Microalgae growth prediction and optimization method and system based on machine learning

Abstract

The invention relates to the technical field of microalgae biological intelligent control, in particular to a microalgae growth prediction and optimization method and system based on machine learning. Based on the characteristic variable random forest model of the microalgae growth parameters, screening out a characteristic variable data set after dimensionality reduction screening which significantly affects the microalgae growth, and performing regression training of potential principal component variables on the characteristic variable data set through a machine learning algorithm to obtain a trained microalgae growth prediction model; acquiring a plurality of actual prediction results under dynamic change of culture conditions by using the trained microalgae growth prediction model, and searching a given parameter regulation space in a genetic algorithm based on the actual prediction results of microalgae growth culture to obtain an optimal culture condition parameter set; and issuing the optimal culture condition parameter set to an actual culture process of microalgae growth. According to the method, microalgae growth dynamics can be accurately described in a multi-dimensional parameter space, and real-time prediction and optimization of the microalgae growth trend and metabolic yield are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of microalgae biological intelligent control, and in particular to a microalgae growth prediction and optimization method and system based on machine learning. Background Art

[0002] Microalgae are photosynthetic autotrophic microorganisms that can convert CO2, light energy and nutrients into biomass and a variety of metabolites. The growth process and metabolic characteristics of microalgae are significantly affected by environmental factors (such as light intensity, photoperiod, temperature, pH value, nutrient concentration, CO2 concentration and stirring speed, etc.). Traditional methods for predicting microalgae growth mostly rely on empirical models or simple linear models, which make it difficult to accurately characterize the impact of multiple factors interacting on the growth rate and metabolite yield of microalgae.

[0003] With the development of data acquisition methods (sensors, online monitoring equipment) and storage technology, people can obtain a large amount of microalgae cultivation process data. However, how to effectively mine the complex nonlinear relationships hidden in these data and predict the growth trend of microalgae in real time, and then optimize the cultivation strategy, is the difficulty and focus of current research. Summary of the invention

[0004] The present invention overcomes the deficiencies of the prior art and provides a method and system for predicting and optimizing microalgae growth based on machine learning.

[0005] To achieve the above object, the technical solution adopted by the present invention is: The first aspect of the present invention provides a method for predicting and optimizing microalgae growth based on machine learning, comprising the following steps: S102: obtaining multidimensional environmental parameters and microalgae growth parameters in the microalgae cultivation process through online monitoring equipment, randomly extracting characteristic samples after preprocessing the multidimensional environmental parameters and microalgae growth parameters, and performing recursive splitting and superposition construction on the characteristic samples to obtain a characteristic variable random forest model of the multidimensional environmental parameters and microalgae growth parameters; S104: Calculating the feature importance of each feature node for the growth of microalgae based on the feature node distribution diagram and node values ​​of the feature variable random forest model, analyzing the feature importance to screen out a feature variable data set after dimensionality reduction screening that significantly affects the growth of microalgae, and performing regression training of potential principal component variables on the feature variable data set through a machine learning algorithm to obtain a trained microalgae growth prediction model; S106: obtaining real-time microalgae growth environment parameters of the actual microalgae cultivation process, predicting the real-time microalgae growth environment parameters through the trained microalgae growth prediction model to obtain actual prediction results, obtaining multiple actual prediction results and expected prediction results under dynamic changes in cultivation conditions, and performing sliding average calculation of the Bezier curve, and updating the learning rate of the prediction model based on the sliding average; S108: Obtain the objective function of microalgae growth and cultivation, set the optimization goal that satisfies the microalgae growth and cultivation plan according to the objective function, and preset the established parameter control space of different growth and cultivation parameters, search the established parameter control space in the genetic algorithm based on the optimization goal, and obtain the optimal cultivation condition parameter set, send the optimal cultivation condition parameter set to the actual cultivation process of microalgae growth, and feed back the latest status data of microalgae growth performance into the prediction model and optimization module to form a dynamic optimization closed loop.

[0006] More specifically, the step S102 includes the following steps: The microalgae cultivation environment is monitored in real time through online and offline sensors installed on the online monitoring equipment to obtain multi-dimensional environmental parameters and microalgae growth parameters during the microalgae cultivation process; Importing a data preprocessing process, performing outlier detection, missing value interpolation and normalization processing on the multidimensional environmental parameters and microalgae growth parameters in the microalgae cultivation process through the data preprocessing process, to obtain preprocessed multidimensional environmental parameters and microalgae growth parameters; Randomly extracting characteristic samples of the pre-processed multidimensional environmental parameters and microalgae growth parameters to construct M Bootstrap characteristic variable sample sets; establishing a corresponding tree root node for each Bootstrap characteristic variable sample set, and presetting the optimal splitting feature and the maximum splitting depth; Based on the best split feature, recursive splitting operation of feature nodes is performed on the Bootstrap feature variable sample set starting from the root node of the tree to determine whether the current split depth reaches the maximum split depth; If it is not reached, continue to recursively split the feature nodes of the Bootstrap feature variable sample set; if it is reached, stop the recursive splitting operation and generate a sub-decision tree for each Bootstrap feature variable sample set; finally, superimpose the sub-decision trees of all Bootstrap feature variable sample sets to obtain a random forest model of feature variables of multidimensional environmental parameters and microalgae growth parameters.

[0007] More specifically, the step S104 includes the following steps: Extracting the characteristic node distribution graph of the characteristic variable random forest model and the node value of each characteristic node, calculating the initial error of the characteristic variable random forest model under the condition of the characteristic node distribution graph based on the node value of each characteristic node, and obtaining the initial mean square error of each characteristic node for the characteristic variable random forest model; Obtain the microalgae growth prediction demand, preset the disturbance threshold of each feature node according to the microalgae growth prediction demand, perturb each feature node in the feature variable random forest model through the disturbance threshold and recalculate the error, and obtain the perturbation mean square error of the feature variable random forest model after perturbing each feature node; Calculating a mean square error increment between the disturbance mean square error and the initial mean square error, and determining the feature importance of each feature node to the growth of microalgae according to the mean square error increment; If the mean square error amplitude is negative, it means that the feature node has little influence on the prediction performance of the feature variable random forest model, that is, the feature importance is low, then the feature node corresponding to the disturbance mean square error is removed; If the mean square error amplitude is positive, it means that the feature node has a greater impact on the prediction performance of the feature variable random forest model, which means that the feature importance is high. Then the feature node corresponding to the perturbation mean square error is extracted and marked as a high-importance feature node; Integrate the data corresponding to all high-importance feature nodes retained in the pre-processed multidimensional environmental parameters and microalgae growth parameters to obtain a feature variable data set after dimensionality reduction screening; A partial least squares regression algorithm is introduced, and latent principal component variable regression training of input feature-output feature is performed on the characteristic variable data set through the partial least squares regression algorithm to obtain a trained microalgae growth prediction model.

[0008] More specifically, the partial least squares regression algorithm is introduced, and the input feature-output feature potential principal component variable regression training of the feature variable data set is performed by the partial least squares regression algorithm to obtain a trained microalgae growth prediction model, which specifically includes the following steps: Taking environmental parameters as input features and microalgae growth trends and metabolic yields as output features, an input feature-output feature covariance matrix is ​​calculated and constructed based on the feature variable data set after dimensionality reduction screening; The partial least squares regression algorithm is introduced, the maximum covariance distance threshold is preset, and the covariance distance between each input feature and each output feature in the input feature-output feature covariance matrix is ​​calculated by the partial least squares regression algorithm to obtain the covariance distance of each matrix node in the input feature-output feature covariance matrix; For each matrix node in the input feature-output feature covariance matrix, if the covariance distance is less than the maximum covariance distance threshold, the matrix node corresponding to the covariance distance is ignored; if the covariance distance is greater than the maximum covariance distance threshold, the matrix node corresponding to the covariance distance is marked and defined as the maximum correlation matrix node; Extract the input features and output features associated with the nodes of the maximum correlation matrix, define them as a set of potential principal component variables, and preset the critical value of the number of potential principal component variables to be extracted; Repeat the above steps of calculating and determining the covariance distance and extracting potential principal component variables from the input feature-output feature covariance matrix. Each time a group of potential principal component variables is extracted, the relationship between the iterative input features and the output features is updated to obtain the current number of extracted potential principal component variables. If the current extraction quantity does not reach the extraction quantity critical value, the covariance distance calculation and determination of the input feature-output feature covariance matrix and the extraction of potential principal component variables are continued; if the current extraction quantity has reached the extraction quantity critical value, the extraction of potential principal component variables is stopped to obtain the extracted potential principal component variable set; The extracted potential principal component variable set is used to construct a regression model between input features and output features in a partial least squares regression algorithm, and a regression training operation is performed on the training set through the regression model to obtain a trained microalgae growth prediction model.

[0009] More specifically, the step S106 includes the following steps: Acquire real-time microalgae growth environment parameters during the actual cultivation process of microalgae, and import the real-time microalgae growth environment parameters into the trained microalgae growth prediction model to perform growth prediction in a future time period, so as to output the actual microalgae growth prediction trend and actual metabolic prediction yield in a specific future time period; Acquire the actual microalgae growth prediction trend and actual metabolic prediction yield output by the microalgae growth prediction model when predicting dynamic changes in culture conditions on a preset culture time sequence, and obtain multiple actual prediction results when culture conditions change dynamically; A Bezier curve method is introduced, and a plurality of the actual prediction results are subjected to curve fitting by the Bezier curve method to obtain a Bezier curve of the actual prediction result, which is defined as a first Bezier curve; the expected prediction result under the dynamic change of the culture condition is simultaneously obtained, and a Bezier curve of the expected prediction result is constructed, which is defined as a second Bezier curve; If the slope deviation value between the slope of the first Bezier curve and the slope of the second Bezier curve is greater than the preset slope deviation value, the historical gradient square of the trained microalgae growth prediction model in the dynamic change process of the culture conditions is tracked based on the slope deviation value and exponentially weighted to obtain a sliding average value; During the tracking process, the current learning rate of the microalgae growth prediction model is obtained, and the current learning rate is adjusted and updated through a sliding average value, so that the trained microalgae growth prediction model can adapt to dynamically changing culture conditions.

[0010] More specifically, the step S108 includes the following steps: Obtaining a microalgae growth and cultivation plan, and obtaining an objective function of microalgae growth and cultivation through the microalgae growth and cultivation plan; wherein the objective function includes maximum biomass output, optimal metabolite synthesis rate, and minimum resource consumption; An optimization goal that satisfies the microalgae growth and cultivation plan is set based on the predicted growth trend of the microalgae in a specific future time period, the predicted metabolic yield, and the objective function; Extracting preset control intervals of different growth and cultivation parameters through the microalgae growth and cultivation plan, and constructing a predetermined parameter control space according to the preset control intervals of the different growth and cultivation parameters; A genetic algorithm is introduced to construct multiple initial genetic populations of a given parameter control space, and the fitness of each growth and cultivation parameter individual in each initial genetic population is evaluated based on the optimization goal to obtain several fitness values; Only the growth and cultivation parameter individuals with the maximum fitness value are extracted, marked as parent individuals, and genetic crossover and mutation operations are performed on the parent individuals to generate new offspring individuals, and the new offspring individuals of each initial genetic population are searched to replace the corresponding parent individuals until the population fitness reaches a preset fitness threshold, and finally the optimal cultivation condition parameter set for microalgae growth is output; Through a closed-loop control strategy, the optimal culture condition parameter set is sent to the actual culture process of microalgae growth to adjust the environmental conditions for microalgae growth in real time, and the microalgae growth performance is monitored in real time to feed the latest status data into the prediction model and optimization module to form a dynamic optimization closed loop.

[0011] More specifically, the real-time monitoring of the growth performance of microalgae and the feeding back of the latest status data into the prediction model and optimization module form a dynamic optimization closed loop, which specifically includes the following steps: Through online monitoring equipment, the growth status of microalgae is continuously monitored to obtain the current growth status variables, and a closed-loop state control model of microalgae growth cultivation-prediction-optimization is established; Introducing a pole placement method, performing a closed-loop feedback calculation of a closed-loop state control model on the current growth state variable through the pole placement method, obtaining a closed-loop feedback gain matrix of the current microalgae growth state, importing the closed-loop feedback gain matrix into the closed-loop state control model for compensation control, and generating a closed-loop feedback controller of the current microalgae growth state; The closed-loop feedback controller of the current microalgae growth state is deployed in the actual microalgae growth cultivation-prediction-optimization process to obtain the latest relevant state data of the current microalgae growth, and finally the latest relevant state data is fed back to the prediction model and optimization module of the microalgae growth prediction model to form a dynamic optimization closed loop.

[0012] A second aspect of the present invention provides a microalgae growth prediction and optimization system based on machine learning, the microalgae growth prediction and optimization system comprising a memory and a processor, the memory storing a microalgae growth prediction and optimization method program based on machine learning, and when the microalgae growth prediction and optimization method program is executed by the processor, any one of the steps of the microalgae growth prediction and optimization method is implemented.

[0013] The present invention solves the technical defects existing in the background technology, and the beneficial technical effects of the present invention are: The multidimensional environmental parameters and microalgae growth parameters in the microalgae cultivation process are pre-processed, and feature samples are randomly extracted and recursively split to obtain a feature variable random forest model of the multidimensional environmental parameters and microalgae growth parameters; based on the feature variable random forest model, the feature importance of each feature node for the microalgae growth is calculated, and the feature variable data set after dimensionality reduction screening that significantly affects the microalgae growth is screened out according to the feature importance, and the feature variable data set is subjected to regression training of potential principal component variables through a machine learning algorithm to obtain a trained microalgae growth prediction model; the trained microalgae growth prediction model is used to predict the real-time microalgae growth environmental parameters to obtain actual prediction results, multiple actual prediction results and expected prediction results under dynamic changes in cultivation conditions are obtained, and the sliding average of the Bezier curve is calculated, and the learning rate of the prediction model is updated based on the sliding average; according to the objective function of microalgae growth and cultivation, an optimization goal that satisfies the microalgae growth and cultivation plan is set, and based on the optimization goal, a predetermined parameter control space is searched in a genetic algorithm to obtain an optimal cultivation condition parameter set, and the optimal cultivation condition parameter set is sent to the actual cultivation process of microalgae growth and the latest status data of microalgae growth performance is fed back into the prediction model and the optimization module to form a dynamic optimization closed loop. By integrating multi-source data with advanced machine learning algorithms, the present invention can accurately depict the growth dynamics of microalgae in a multidimensional parameter space, and achieve real-time prediction and optimization of growth trends and metabolic yields. It can effectively improve the efficiency of microalgae cultivation, reduce operating costs, achieve automatic or semi-automatic optimization of cultivation conditions, and improve the yield of microalgae and the synthesis efficiency of target metabolites. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, drawings of other embodiments can be obtained based on these drawings without paying creative work.

[0015] Figure 1 A first method flow chart of a method for predicting and optimizing microalgae growth based on machine learning is shown; Figure 2 A second method flow chart of a method for predicting and optimizing microalgae growth based on machine learning is shown; Figure 3 A system framework diagram of a microalgae growth prediction and optimization system based on machine learning is shown. DETAILED DESCRIPTION

[0016] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0017] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0018] The first aspect of the present invention provides a method for predicting and optimizing microalgae growth based on machine learning, such as Figure 1 As shown, the following steps are included: S102: obtaining multidimensional environmental parameters and microalgae growth parameters in the microalgae cultivation process through online monitoring equipment, randomly extracting characteristic samples after preprocessing the multidimensional environmental parameters and microalgae growth parameters, and performing recursive splitting and superposition construction on the characteristic samples to obtain a characteristic variable random forest model of the multidimensional environmental parameters and microalgae growth parameters; S104: Calculating the feature importance of each feature node for the growth of microalgae based on the feature node distribution diagram and node values ​​of the feature variable random forest model, analyzing the feature importance to screen out a feature variable data set after dimensionality reduction screening that significantly affects the growth of microalgae, and performing regression training of potential principal component variables on the feature variable data set through a machine learning algorithm to obtain a trained microalgae growth prediction model; S106: obtaining real-time microalgae growth environment parameters of the actual microalgae cultivation process, predicting the real-time microalgae growth environment parameters through the trained microalgae growth prediction model to obtain actual prediction results, obtaining multiple actual prediction results and expected prediction results under dynamic changes in cultivation conditions, and performing sliding average calculation of the Bezier curve, and updating the learning rate of the prediction model based on the sliding average; S108: Obtain the objective function of microalgae growth and cultivation, set the optimization goal that satisfies the microalgae growth and cultivation plan according to the objective function, and preset the established parameter control space of different growth and cultivation parameters, search the established parameter control space in the genetic algorithm based on the optimization goal, and obtain the optimal cultivation condition parameter set, send the optimal cultivation condition parameter set to the actual cultivation process of microalgae growth, and feed back the latest status data of microalgae growth performance into the prediction model and optimization module to form a dynamic optimization closed loop.

[0019] More specifically, the step S102 includes the following steps: The microalgae cultivation environment is monitored in real time through online and offline sensors installed on the online monitoring equipment to obtain multi-dimensional environmental parameters and microalgae growth parameters during the microalgae cultivation process; Importing a data preprocessing process, performing outlier detection, missing value interpolation and normalization processing on the multidimensional environmental parameters and microalgae growth parameters in the microalgae cultivation process through the data preprocessing process, to obtain preprocessed multidimensional environmental parameters and microalgae growth parameters; Randomly extracting characteristic samples of the pre-processed multidimensional environmental parameters and microalgae growth parameters to construct M Bootstrap characteristic variable sample sets; establishing a corresponding tree root node for each Bootstrap characteristic variable sample set, and presetting the optimal splitting feature and the maximum splitting depth; Based on the best split feature, recursive splitting operation of feature nodes is performed on the Bootstrap feature variable sample set starting from the root node of the tree to determine whether the current split depth reaches the maximum split depth; If it is not reached, continue to recursively split the feature nodes of the Bootstrap feature variable sample set; if it is reached, stop the recursive splitting operation and generate a sub-decision tree for each Bootstrap feature variable sample set; finally, superimpose the sub-decision trees of all Bootstrap feature variable sample sets to obtain a random forest model of feature variables of multidimensional environmental parameters and microalgae growth parameters.

[0020] It should be noted that, since the multidimensional environmental parameters and microalgae growth parameters in the microalgae cultivation process contain different characteristics that affect the growth of microalgae, if the multidimensional environmental parameters and microalgae growth parameters obtained by pretreatment are directly imported into the machine learning algorithm for training, it may cause the machine learning algorithm to perform too many weight calculations on some characteristic variables that have a lower impact on the growth of microalgae when training data, which may lead to data redundancy and the risk of overfitting. This makes the microalgae growth prediction model obtained by training have a large prediction error for the actual environmental parameters, causing the prediction accuracy and performance of the microalgae growth prediction model to drop sharply, which is not conducive to the optimization of the actual microalgae growth and cultivation. In this regard, this method randomly extracts characteristic samples of pretreated multidimensional environmental parameters and microalgae growth parameters to construct a characteristic variable sample set with different randomness affecting the growth of microalgae, increase the diversity of prediction model sample training, reduce the correlation of different characteristic variables, and thus improve the integration effect of subsequent feature importance screening. Then, based on the preset optimal splitting feature of each Bootstrap feature variable sample set, the recursive splitting operation of the feature node is performed from the root node to construct a sub-decision tree of each Bootstrap feature variable sample set. By constructing a sub-decision tree representing the randomness of the importance of different microalgae growth features, the fitting ability of the training data can be enhanced, so that the feature importance screening of the random forest model has a higher generalization ability, and the subsequent microalgae growth-influencing significant features based on feature importance screening do not have data training overfitting, thereby improving the overall performance of the microalgae growth prediction model. Finally, all sub-decision trees are superimposed to form a random forest model that expresses the randomness of the feature variables of multidimensional environmental parameters and microalgae growth parameters. The random forest model can make the screening of significant features affecting microalgae growth based on feature importance have a better integration effect, improve the training accuracy of the machine learning algorithm for the feature variables that significantly affect microalgae growth, and reduce the prediction error of the prediction model.

[0021] It should be noted that the multidimensional environmental parameters include light intensity, temperature, pH value, CO2 concentration, nutrient concentration and stirring rate. Microalgae growth parameters include cell density, chlorophyll content, biomass concentration and specific metabolite concentration. This method can provide a data feature integration platform for subsequent machine learning based on feature importance to screen significant features affecting microalgae growth, reduce the feature screening phenomenon that traditional machine learning algorithms may overfit a single decision tree, and greatly improve the training sensitivity of data features that affect microalgae growth, so as to reduce data redundancy and overfitting risks, and further improve the performance of the prediction model. Bootstrap is a commonly used front-end framework that contains many class names (i.e., feature variables) for rapid construction and design of responsive web pages.

[0022] More specifically, the step S104 includes the following steps: Extracting the characteristic node distribution graph of the characteristic variable random forest model and the node value of each characteristic node, calculating the initial error of the characteristic variable random forest model under the condition of the characteristic node distribution graph based on the node value of each characteristic node, and obtaining the initial mean square error of each characteristic node for the characteristic variable random forest model; Obtain the microalgae growth prediction demand, preset the disturbance threshold of each feature node according to the microalgae growth prediction demand, perturb each feature node in the feature variable random forest model through the disturbance threshold and recalculate the error, and obtain the perturbation mean square error of the feature variable random forest model after perturbing each feature node; Calculating a mean square error increment between the disturbance mean square error and the initial mean square error, and determining the feature importance of each feature node to the growth of microalgae according to the mean square error increment; If the mean square error amplitude is negative, it means that the feature node has little influence on the prediction performance of the feature variable random forest model, that is, the feature importance is low, then the feature node corresponding to the disturbance mean square error is removed; If the mean square error amplitude is positive, it means that the feature node has a greater impact on the prediction performance of the feature variable random forest model, which means that the feature importance is high. Then the feature node corresponding to the perturbation mean square error is extracted and marked as a high-importance feature node; Integrate the data corresponding to all high-importance feature nodes retained in the pre-processed multidimensional environmental parameters and microalgae growth parameters to obtain a feature variable data set after dimensionality reduction screening; A partial least squares regression algorithm is introduced, and latent principal component variable regression training of input feature-output feature is performed on the characteristic variable data set through the partial least squares regression algorithm to obtain a trained microalgae growth prediction model.

[0023] It should be noted that since the feature node distribution diagram of the feature variable random forest model and the node value of each feature node reflect the potential training level and weight of each feature data in machine learning training, for the screening of the features that significantly affect the growth of microalgae integrated in the feature variable random forest model, this method reduces the dimension by applying a perturbed feature value to each feature to evaluate its importance. First, the initial error of the feature variable random forest model under the condition of the feature node distribution diagram is calculated based on the node value of each feature node, so as to obtain an initial mean square error for the model prediction performance when each feature node is not perturbed. The initial mean square error is a judgment benchmark; then, each feature node is perturbed, that is, the value of the feature is disrupted, and the mean square error of the model is recalculated to obtain a perturbed mean square error. The mean square error increment between the perturbed mean square error and the initial mean square error can express the feature importance of each feature node to the growth of microalgae. Among them, if the mean square error amplitude is displayed as a negative value, it means that the feature node has little effect on the prediction performance of the feature variable random forest model, that is, the feature importance is low; if the mean square error amplitude is displayed as a positive value, it means that the feature node has a greater impact on the prediction performance of the feature variable random forest model, which means that the feature importance is high. Finally, the partial least squares regression algorithm is introduced to perform latent principal component variable regression training of input features-output features on the feature variable data set to generate a microalgae growth prediction model. This method can screen out feature variables that have a significant impact on microalgae growth based on the feature importance of random forests, so as to reduce data redundancy and overfitting risks, improve the training efficiency of the prediction model, and effectively optimize the model's prediction performance for microalgae growth.

[0024] More specifically, the partial least squares regression algorithm is introduced, and the input feature-output feature potential principal component variable regression training of the feature variable data set is performed by the partial least squares regression algorithm to obtain a trained microalgae growth prediction model, which specifically includes the following steps: Taking environmental parameters as input features and microalgae growth trends and metabolic yields as output features, an input feature-output feature covariance matrix is ​​calculated and constructed based on the feature variable data set after dimensionality reduction screening; The partial least squares regression algorithm is introduced, the maximum covariance distance threshold is preset, and the covariance distance between each input feature and each output feature in the input feature-output feature covariance matrix is ​​calculated by the partial least squares regression algorithm to obtain the covariance distance of each matrix node in the input feature-output feature covariance matrix; For each matrix node in the input feature-output feature covariance matrix, if the covariance distance is less than the maximum covariance distance threshold, the matrix node corresponding to the covariance distance is ignored; if the covariance distance is greater than the maximum covariance distance threshold, the matrix node corresponding to the covariance distance is marked and defined as the maximum correlation matrix node; Extract the input features and output features associated with the nodes of the maximum correlation matrix, define them as a set of potential principal component variables, and preset the critical value of the number of potential principal component variables to be extracted; Repeat the above steps of calculating and determining the covariance distance and extracting potential principal component variables from the input feature-output feature covariance matrix. Each time a group of potential principal component variables is extracted, the relationship between the iterative input features and the output features is updated to obtain the current number of extracted potential principal component variables. If the current extraction quantity does not reach the extraction quantity critical value, the covariance distance calculation and determination of the input feature-output feature covariance matrix and the extraction of potential principal component variables are continued; if the current extraction quantity has reached the extraction quantity critical value, the extraction of potential principal component variables is stopped to obtain the extracted potential principal component variable set; The extracted potential principal component variable set is used to construct a regression model between input features and output features in a partial least squares regression algorithm, and a regression training operation is performed on the training set through the regression model to obtain a trained microalgae growth prediction model.

[0025] It should be noted that the growth trend of microalgae and metabolic yield indicators are crucial for the subsequent precise cultivation of microalgae. Therefore, for the prediction of microalgae growth, it is usually achieved in the form of existing environmental parameters as input features and microalgae growth trends and metabolic yields as output features. However, it is difficult for existing prediction models to achieve correlation screening and training between existing environmental parameters and the required microalgae growth trends and metabolic yields. This makes it difficult to make a decision when analyzing a certain output feature to know which input features may have necessary correlations. For example, the growth trend of microalgae and metabolic yield may be affected by the temperature in the input feature, which can be optimized from the temperature aspect. The above problems make it impossible to make corresponding precise improvements and optimizations in the subsequent microalgae growth and cultivation. Therefore, this method introduces the partial least squares regression algorithm to train the feature variable data set after dimension reduction and screening. The core idea of ​​the partial least squares regression algorithm is to find a set of new principal component variables, and these principal component variables have the largest variance in the independent variable space. At the same time, these principal component variables also maximize the correlation with the dependent variable as much as possible. Therefore, the input feature-output feature covariance matrix must be calculated and constructed by the feature variable data set after dimension reduction and screening; then the covariance distance between each input feature and each output feature is calculated. If the covariance distance is less than the maximum covariance distance threshold, it means that the correlation between a certain input feature and the output feature on the matrix node is low or even does not exist closely, so the matrix node corresponding to the covariance distance is ignored; if the covariance distance is greater than the maximum covariance distance threshold, it means that the correlation between a certain input feature and the output feature on the matrix node reaches the maximum value, so the matrix node is marked as the maximum correlation matrix node. On the one hand, it can ensure the characteristic prediction correlation between environmental parameters and the trend of microalgae growth and metabolic yield during data training, which is convenient for the subsequent precise optimization of microalgae growth and cultivation; on the other hand, it can reduce the complexity of the prediction model and avoid the overfitting problem caused by too high a dimension.

[0026] It should be noted that each time a set of potential principal component variables is extracted, the relationship between the input features and the output features is updated and iterated until the critical value of the extraction quantity is reached. This can ensure that the model can fully capture the important information in the data, while reducing the introduction of redundant principal components, and optimizing the prediction performance and generalization ability of the model. Finally, the input features and output features associated with the extracted maximum correlation matrix nodes are used as a set of potential principal component variables to establish a regression model between the independent variables and the dependent variables. The microalgae growth prediction model can be generated by performing regression training operations on the training set through the regression model. Through this method, the correlation between the input features and the output features can be maximized and analyzed during the training of the existing environmental parameters by the machine learning algorithm, so as to train an efficient and accurate microalgae growth prediction model by dimensionality reduction regression, so as to achieve a high correlation between the prediction result features and the existing data features, provide traceability data for the subsequent cultivation optimization of microalgae growth, and improve the accuracy of cultivation optimization. Among them, the partial least squares regression algorithm is the most representative statistical method in the machine learning algorithm and is often used for high-dimensional data or multicollinear data. The partial least squares regression algorithm can simultaneously consider the maximum information correlation of the independent variables and the dependent variables, so it has a good prediction training ability.

[0027] More specifically, the step S106 is as follows: Figure 2 As shown, the specific steps include: S202: Acquire real-time microalgae growth environment parameters of the actual microalgae cultivation process, and import the real-time microalgae growth environment parameters into the trained microalgae growth prediction model to perform growth prediction in a future time period, so as to output the actual microalgae growth prediction trend and actual metabolic prediction yield in a specific future time period; S204: obtaining the actual microalgae growth prediction trend and the actual metabolic prediction yield output by the microalgae growth prediction model when predicting dynamic changes in culture conditions on a preset culture time sequence, and obtaining multiple actual prediction results when culture conditions change dynamically; S206: Introducing a Bezier curve method, performing curve fitting on the plurality of actual prediction results through the Bezier curve method, obtaining a Bezier curve of the actual prediction result, which is defined as a first Bezier curve; synchronously obtaining an expected prediction result under dynamic changes in culture conditions, constructing a Bezier curve of the expected prediction result, which is defined as a second Bezier curve; S208: If the slope deviation value between the slope of the first Bezier curve and the slope of the second Bezier curve is greater than a preset slope deviation value, tracking the historical gradient square of the trained microalgae growth prediction model in the dynamic change process of the culture conditions based on the slope deviation value and performing exponential weighting to obtain a sliding average value; S210: During the tracking process, the current learning rate of the microalgae growth prediction model is obtained, and the current learning rate is adjusted and updated by a sliding average, so that the trained microalgae growth prediction model can adapt to dynamically changing culture conditions.

[0028] It should be noted that since the growth and cultivation of microalgae is a relatively dynamic process, if the microalgae growth prediction model fails to respond in time, the learning rate of the current prediction model may not be able to keep up with the dynamic changes in the culture conditions, making the prediction model unable to adapt to the dynamic changes in the culture conditions of microalgae growth, resulting in a lag in the prediction results compared to the dynamic changes in the culture conditions, and a deviation in the prediction results. Therefore, it is necessary to perform a regular update on the microalgae growth prediction model. In view of this, the present method obtains the actual prediction results of multiple dynamic changes in culture conditions through the microalgae growth prediction model, and obtains the expected prediction results under the dynamic changes in culture conditions, and then constructs the respective Bezier curves of the two, namely the first Bezier curve and the second Bezier curve, respectively, using the Bezier curve method. The Bezier curve method can accurately reflect the fitting variables and errors between the two sets of data through good continuity and smoothness curve slopes, thereby making the hysteresis amplitude deviation positioning between the actual prediction results and the expected prediction results more efficient and accurate, further optimizing the update rate of the prediction model, saving cumbersome and unnecessary calculation steps, and reducing the update error rate of the prediction model. Among them, if the slope deviation value between the slope of the first Bezier curve and the slope of the second Bezier curve is greater than the preset slope deviation value, it means that the current microalgae growth prediction model has a large lag deviation phenomenon in its prediction performance for the dynamic changes in culture conditions. Therefore, it is necessary to track and calculate the sliding average of the historical gradient square of the microalgae growth prediction model in the process of dynamic changes in culture conditions based on the slope deviation value, and adjust and update the learning rate of the current microalgae growth prediction model through the sliding average to achieve the effect of different online incremental learning of the prediction model. This method can track and update the learning rate of the current microalgae growth prediction model based on the lag deviation of the prediction results under the dynamic changes in culture conditions, so that the trained microalgae growth prediction model can adapt to the dynamically changing culture conditions, so as to ensure that the microalgae growth prediction model can adapt to the dynamically changing culture conditions.

[0029] More specifically, the step S108 includes the following steps: Obtaining a microalgae growth and cultivation plan, and obtaining an objective function of microalgae growth and cultivation through the microalgae growth and cultivation plan; wherein the objective function includes maximum biomass output, optimal metabolite synthesis rate, and minimum resource consumption; An optimization goal that satisfies the microalgae growth and cultivation plan is set based on the predicted growth trend of the microalgae in a specific future time period, the predicted metabolic yield, and the objective function; Extracting preset control intervals of different growth and cultivation parameters through the microalgae growth and cultivation plan, and constructing a predetermined parameter control space according to the preset control intervals of the different growth and cultivation parameters; A genetic algorithm is introduced to construct multiple initial genetic populations of a given parameter control space, and the fitness of each growth and cultivation parameter individual in each initial genetic population is evaluated based on the optimization goal to obtain several fitness values; Only the growth and cultivation parameter individuals with the maximum fitness value are extracted, marked as parent individuals, and genetic crossover and mutation operations are performed on the parent individuals to generate new offspring individuals, and the new offspring individuals of each initial genetic population are searched to replace the corresponding parent individuals until the population fitness reaches a preset fitness threshold, and finally the optimal cultivation condition parameter set for microalgae growth is output; Through a closed-loop control strategy, the optimal culture condition parameter set is sent to the actual culture process of microalgae growth to adjust the environmental conditions for microalgae growth in real time, monitor the growth performance of microalgae, and feed the latest status data into the prediction model and optimization module to form a dynamic optimization closed loop.

[0030] It should be noted that after obtaining the prediction results of microalgae growth trend and metabolic yield, further optimization decisions can be made on the culture conditions based on the prediction results. However, it is difficult for traditional prediction models to automatically plan and search for the optimal culture conditions for microalgae growth based on the prediction results. Manual optimization calculations are still required with the help of human intervention. The steps are cumbersome and time-consuming, which greatly reduces the rate of microalgae growth and culture optimization. There may be large human intervention errors, which reduces the accuracy of microalgae growth and culture optimization, making it difficult for the final cultured microalgae samples to meet the expected culture requirements. Therefore, it is necessary to give the prediction model a decision-making mechanism that automatically searches for the optimal culture conditions based on the prediction results. In this regard, this method sets an optimization goal that satisfies the microalgae growth and cultivation plan based on the prediction results and the objective function, and constructs a given parameter control space of different growth and cultivation parameters through the microalgae growth and cultivation plan, and then uses a genetic algorithm to calculate the fitness of the optimization goal in the initial genetic population of the given parameter control space to perform genetic crossover and mutation analysis of the offspring individuals on the parent individuals, so that the optimal culture condition parameter set for microalgae growth can be searched out, and finally the optimal culture condition parameter set is applied to the actual cultivation process of microalgae growth to adjust the environmental conditions for microalgae growth in real time, and monitor the growth performance of microalgae to feed back the latest status data into the prediction model and optimization module. Through this method, the optimal culture condition parameters for microalgae growth can be searched based on the prediction results, thereby realizing real-time optimization of microalgae growth trends and metabolic yields, which can effectively improve microalgae cultivation efficiency, reduce operating costs, increase the yield of target microalgae products, and greatly improve economic benefits.

[0031] More specifically, the real-time monitoring of the growth performance of microalgae and the feeding back of the latest status data into the prediction model and optimization module form a dynamic optimization closed loop, which specifically includes the following steps: Through online monitoring equipment, the growth status of microalgae is continuously monitored to obtain the current growth status variables, and a closed-loop state control model of microalgae growth cultivation-prediction-optimization is established; Introducing a pole placement method, performing a closed-loop feedback calculation of a closed-loop state control model on the current growth state variable through the pole placement method, obtaining a closed-loop feedback gain matrix of the current microalgae growth state, importing the closed-loop feedback gain matrix into the closed-loop state control model for compensation control, and generating a closed-loop feedback controller of the current microalgae growth state; The closed-loop feedback controller of the current microalgae growth state is deployed in the actual microalgae growth cultivation-prediction-optimization process to obtain the latest relevant state data of the current microalgae growth, and finally the latest relevant state data is fed back to the prediction model and optimization module of the microalgae growth prediction model to form a dynamic optimization closed loop.

[0032] It should be noted that since the growth of microalgae is a dynamic process, its growth state is constantly changing, and the cultivation and prediction of microalgae growth to the optimization of growth and cultivation is a relatively complete closed-loop systematic process. Therefore, this method constructs a closed-loop state control model of the process, and then uses the pole configuration method to perform closed-loop control feedback on the current growth state variable, so as to obtain some performance parameter gains required for feedback of the current growth state, that is, the closed-loop feedback gain matrix; then the closed-loop feedback gain matrix is ​​injected into the closed-loop state control model for compensation control calculation, so as to construct a closed-loop feedback controller of the current microalgae growth state. The closed-loop feedback controller can ensure the stability of the closed-loop system through the feedback gain of the current growth state, so that the prediction model is more accurate and stable for the dynamic optimization of microalgae growth and cultivation. Finally, the closed-loop feedback controller is deployed in the actual microalgae growth and cultivation-prediction-optimization process to obtain the latest relevant state data of the current microalgae growth, and the latest relevant state data is fed back to the prediction model and optimization module of the microalgae growth prediction model. This method can form a dynamic optimization closed loop for the prediction and optimization of microalgae growth and cultivation, so as to improve the prediction model and optimize the cultivation strategy in real time according to the current growth status of the microalgae, thereby realizing automatic or semi-automatic optimization of the cultivation conditions, which is more effective in improving the microalgae yield and the synthesis efficiency of the target metabolites compared with traditional prediction and optimization methods.

[0033] The second aspect of the present invention provides a microalgae growth prediction and optimization system based on machine learning, such as Figure 3As shown, the microalgae growth prediction and optimization system includes a memory 31 and a processor 32. The memory 31 stores a microalgae growth prediction and optimization method program based on machine learning. When the microalgae growth prediction and optimization method program is executed by the processor 32, any one of the steps of the microalgae growth prediction and optimization method is implemented.

[0034] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting and optimizing microalgae growth based on machine learning, characterized in that: The following steps are involved: S102: obtaining multidimensional environmental parameters and microalgae growth parameters in the microalgae cultivation process through online monitoring equipment, randomly extracting characteristic samples after preprocessing the multidimensional environmental parameters and microalgae growth parameters, and performing recursive splitting and superposition construction on the characteristic samples to obtain a characteristic variable random forest model of the multidimensional environmental parameters and microalgae growth parameters; S104: Calculating the feature importance of each feature node for the growth of microalgae based on the feature node distribution diagram and node values ​​of the feature variable random forest model, analyzing the feature importance to screen out a feature variable data set after dimensionality reduction screening that significantly affects the growth of microalgae, and performing regression training of potential principal component variables on the feature variable data set through a machine learning algorithm to obtain a trained microalgae growth prediction model; S106: obtaining real-time microalgae growth environment parameters of the actual microalgae cultivation process, predicting the real-time microalgae growth environment parameters through the trained microalgae growth prediction model to obtain actual prediction results, obtaining multiple actual prediction results and expected prediction results under dynamic changes in cultivation conditions, and performing sliding average calculation of the Bezier curve, and updating the learning rate of the prediction model based on the sliding average; S108: Obtain the objective function of microalgae growth and cultivation, set the optimization goal that satisfies the microalgae growth and cultivation plan according to the objective function, and preset the established parameter control space of different growth and cultivation parameters, search the established parameter control space in the genetic algorithm based on the optimization goal, and obtain the optimal cultivation condition parameter set, send the optimal cultivation condition parameter set to the actual cultivation process of microalgae growth, and feed back the latest status data of microalgae growth performance into the prediction model and optimization module to form a dynamic optimization closed loop.

2. The method for predicting and optimizing microalgae growth based on machine learning according to claim 1, characterized in that: The step S102 specifically includes the following steps: The microalgae cultivation environment is monitored in real time through online and offline sensors installed on the online monitoring equipment to obtain multi-dimensional environmental parameters and microalgae growth parameters during the microalgae cultivation process; Importing a data preprocessing process, performing outlier detection, missing value interpolation and normalization processing on the multidimensional environmental parameters and microalgae growth parameters in the microalgae cultivation process through the data preprocessing process, to obtain preprocessed multidimensional environmental parameters and microalgae growth parameters; Randomly extracting characteristic samples of the pre-processed multidimensional environmental parameters and microalgae growth parameters to construct M Bootstrap characteristic variable sample sets; establishing a corresponding tree root node for each Bootstrap characteristic variable sample set, and presetting the optimal splitting feature and the maximum splitting depth; Based on the best split feature, recursive splitting operation of feature nodes is performed on the Bootstrap feature variable sample set starting from the root node of the tree to determine whether the current split depth reaches the maximum split depth; If it is not reached, continue to recursively split the feature nodes of the Bootstrap feature variable sample set; if it is reached, stop the recursive splitting operation and generate a sub-decision tree for each Bootstrap feature variable sample set; finally, superimpose the sub-decision trees of all Bootstrap feature variable sample sets to obtain a random forest model of feature variables of multidimensional environmental parameters and microalgae growth parameters.

3. The method for predicting and optimizing microalgae growth based on machine learning according to claim 1, characterized in that: The step S104 specifically includes the following steps: Extracting the characteristic node distribution graph of the characteristic variable random forest model and the node value of each characteristic node, calculating the initial error of the characteristic variable random forest model under the condition of the characteristic node distribution graph based on the node value of each characteristic node, and obtaining the initial mean square error of each characteristic node for the characteristic variable random forest model; Obtain the microalgae growth prediction demand, preset the disturbance threshold of each feature node according to the microalgae growth prediction demand, perturb each feature node in the feature variable random forest model through the disturbance threshold and recalculate the error, and obtain the perturbation mean square error of the feature variable random forest model after perturbing each feature node; Calculating a mean square error increment between the disturbance mean square error and the initial mean square error, and determining the feature importance of each feature node to the growth of microalgae according to the mean square error increment; If the mean square error amplitude is negative, it means that the feature node has little influence on the prediction performance of the feature variable random forest model, that is, the feature importance is low, then the feature node corresponding to the disturbance mean square error is removed; If the mean square error amplitude is positive, it means that the feature node has a greater impact on the prediction performance of the feature variable random forest model, which means that the feature importance is high. Then the feature node corresponding to the perturbation mean square error is extracted and marked as a high-importance feature node; Integrate the data corresponding to all high-importance feature nodes retained in the pre-processed multidimensional environmental parameters and microalgae growth parameters to obtain a feature variable data set after dimensionality reduction screening; A partial least squares regression algorithm is introduced, and latent principal component variable regression training of input feature-output feature is performed on the characteristic variable data set through the partial least squares regression algorithm to obtain a trained microalgae growth prediction model.

4. The method for predicting and optimizing microalgae growth based on machine learning according to claim 3, characterized in that: The partial least squares regression algorithm is introduced, and the input feature-output feature potential principal component variable regression training of the feature variable data set is performed by the partial least squares regression algorithm to obtain a trained microalgae growth prediction model, which specifically includes the following steps: Taking environmental parameters as input features and microalgae growth trends and metabolic yields as output features, an input feature-output feature covariance matrix is ​​calculated and constructed based on the feature variable data set after dimension reduction screening; The partial least squares regression algorithm is introduced, the maximum covariance distance threshold is preset, and the covariance distance between each input feature and each output feature in the input feature-output feature covariance matrix is ​​calculated by the partial least squares regression algorithm to obtain the covariance distance of each matrix node in the input feature-output feature covariance matrix; For each matrix node in the input feature-output feature covariance matrix, if the covariance distance is less than the maximum covariance distance threshold, the matrix node corresponding to the covariance distance is ignored; if the covariance distance is greater than the maximum covariance distance threshold, the matrix node corresponding to the covariance distance is marked and defined as the maximum correlation matrix node; Extract the input features and output features associated with the nodes of the maximum correlation matrix, define them as a set of potential principal component variables, and preset the critical value of the number of potential principal component variables to be extracted; Repeat the above steps of calculating and determining the covariance distance and extracting potential principal component variables from the input feature-output feature covariance matrix. Each time a group of potential principal component variables is extracted, the relationship between the iterative input features and the output features is updated to obtain the current number of extracted potential principal component variables. If the current extraction quantity does not reach the extraction quantity critical value, the covariance distance calculation and determination of the input feature-output feature covariance matrix and the extraction of potential principal component variables are continued; if the current extraction quantity has reached the extraction quantity critical value, the extraction of potential principal component variables is stopped to obtain the extracted potential principal component variable set; The extracted potential principal component variable set is used to construct a regression model between input features and output features in a partial least squares regression algorithm, and a regression training operation is performed on the training set through the regression model to obtain a trained microalgae growth prediction model.

5. The method for predicting and optimizing microalgae growth based on machine learning according to claim 1, characterized in that: The step S106 specifically includes the following steps: Acquire real-time microalgae growth environment parameters during the actual cultivation process of microalgae, and import the real-time microalgae growth environment parameters into the trained microalgae growth prediction model to perform growth prediction in a future time period, so as to output the actual microalgae growth prediction trend and actual metabolic prediction yield in a specific future time period; Acquire the actual microalgae growth prediction trend and actual metabolic prediction yield output by the microalgae growth prediction model when predicting dynamic changes in culture conditions on a preset culture time sequence, and obtain multiple actual prediction results when culture conditions change dynamically; A Bezier curve method is introduced, and a plurality of the actual prediction results are subjected to curve fitting by the Bezier curve method to obtain a Bezier curve of the actual prediction result, which is defined as a first Bezier curve; the expected prediction result under the dynamic change of the culture condition is simultaneously obtained, and a Bezier curve of the expected prediction result is constructed, which is defined as a second Bezier curve; If the slope deviation value between the slope of the first Bezier curve and the slope of the second Bezier curve is greater than the preset slope deviation value, the historical gradient square of the trained microalgae growth prediction model in the dynamic change process of the culture conditions is tracked based on the slope deviation value and exponentially weighted to obtain a sliding average value; During the tracking process, the current learning rate of the microalgae growth prediction model is obtained, and the current learning rate is adjusted and updated through a sliding average value, so that the trained microalgae growth prediction model can adapt to dynamically changing culture conditions.

6. The method for predicting and optimizing microalgae growth based on machine learning according to claim 1, characterized in that: The step S108 specifically includes the following steps: Obtaining a microalgae growth and cultivation plan, and obtaining an objective function of microalgae growth and cultivation through the microalgae growth and cultivation plan; wherein the objective function includes maximum biomass output, optimal metabolite synthesis rate, and minimum resource consumption; An optimization goal that satisfies the microalgae growth and cultivation plan is set based on the predicted growth trend of the microalgae in a specific future time period, the predicted metabolic yield, and the objective function; Extracting preset control intervals of different growth and cultivation parameters through the microalgae growth and cultivation plan, and constructing a predetermined parameter control space according to the preset control intervals of the different growth and cultivation parameters; A genetic algorithm is introduced to construct multiple initial genetic populations of a given parameter control space, and the fitness of each growth and cultivation parameter individual in each initial genetic population is evaluated based on the optimization goal to obtain several fitness values; Only the growth and cultivation parameter individuals with the maximum fitness value are extracted, marked as parent individuals, and genetic crossover and mutation operations are performed on the parent individuals to generate new offspring individuals, and the new offspring individuals of each initial genetic population are searched to replace the corresponding parent individuals until the population fitness reaches a preset fitness threshold, and finally the optimal cultivation condition parameter set for microalgae growth is output; Through a closed-loop control strategy, the optimal culture condition parameter set is sent to the actual culture process of microalgae growth to adjust the environmental conditions for microalgae growth in real time, and the microalgae growth performance is monitored in real time to feed the latest status data into the prediction model and optimization module to form a dynamic optimization closed loop.

7. The method for predicting and optimizing microalgae growth based on machine learning according to claim 6, characterized in that: The real-time monitoring of the growth performance of microalgae and the feeding back of the latest status data into the prediction model and optimization module form a dynamic optimization closed loop, specifically including the following steps: Through online monitoring equipment, the growth status of microalgae is continuously monitored to obtain the current growth status variables, and a closed-loop state control model of microalgae growth cultivation-prediction-optimization is established; Introducing a pole placement method, performing a closed-loop feedback calculation of a closed-loop state control model on the current growth state variable through the pole placement method, obtaining a closed-loop feedback gain matrix of the current microalgae growth state, importing the closed-loop feedback gain matrix into the closed-loop state control model for compensation control, and generating a closed-loop feedback controller of the current microalgae growth state; The closed-loop feedback controller of the current microalgae growth state is deployed in the actual microalgae growth cultivation-prediction-optimization process to obtain the latest relevant state data of the current microalgae growth, and finally the latest relevant state data is fed back to the prediction model and optimization module of the microalgae growth prediction model to form a dynamic optimization closed loop.

8. A microalgae growth prediction and optimization system based on machine learning, characterized in that: The microalgae growth prediction and optimization system includes a memory and a processor, wherein the memory stores a microalgae growth prediction and optimization method program based on machine learning, and when the microalgae growth prediction and optimization method program is executed by the processor, the microalgae growth prediction and optimization method steps according to any one of claims 1 to 7 are implemented.

Images