A multi-factor integrated Pleurotus geesteranus growth environment control system and use method

Abstract

The present invention relates to the field of electronic digital data processing, and specifically to a multi-factor integrated Pleurotus geesteranus growth environment control system and a use method thereof. The system includes a storage module, a Pleurotus geesteranus multi-factor data acquisition module, a Pleurotus geesteranus multi-factor data feature extraction module, a Pleurotus geesteranus multi-factor data feature dimension reduction module, a Pleurotus geesteranus growth environment control prediction module, and an execution module. The existing Pleurotus geesteranus growth environment control system has the problems of being unable to effectively integrate multi-factor data, unable to perform data mining on multi-factor data, and insufficiently accurate judgment of multi-factor data. The multi-factor integrated Pleurotus geesteranus growth environment control system and a use method thereof provided by the present invention can effectively integrate multi-factor data, perform data mining on multi-factor data, and relatively accurately judge multi-factor data.

Description

Technical Field

[0001] The present invention relates to the field of electronic digital data processing, and in particular to a multi-factor integrated Pleurotus geesteranus growth environment control system and a use method thereof. Background Art

[0002] Pleurotus geesteranus is a fungus of the genus Pleurotus in the family Pleurotus. It has a delicious taste and is rich in nutrients. It is an important edible fungus that is widely cultivated. The growth environment of Pleurotus geesteranus has a decisive influence on its yield and quality, so precise growth environment regulation is usually required during its growth. Traditional methods of regulating the growth environment of Pleurotus geesteranus rely on experience-based regulation. After manually reading temperature, humidity and other data, corresponding heating, humidification and other operations are performed. There is a lack of systematic and automated control methods.

[0003] Afterwards, people developed the existing Pleurotus geesteranus growth environment control system. The main working process of the existing Pleurotus geesteranus growth environment control system is: collecting the growth environment data of Pleurotus geesteranus, processing the data, judging the processed data and obtaining specific system operation instructions, executing the system operation instructions to adjust the growth environment of Pleurotus geesteranus, and being able to systematically and automatically control the growth environment of Pleurotus geesteranus. However, the existing Pleurotus geesteranus growth environment control system mainly has the following three problems:

[0004] (1) The existing Pleurotus geesteranus growth environment control system cannot effectively integrate multi-factor data. The collection and processing of environmental data by the existing Pleurotus geesteranus growth environment control system is often limited to a single or a small number of environmental factors. However, the growth environment data of Pleurotus geesteranus is multi-factor data, because it is not only affected by a single or a small number of environmental factors, but also by multiple environmental factors such as temperature, humidity, carbon dioxide concentration, and oxygen concentration. Therefore, the existing Pleurotus geesteranus growth environment control system has a low level of intelligence, and it is difficult to accurately control the growth environment in real time, which affects the growth effect of Pleurotus geesteranus.

[0005] (2) The existing Pleurotus geesteranus growth environment control system is unable to perform data mining on multi-factor data. As the number of influencing factors involved in the environmental control system increases, the growth environment data collected will have multi-dimensional and multi-type characteristics. The existing feature extraction and feature dimensionality reduction methods show obvious limitations when dealing with these high-dimensional and complex data, which can easily lead to the loss of growth environment data during the processing process or fail to fully retain the inherent characteristics of the growth environment data.

[0006] (3) The existing Pleurotus geesteranus growth environment control system is not accurate enough in judging multi-factor data. The existing Pleurotus geesteranus growth environment control system mainly uses a classifier to classify and process the growth environment data and annotate the data with specific system operation instructions. However, the existing Pleurotus geesteranus growth environment control system has insufficient classification and processing capabilities under the influence of multiple factors, making it difficult to output operation instructions efficiently and accurately.

[0007] Therefore, the existing Pleurotus geesteranus growth environment control system has the problems of being unable to effectively integrate multi-factor data, unable to perform data mining on multi-factor data, and the inaccurate judgment of multi-factor data. Summary of the invention

[0008] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a Pleurotus geesteranus growth environment regulation system that can effectively integrate multi-factor data, perform data mining on multi-factor data and make relatively accurate judgments on multi-factor data, as well as a method for using the Pleurotus geesteranus growth environment regulation system that integrates multi-factors.

[0009] In order to solve the above technical problems, the present invention provides a multi-factor integrated Pleurotus geesteranus growth environment control system, including a storage module for storing data;

[0010] The multi-factor data acquisition module of Pleurotus geesteranus is used to collect the growth environment data of Pleurotus geesteranus and send it to the storage module for storage. The growth environment data includes environmental sensor data, system operation data, growth feedback data and external influence data. The growth environment data includes multiple attributes.

[0011] The Pleurotus geesteranus multi-factor data feature extraction module is used to perform feature mining on the growth environment data with the number of attributes less than or equal to 50 collected by the Pleurotus geesteranus multi-factor data collection module to obtain feature-mined data; the Pleurotus geesteranus multi-factor data feature extraction module includes a feature extraction model of a neural network based on adaptive biological influence optimization;

[0012] The Pleurotus geesteranus multi-factor data feature dimensionality reduction module is used to perform feature mining on the growth environment data with the number of data attributes collected by the Pleurotus geesteranus multi-factor data collection module being greater than 50 to obtain feature-mined data; the Pleurotus geesteranus multi-factor data feature dimensionality reduction module includes a feature dimensionality reduction model based on an autoencoder algorithm combined with a Riemann geometry optimization algorithm;

[0013] The Pleurotus geesteranus growth environment regulation prediction module is used to generate corresponding system operation instructions according to the data after feature mining; the Pleurotus geesteranus growth environment regulation prediction module includes a classifier based on the nuclear extreme learning machine classification algorithm optimized by quantum Hamilton;

[0014] The execution module is used to control the growth environment parameters of Pleurotus geesteranus and receive and execute system operation instructions to adjust the growth environment parameters.

[0015] The present invention provides a method for using a Pleurotus geesteranus growth environment regulation system based on the above-mentioned Pleurotus geesteranus growth environment regulation system integrating multiple factors, including:

[0016] S1. Collect historical growth environment data of Pleurotus geesteranus, where the growth environment data includes environmental sensor data, system operation data, growth feedback data, and external influence data, and each of the environmental sensor data, system operation data, growth feedback data, and external influence data includes multiple attributes;

[0017] S2. Manually label the historical growth environment data and store them in a storage module, where the labeling category is system operation instructions;

[0018] S3, inputting the historical growth environment data with the number of attributes less than or equal to 50 into the Pleurotus geesteranus multi-factor data feature extraction module, training the feature extraction model of the Pleurotus geesteranus multi-factor data feature extraction module based on the neural network of adaptive biological impact optimization, obtaining the output data of the neural network, and storing it in the storage module;

[0019] S4, inputting the growth environment data with the number of attributes greater than 50 into the Pleurotus geesteranus multi-factor data feature dimensionality reduction module, training the feature dimensionality reduction model of the Pleurotus geesteranus multi-factor data feature dimensionality reduction module based on the autoencoder algorithm combined with the Riemann geometry optimization algorithm, obtaining the output data of the autoencoder, and storing it in the storage module;

[0020] S5, inputting the output data of the neural network and the output data of the autoencoder into the Pleurotus geesteranus growth environment regulation prediction module, and training the classifier of the Pleurotus geesteranus growth environment regulation prediction module based on the quantum Hamiltonian optimized nuclear extreme learning machine classification algorithm;

[0021] S6, collecting new growth environment data through the Pleurotus geesteranus multi-factor data collection module and storing it in the storage module;

[0022] S7, determining the number of attributes of the new growth environment data; when the number of attributes is less than or equal to 50, inputting the new growth environment data into the Pleurotus geesteranus multi-factor data feature extraction module for feature mining; when the number of attributes is greater than 50, inputting the new growth environment data into the Pleurotus geesteranus multi-factor data feature dimension reduction module for feature mining; after completing the feature mining of the new growth environment data, obtaining the data after feature mining;

[0023] S8. Input the feature-mined data into the Pleurotus geesteranus growth environment regulation and prediction module for marking, obtain system operation instructions corresponding to the feature-mined data, send the system operation instructions to the execution module, and the execution module adjusts the growth environment parameters of Pleurotus geesteranus according to the system operation instructions.

[0024] As a further improvement of the present invention: In S1:

[0025] Environmental sensor data includes three attributes: temperature ,humidity and carbon dioxide concentration ;

[0026] System operation data includes four attributes: Geothermal heating , water cooling , hydration and humidification and oxygenation The operational status of the system;

[0027] Growth feedback data includes four attributes: mycelium growth rate , fruiting body differentiation speed ,Yield Data and quality data;

[0028] External impact data includes three attributes: seasonal changes , outside air temperature and humidity .

[0029] As a further improvement of the present invention: the system operation instructions in S2 and S8 include heating on, heating off, cooling on, cooling off, humidification on, humidification off, ventilation on, ventilation off and unspecified specific operations.

[0030] As a further improvement of the present invention: the feature extraction model of the neural network based on adaptive biological influence optimization for training the Pleurotus geesteranus multi-factor data feature extraction module in S3 includes:

[0031] S301, initialization setting of neural network parameters, including neural network weights Bias with Neural Networks , the initialization method is expressed as:

[0032] (1)

[0033] (2)

[0034] In formula (1) and formula (2), is the initial weight of the neural network, is a nonlinear transformation function, is the initialization random seed parameter, is the input dimension of the neural network, is the initial bias of the neural network;

[0035] S302. According to the characteristics and objectives of the training data, a dynamically changing environmental factor is simulated. The environmental factor affects the adaptability requirements of the neural network parameters, and the environmental factor is dynamically generated based on the statistical characteristics of the training data characteristics. The simulation is performed by evaluating the coefficient of variation and skewness of the input characteristics, which is expressed as:

[0036] (5)

[0037] (6)

[0038] (7)

[0039] In formula (5), formula (6) and formula (7), For environmental factors, is the first hyperparameter to adjust the sensitivity of the environmental simulation. is the coefficient of variation calculation function, is the feature input to the neural network, is the second hyperparameter that adjusts the sensitivity of the environmental simulation. is the skewness calculation function, Features The standard deviation of Features The mean of

[0040] S303, define the initial fitness evaluation function, which considers the prediction accuracy and parameter adaptability of the model to ensure that the network performance is evaluated from multiple dimensions. The calculation method is:

[0041] (8)

[0042] (9)

[0043] In formula (8) and formula (9), is the fitness function of the neural network, is the weight of the neural network, is the bias of the neural network, For environmental factors, is the cross entropy loss function, is the regularization coefficient of the neural network, for norm, is the weight change of the neural network, is the fitness adaptive regularization factor, is the jth weight of the neural network, is the change of the neural network in the previous iteration;

[0044] S304, the weight and bias of the neural network are adjusted based on the dynamic adjustment factor, and the adjustment method is expressed as:

[0045] (12)

[0046] (13)

[0047] (14)

[0048] In formula (12), formula (13) and formula (14), is the nonlinear adjustment factor parameter, is a hyperparameter that adjusts the intensity of the response, is a nonlinear response function, is the fitness function of the neural network, is the weight of the neural network, is the bias of the neural network, For environmental factors, is random noise, is the parameter update operation, is the learning rate of the neural network, is the gradient of the neural network’s fitness function with respect to the weight parameters, is the gradient of the fitness function of the neural network with respect to the bias parameter;

[0049] S305, repeating S302-S304 until the preset stop iteration condition is met, which means that the model training is completed.

[0050] As a further improvement of the present invention: In S303, the fitness adaptive regularization factor is calculated based on the activity of the feature, which is expressed as:

[0051] (11)

[0052] In formula (11), is the fitness adaptive regularization factor, is a hyperparameter that adjusts the strength of the regularization baseline, is the activity calculation function of the feature, is the weight of the neural network.

[0053] As a further improvement of the present invention: the feature dimensionality reduction model based on the autoencoder algorithm combined with the Riemann geometry optimization algorithm for training the Pleurotus geesteranus multi-factor data feature dimensionality reduction module in S4 includes:

[0054] S401, define the structure of the autoencoder, the input layer of the autoencoder receives multi-dimensional feature data, the number of neurons in the output layer of the autoencoder is 50, and define the transformation method of the input to each layer as:

[0055] (16)

[0056] In formula (16), For the autoencoder The linear transformation output of the layer; For the autoencoder The weight matrix of the layer; For the autoencoder The output of the layer, i.e., the output of the autoencoder The input of the layer; For the autoencoder The bias vector of the layer;

[0057] S402, data is transmitted through each hidden layer of the autoencoder, and the data flow path is adjusted in each hidden layer using the Riemann geometry optimization algorithm to ensure that the data flow conforms to the properties of its characteristic intrinsic distribution, expressed as:

[0058] (17)

[0059] In formula (17), After applying the activation function to the autoencoder Layer output, For the autoencoder The linear transformation output of the layer; yes of Norm, representation The module length; To adjust the Riemannian geometry parameters of layer activation strengths; is the hyperbolic tangent function;

[0060] S403, use reconstruction loss and Riemann distance to jointly evaluate the difference between the output data and the original data to ensure that the essential characteristics of the data are preserved during the dimensionality reduction process. The loss function of the autoencoder The calculation method is:

[0061] (20)

[0062] In formula (20), mrc is the number of samples input into the autoencoder in the current batch, is the data vector of the i-th training sample, is the reconstructed output of the i-th autoencoder, To control the regularization coefficient of the Riemann distance, is the Riemann distance, is the noise penalty loss;

[0063] S404, the gradient of the loss function to each parameter is calculated by the back propagation algorithm, and the parameters of each layer of the autoencoder are updated. The back propagation uses the chain rule to calculate the gradient, and the way to update the autoencoder weights and biases is expressed as:

[0064] (25)

[0065] (26)

[0066] In formula (25) and formula (26), is the update amount of the autoencoder weight, is the learning rate of the autoencoder, is the symbol of partial derivative, is the partial derivative of the loss function of the autoencoder with respect to the autoencoder weights, is the weight decay rate of the autoencoder, is the update amount of the autoencoder bias, is the partial derivative of the autoencoder loss function with respect to the autoencoder bias;

[0067] S405. Repeat iterations S402 to S404 until a preset stop iteration condition is met, indicating that the model training is completed.

[0068] As a further improvement of the present invention: of Norm The calculation method is:

[0069] (18)

[0070] In formula (18), for The dimension of the vector, express The i-th component of the vector.

[0071] As a further improvement of the present invention: the partial derivative of the loss function of the autoencoder with respect to the weight in S404 The calculation method is:

[0072] (27)

[0073] In formula (27), mrc is the number of samples input into the autoencoder in the current batch, After applying the activation function to the autoencoder Layer output, is the data vector of the i-th training sample, yes The transpose of For the autoencoder The output of the layer, i.e., the output of the autoencoder The input of the layer; is the regularization parameter of the autoencoder, is the partial derivative of the Riemann distance with respect to the weights of the autoencoder.

[0074] As a further improvement of the present invention: the classifier of the quantum Hamiltonian optimized nuclear extreme learning machine classification algorithm for training the Pleurotus geesteranus growth environment control prediction module in S5 includes:

[0075] S501, initialize the kernel function of the kernel extreme learning machine, and define the calculation method of the kernel function as:

[0076] (29)

[0077] In formula (29), is the kernel function of the kernel extreme learning machine, is the width parameter of the kernel function of the kernel extreme learning machine, Dynamically adjust the function for the kernel function, is the i-th sample input to the kernel extreme learning machine, is the jth sample input to the kernel extreme learning machine;

[0078] S502, mapping the data input to the kernel extreme learning machine to the high-dimensional feature space through the selected kernel function, and performing nonlinear mapping of the data, which is expressed as:

[0079] (31)

[0080] In formula (31), is the high-dimensional feature space mapping function, represents the data input to the kernel extreme learning machine, is the kernel width parameter of the kernel extreme learning machine, is the i-th sample input to the kernel extreme learning machine;

[0081] S503, use the quantum Hamiltonian optimization algorithm to optimize the kernel parameters again to adapt to the specific data distribution, use the probability and parallelism of quantum computing to search the parameter space, find the kernel parameters that best suit the current data distribution, and solve the calculation method for the optimal kernel width parameter and output layer weight as follows:

[0082] (32)

[0083] In formula (32), Optimize the function for the quantum Hamiltonian, is the kernel width parameter of the kernel extreme learning machine, is the output layer weight of the kernel extreme learning machine, To obtain the minimum function, is the target output vector, is the data matrix of the high-dimensional feature space, is the regularization parameter of the kernel extreme learning machine;

[0084] S504. According to the mapped data and target output, the distribution of the data in the kernel space is further adjusted using the manifold mapping strategy, and then the weight of the output layer is calculated using the improved regularized least squares method. The output layer weight of the kernel extreme learning machine is calculated by minimizing the regularized loss function, which is expressed as:

[0085] (33)

[0086] In formula (33), is the output layer weight of the kernel extreme learning machine, express The transpose of is the data matrix of the high-dimensional feature space, is the regularization parameter of the kernel extreme learning machine, I is the identity matrix;

[0087] S505. After the output layer weight calculation of the kernel extreme learning machine is completed, it means that the training of the kernel extreme learning machine is completed.

[0088] The beneficial effects of the present invention are as follows: the Pleurotus geesteranus growth environment control system and use method provided by the present invention can effectively integrate multi-factor data, can perform data mining on multi-factor data, and can make relatively accurate judgments on multi-factor data.

[0089] The growth environment data environmental sensor data, system operation data, growth feedback data and external influence data used in the present invention have multiple data sources and are multi-factor data that integrates multiple environmental factors, rather than relying solely on a single environmental factor, thereby making the acquisition of growth environment data more comprehensive.

[0090] When the number of attributes of the growth environment data is less than or equal to 50, the present invention uses a neural network optimized by adaptive biological influence to extract features from the growth environment data, simulates the adaptive behavior of organisms under environmental pressure, and can self-adjust parameter update strategies according to changes in input data features, avoiding problems such as gradient vanishing and gradient explosion, improving the stability and training efficiency of feature extraction, and dynamically adjusting the weights and biases of the neural network, thereby achieving efficient feature extraction and model stability. When the number of attributes of the growth environment data is greater than 50, the high-dimensional data processing problem is solved by the feature dimension reduction module, and the feature dimension reduction is effectively reduced by using the autoencoder combined with the Riemann geometry optimization algorithm to effectively reduce the dimension of the high-dimensional feature. At the same time, the weights and biases are adjusted by Riemann geometry, so that the data after dimension reduction maintains the intrinsic geometric structure, and the generalization ability and dimension reduction accuracy of the system in the high-dimensional data environment are improved. The autoencoder maintains the intrinsic geometric structure of the data through the Riemann geometry optimization algorithm, ensures that the data after dimension reduction still has the key features of the original data, effectively reduces redundant information, and makes the model more accurate and efficient when processing high-dimensional data. Thus, the present invention can perform data mining on multi-factor data.

[0091] The Pleurotus geesteranus growth environment control prediction module of the present invention classifies the data after feature mining through the kernel extreme learning machine classification algorithm based on quantum Hamiltonian optimization, and adjusts the kernel function parameters and the output layer weight through the quantum Hamiltonian optimization strategy, so that the global optimal solution can be found in the complex parameter space, and the classification accuracy of the model when processing complex growth environment control data is improved. The global optimal solution is searched in the complex parameter space through the quantum Hamiltonian optimization strategy, and the decision accuracy of the classifier is improved, so that the system can quickly and accurately output the system operation instructions for the growth environment, and the real-time and accuracy of the Pleurotus geesteranus growth environment control is guaranteed. Therefore, the judgment of multi-factor data is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 This is a principle block diagram of the Pleurotus geesteranus growth environment control system integrating multiple factors provided by the present invention. DETAILED DESCRIPTION

[0093] The specific implementation modes of the present invention will be further described in detail below in conjunction with the accompanying drawings.

[0094] like Figure 1 As shown, a multi-factor integrated Pleurotus geesteranus growth environment control system provided by the present invention includes a storage module for storing data;

[0095] The multi-factor data acquisition module of Pleurotus geesteranus is used to collect the growth environment data of Pleurotus geesteranus and send it to the storage module for storage. The growth environment data includes environmental sensor data, system operation data, growth feedback data and external influence data. The growth environment data includes multiple attributes.

[0096] The Pleurotus geesteranus multi-factor data feature extraction module is used to perform feature mining on the growth environment data with the number of attributes less than or equal to 50 collected by the Pleurotus geesteranus multi-factor data collection module to obtain feature-mined data; the Pleurotus geesteranus multi-factor data feature extraction module includes a feature extraction model of a neural network based on adaptive biological influence optimization;

[0097] The Pleurotus geesteranus multi-factor data feature dimensionality reduction module is used to perform feature mining on the growth environment data with the number of data attributes collected by the Pleurotus geesteranus multi-factor data collection module being greater than 50 to obtain feature-mined data; the Pleurotus geesteranus multi-factor data feature dimensionality reduction module includes a feature dimensionality reduction model based on an autoencoder algorithm combined with a Riemann geometry optimization algorithm;

[0098] The Pleurotus geesteranus growth environment regulation prediction module is used to generate corresponding system operation instructions according to the data after feature mining; the Pleurotus geesteranus growth environment regulation prediction module includes a classifier based on the nuclear extreme learning machine classification algorithm optimized by quantum Hamilton;

[0099] The execution module is used to control the growth environment parameters of Pleurotus geesteranus and receive and execute system operation instructions to adjust the growth environment parameters.

[0100] The present invention provides a method for using a multi-factor integrated Pleurotus geesteranus growth environment control system, including:

[0101] Part 1 (S1~S5): Training the Growth Environment Control System of Pleurotus geesteranus

[0102] S1. Collect historical growth environment data of Pleurotus geesteranus. The growth environment data includes environmental sensor data, system operation data, growth feedback data and external influence data. The environmental sensor data, system operation data, growth feedback data and external influence data all include multiple attributes; therefore, the growth environment data includes multiple attributes.

[0103] Environmental sensor data is data collected in real time by sensors and includes three attributes: temperature ,humidity and carbon dioxide concentration ;

[0104] System operation data includes four attributes: Geothermal heating , water cooling , hydration and humidification and oxygenation The operational status of the system;

[0105] Growth feedback data includes four attributes: mycelium growth rate , fruiting body differentiation speed ,Yield Data and quality data;

[0106] External impact data includes three attributes: seasonal changes , outside air temperature and humidity .

[0107] S2. Manually label the historical growth environment data and store them in a storage module, where the labeling category is system operation instructions;

[0108] S3, inputting the historical growth environment data with the number of attributes less than or equal to 50 into the Pleurotus geesteranus multi-factor data feature extraction module, training the feature extraction model of the Pleurotus geesteranus multi-factor data feature extraction module based on the neural network of adaptive biological impact optimization, obtaining the output data of the neural network, and storing it in the storage module;

[0109] The present invention adopts a neural network feature extraction model based on adaptive biological impact optimization. Inspired by the adaptive behavior of the biological world and the self-regulation mechanism of the ecosystem, the adaptive biological impact optimization strategy imitates the adaptive changes of organisms under environmental pressure, specifically the regulatory mechanism of gene expression and the balance of population dynamics, to guide the optimization process of neural network weights and biases, so that the neural network can automatically adjust its parameter update strategy according to the changes in training data. The feature extraction model of the neural network based on adaptive biological impact optimization for training the multi-factor data feature extraction module of Pleurotus geesteranus includes:

[0110] S301, initialization setting of neural network parameters, including neural network weights Bias with Neural Networks , the initialization method is expressed as:

[0111] (1)

[0112] (2)

[0113] In formula (1) and formula (2), is the initial weight of the neural network, is a nonlinear transformation function, is the initialization random seed parameter, is the input dimension of the neural network, is the initial bias of the neural network;

[0114] The nonlinear transformation function is adaptively adjusted based on the input dimension and is calculated as:

[0115] (3)

[0116] (4)

[0117] In formula (3) and formula (4), is a nonlinear transformation function, is the initialization random seed parameter, is the input dimension of the neural network, is the weight parameter of the nonlinear transformation function, Indicates random generation The value within the interval.

[0118] S302. According to the characteristics and objectives of the training data, a dynamically changing environmental factor is simulated. The environmental factor affects the adaptability requirements of the neural network parameters, and the environmental factor is dynamically generated based on the statistical characteristics of the training data characteristics. The simulation is performed by evaluating the coefficient of variation and skewness of the input characteristics, which is expressed as:

[0119] (5)

[0120] (6)

[0121] (7)

[0122] In formula (5), formula (6) and formula (7), For environmental factors; is the first hyperparameter to adjust the sensitivity of the environmental simulation. Set to 0.5; is the coefficient of variation calculation function, is the feature input to the neural network; is the second hyperparameter that adjusts the sensitivity of the environmental simulation. Set to 0.3; is the skewness calculation function, Features The standard deviation of Features The mean of

[0123] S303, define the initial fitness evaluation function, which considers the prediction accuracy and parameter adaptability of the model to ensure that the network performance is evaluated from multiple dimensions. The calculation method is:

[0124] (8)

[0125] (9)

[0126] In formula (8) and formula (9), is the fitness function of the neural network, is the weight of the neural network, is the bias of the neural network, For environmental factors, is the cross entropy loss function; is the regularization coefficient of the neural network, Set to 0.05; for norm, is the weight change of the neural network, is the fitness adaptive regularization factor, is the jth weight of the neural network, is the change of the neural network in the previous iteration;

[0127] The fitness adaptive regularization factor automatically adjusts the coefficient of the regularization term by analyzing the scale sensitivity of the weight, thereby enhancing the model's adaptability to features of different scales. The scale sensitivity index of the weight is defined to measure the activity of features of different scales in the weight. The calculation method is:

[0128] (10)

[0129] In formula (10), is the activity calculation function of the feature, is the weight of the neural network, is the total number of scales considered, obtained through the coefficients of the wavelet transform, is the weight The component at the kth scale, is the kth scale.

[0130] The fitness adaptive regularization factor is calculated based on the activity of the feature, expressed as:

[0131] (11)

[0132] In formula (11), is the fitness adaptive regularization factor; is a hyperparameter that adjusts the strength of the regularization baseline, Set to 0.2; is the activity calculation function of the feature, is the weight of the neural network.

[0133] S304, the weight and bias of the neural network are adjusted based on the dynamic adjustment factor, and the adjustment method is expressed as:

[0134] (12)

[0135] (13)

[0136] (14)

[0137] In formula (12), formula (13) and formula (14), is the nonlinear adjustment factor parameter, is a hyperparameter that adjusts the intensity of the response, is a nonlinear response function, is the fitness function of the neural network, is the weight of the neural network, is the bias of the neural network, For environmental factors, is random noise, It is the parameter update operation; is the learning rate of the neural network, Set to 0.01; is the gradient of the neural network’s fitness function with respect to the weight parameters, is the gradient of the fitness function of the neural network with respect to the bias parameter;

[0138] Assume that the input of the nonlinear response function is , calculated as:

[0139] (15)

[0140] In formula (15), is a nonlinear response function, is the input of the nonlinear response function.

[0141] S305, repeating iterations S302 to S304 until a preset stop iteration condition is met, indicating that the model training is completed. The preset stop iteration condition is reaching a preset maximum number of iterations, and the preset maximum number of iterations is set to 1000 times.

[0142] S4, inputting the growth environment data with the number of attributes greater than 50 into the Pleurotus geesteranus multi-factor data feature dimensionality reduction module, training the feature dimensionality reduction model of the Pleurotus geesteranus multi-factor data feature dimensionality reduction module based on the autoencoder algorithm combined with the Riemann geometry optimization algorithm, obtaining the output data of the autoencoder, and storing it in the storage module;

[0143] The feature dimensionality reduction model based on the autoencoder algorithm combined with the Riemann geometry optimization algorithm for training the multi-factor data feature dimensionality reduction module of Pleurotus geesteranus in S4 includes:

[0144] S401, define the structure of the autoencoder, the input layer of the autoencoder receives multi-dimensional feature data, and the number of neurons in the output layer of the autoencoder is 50, that is, after the data with a dimension higher than 50 is input into the autoencoder, the feature dimension obtained by dimensionality reduction is unified to 50. The hidden layer of the autoencoder adjusts the weights and biases to adapt to the intrinsic structure of the data, and defines the transformation method of the input to each layer as:

[0145] (16)

[0146] In formula (16), For the autoencoder The linear transformation output of the layer; For the autoencoder The weight matrix of the layer; For the autoencoder The output of the layer, i.e., the output of the autoencoder The input of the layer; For the autoencoder The bias vector of the layer;

[0147] S402, data is transmitted through each hidden layer of the autoencoder, and the data flow path is adjusted in each hidden layer using the Riemann geometry optimization algorithm to ensure that the data flow conforms to the properties of its characteristic intrinsic distribution. The input of the activation function is the ratio of the modulus of the data to the Riemann geometry parameter, which is processed by the hyperbolic tangent function, and the output is a vector that is consistent with the direction of the data but normalized, considering not only the modulus of the data but also the direction of the data, so that the model can more accurately capture and express the intrinsic geometric structure of the data. The calculation method is expressed as:

[0148] (17)

[0149] In formula (17), After applying the activation function to the autoencoder Layer output, For the autoencoder The linear transformation output of the layer; yes of Norm, representation The module length; To adjust the Riemannian geometry parameters of layer activation strengths; is the hyperbolic tangent function;

[0150] of Norm In Riemannian geometry, it is used to measure the length of a vector, which helps to understand the relative position of data points in high-dimensional space. The calculation method is:

[0151] (18)

[0152] In formula (18), for The dimension of the vector, express The i-th component of the vector.

[0153] Riemannian geometry parameters are based on The standard deviation and adjustment factor are adjusted to reflect the distribution characteristics of the data. Its function is to adjust the response intensity of the activation function, which helps the model adapt to the inherent structure of the data. The calculation method is:

[0154] (19)

[0155] In formula (19), is the standard deviation function; For the autoencoder The linear transformation output of the layer, yes The standard deviation of is the Riemann geometry adjustment factor, which controls the sensitivity of compression.

[0156] S403, use reconstruction loss and Riemann distance to jointly evaluate the difference between the output data and the original data to ensure that the essential characteristics of the data are preserved during the dimensionality reduction process. The loss function of the autoencoder The calculation method is:

[0157] (20)

[0158] In formula (20), mrc is the number of samples input into the autoencoder in the current batch, is the data vector of the i-th training sample, is the reconstructed output of the i-th autoencoder, To control the regularization coefficient of the Riemann distance, is the Riemann distance, is the noise penalty loss;

[0159] The growth environment data of Pleurotus geesteranus is often affected by various uncontrolled factors, such as equipment errors, environmental interference, etc. These factors may cause noise in the data. The traditional autoencoder may not be able to effectively extract useful features when processing noisy data. The present invention uses the noise penalty loss to improve the robustness of the model. The calculation method is:

[0160] (twenty one)

[0161] In formula (21), is the noise penalty loss, is the output of the autoencoder, is the input of the autoencoder, To punish noise, is the regularization coefficient to control the influence of Riemann distance;

[0162] Penalty noise simulates the random interference that may be encountered in actual operation and is calculated as:

[0163] (twenty two)

[0164] In formula (22), is the noise level that is dynamically adjusted based on the data, represents a normal distribution with a mean of 0 and a variance of the unit matrix; ~ represents compliance with a specific distribution.

[0165] The noise level is automatically determined based on the local variance of the autoencoder’s input data, ensuring that the addition of noise does not completely cover the true characteristics of the data, expressed as:

[0166] (twenty three)

[0167] In formula (23), is the noise level that is dynamically adjusted based on the data, is the coefficient that controls the noise intensity, Set to 0.01; is the variance of the input data; is the variance function.

[0168] Riemann distance is used to measure the distance between two data points on a manifold. It can capture the intrinsic geometric structure of the manifold and better reflect the true similarity of the data than the traditional Euclidean distance. The calculation method is:

[0169] (twenty four)

[0170] In formula (24), is the Riemann distance, is the data vector of the training sample, is the reconstructed output of the autoencoder, is the inverse hyperbolic cosine function, which is used to calculate the Riemann distance between two data points.

[0171] S404, the gradient of the loss function to each parameter is calculated by the back propagation algorithm, and the parameters of each layer of the autoencoder are updated. The back propagation uses the chain rule to calculate the gradient, and the way to update the autoencoder weights and biases is expressed as:

[0172] (25)

[0173] (26)

[0174] In formula (25) and formula (26), is the update amount of the autoencoder weight, is the learning rate of the autoencoder, is the symbol of partial derivative, is the partial derivative of the loss function of the autoencoder with respect to the autoencoder weights, is the weight decay rate of the autoencoder, used to prevent overfitting, is the update amount of the autoencoder bias, is the partial derivative of the autoencoder loss function with respect to the autoencoder bias;

[0175] Partial derivative of the loss function of the autoencoder with respect to the weights The calculation method is:

[0176] (27)

[0177] In formula (27), mrc is the number of samples input into the autoencoder in the current batch, After applying the activation function to the autoencoder Layer output, is the data vector of the i-th training sample, yes The transpose of For the autoencoder The output of the layer, i.e., the output of the autoencoder The input of the layer; is the regularization parameter of the autoencoder, Set to 0.3; is the partial derivative of the Riemann distance with respect to the weights of the autoencoder.

[0178] To adapt to different data characteristics, dynamically adjust the regularization parameters of the autoencoder , expressed as:

[0179] (28)

[0180] In formula (28), is the regularization parameter of the initial autoencoder, Set to 0.5; yes The variance of is the variance scaling parameter, Set to 0.1.

[0181] S405, repeating iterations S402 to S404 until a preset stop iteration condition is met, indicating that the model training is completed. The preset stop iteration condition is reaching a preset maximum number of iterations, and the preset maximum number of iterations is set to 1000 times.

[0182] S5, inputting the output data of the neural network and the output data of the autoencoder into the Pleurotus geesteranus growth environment regulation prediction module, and training the classifier of the Pleurotus geesteranus growth environment regulation prediction module based on the quantum Hamiltonian optimized nuclear extreme learning machine classification algorithm;

[0183] The classifier of the quantum Hamiltonian optimized nuclear extreme learning machine classification algorithm for training the Pleurotus geesteranus growth environment regulation prediction module includes:

[0184] S501, initialize the kernel function of the kernel extreme learning machine and define the kernel function The calculation method is:

[0185] (29)

[0186] In formula (29), is the kernel function of the kernel extreme learning machine; is the width parameter of the kernel function of the kernel extreme learning machine, Set to 0.3; Dynamically adjust the function for the kernel function, is the i-th sample input to the kernel extreme learning machine, is the jth sample input to the kernel extreme learning machine;

[0187] The kernel function dynamic adjustment function automatically adjusts the shape and parameters of the kernel function according to the changes in data distribution during the kernel extreme learning machine training process. The kernel function dynamic adjustment function is adjusted according to the distribution density of data points and the width parameter of the kernel function of the kernel extreme learning machine. Suppose the input of the kernel function dynamic adjustment function is , then the kernel function dynamically adjusts the function The calculation method is:

[0188] (30)

[0189] In formula (30), Dynamically adjust functions for kernel functions; is the adjusted sensitivity coefficient, Set to 0.1; Indicates at point The data density at It is the maximum data density.

[0190] S502, mapping the data input to the kernel extreme learning machine to the high-dimensional feature space through the selected kernel function, and performing nonlinear mapping of the data, which is expressed as:

[0191] (31)

[0192] In formula (31), is the high-dimensional feature space mapping function, represents the data input to the kernel extreme learning machine, is the kernel width parameter of the kernel extreme learning machine, is the i-th sample input to the kernel extreme learning machine;

[0193] S503. Use the quantum Hamiltonian optimization algorithm to optimize the kernel parameters again to adapt to the specific data distribution. Use the probability and parallelism of quantum computing to search the parameter space and find the kernel parameters that best suit the current data distribution. In the quantum Hamiltonian optimization strategy, the goal of quantum Hamiltonian optimization is to minimize a specific energy function and solve the optimal kernel width parameter and output layer weight. The calculation method is:

[0194] (32)

[0195] In formula (32), The quantum Hamiltonian optimization function is a quantum Hamiltonian optimization algorithm that explores the parameter space by simulating the Hamiltonian of a quantum system. In quantum computing, the Hamiltonian describes the total energy of the system, and the quantum Hamiltonian optimization algorithm uses the evolution of the quantum state to search for the configuration with the lowest energy, which corresponds to the process of finding the minimum value of the loss function in the above calculation. is the kernel width parameter of the kernel extreme learning machine, is the output layer weight of the kernel extreme learning machine, To obtain the minimum function, is the target output vector, It is a data matrix in high-dimensional feature space; is the regularization parameter of the kernel extreme learning machine, Set to 0.2;

[0196] S504: According to the mapped data and target output, the distribution of data in the kernel space is further adjusted using the manifold mapping strategy, and then the weight of the output layer is calculated using the improved regularized least squares method. The output layer weight of the kernel extreme learning machine is calculated by minimizing the regularized loss function. , expressed as:

[0197] (33)

[0198] In formula (33), is the output layer weight of the kernel extreme learning machine, express The transpose of is the data matrix of the high-dimensional feature space, is the regularization parameter of the kernel extreme learning machine, I is the identity matrix;

[0199] Data matrix in high-dimensional feature space The calculation method is:

[0200] (34)

[0201] In formula (34), is the i-th vector in the data matrix of the high-dimensional feature space, is the first vector in the data matrix of the high-dimensional feature space, is the second vector in the data matrix of the high-dimensional feature space, is the first vectors, is the number of vectors in the data matrix of the high-dimensional feature space.

[0202] The i-th vector in the data matrix of the high-dimensional feature space The calculation method is:

[0203] (35)

[0204] In formula (35), is the i-th sample input to the kernel extreme learning machine, is the kernel width parameter of the kernel extreme learning machine, is the jth sample input to the kernel extreme learning machine.

[0205] S505. After the output layer weight calculation of the kernel extreme learning machine is completed, it means that the training of the kernel extreme learning machine is completed.

[0206] Part 2 (S6~S8): Application of Pleurotus geesteranus growth environment control system

[0207] S6, collecting new growth environment data through the Pleurotus geesteranus multi-factor data collection module and storing it in the storage module;

[0208] S7, determining the number of attributes of the new growth environment data;

[0209] When the number of attributes is less than or equal to 50, the new growth environment data is input into the Pleurotus geesteranus multi-factor data feature extraction module for feature mining. In the prior art, some solutions use neural networks for feature extraction. In some neural network structures, problems such as gradient disappearance, gradient explosion, or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model.

[0210] When the number of attributes is greater than 50, the new growth environment data is input into the Pleurotus geesteranus multi-factor data feature dimension reduction module for feature mining; after the feature mining of the new growth environment data is completed, the feature mined data is obtained; the autoencoder algorithm is used as feature dimension reduction to effectively reduce the dimension of high-dimensional feature data while maintaining the essential characteristics of the data. Through mapping, the autoencoder can be more accurate in processing the intrinsic geometric structure of the data, and has good generalization ability and accuracy for processing complex Pleurotus geesteranus growth environment data.

[0211] Since the collected data attributes may be correlated or uncorrelated, and the collected data attributes may be redundant or invalid, it is necessary to perform feature mining on the collected data.

[0212] S8, input the data after feature mining into the Pleurotus geesteranus growth environment regulation prediction module for annotation, obtain the system operation instruction corresponding to the data after feature mining, send the system operation instruction to the execution module, and the execution module adjusts the growth environment parameters of Pleurotus geesteranus according to the system operation instruction. The classifier of the Pleurotus geesteranus growth environment regulation prediction module is based on the quantum Hamiltonian optimized nuclear extreme learning machine classification algorithm. On the basis of the nuclear extreme learning machine, the quantum Hamiltonian optimization strategy is used to optimize the parameter selection of the kernel function and the weight adjustment of the output layer. The quantum Hamiltonian optimization strategy uses the principles of quantum mechanics to perform optimization calculations, and can find the global optimal solution in the complex parameter space, thereby effectively improving the classification accuracy and the generalization ability of the model.

[0213] In S2 and S6, the data storage format is structured vector data. The data is stored in a matrix form, where each row represents the observation data at a time point, and each column corresponds to a specific data attribute. This is only one data format and type. In actual applications, the attributes of data are usually more than 10 attributes, and the number of data attributes may reach dozens or even hundreds.

[0214] The system operation instructions in the present invention include heating on, heating off, cooling on, cooling off, humidification on, humidification off, ventilation on, ventilation off, and unspecified specific operation, a total of nine categories. When the system operation instruction is "unspecified specific operation", the operator needs to manually adjust the growth environment of Pleurotus geesteranus according to expert experience. The specific content of expert experience is as follows:

[0215] (1) Environmental control methods for Pleurotus geesteranus cultivation (mycelium growth stage)

[0216] During the low temperature period in winter and spring, when the temperature in the mushroom shed is below 20℃, the geothermal heating system is activated. The temperature in the shed is sensed by the temperature sensor in the shed. When the temperature in the shed is lower than the lowest value of the set range, the controller automatically turns on the heating; when it reaches the upper limit of the range, it automatically stops working. Through the controller, the temperature in the shed is automatically maintained in an appropriate range, ensuring that the mycelium of Pleurotus geesteranus grows under the most suitable temperature conditions during the low temperature period in winter and spring.

[0217] During the high temperature period in summer and autumn, when the temperature in the mushroom shed is higher than 28℃, the temperature control pipe network in the water cooling system is activated. When the temperature in the shed is higher than the upper limit of the set interval, the controller automatically starts cooling; it automatically stops working when it reaches the lowest value of the interval; when the temperature in the mushroom shed is higher than 35℃, the refrigeration bud-inducing system is turned on at the same time. When the temperature in the shed is higher than the upper limit of the set interval, the controller automatically starts cooling, and it automatically stops working when it reaches the lowest value of the interval. Through the controller, the temperature in the shed is automatically maintained in an appropriate range to ensure that the mycelium of Pleurotus geesteranus grows under the most suitable temperature conditions during the high temperature period in summer and autumn.

[0218] (2) Environmental control methods for Pleurotus geesteranus fruiting (fruiting body differentiation and growth stage)

[0219] During the fruiting body differentiation stage of Pleurotus geesteranus, the refrigeration and bud-inducing system of the mushroom shed where the fruiting is planned to be arranged is activated. When the temperature in the shed is higher than the upper limit of the set range, the controller automatically starts cooling; when it reaches the lowest value of the range, it automatically stops working. The temperature of the refrigeration range is maintained for 5 hours, and the time from opening to closing of the refrigeration and bud-inducing system is set in the range of 13 to 15 hours depending on the temperature.

[0220] During the growth stage of Pleurotus geesteranus fruiting bodies, temperature, humidity and CO2 concentration are key factors affecting yield and quality. The three factors interact with each other and need to be comprehensively considered and scientifically regulated according to the different needs of the fruiting bodies in different growth stages.

Claims

1. A multi-factor edible mushroom growth environment control system, characterized in that: A storage module is included for storing data; A multi-factor data collection module for Pleurotus geesteranus, used for collecting growth environment data of Pleurotus geesteranus and sending it to the storage module for storage, wherein the growth environment data includes environmental sensor data, system operation data, growth feedback data and external influence data, and the growth environment data includes multiple attributes; A Pleurotus geesteranus multi-factor data feature extraction module is used to perform feature mining on the growth environment data with the number of attributes collected by the Pleurotus geesteranus multi-factor data collection module being less than or equal to 50 to obtain feature-mined data; the Pleurotus geesteranus multi-factor data feature extraction module includes a feature extraction model of a neural network based on adaptive biological influence optimization; The Pleurotus geesteranus multi-factor data feature dimensionality reduction module is used to perform feature mining on the growth environment data whose data attributes are greater than 50 and collected by the Pleurotus geesteranus multi-factor data collection module to obtain feature-mined data; the Pleurotus geesteranus multi-factor data feature dimensionality reduction module includes a feature dimensionality reduction model based on an autoencoder algorithm combined with a Riemann geometry optimization algorithm; Training the feature dimensionality reduction model based on the autoencoder algorithm combined with the Riemannian geometry optimization algorithm includes: Define the structure of the autoencoder. The input layer of the autoencoder receives multi-dimensional feature data. The number of neurons in the output layer of the autoencoder is 50. Define the transformation method of the input to each layer as follows: （16） In formula (16), For the autoencoder The linear transformation output of the layer; For the autoencoder The weight matrix of the layer; For the autoencoder The output of the layer, i.e., the output of the autoencoder The input of the layer; For the autoencoder The bias vector of the layer; The data is passed through each hidden layer of the autoencoder, and the data flow path is adjusted using the Riemann geometry optimization algorithm at each hidden layer to ensure that the data flow conforms to the properties of its characteristic intrinsic distribution, expressed as: （17） In formula (17), After applying the activation function to the autoencoder Layer output, For the autoencoder The linear transformation output of the layer; yes of Norm, representation The module length; To adjust the Riemannian geometry parameters of layer activation strengths; is the hyperbolic tangent function; The reconstruction loss and Riemann distance are used to jointly evaluate the difference between the output data and the original data to ensure that the essential characteristics of the data are preserved during the dimensionality reduction process. The loss function of the autoencoder is The calculation method is: （20） In formula (20), mrc is the number of samples input into the autoencoder in the current batch, is the data vector of the i-th training sample, is the reconstructed output of the i-th autoencoder, To control the regularization coefficient of the Riemann distance, is the Riemann distance, is the noise penalty loss; The back propagation algorithm is used to calculate the gradient of the loss function to each parameter, and the parameters of each layer of the autoencoder are updated. Back propagation uses the chain rule to calculate the gradient and update the autoencoder weights and biases as follows: （25） （26） In formula (25) and formula (26), is the update amount of the autoencoder weight, is the learning rate of the autoencoder, is the symbol of partial derivative, is the partial derivative of the loss function of the autoencoder with respect to the autoencoder weights, is the weight decay rate of the autoencoder, is the update amount of the autoencoder bias, is the partial derivative of the autoencoder loss function with respect to the autoencoder bias; Repeat the iteration until the preset stop iteration condition is met, which means the model training is completed; A Pleurotus geesteranus growth environment control prediction module is used to generate corresponding system operation instructions according to the feature-mined data; the Pleurotus geesteranus growth environment control prediction module includes a classifier based on a quantum Hamiltonian optimized nuclear extreme learning machine classification algorithm; Training the classifier of the quantum Hamiltonian optimized nuclear extreme learning machine classification algorithm includes: Initialize the kernel function of the kernel extreme learning machine and define the calculation method of the kernel function as: （29） In formula (29), is the kernel function of the kernel extreme learning machine, is the width parameter of the kernel function of the kernel extreme learning machine, Dynamically adjust the function for the kernel function, is the i-th sample input to the kernel extreme learning machine, is the jth sample input to the kernel extreme learning machine; The data input to the kernel extreme learning machine is mapped to the high-dimensional feature space through the selected kernel function, and the nonlinear mapping of the data is performed, which is expressed as: （31） In formula (31), is the high-dimensional feature space mapping function, represents the data input to the kernel extreme learning machine, is the kernel width parameter of the kernel extreme learning machine, is the i-th sample input to the kernel extreme learning machine; The quantum Hamiltonian optimization algorithm is used to optimize the kernel parameters again to adapt to the specific data distribution. The probability and parallelism of quantum computing are used to search the parameter space to find the kernel parameters that best suit the current data distribution. The calculation method for solving the optimal kernel width parameter and output layer weight is: （32） In formula (32), Optimize the function for the quantum Hamiltonian, is the kernel width parameter of the kernel extreme learning machine, is the output layer weight of the kernel extreme learning machine, To obtain the minimum function, is the target output vector, is the data matrix of the high-dimensional feature space, is the regularization parameter of the kernel extreme learning machine; According to the mapped data and target output, the manifold mapping strategy is used to further adjust the distribution of data in the kernel space, and then the weight of the output layer is calculated using the improved regularized least squares method. By minimizing the regularized loss function, the output layer weight of the kernel extreme learning machine is calculated, which is expressed as: （33） In formula (33), is the output layer weight of the kernel extreme learning machine, express The transpose of is the data matrix of the high-dimensional feature space, is the regularization parameter of the kernel extreme learning machine, I is the identity matrix; Once the output layer weight calculation of the kernel extreme learning machine is completed, the kernel extreme learning machine training is completed; The execution module is used to control the growth environment parameters of Pleurotus geesteranus and receive and execute system operation instructions to adjust the growth environment parameters.

2. A method for using a Pleurotus geesteranus growth environment regulation system integrating multiple factors based on the Pleurotus geesteranus growth environment regulation system integrating multiple factors according to claim 1, characterized in that: include: S1. Collect historical growth environment data of Pleurotus geesteranus, wherein the growth environment data includes environmental sensor data, system operation data, growth feedback data, and external influence data, and each of the environmental sensor data, system operation data, growth feedback data, and external influence data includes multiple attributes; S2, manually labeling the historical growth environment data and storing it in a storage module, wherein the labeling category of the labeling is a system operation instruction; S3, inputting the historical growth environment data with the number of attributes less than or equal to 50 into the Pleurotus geesteranus multi-factor data feature extraction module, training the feature extraction model of the Pleurotus geesteranus multi-factor data feature extraction module based on the neural network of adaptive biological impact optimization, obtaining the output data of the neural network, and storing it in the storage module; S4, inputting the growth environment data with the number of attributes greater than 50 into the Pleurotus geesteranus multi-factor data feature dimensionality reduction module, training the feature dimensionality reduction model of the Pleurotus geesteranus multi-factor data feature dimensionality reduction module based on the autoencoder algorithm combined with the Riemann geometry optimization algorithm, obtaining the output data of the autoencoder, and storing it in the storage module; The feature dimensionality reduction model based on the autoencoder algorithm combined with the Riemann geometry optimization algorithm for training the Pleurotus geesteranus multi-factor data feature dimensionality reduction module in S4 includes: S401, define the structure of the autoencoder, the input layer of the autoencoder receives multi-dimensional feature data, the number of neurons in the output layer of the autoencoder is 50, and define the transformation method of the input to each layer as: （16） In formula (16), For the autoencoder The linear transformation output of the layer; For the autoencoder The weight matrix of the layer; For the autoencoder The output of the layer, i.e., the output of the autoencoder The input of the layer; For the autoencoder The bias vector of the layer; S402, data is transmitted through each hidden layer of the autoencoder, and the data flow path is adjusted in each hidden layer using the Riemann geometry optimization algorithm to ensure that the data flow conforms to the properties of its characteristic intrinsic distribution, expressed as: （17） In formula (17), After applying the activation function to the autoencoder Layer output, For the autoencoder The linear transformation output of the layer; yes of Norm, representation The module length; To adjust the Riemannian geometry parameters of layer activation strengths; is the hyperbolic tangent function; S403, use reconstruction loss and Riemann distance to jointly evaluate the difference between the output data and the original data to ensure that the essential characteristics of the data are preserved during the dimensionality reduction process. The loss function of the autoencoder The calculation method is: （20） In formula (20), mrc is the number of samples input into the autoencoder in the current batch, is the data vector of the i-th training sample, is the reconstructed output of the i-th autoencoder, To control the regularization coefficient of the Riemann distance, is the Riemann distance, is the noise penalty loss; S404, the gradient of the loss function to each parameter is calculated by the back propagation algorithm, and the parameters of each layer of the autoencoder are updated. The back propagation uses the chain rule to calculate the gradient, and the way to update the autoencoder weights and biases is expressed as: （25） （26） In formula (25) and formula (26), is the update amount of the autoencoder weight, is the learning rate of the autoencoder, is the symbol of partial derivative, is the partial derivative of the loss function of the autoencoder with respect to the autoencoder weights, is the weight decay rate of the autoencoder, is the update amount of the autoencoder bias, is the partial derivative of the autoencoder loss function with respect to the autoencoder bias; S405, repeating S402 to S404 until the preset stop iteration condition is met, which means that the model training is completed; S5, inputting the output data of the neural network and the output data of the autoencoder into a Pleurotus geesteranus growth environment regulation prediction module, and training a classifier of the Pleurotus geesteranus growth environment regulation prediction module based on a quantum Hamiltonian optimized nuclear extreme learning machine classification algorithm; The classifier of the quantum Hamiltonian optimized nuclear extreme learning machine classification algorithm for training the Pleurotus geesteranus growth environment control prediction module in S5 includes: S501, initialize the kernel function of the kernel extreme learning machine, and define the calculation method of the kernel function as: （29） In formula (29), is the kernel function of the kernel extreme learning machine, is the width parameter of the kernel function of the kernel extreme learning machine, Dynamically adjust the function for the kernel function, is the i-th sample input to the kernel extreme learning machine, is the jth sample input to the kernel extreme learning machine; S502, mapping the data input to the kernel extreme learning machine to the high-dimensional feature space through the selected kernel function, and performing nonlinear mapping of the data, which is expressed as: （31） In formula (31), is the high-dimensional feature space mapping function, represents the data input to the kernel extreme learning machine, is the kernel width parameter of the kernel extreme learning machine, is the i-th sample input to the kernel extreme learning machine; S503, use the quantum Hamiltonian optimization algorithm to optimize the kernel parameters again to adapt to the specific data distribution, use the probability and parallelism of quantum computing to search the parameter space, find the kernel parameters that best suit the current data distribution, and solve the calculation method for the optimal kernel width parameter and output layer weight as follows: （32） In formula (32), Optimize the function for the quantum Hamiltonian, is the kernel width parameter of the kernel extreme learning machine, is the output layer weight of the kernel extreme learning machine, To obtain the minimum function, is the target output vector, is the data matrix of the high-dimensional feature space, is the regularization parameter of the kernel extreme learning machine; S504. According to the mapped data and target output, the distribution of the data in the kernel space is further adjusted using the manifold mapping strategy, and then the weight of the output layer is calculated using the improved regularized least squares method. The output layer weight of the kernel extreme learning machine is calculated by minimizing the regularized loss function, which is expressed as: （33） In formula (33), is the output layer weight of the kernel extreme learning machine, express The transpose of is the data matrix of the high-dimensional feature space, is the regularization parameter of the kernel extreme learning machine, I is the identity matrix; S505, after the output layer weight calculation of the kernel extreme learning machine is completed, it means that the training of the kernel extreme learning machine is completed; S6, collecting new growth environment data through the Pleurotus geesteranus multi-factor data collection module and storing it in the storage module; S7, determining the number of attributes of the new growth environment data; when the number of attributes is less than or equal to 50, inputting the new growth environment data into the Pleurotus geesteranus multi-factor data feature extraction module for feature mining; when the number of attributes is greater than 50, inputting the new growth environment data into the Pleurotus geesteranus multi-factor data feature dimension reduction module for feature mining; after completing the feature mining of the new growth environment data, obtaining the feature-mined data; S8, input the feature-mined data into the Pleurotus geesteranus growth environment regulation and prediction module for marking, obtain system operation instructions corresponding to the feature-mined data, send the system operation instructions to the execution module, and the execution module adjusts the growth environment parameters of Pleurotus geesteranus according to the system operation instructions.

3. The method for using the Pleurotus geesteranus growth environment control system integrating multiple factors according to claim 2, characterized in that: In S1: The environmental sensor data includes three attributes: temperature ,humidity and carbon dioxide concentration ; The system operation data includes four attributes: geothermal heating , water cooling , hydration and humidification and oxygenation The operational status of the system; The growth feedback data includes four attributes: mycelium growth rate , fruiting body differentiation speed ,Yield Data and quality data; The external impact data includes three attributes: seasonal changes , outside air temperature and humidity .

4. The method for using the Pleurotus geesteranus growth environment control system integrating multiple factors according to claim 2, characterized in that: The system operation instructions in S2 and S8 include heating on, heating off, cooling on, cooling off, humidification on, humidification off, ventilation on, ventilation off, and unspecified specific operations.

5. The method for using the Pleurotus geesteranus growth environment control system integrating multiple factors according to claim 2, characterized in that: The feature extraction model of the neural network based on adaptive biological influence optimization for training the multi-factor data feature extraction module of Pleurotus geesteranus in S3 includes: S301, initialization setting of neural network parameters, including neural network weights Bias with Neural Networks , the initialization method is expressed as: （1） （2） In formula (1) and formula (2), is the initial weight of the neural network, is a nonlinear transformation function, is the initialization random seed parameter, is the input dimension of the neural network, is the initial bias of the neural network; S302, according to the characteristics and objectives of the training data, simulate a dynamically changing environmental factor, the environmental factor affects the adaptability requirements of the neural network parameters, and the environmental factor is dynamically generated based on the statistical characteristics of the training data characteristics, and the simulation is performed by evaluating the coefficient of variation and skewness of the input features, which is expressed as: （5） （6） （7） In formula (5), formula (6) and formula (7), For environmental factors, is the first hyperparameter to adjust the sensitivity of the environmental simulation. is the coefficient of variation calculation function, is the feature input to the neural network, is the second hyperparameter that adjusts the sensitivity of the environmental simulation. is the skewness calculation function, Features The standard deviation of Features The mean of S303, define an initial fitness evaluation function, which considers the prediction accuracy and parameter adaptability of the model to ensure that the network performance is evaluated from multiple dimensions. The calculation method is: （8） （9） In formula (8) and formula (9), is the fitness function of the neural network, is the weight of the neural network, is the bias of the neural network, For environmental factors, is the cross entropy loss function, is the regularization coefficient of the neural network, for norm, is the weight change of the neural network, is the fitness adaptive regularization factor, is the jth weight of the neural network, is the change of the neural network in the previous iteration; S304, the weight and bias of the neural network are adjusted based on the dynamic adjustment factor, and the adjustment method is expressed as: （12） （13） （14） In formula (12), formula (13) and formula (14), is the nonlinear adjustment factor parameter, is a hyperparameter that adjusts the intensity of the response, is a nonlinear response function, is the fitness function of the neural network, is the weight of the neural network, is the bias of the neural network, For environmental factors, is random noise, is the parameter update operation, is the learning rate of the neural network, is the gradient of the neural network’s fitness function with respect to the weight parameters, is the gradient of the fitness function of the neural network with respect to the bias parameter; S305, repeating S302-S304 until the preset stop iteration condition is met, which means that the model training is completed.

6. The method for using the multi-factor integrated Pleurotus geesteranus growth environment control system according to claim 5, characterized in that: In S303, the fitness adaptive regularization factor is calculated based on the activity of the feature, which is expressed as: （11） In formula (11), is the fitness adaptive regularization factor, is a hyperparameter that adjusts the strength of the regularization baseline, is the activity calculation function of the feature, is the weight of the neural network.

7. The method for using the multi-factor integrated Pleurotus geesteranus growth environment control system according to claim 2, characterized in that: In S402 of Norm The calculation method is: （18） In formula (18), for The dimension of the vector, express The i-th component of the vector.

8. The method for using the Pleurotus geesteranus growth environment control system integrating multiple factors according to claim 2, characterized in that: The partial derivative of the loss function of the autoencoder in S404 with respect to the weight The calculation method is: （27） In formula (27), mrc is the number of samples input into the autoencoder in the current batch, After applying the activation function to the autoencoder Layer output, is the data vector of the i-th training sample, yes The transpose of For the autoencoder The output of the layer, i.e., the output of the autoencoder The input of the layer; is the regularization parameter of the autoencoder, is the partial derivative of the Riemann distance with respect to the weights of the autoencoder.

Images