Microalgae growth curve prediction method based on transfer learning under small sample condition

Abstract

The invention discloses an innovative microalgae growth curve prediction method, which aims at a microalgae growth curve prediction problem under small sample data, combines a Box-Behnken design method, selects key environmental factors, collects microalgae growth curve samples under minimum environmental diversity, and incorporates light and dark period variables, so as to predict the microalgae growth curve. A fine microalgae growth kinetic model is constructed to simulate growth change under a natural light condition, and a large amount of high-quality simulation data is generated. The advanced Twood-Stage TrAdaboost. R2 algorithm is used for merging simulation output and scarce experiment observation data for LSTM (Long Short Term Memory) model training, and aims to enhance the capability of predicting the growth trend of microalgae in a complex illumination condition by the model and reduce the dependence of a deep learning model on large-scale field measurement data at the same time. According to the method, a new strategy for optimizing microalgae culture parameters and accurately controlling is provided for the industry, and productivity increase and efficiency improvement of the microalgae industry are powerfully promoted.

Description

Technical Field

[0001] The present invention relates to the intersection of biotechnology and deep learning, and specifically is a method for predicting the growth curve of microalgae by using transfer learning technology and deep learning technology under small sample data conditions. Background Art

[0002] Microalgae are considered as a highly promising renewable resource due to their amazing growth rate, efficient photosynthesis capacity (nearly ten times higher than terrestrial plants), and ability to synthesize important biological compounds. Based on these unique advantages, microalgae have been widely used in the production of high value-added products such as fuels, cosmetics, and nutritional products, as well as in wastewater treatment and carbon emission mitigation. The growth curve of microalgae is a comprehensive indicator to measure its population growth, reflecting the ability of microalgae to adapt to and survive in a specific environment. Accurately predicting the growth curve of microalgae is of great significance for specifying culture plans, optimizing the culture environment, and controlling microalgae process conditions.

[0003] The traditional microalgae industry mainly relies on experience summary, simple kinetic models or single machine learning models to predict the growth status of microalgae. However, these methods have great limitations, including poor generalization performance and high requirements for previous historical data. Therefore, accurate simulation of microalgae growth still faces many challenges. The growth of algae is affected by many factors (such as light intensity, culture medium concentration, pH value, temperature and carbon dioxide concentration). These factors affect microbial growth through complex metabolic mechanisms. Therefore, it is extremely challenging to capture this multi-stage nonlinear growth pattern.

[0004] Existing kinetic models, such as the Monod model, the Droop model, and the Han model, have problems such as difficulty in parameter estimation and large demand for experimental data. With the development of deep learning technology, data-driven models are gradually applied to the field of biochemical engineering. Among them, the LSTM model has shown the potential for wide application in the prediction of biological growth status due to its excellent time series data processing capabilities and long-term dependency capture capabilities. However, as a deep learning model, the accuracy of the LSTM model is highly dependent on data quality. The scarcity of experimental data and measurement errors, as well as the high cost and long cycle of biological experiments and detection, have jointly restricted the application of the LSTM model in microalgae growth prediction. Summary of the invention

[0005] The core purpose of the present invention is to overcome the limitations of traditional growth dynamics models and LSTM models in view of the many challenges faced in the task of microalgae growth prediction, such as scarce data, complex nonlinear relationships, and numerous influencing factors. The present invention is committed to solving the problems caused by large prediction deviations of microalgae growth curves, high experimental costs, and insufficient model generalization capabilities. To this end, the present invention proposes an innovative solution, namely a dynamics-assisted microalgae growth curve prediction method combined with transfer learning. This method significantly improves the accuracy of prediction results under small sample conditions by integrating a small amount of historical data and simulated data. The microalgae growth curve is predicted by combining the advantages of the dynamic model and the LSTM model. This method is expected to break through the bottleneck of the prior art, effectively reduce prediction deviations, and reduce experimental costs.

[0006] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0007] A method for predicting microalgae growth curve based on transfer learning under small sample conditions, the method comprising the following steps:

[0008] Step S1, using the Box-Behnken method to prepare the culture of the training group and the fitting group, and obtaining the experimental data of {key environmental influencing factors, biomass concentration value of the long-time series liquid microalgae system} of the training group and the fitting group, wherein the biomass concentration value of the long-time series liquid microalgae system represents the growth curve of the microalgae;

[0009] Step S2, considering the light and dark cycle, constructing a microalgae growth dynamics model based on the logistic model; using the model to perform iterative learning and training with the fitting group data in the experimental data as input to obtain an ideal model of microalgae growth dynamics;

[0010] Step S3, using the key environmental influencing factors of the training group in the test data as input, using the ideal model of microalgae growth dynamics to simulate and generate dense dynamics simulation data X″ of the biomass concentration value of the microalgae system, which is used to characterize the dynamic growth curve and output it, so as to achieve generalization and enhancement of the data volume;

[0011] Step S4, using the LSTM model as the prediction model, integrating {key environmental influencing factors, microalgae system biomass concentration values} of the test data X′ and the simulation data X″, and using the Two-Stage TrAdaboost.R2 method to iteratively train the fitting learning relationship of the prediction model to obtain an ideal prediction model, so that the ideal prediction model is used to predict the output of the corresponding microalgae system biomass concentration values ​​for the input key environmental influencing factors;

[0012] Step S5, collecting key environmental influencing factors of the water samples to be tested in a long time series, inputting the trained ideal prediction model, automatically predicting and outputting the biomass concentration value of the microalgae system in a long time series, and characterizing the growth trend of the microalgae in a long time series.

[0013] The acquisition of test data in step S1 includes:

[0014] Step S11, collecting the long-time series OD values ​​of the liquid microalgae system of the training group and the fitting group;

[0015] Step S12, fitting the calibration curve and regression relationship F between the OD value and dry weight of microalgae using the fitting group data;

[0016] Step S13, using F to calibrate the microalgae OD value of the training group, obtaining its corresponding calibrated biomass concentration value X′, and establishing {key environmental influencing factors, biomass concentration of long-term liquid microalgae system} for the training group and the fitting group.

[0017] The step S11 comprises:

[0018] Step S11a, using culture medium and combining the variable range of key environmental factors to prepare multiple training groups of microalgae culture solutions, and collecting OD values ​​of long-term liquid microalgae systems under multiple groups of different variable environments by absorbance method;

[0019] Step S11b: preparing a fitting group of microalgae culture solutions using culture medium and in combination with standard key environmental factors, and collecting OD values ​​of a long-time series of liquid microalgae systems under a standard variable environment using an absorbance method.

[0020] The key environmental factors include nitrogen concentration, phosphorus concentration, light intensity and light-dark ratio; the nitrogen concentration range is 335 mg·L -1 -385mg·L -1 , phosphorus concentration range is 6.4 mg·L -1 -25mg·L -1 , light intensity range: 26 μmol·m -2 ·s -1 -78 μmol·m -2 ·s -1 , the light-dark ratio range is 12:12-24:0;

[0021] The method of using culture medium and combining standard key environmental factors to prepare multiple matching groups of microalgae culture solutions includes: setting the nitrogen concentration range to 335 mg·L -1 -385mg·L -1 , phosphorus concentration range is 6.4 mg·L -1 -25mg·L -1 , light-dark ratio of 14:10, light intensity of 26 μmol·m -2·s -1 The culture was carried out for 120 hours until the microalgae culture solution entered a stationary phase.

[0022] The microalgae growth curve data is collected by using an ultraviolet-visible spectrophotometer to detect the microalgae absorbance data OD at a wavelength of 750nm according to the sampling interval. 750 ; The microalgae growth curve data under the standard variable environment is the OD value detected every 4 hours.

[0023] The calibration curve and regression relationship between the OD value and dry weight of the fitted microalgae are as follows: preparing a microalgae culture solution of the fitted group into a microalgae dry powder, calculating the microalgae biomass concentration X according to the mass and volume of the dry powder, and characterizing the long-term growth curve data of the solid microalgae of the fitted group; and obtaining a calibration curve and regression relationship F between the OD value and the dry weight of the microalgae according to the OD value and the microalgae biomass concentration X at the corresponding OD.

[0024] Step S2 includes: step S21, establishing a microalgae growth dynamics model including sequentially interlacing and connecting a plurality of light cycle models and dark cycle models;

[0025] a) Taking the key environmental factors affecting microalgae cultivation as input variables and the microalgae biomass concentration as output variables, a growth dynamics loop model is constructed, as follows: Introducing factors related to light, nitrogen, and phosphorus, and acting on the maximum specific growth rate μ in the logistic model in the form of a multiplication model max , as shown in equations (1)-(2); the light intensity is modeled in the form of the Aiba equation, as shown in equation (3); the nitrogen and phosphorus elements are modeled in the form of the Andrews equation, as shown in equations (4)-(5); the attenuation law of light propagating along the medium is described according to the Lambert-Beer law, as shown in equation (6);

[0026]

[0027] μ max =μ m ·f(I)·f(N)·f(P) (2)

[0028]

[0029]

[0030] I=I0exp(-K a XZ) (6)

[0031] Where X(t) is the biomass concentration of microalgae at time t, in g·L -1 ;μ max Represents the maximum specific growth rate, unit: h -1 ;X maxThe maximum biomass concentration that the environment can accommodate, in g L -1 ;μ d Represents the attenuation coefficient, which indicates the rate of decay of the microalgae cell population in the environment, in h -1 ;

[0032] Where f(I), f(N) and f(P) represent the changes of light, nitrogen and phosphorus respectively; m is the specific growth rate constant, unit is h -1 ; I represents light intensity, unit: μmol·m -2 ·s -1 ; K I,i The photoinhibition constant is used to indicate the phenomenon that excessive light intensity inhibits photosynthesis. The unit is μmol·m -2 ·s -1 ; K I,s Light saturation constant, used to indicate that higher light intensity does not significantly improve photosynthesis, unit: μmol·m -2 ·s -1 ; N and P are the concentrations of nitrogen and phosphorus, in mg·L -1 ; K N,i ,K P,i They are the inhibition constants of nitrogen and phosphorus, in mg·L -1 ; K N,s ,K P,s They are the saturation constants of nitrogen and phosphorus, in mg·L -1 ;

[0033] Where I0 represents the incident light intensity in μmol·m -2 ·s -1 , Ka is the light attenuation coefficient, unit is L·m -1 ·g -1 , Z is the distance behind the irradiated area, in m -1 ;

[0034] In order to reduce the influence of spatial dimension on the solution of the equation, the ten-step trapezoidal rule is used to transform equation (6) into equation (7):

[0035]

[0036] The substrate consumption rate can be calculated using the mass conservation equation:

[0037]

[0038] Among them, Y XN ,Y XP is the yield coefficient of biomass production relative to the substrate; m N ,m Pis the specific consumption rate maintained by the cell.

[0039] b) Consider the light cycle process and the dark cycle process separately, and use formulas (1)-(5), (7)-(10) as the light cycle model;

[0040] Formulas (1), (4), (5), (9)-(11) are used as dark cycle models; and the maximum specific growth rate is expressed as follows:

[0041] μ max =μ m ·f(N)·f(P) (12)

[0042] Step S2 also includes: step S22, using the model to perform iterative learning with key environmental influencing factors as input, including adopting a two-stage parameter fitting method and fitting parameters in combination with a differential evolution algorithm; including the following steps:

[0043] Step S221, setting the light-dark ratio, dividing the fitting group data into light cycle part data and dark cycle part data; each part of the data is {key environmental influencing factors, long-time series liquid microalgae system concentration};

[0044] Step S222, initializing the light cycle model parameters, fixing the dark cycle model parameters, using the light cycle partial data in step S221, only performing parameter fitting on the light cycle model, and obtaining the kinetic parameters of the light cycle model;

[0045] Step S223, fixing the kinetic parameters of the light cycle model obtained in step S222, and using the dark cycle partial data in step S221 to perform parameter fitting on the dark cycle model to obtain the kinetic parameters of the dark cycle model.

[0046] The step S4 comprises the following steps:

[0047] Step S41, splicing the experimental data and dynamic simulation data under the training set culture conditions, wherein the experimental data is used as the target domain data T in transfer learning target =(X exp ,Y exp ), the simulation data is used as the source domain data T source =(X model ,Y model ); splicing the two is T = {(X model ,X exp ),(Y model ,Y exp )};

[0048] Step S42: construct an LSTM model as a base learner and use the Two-Stage TrAdaboost.R2 method for training.

[0049] The Two-Stage TrAdaBoost.R2 algorithm in step S42 includes the following steps:

[0050] Step S421, set the number of iterations S, t=1, ..., S, and set the initial sample weight:

[0051] Step S422: Update the target domain instance according to the weight update strategy of the Adaboost.R2 algorithm; keep the source domain weight unchanged, and gradually reduce the weight of all target data sets T source The weights of the samples in the training set T and their proportion in the total weight ω1, and the optimal solution of the weight ratio is obtained through cross-validation; source The weights of the samples in the training set are updated by calling the base regressor (LSTM) to obtain a learner on the combined training set T, and then calculating the weight adjustment error e for each sample. i t , downgrade according to the error value;

[0052] Step S423: Based on the determined weight ratio, update the target domain instance according to the weight update strategy of the Adaboost.R2 algorithm; perform weighted processing on the data points with large errors; in this stage, the auxiliary training set T source The weights of the samples in the target training set T remain unchanged; through cross-validation, find target The model with the smallest regression error value is taken as the final training result;

[0053] The weight update rules are as follows:

[0054]

[0055] Z t is the normalization constant, β t The goal of the setting is to balance the proportion of source domain data and the overall weight ω1 through binary search. The goal of the search is to make the overall weight of the target domain equal to This target is not set for individual sample weights.

[0056] Step S424: output the model with the smallest error t , t = argmin(error t ).

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] (1) Based on the logistic model, factors affecting the growth of microalgae, such as light intensity, nitrogen concentration, and phosphorus concentration, were incorporated into the mathematical model in the form of a multiplication model;

[0059] (2) Use a multi-stage modeling approach to effectively simulate the growth process of microalgae under the influence of light-dark cycles and improve the prediction accuracy of the kinetic model under natural light-dark cycle environments;

[0060] (3) The mathematical models of the light cycle and dark cycle were fitted separately by a two-stage parameter fitting method, which solved the problem that all models were non-differentiable due to multi-stage modeling, resulting in the inability to fit the kinetic parameters by conventional methods;

[0061] (4) Since the state at any time during the growth of microalgae is related to the previous state, the LSTM model is used as the prediction model to effectively capture the temporal relationship of microalgae growth and improve the prediction accuracy of the model.

[0062] (5) The present invention cleverly utilizes the transfer learning strategy of the Two-Stage TrAdaBoost.R2 algorithm to extract valuable information from the rich simulation data generated by the kinetic model, supplement and amplify the value of the limited actual microalgae cultivation experimental data, thereby significantly improving the prediction accuracy of the microalgae growth dynamics without relying heavily on new experimental data. This method not only optimizes the efficiency of data utilization, but also enhances the generalization ability of the model for unseen data by virtue of the transferred knowledge, effectively alleviates the problem of data acquisition difficulties in the application of deep learning models in the industry, and expands the prediction boundary of the model under large data volumes. Therefore, the present invention not only improves the prediction effect, but also greatly alleviates the time and economic cost pressure of biological experiments, and provides an efficient and economical solution for the optimization of microalgae culture conditions and the prediction of biomass production. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a work flow chart of the present invention;

[0064] Figure 2 Run the process for multi-stage models;

[0065] Figure 3 The process of training the model using the Two-stage TrAdaboost.R2 algorithm;

[0066] Figure 4 The prediction effect diagram of microalgae growth curve under different situations of the present invention; DETAILED DESCRIPTION

[0067] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation method of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the invention, so the present invention is not limited by the specific implementation disclosed below.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0069] (1) Based on experience, the microalgae growth curve was obtained as the fitting group data: the designed nitrogen concentration was 335 mg·L -1 , phosphorus concentration is 6.4 mg L -1 , the light intensity range is 26 μmol·m -2 ·s -1 , the light-dark ratio was 14:10. The OD value of the fitting group data was detected every 4 hours using the absorbance method, and the sample was made into algae powder, and the corresponding dry weight was obtained as the biomass concentration value:

[0070] Take two 2 ml centrifuge tubes, dry them in an oven at 80 °C for 24 hours, and weigh the centrifuge tube weight m1 on a high-precision balance. At the same time, take 4 ml of algae solution and put it into a centrifuge tube, centrifuge it at 3000 r / min for 13 minutes, wash it with ultrapure water, repeat three times, and absorb the supernatant. Dry it in an oven at 80 °C for 24 hours, and weigh the centrifuge tube and algae powder weight m2 on a high-precision balance. Calculate the dry weight of microalgae at the corresponding OD according to formula (1).

[0071]

[0072] In formula (1), X is the biomass concentration of microalgae, in mg·L -1 .

[0073] The regression relationship between the OD value and the biomass concentration value is calibrated. The OD value is then uniformly converted into the biomass concentration of the microalgae system through this relationship;

[0074] (2) According to the Box-Behnken design method, the initial microalgae growth curve data were obtained as the source data of the training group and the fitting group data: nitrogen concentration, phosphorus concentration, light intensity and light-dark ratio were selected as key environmental factors, and the variable range was designed based on literature and experience. The nitrogen concentration range was designed to be 335 mg·L -1 -385mg·L -1, phosphorus concentration range is 6.4 mg L -1 -25mg·L -1 , the light intensity range is 26 μmol·m -2 ·s -1 -78 μmol m -2 ·s -1 , the light-dark ratio range is 12:12-24:0;

[0075] (3) Considering the light-dark cycle, a microalgae growth dynamics model based on the logistic model was constructed;

[0076] The microalgae cultivation involved in the present invention is applied in an industrial environment that controls the light-dark ratio. Therefore, the light-dark ratio needs to be incorporated into the modeling so that the subsequent transfer learning model can learn the dynamic process of the model under the light-dark cycle in a small sample. The light cycle model is as follows:

[0077]

[0078] μ max =μ m ·f(I)·f(N)·f(P) (3)

[0079]

[0080] Compared with the light cycle model, the dark cycle model cancels the influence of light intensity, which is only reflected in the maximum specific growth rate formula:

[0081] μ max =μ m ·f(N)·f(P) (10)

[0082] The model operation mechanism is to alternately operate the light cycle model and the dark cycle model, and the output of the former model serves as the input of the latter model. For detailed operation process, please refer to Figure 2 .

[0083] In view of the fact that many empirical parameters in the kinetic model are difficult to determine directly, the parameter calibration link driven by experimental data is particularly important. Considering that the model construction involves the dual mechanism of light cycle and dark cycle, the present invention adopts the differential evolution algorithm as the parameter fitting optimization method. The differential evolution algorithm is an efficient non-gradient optimization strategy that aims to accurately optimize these key parameters by minimizing the mean square error between the experimental observation value and the model simulation output. In order to further improve the accuracy and efficiency of parameter estimation, the present invention implements a parameter optimization strategy carried out in two stages to ensure that the model can more closely reflect the actual situation.

[0084] (4) Based on the designed experiment, use the kinetic model to generate complete kinetic curve data under corresponding conditions;

[0085] (5) Use the Two-Stage TrAdaboost.R2 method to integrate experimental data and simulation data to train the LSTM model;

[0086] (5.1) Splice the experimental data and dynamic simulation data under the training set culture conditions. The experimental data is used as the target domain data T in transfer learning. target =(X exp ,Y exp ), the simulation data is derived from the data T source =(X model ,Y model ); splicing the two is T = {(X model ,X exp ),(Y model ,Y exp )}.

[0087] (5.2) Use LSTM model as base learner. Use Two-Stage TrAdaboost.R2 method for training;

[0088] In order to compensate for the impact of data quality on model training in basic LSTM model training, the present invention adopts the idea of ​​transfer learning and uses the Two-Stage TrAdaBoost.R2 algorithm to fuse a large amount of simulation data with experimental data that is similar to the actual data distribution. The Two-Stage TrAdaBoost.R2 algorithm in step (5.2) includes the following steps. The specific process is shown in Figure 3 .

[0089] (5.2.1) Set the number of iterations S, t = 1, ..., S, and set the initial sample weights:

[0090] (5.2.2) Update the target domain instance according to the weight update strategy of the Adaboost.R2 algorithm. Keep the first n weights, that is, the source domain weights unchanged, and gradually reduce all source domain data sets T source The weights of the samples in the training set T and their proportion in the total weight ω1 are calculated, and the optimal solution of the weight ratio is obtained through cross-validation. This stage aims to determine the target training set T target The appropriate sum of weights for the instances in the target training set T target The amount of data is much smaller than the auxiliary training set T source The target training set T target The initial weight is too low, and the auxiliary training set T sourceThe weights of the samples in the training set are updated by calling the base regressor to obtain a learner on the combined training set T, and then calculating the weight adjustment error of each sample. Downgrade according to the error value;

[0091] (5.2.3) Based on the determined weight ratio, the target domain instance is updated according to the weight update strategy of the Adaboost.R2 algorithm. Weight distribution, that is, weighted processing of data points with large errors. In this stage, the auxiliary training set T source The weights of the samples in the target training set T remain unchanged; only the hypotheses generated in this stage will be stored and used to determine the output of the resulting model; through cross-validation, find target The model with the smallest regression error value is taken as the final training result. The weight update rule is as follows:

[0092]

[0093] Z t is the normalization constant, β t The goal of the setting is to balance the proportion of source domain data and the overall weight ω1 through binary search. The goal of the search is to make the overall weight of the target domain equal to This target is not set for individual sample weights.

[0094] (5.2.4) The model with the smallest output error, model t , t = argmin(error t ).

[0095] (6) Model prediction: Use the trained LSTM model to predict new input data (including light intensity, light-dark ratio, N element concentration, and P element concentration) to obtain the prediction results of the microalgae growth curve.

[0096] The training group data is split according to the 5-fold validation and 2-fold validation methods, while ensuring that the test set is the same, so that the model is trained under two training sets of different sizes. The former is used as a case where the data set is relatively sufficient, and the latter is used as a case where the sample size is small. The unmodified LSTM and the Trans-LSTM that has been improved by the method of the present invention are selected for result comparison, and a model improvement rate index IMR is defined to characterize the ratio of the performance improvement of our method on the model under the same training data. The key environmental factors in the test set are used as inputs to the four trained models to obtain the predicted growth curve results, such as Figure 4As shown in the figure, (a) (b) (c) (d) (e) are the prediction results of the trained model under five typical growth conditions. Figure (a) is the case of low light intensity and low nutrient element concentration; Figure (b) is the case of high light intensity and low light-dark ratio; Figure (c) is the case of medium light intensity; Figure (d) is the case of high nutrient element concentration and no dark cycle; Figure (e) is the case of moderate all key factors. It can be seen that in most cases, the original LSTM model and the Trans-LSTM model can well predict the growth curve of microalgae, which is due to the time series prediction characteristics of LSTM. However, in the case of Figure (a), although the LSTM model under small sample conditions effectively predicts the growth trend of microalgae, the accuracy is insufficient. In the case of Figure (b), the LSTM model obviously failed to learn the characteristics of the fluctuation of biomass concentration of microalgae in the late growth period when the light-dark ratio is low, and the Trans-LSTM model can capture such features because it has a large amount of relevant simulation data as an enhancement means. The overall performance is shown in Table 1. The results show that the present invention can effectively improve the prediction accuracy of the microalgae growth curve under small sample conditions.

[0097]

[0098] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting microalgae growth curve based on transfer learning under small sample conditions, characterized by: The method comprises the following steps: Step S1, using the Box-Behnken method to prepare the culture of the training group and the fitting group, and obtaining the experimental data of {key environmental influencing factors, biomass concentration value of the long-time series liquid microalgae system} of the training group and the fitting group, wherein the biomass concentration value of the long-time series liquid microalgae system represents the growth curve of the microalgae; Step S2, considering the light and dark cycle, constructing a microalgae growth dynamics model based on the logistic model; using the model to perform iterative learning and training with the fitting group data in the experimental data as input to obtain an ideal model of microalgae growth dynamics; Step S3, using the key environmental influencing factors of the training group in the test data as input, using the ideal model of microalgae growth dynamics to simulate and generate dense dynamics simulation data X″ of the biomass concentration value of the microalgae system, which is used to characterize the dynamic growth curve and output it, so as to achieve generalization and enhancement of the data volume; Step S4, using the LSTM model as the prediction model, integrating {key environmental influencing factors, microalgae system biomass concentration values} of the test data X′ and the simulation data X″, and using the Two-Stage TrAdaboost.R2 method to iteratively train the fitting learning relationship of the prediction model to obtain an ideal prediction model, so that the ideal prediction model is used to predict the output of the corresponding microalgae system biomass concentration values ​​for the input key environmental influencing factors; Step S5, collecting key environmental influencing factors of the water samples to be tested in a long time series, inputting the trained ideal prediction model, automatically predicting and outputting the biomass concentration value of the microalgae system in a long time series, and characterizing the growth trend of the microalgae in a long time series.

2. The microalgae growth curve prediction method based on transfer learning under small sample conditions according to claim 1, characterized in that: The acquisition of test data in step S1 includes: Step S11, collecting the long-time series OD values ​​of the liquid microalgae system of the training group and the fitting group; Step S12, fitting the calibration curve and regression relationship F between the OD value and dry weight of microalgae using the fitting group data; Step S13, using F to calibrate the microalgae OD value of the training group, obtaining its corresponding calibrated biomass concentration value X′, and establishing {key environmental influencing factors, biomass concentration of long-term liquid microalgae system} for the training group and the fitting group.

3. The microalgae growth curve prediction method based on transfer learning under small sample conditions according to claim 1, characterized in that: The step S11 comprises: Step S11a, using culture medium and combining the variable range of key environmental factors to prepare multiple training groups of microalgae culture solutions, and collecting OD values ​​of long-term liquid microalgae systems under multiple groups of different variable environments by absorbance method; Step S11b: preparing a fitting group of microalgae culture solutions using culture medium and in combination with standard key environmental factors, and collecting OD values ​​of a long-time series of liquid microalgae systems under a standard variable environment using an absorbance method.

4. The microalgae growth curve prediction method based on transfer learning under small sample conditions according to claim 3, characterized in that: The key environmental factors include nitrogen concentration, phosphorus concentration, light intensity and light-dark ratio; the nitrogen concentration range is 335 mg·L -1 -385mg·L -1 , phosphorus concentration range is 6.4 mg·L -1 -25mg·L -1 , light intensity range: 26 μmol·m -2 ·s -1 -78 μmol·m -2 ·s -1 , the light-dark ratio range is 12:12-24:0; The method of using culture medium and combining standard key environmental factors to prepare multiple matching groups of microalgae culture solutions includes: setting the nitrogen concentration range to 335 mg·L -1 -385mg·L -1 , phosphorus concentration range is 6.4 mg·L -1 -25mg·L -1 , light-dark ratio of 14:10, light intensity of 26 μmol·m -2 ·s -1 The culture was carried out for 120 hours until the microalgae culture solution entered a stationary phase.

5. The method for predicting microalgae growth curve based on transfer learning under small sample conditions according to claim 3, characterized in that: The microalgae growth curve data is collected by using an ultraviolet-visible spectrophotometer to detect the microalgae absorbance data OD at a wavelength of 750nm according to the sampling interval. 750 ; The microalgae growth curve data under the standard variable environment is the OD value detected every 4 hours.

6. The method for predicting microalgae growth curve based on transfer learning under small sample conditions according to claim 3, characterized in that: The calibration curve and regression relationship between the OD value and dry weight of the fitted microalgae are as follows: preparing a microalgae culture solution of the fitted group into a microalgae dry powder, calculating the microalgae biomass concentration X according to the mass and volume of the dry powder, and characterizing the long-term growth curve data of the solid microalgae of the fitted group; and obtaining a calibration curve and regression relationship F between the OD value and the dry weight of the microalgae according to the OD value and the microalgae biomass concentration X at the corresponding OD.

7. The method for predicting microalgae growth curve based on transfer learning under small sample conditions according to claim 1, characterized in that: Step S2 includes: step S21, establishing a microalgae growth dynamics model including sequentially interlacing and connecting a plurality of light cycle models and dark cycle models; a) Taking the key environmental factors affecting microalgae cultivation as input variables and the microalgae biomass concentration as output variables, a growth dynamics loop model is constructed, as follows: Introducing factors related to light, nitrogen, and phosphorus, and acting on the maximum specific growth rate μ in the logistic model in the form of a multiplication model max , as shown in equations (1)-(2); the light intensity is modeled in the form of the Aiba equation, as shown in equation (3); the nitrogen and phosphorus elements are modeled in the form of the Andrews equation, as shown in equations (4)-(5); the attenuation law of light propagating along the medium is described according to the Lambert-Beer law, as shown in equation (6); m max =μ m ·f(I)·f(N)·f(P) (2) I=I0exp(-K a XZ) (6) Where X(t) is the biomass concentration of microalgae at time t, in g·L -1 ;μ max Represents the maximum specific growth rate, unit: h -1 ;X max is the maximum biomass concentration that the environment can accommodate, in g L -1 ;μ d Represents the attenuation coefficient, which indicates the rate of decay of the microalgae cell population in the environment, in h -1 ; Where f(I), f(N) and f(P) represent the changes of light, nitrogen and phosphorus respectively; m is the specific growth rate constant, unit is h -1 ; I represents light intensity, unit: μmol·m -2 ·s -1 ; K I,i The photoinhibition constant is used to indicate the phenomenon that excessive light intensity inhibits photosynthesis. The unit is μmol·m -2 ·s -1 ; K I,s Light saturation constant, used to indicate that higher light intensity does not significantly improve photosynthesis, unit: μmol·m -2 ·s -1 ; N and P are the concentrations of nitrogen and phosphorus, in mg·L -1 ; K N,i ,K P,i They are the inhibition constants of nitrogen and phosphorus, in mg·L -1 ; K N,s ,K P,s They are the saturation constants of nitrogen and phosphorus, in mg·L -1 ; Where I0 represents the incident light intensity in μmol·m -2 ·s -1 , Ka is the light attenuation coefficient, unit is L·m -1 ·g -1 , Z is the distance behind the irradiated area, in m -1 ; In order to reduce the influence of spatial dimension on the solution of the equation, the ten-step trapezoidal rule is used to transform equation (6) into equation (7): The substrate consumption rate can be calculated using the mass conservation equation: Among them, Y X / N ,Y XP is the yield coefficient of biomass production relative to the substrate; m N ,m P is the specific consumption rate maintained by the cell. b) Consider the light cycle process and the dark cycle process separately, and use formulas (1)-(5), (7)-(10) as the light cycle model; Formulas (1), (4), (5), (9)-(11) are used as dark cycle models; and the maximum specific growth rate is expressed as follows: m max =μ m ·f(N)·f(P) (12) 8. The method for predicting microalgae growth curve based on transfer learning under small sample conditions according to claim 1, characterized in that: Step S2 also includes: step S22, using the model to perform iterative learning with key environmental influencing factors as input, including adopting a two-stage parameter fitting method and fitting parameters in combination with a differential evolution algorithm; including the following steps: Step S221, setting the light-dark ratio, dividing the fitting group data into light cycle part data and dark cycle part data; each part of the data is {key environmental influencing factors, long-time series liquid microalgae system concentration}; Step S222, initializing the light cycle model parameters, fixing the dark cycle model parameters, using the light cycle partial data in step S221, only performing parameter fitting on the light cycle model, and obtaining the kinetic parameters of the light cycle model; Step S223, fixing the kinetic parameters of the light cycle model obtained in step S222, and using the dark cycle partial data in step S221 to perform parameter fitting on the dark cycle model to obtain the kinetic parameters of the dark cycle model.

9. The method for predicting microalgae growth curve based on transfer learning under small sample conditions according to claim 1, characterized in that: The step S4 comprises the following steps: Step S41, splicing the experimental data and dynamic simulation data under the training set culture conditions, wherein the experimental data is used as the target domain data T in transfer learning target =(X exp ,Y exp ), the simulation data is used as the source domain data T source =(X model ,Y model ); splicing the two is T = {(X model ,X exp ),(Y model ,Y exp )}; Step S42: construct an LSTM model as a base learner and use the Two-Stage TrAdaboost.R2 method for training.

10. The method for predicting microalgae growth curve based on transfer learning under small sample conditions according to claim 9, characterized in that: The Two-Stage TrAdaBoost.R2 algorithm in step S42 includes the following steps: Step S421, set the number of iterations S, t=1, ..., S, and set the initial sample weight: Step S422: Update the target domain instance according to the weight update strategy of the Adaboost.R2 algorithm; keep the source domain weight unchanged, and gradually reduce the weight of all target data sets T source The weights of the samples in the training set T and their proportion in the total weight ω1, and the optimal solution of the weight ratio is obtained through cross-validation; source The weights of the samples in the training set are updated by calling the base regressor (LSTM) to obtain a learner on the combined training set T, and then calculating the weight adjustment error e for each sample. i t , downgrade according to the error value; Step S423: Based on the determined weight ratio, update the target domain instance according to the weight update strategy of the Adaboost.R2 algorithm; perform weighted processing on the data points with large errors; in this stage, the auxiliary training set T source The weights of the samples in the target training set T remain unchanged; through cross-validation, find target The model with the smallest regression error value is taken as the final training result; The weight update rules are as follows: Z t is the normalization constant, β t The goal of the setting is to balance the proportion of source domain data and the overall weight ω1 through binary search. The goal of the search is to make the overall weight of the target domain equal to This target is not set for individual sample weights. Step S424: output the model with the smallest error t , t = argmin(error t ).

Images