) neural network model was proposed. Compared with traditional RNN, the LSTM neural network model can effectively utilize long-distance time information. Summary of the invention

[0004] In order to overcome the above-mentioned shortcomings and deficiencies of the prior art, the object of the present invention is to provide a training method for a greenhouse growth environment model of golden ear based on LSTM neural network

[0005] Another object of the present invention is to provide a method for predicting the growth environment of golden ear greenhouse based on LSTM neural network.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A training method for a golden ear greenhouse growth environment model based on an LSTM neural network, comprising:

[0008] Collect the original growth environment variable data of golden ear in the greenhouse;

[0009] After preprocessing the original growth environment variable data of golden ear, the data set was divided according to the growth stage of golden ear;

[0010] The modulated Adam optimizer was used to train the greenhouse growth environment model of golden ear based on LSTM neural network;

[0011] Based on the Bayesian optimization algorithm, the hyperparameters of the golden ear greenhouse growth environment model were optimized to determine the optimal golden ear greenhouse growth environment model for each growth stage of the golden ear.

[0012] Furthermore, the growth stages of the golden ear include the opening stage, differentiation stage, color change stage and harvesting stage. The data set is divided according to the growth stage, specifically as follows:

[0013] The environmental variables during the growth of golden ear were collected at equal time intervals to obtain the environmental variables at the first n moments and the environmental variables at the n+1th moment, and the corresponding data sets were divided according to the growth stages.

[0014] Furthermore, the golden ear greenhouse growth environment model includes four layers, namely, input layer, LSTM layer, fully connected layer and output layer. The input of the golden ear greenhouse growth environment module is the environmental variables of the previous n moments, and the output is the environmental variables of the future n+1th moment.

[0015] Furthermore, the method of using the modulated Adam optimizer to train the golden ear greenhouse growth environment model based on the LSTM neural network is specifically as follows:

[0016] Initialize parameters η1, η2, initialize time step t = 0, initial parameter vector θ0, first-order momentum M0 = 0, second-order momentum P0 = 0, initialize modulation parameters: initial value λ0 of decay learning rate, learning rate reduction factor k, learning rate reduction interval Q and weight decay rate γ;

[0017] Calculate the first-order momentum M at the current time t t and the second-order momentum P t :

[0018] Calculate the deviation correction of the first-order momentum and second-order momentum at the current time t

[0019] Update the decay learning rate λ at the current time t t ;

[0020] Update the parameter vector θ at the current time t t ;

[0021] When |θ t -θ t-1|≤ε or the maximum number of iterations is reached, the optimization process ends, and ε is the minimum parameter change threshold.

[0022] Furthermore, the hyperparameters of the golden ear greenhouse growth environment model were optimized based on the Bayesian optimization algorithm to determine the optimal golden ear greenhouse growth environment model for each growth stage of the golden ear, specifically:

[0023] Determine the hyperparameters of the golden ear greenhouse growth environment model that need to be optimized, and establish the objective function between the hyperparameters and the test set;

[0024] Define the parameter bounds of the hyperparameters that need to be optimized;

[0025] The Bayesian optimization algorithm is iterated, and the local optimal solution of the objective function is used as the global optimal solution at each iteration to obtain the hyperparameters of the golden ear greenhouse growth environment model at the global optimal moment;

[0026] If the current number of iterations reaches the maximum number of iterations, the optimization process is exited and the current global optimal golden ear greenhouse growth environment model hyperparameters are used as the final golden ear greenhouse growth environment model parameters. Otherwise, the number of iterations increases and returns to the parameter boundary step where the hyperparameters that need to be optimized need to be defined.

[0027] Furthermore, the objective function is a root mean square error function.

[0028] Furthermore, the original growth environment variable data of the golden ear include temperature, humidity, light intensity and carbon dioxide concentration.

[0029] A prediction method for a golden ear greenhouse growth environment model based on an LSTM neural network, comprising:

[0030] Using smart sensors in the greenhouse to collect data on variables in the golden ear growth environment;

[0031] Preprocess the collected environmental variable data and normalize them;

[0032] The normalized data is input into the optimal golden ear greenhouse growth environment model of the corresponding growth stage to obtain the environmental variables at the future n+1th moment. The optimal golden ear greenhouse growth environment model is obtained based on the training method of the golden ear greenhouse growth environment model based on the LSTM neural network.

[0033] A prediction system for a golden ear greenhouse growth environment model based on an LSTM neural network, comprising:

[0034] Data acquisition module: using smart sensors in the greenhouse to collect data on the growth environment variables of golden ear mushrooms;

[0035] Data preprocessing module: preprocess the collected environmental variable data and normalize them;

[0036] Data prediction module: input the normalized data into the optimal golden ear greenhouse growth environment model of the corresponding growth stage to obtain the environmental variables at the future n+1th moment, wherein the optimal golden ear greenhouse growth environment model is obtained based on the training method of the golden ear greenhouse growth environment model based on the LSTM neural network.

[0037] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method for predicting the golden ear greenhouse growth environment model is implemented.

[0038] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0039] (1) The golden ear greenhouse growth environment model based on LSTM neural network proposed in the present invention combines deep learning technology with agricultural Internet of Things technology, and uses the LSTM neural network model to model the golden ear greenhouse growth environment according to the different growth stages of the golden ear. This can help the greenhouse control system provide a more suitable growth environment for the growth of the golden ear and increase the yield of the golden ear.

[0040] (2) For the collected dataset of environmental variables of the greenhouse growth of golden ear, considering that the growth stages of golden ear are different, the temporal characteristics of the environmental variables of golden ear growth are also different. Therefore, the dataset is divided according to the growth stage of golden ear. The LSTM neural network model training is performed separately for the datasets of different growth stages, which improves the efficiency of model training and ensures the training accuracy of the model.

[0041] (3) In the modeling method of the golden ear greenhouse growth environment based on the LSTM neural network proposed in the present invention, the Bayesian optimization algorithm is used to optimize the hyperparameters of the LSTM neural network model to obtain a set of optimal hyperparameter solutions, thereby avoiding a large amount of repeated parameter adjustment work and improving the efficiency of model training and the accuracy of the model.

[0042] (4) During the training process of the LSTM neural network model, the traditional Adam optimizer is modulated to improve the convergence of the model training process and avoid overfitting. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of a training method of a golden ear greenhouse growth environment model based on an LSTM neural network of the present invention;

[0044] Figure 2 This is a schematic diagram of prediction using the golden ear greenhouse growth environment model in the present invention;

[0045] Figure 31 is a comparison chart of the model prediction and actual data of the golden ear opening period in the present invention, wherein (a) is temperature; (b) is humidity; (c) is carbon dioxide concentration;

[0046] Figure 4 1 is a comparison diagram of the model prediction and actual data of Tremella fuciformis differentiation period in the present invention, wherein (a) is temperature; (b) is humidity; (c) is carbon dioxide concentration;

[0047] Figure 5 1 is a comparison chart of the model prediction and actual data of the color change period of golden ear in the present invention, wherein (a) is temperature; (b) is humidity; (c) is carbon dioxide concentration;

[0048] Figure 6 1 is a comparison chart of the model prediction and actual data of the Tremella fuciformis harvesting period in the present invention, wherein (a) is temperature; (b) is humidity; and (c) is carbon dioxide concentration. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0050] Example 1

[0051] like Figure 1 As shown, a training method for a golden ear greenhouse growth environment model based on an LSTM neural network includes the following steps:

[0052] S1 uses smart sensors in the greenhouse to collect data on the original growth environment variables of golden ear;

[0053] S2 preprocesses the data of the environmental variables of the golden ear growth, preprocesses the collected raw environmental variable data, removes abnormal data, normalizes the raw data, constructs a data set, and divides the data sets into data sets corresponding to different stages according to the growth stage of the golden ear, and divides each data set into a training set and a test set;

[0054] Further explanation:

[0055] The growth stages of golden ear include opening stage, differentiation stage, color change stage and harvest stage. The data set is divided into data sets of different growth stages according to the growth stage, and each data set is then divided into a training set and a test set, so as to model the different growth stages of golden ear. The methods for dividing the data set include:

[0056] 1) The environmental variables of the golden ear growth process are collected at equal time intervals to obtain the environmental variables of the first n moments and the environmental variables of the n+1th moment, and the collected golden ear growth environmental variables are divided into data sets according to the growth stage.

[0057] 2) In the data set of each growth stage of golden ear, the growth environment variables of the golden ear at the first n moments are used as the input data of the training set and the test set, and the growth environment variables of the golden ear at the n+1th moment are used as the output data of the training set and the test set.

[0058] S3 creates a golden ear greenhouse growth environment model based on the LSTM neural network. It consists of four layers: input layer, LSTM layer, fully connected layer, and output layer. The input is the environmental variables of the previous n moments, and the output is the environmental variables of the future n+1 moment.

[0059] In view of the time series correlation characteristics of greenhouse environmental variables, the LSTM neural network is used to create a golden ear greenhouse growth environment model. The LSTM neural network is composed of LSTM cells. Each LSTM cell is equivalent to a neuron and has the function of memorizing and transmitting information. LSTM cells mainly include forget gates, input gates, and output gates to memorize and transmit information. t Determines how much of the cell state at the previous moment to retain, and its expression is:

[0060] L t =σ(w l *[H t-1 ,I t ]+b l ) (1)

[0061] Where t represents the current time, L t is the output of the forget gate at the current moment, σ(·) is the sigmoid function, and w l is the weight vector of the forget gate input, H t-1 is the output of the LSTM cell at the previous moment, I t is the input of the LSTM neural network at the current moment, b l is the bias coefficient of the forget gate.

[0062] Input gate U t Determines how much of the current input to keep in the current state. Its expression is:

[0063]

[0064] where w u ,w s is the weight vector of the input gate, b u ,b s is the bias coefficient of the input gate, St It is the state information of the LSTM cell at the current moment.

[0065] The output gate determines the output of the LSTM cell at the current moment, and the expression is:

[0066]

[0067] where w o is the weight vector of the input gate, b o is the bias coefficient of the input gate.

[0068] S4 uses a modulated Adam optimizer to train the golden ear greenhouse growth environment model.

[0069] The traditional Adam optimizer may have overfitting problems caused by insufficient convergence and insufficient weight decay. Therefore, the traditional Adam optimizer is modulated. Compared with the traditional Adam optimizer, the modulated Adam optimizer introduces two parameters γ and λ. t The introduction of the weight decay rate γ avoids the overfitting problem caused by insufficient weight decay during the training of the LSTM neural network model. t The introduction of improves the convergence of LSTM neural network model training.

[0070] The optimization process of modulating the Adam optimizer is as follows:

[0071] S41, initialization parameters η1, η2, initialization time step t = 0, initial parameter vector θ0, first-order momentum M0 = 0, second-order momentum P0 = 0. Initialization modulation parameters: initial value λ0 of decay learning rate, learning rate reduction factor k, learning rate reduction interval Q and weight decay rate γ, the weight decay rate is a constant, obtained by iteration according to the Bayesian optimization algorithm;

[0072] S42. Calculate the first-order momentum M at the current time t t and the second-order momentum P t :

[0073]

[0074] where Δ t is the gradient of the objective function at the current time t.

[0075] S43, calculate the deviation correction of the first-order momentum and the second-order momentum at the current time t

[0076]

[0077] S44, update the decay learning rate λ at the current time t t :

[0078]

[0079] S45, update the parameter vector θ at the current time t t :

[0080]

[0081] S46, when |θ t -θ t-1 |≤ε or the maximum number of iterations is reached, the optimization process ends, ε is the minimum parameter change threshold. Otherwise, the current time t=t+1, and the process returns to step S42.

[0082] S5 is based on the Bayesian optimization algorithm for hyperparameter optimization.

[0083] The Bayesian optimization algorithm is used to optimize the hyperparameters of the LSTM neural network greenhouse growth environment model of the golden ear in each growth period of the golden ear, and the optimal LSTM neural network greenhouse growth environment model of the golden ear in each growth period of the golden ear is obtained. The steps for optimizing the hyperparameters of the Bayesian optimization algorithm are as follows:

[0084] S51, determine the hyperparameters of the LSTM neural network that need to be optimized, establish an objective function between the hyperparameters and the test set error, and select the root mean square error function as the objective function;

[0085] Hyperparameters include the number of LSTM cells C in the LSTM layer n , modulate the weight decay rate γ, initial learning rate λ0 and learning rate reduction factor k of the Adam optimizer.

[0086] S52. Define the parameter boundary of the hyperparameter to be optimized; the parameter boundary of the hyperparameter is defined as C n ∈[10,200],γ∈[0,0.1],λ0∈[0.001,0.1],k∈[0.1,0.5];

[0087] S53, performing the Bayesian optimization algorithm iteration, taking the local optimal solution of the objective function as the global optimal solution at the current moment in each iteration, and obtaining the hyperparameters of the LSTM neural network that is the global optimal solution at the current moment;

[0088] S54, if the current number of iterations reaches the maximum number of iterations, then exit the optimization process and use the current global optimal LSTM neural network hyperparameters as the final LSTM neural network model parameters. Otherwise, increase the number of iterations and return to step S52.

[0089] S6 uses the test set to verify the trained optimal LSTM neural network model.

[0090] Further explanation:

[0091] The input vector and output vector of the golden ear greenhouse growth environment model of the LSTM neural network are:

[0092]

[0093] where v i is a vector of greenhouse growth environment variables for golden ear.

[0094] At this time, the expression of the objective function between the hyperparameters selected by the Bayesian optimization algorithm in step S5 and the test set error is as follows:

[0095]

[0096] Where ω is the hyperparameter vector to be optimized, x i is the LSTM neural network input vector of the test set, y i is the LSTM neural network output vector of the test set, f lstm (·) is the LSTM neural network model expression, and M is the number of samples in the test set.

[0097] like Figure 2 As shown, this embodiment provides a prediction method for a golden ear greenhouse growth environment model based on an LSTM neural network, including:

[0098] The intelligent sensors in the golden ear greenhouse system are used to collect the variable data of the golden ear growth environment; the variable data include temperature, humidity, carbon dioxide concentration and light intensity.

[0099] Preprocess the collected environmental variable data and normalize them;

[0100] The normalized data is input into the optimal golden ear greenhouse growth environment model of the corresponding growth stage to obtain the environmental variables at the future n+1th moment. The optimal golden ear greenhouse growth environment model is obtained based on the training method of the golden ear greenhouse growth environment model based on the LSTM neural network.

[0101] Based on the prediction method of the golden ear greenhouse growth environment model of the above LSTM neural network, the following implementation plan is proposed:

[0102] In order to verify the effectiveness of the proposed LSTM neural network modeling method, the greenhouse environmental variable dataset collected during a complete golden ear growth cycle from June 27, 2023 to August 27, 2023 was used for LSTM neural network modeling, and the performance of the proposed golden ear greenhouse growth environment model based on LSTM neural network was tested.

[0103] (1) Experimental data collection and data set division

[0104] According to the data collection work described in step S1 and the data set division work described in step S2, this example is modeled based on the LSTM neural network model for the environmental variables collected for a complete golden ear growth cycle in the greenhouse. The collected greenhouse environmental variables include temperature, humidity and carbon dioxide concentration. Since the light intensity variable data is not collected, the weight of the light intensity variable is reset to 0 during the LSTM neural network training process. The time range of the collected greenhouse environmental variables is from 19:08:24 on June 27, 2023 to 9:53:14 on August 27, 2023, and the time interval of collection is about 1min. After expert calibration, June 29 to July 4 is the opening period of golden ear growth, July 5 to July 19 is the differentiation period of golden ear growth, July 20 to August 3 is the color change period of golden ear growth, and August 4 to August 27 is the harvest period of golden ear growth. According to the growth stages of golden ear calibrated by experts, the greenhouse environmental variable data set of each growth stage of golden ear was obtained, and the data set was divided according to the ratio of training set to test set 7:3 to obtain training set and test set of golden ear opening period, differentiation period, color change period and harvest period respectively.

[0105] (2) Parameter setting of the golden ear greenhouse growth environment model and hyperparameter optimization based on the Bayesian optimization algorithm

[0106] For the construction of the LSTM neural network model described in step S3, a 4-layer LSTM neural network model is built, which includes an input layer, an LSTM layer, a fully connected layer, and an output layer. The input of the LSTM neural network is a 10×4 historical environmental variable, which represents the greenhouse environmental variables in the previous 10 minutes. The output of the LSTM neural network is a 1×4 environmental variable prediction vector, which represents the predicted greenhouse environmental variables in the 11th minute in the future.

[0107] For the Bayesian optimization algorithm described in step S5, set the maximum number of training times IT for the LSTM neural network model max =1000, the learning rate drop interval of the Adam optimizer is P=800. Select the LSTM neural network hyperparameters to be optimized and perform Bayesian optimization algorithm optimization, including: the number of LSTM cells C in the LSTM layer n , modulate the weight decay rate γ of the Adam optimizer, modulate the initial learning rate λ0 of the Adam optimizer, and modulate the learning rate reduction factor k of the Adam optimizer. These four hyperparameters are used as the optimization parameters of the Bayesian optimization algorithm, and the parameter boundary C is set n ∈[10,200],γ∈[0,0.1],λ0∈[0.001,0.1],k∈[0.1,0.5].

[0108] (3) Modeling results of the greenhouse growth environment model of golden ear

[0109] The results of modeling the growth environment of golden ear during the opening, differentiation, color change and harvesting periods are as follows: Figure 3-Figure 6 As shown in the figure, it can be seen that the established golden ear greenhouse growth environment model based on LSTM neural network can well fit the actual greenhouse environmental variable changes, indicating the effectiveness of this method. In order to express the accuracy of the golden ear greenhouse growth environment modeling method based on LSTM neural network, the following mean absolute percentage error (MAPE) is used as the evaluation criterion:

[0110]

[0111] In the above example, the average absolute percentage error of greenhouse environmental variables during the golden ear opening period is: temperature 0.4584%, humidity 0.3892%, carbon dioxide concentration 0.7889%; the average absolute percentage error of greenhouse environmental variables during the golden ear differentiation period is: temperature 0.2680%, humidity 0.2944%, carbon dioxide concentration 0.7913%; the average absolute percentage error of greenhouse environmental variables during the golden ear color change period is: temperature 0.3775%, humidity 0.2646%, carbon dioxide concentration 0.7048%; the average absolute percentage error of greenhouse environmental variables during the golden ear harvest period is: temperature 0.3484%, humidity 0.2417%, carbon dioxide concentration 0.8475%. It can be seen that the golden ear greenhouse growth environment modeling method based on LSTM neural network has high accuracy and meets the accuracy requirements.

[0112] Example 2

[0113] This embodiment 2 provides a prediction system for a golden ear greenhouse growth environment model based on an LSTM neural network, including:

[0114] Data acquisition module: using smart sensors in the greenhouse to collect data on the growth environment variables of golden ear mushrooms;

[0115] Data preprocessing module: preprocess the collected environmental variable data and normalize them;

[0116] Data prediction module: input the normalized data into the optimal golden ear greenhouse growth environment model of the corresponding growth stage to obtain the environmental variables at the future n+1th moment. The optimal golden ear greenhouse growth environment model is obtained by the training method of the golden ear greenhouse growth environment model based on the LSTM neural network.

[0117] Example 3

[0118] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a method for predicting a golden ear greenhouse growth environment model when executing the program.

[0119] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principles of the present invention shall be equivalent replacement modes and shall be included in the protection scope of the present invention.

Claims

1. A training method for a golden ear greenhouse growth environment model based on an LSTM neural network, characterized in that: include: Collect the original growth environment variable data of golden ear in the greenhouse; After preprocessing the original growth environment variable data of golden ear, the data set was divided according to the growth stage of golden ear; The modulated Adam optimizer was used to train the greenhouse growth environment model of golden ear based on LSTM neural network; Based on the Bayesian optimization algorithm, the hyperparameters of the golden ear greenhouse growth environment model were optimized to determine the optimal golden ear greenhouse growth environment model for each growth stage of the golden ear.

2. The training method according to claim 1, characterized in that: The growth stages of golden ear mushrooms include opening stage, differentiation stage, color change stage and harvesting stage. The data set is divided according to the growth stages, specifically as follows: The environmental variables during the growth of golden ear were collected at equal time intervals to obtain the environmental variables at the first n moments and the environmental variables at the n+1th moment, and the corresponding data sets were divided according to the growth stages.

3. The training method according to claim 1, characterized in that: The golden ear greenhouse growth environment model includes four layers, namely, input layer, LSTM layer, fully connected layer and output layer. The input of the golden ear greenhouse growth environment module is the environmental variables of the previous n moments, and the output is the environmental variables of the future n+1th moment.

4. The training method according to any one of claims 1 to 3, characterized in that: The method of using the modulated Adam optimizer to train the golden ear greenhouse growth environment model based on the LSTM neural network is as follows: Initialize parameters η1, η2, initialize time step t = 0, initial parameter vector θ0, first-order momentum M0 = 0, second-order momentum P0 = 0, initialize modulation parameters: initial value λ0 of decay learning rate, learning rate reduction factor k, learning rate reduction interval Q and weight decay rate γ; Calculate the first-order momentum M at the current time t t and the second-order momentum P t : Calculate the deviation correction of the first-order momentum and second-order momentum at the current time t Update the decay learning rate λ at the current time t t ; Update the parameter vector θ at the current time t t ; When |θ t -θ t-1 |≤ε or the maximum number of iterations is reached, the optimization process ends, and ε is the minimum parameter change threshold.

5. The training method according to any one of claims 1 to 3, characterized in that: Based on the Bayesian optimization algorithm, the hyperparameters of the golden ear greenhouse growth environment model are optimized to determine the optimal golden ear greenhouse growth environment model for each growth stage of the golden ear. Specifically, Determine the hyperparameters of the golden ear greenhouse growth environment model that need to be optimized, and establish the objective function between the hyperparameters and the test set; Define the parameter bounds of the hyperparameters that need to be optimized; The Bayesian optimization algorithm is iterated, and the local optimal solution of the objective function is used as the global optimal solution at each iteration to obtain the hyperparameters of the golden ear greenhouse growth environment model at the global optimal moment; If the current number of iterations reaches the maximum number of iterations, the optimization process is exited and the current global optimal golden ear greenhouse growth environment model hyperparameters are used as the final golden ear greenhouse growth environment model parameters. Otherwise, the number of iterations increases and returns to the parameter boundary step where the hyperparameters that need to be optimized need to be defined.

6. The method for training the golden ear greenhouse growth environment model according to claim 5, characterized in that: The objective function is a root mean square error function.

7. The method for training the golden ear greenhouse growth environment model according to claim 1, characterized in that: The original growth environment variable data of the golden ear include temperature, humidity, light intensity and carbon dioxide concentration.

8. A prediction method for the golden ear greenhouse growth environment model based on LSTM neural network, characterized in that: include: Using smart sensors in the greenhouse to collect data on variables in the golden ear growth environment; Preprocess the collected environmental variable data and normalize them; The normalized data is input into the optimal golden ear greenhouse growth environment model of the corresponding growth stage to obtain the environmental variables at the future n+1th moment, wherein the optimal golden ear greenhouse growth environment model is obtained based on the training method of the golden ear greenhouse growth environment model based on the LSTM neural network described in any one of claims 1-7.

9. A prediction system for the golden ear greenhouse growth environment model based on LSTM neural network, characterized in that: include: Data acquisition module: using smart sensors in the greenhouse to collect data on the growth environment variables of golden ear mushrooms; Data preprocessing module: preprocess the collected environmental variable data and normalize them; Data prediction module: input the normalized data into the optimal golden ear greenhouse growth environment model of the corresponding growth stage to obtain the environmental variables at the future n+1th moment, wherein the optimal golden ear greenhouse growth environment model is obtained based on the training method of the golden ear greenhouse growth environment model based on the LSTM neural network described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the golden ear greenhouse growth environment model prediction method as described in claim 8 is implemented.

Images