Economic insect breeding method for optimizing feed nutrition ratio based on machine learning

Abstract

The invention relates to the technical field of insect breeding, and discloses an economic insect breeding method for optimizing a feed nutrition ratio based on machine learning, which comprises the following steps: constructing a training data set by collecting feed nutrition components of target economic insects and average weight data of larvae, and performing offline training on five machine learning models respectively, the model with the highest performance index is an economic insect feed nutrition evaluation model; and then randomly generating at least 30000 feed formulas, sequentially inputting the feed formulas into the offline trained model to obtain the predicted weight increment of the insect larvae, selecting the feed formula with the predicted weight increment TOP5 of the insect larvae for actual breeding, and taking the feed formula with the highest average weight increment of the larvae after 10 days of breeding as the optimal feed nutrition formula. According to the method, data collection and machine learning are added in the economic insect breeding process to analyze the optimal nutritional components of the insects, the feed nutritional components can be continuously optimized in the breeding process, and the economic insect breeding output and the feed conversion efficiency are improved.

Description

Technical Field

[0001] The present invention belongs to the technical fields of deep learning networks and insect breeding, and specifically relates to an economic insect breeding method for optimizing feed nutrient ratio based on machine learning. Background Art

[0002] In recent years, the animal husbandry and agriculture in China have developed continuously and steadily, with a significant improvement in the scale level. A large amount of breeding waste and agricultural waste have not been effectively treated and utilized. The waste causes greenhouse gas emissions, pollutes water resources, and creates breeding grounds for pests and pathogens, directly affecting public health and threatening environmental sustainability.

[0003] Using environmental insects to convert agricultural and urban waste is an emerging means of recycling organic waste. Since resource insects such as black soldier flies, yellow mealworms, and house flies can utilize urban waste such as food waste, as well as organic waste such as mushroom residue, distillers' grains, wheat bran, and soybean residue to complete their life cycles, their larvae and pupae can be further processed into high-quality insect protein and fat resources, which can be widely used in animal husbandry and industry. Therefore, using organic waste to raise resource insects can effectively realize the recycling of waste and the reuse of high-quality resources, which is conducive to promoting the resource utilization of livestock and poultry breeding waste and agricultural waste and is related to the effective supply of livestock products and agricultural products.

[0004] However, at present, the breeding of economic insects such as black soldier flies and yellow mealworms is still in the stage of "small workshop" breeding, and there are no clear requirements for the conversion efficiency of insects to organic waste and the yield of products such as protein and insect oil. At the same time, most of the materials used to raise such insects in the current market are waste with relatively low prices and stable acquisition channels. Such materials generally have the problem of unbalanced nutrition (for example, crop straw contains a large amount of indigestible cellulose), and the remaining nutrients in them are difficult to be effectively utilized by insects, resulting in relatively low larval yield and weight gain. These situations all lead to relatively low efficiency in the current resource insect breeding industry.

[0005] For the correlation analysis between the nutrient components contained in the waste used to raise economic insects and the larval yield and weight gain, Pearson correlation analysis method and response surface analysis (RSA) method are mostly used. By studying experimental variables (such as nutrient components), the influence on the results (such as larval growth) is obtained. The response surface analysis method is suitable for quadratic relationships, with poor complex non-linear fitting, high experimental costs, results limited by the variable range, and possible overfitting. Pearson correlation analysis is only applicable to linear relationships, cannot identify causal relationships, is easily affected by outliers, and cannot analyze interaction effects.

[0006] Therefore, there is an urgent need for the current organic waste conversion industry of economic insects to have a precise breeding strategy based on breeding data and combined with machine learning models to assist breeders in systematically, conveniently, and scientifically controlling the nutritional components of insect feed. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method for breeding economic insects that optimizes the nutritional ratio of feed based on machine learning. By constructing an economic insect feed nutrition evaluation model through machine learning algorithms, the optimal ratio of nutritional components in economic insect feed is obtained for the precise breeding of economic insects.

[0008] To solve the above technical problem, the present invention provides a method for breeding economic insects that optimizes the nutritional ratio of feed based on machine learning, including the following steps:

[0009] Step S1: Construct a training data set by collecting the feed nutritional components of target economic insects and the data of the average larval weight.

[0010] Step S2: Offline train five machine learning models respectively using the training data set, and select the model with the highest performance index as the economic insect feed nutrition evaluation model; the machine learning models include binary decision tree, random forest, extreme gradient boosting tree, support vector regression, and linear regression.

[0011] Step S3: Randomly generate at least 30,000 feed formulations, and then input them into the offline-trained economic insect feed nutrition evaluation model in sequence to obtain at least 30,000 predicted weight increments of insect larvae; subsequently, select 5 groups of feed formulations corresponding to the TOP5 of the predicted weight increments of insect larvae to configure artificial feed for economic insects for the breeding of target economic insects, and take the feed formulation with the highest average weight increment of target economic insect larvae after 10 days of breeding as the best feed nutrition formulation.

[0012] As an improvement of the method for breeding economic insects that optimizes the nutritional ratio of feed based on machine learning of the present invention:

[0013] The process of constructing the training data set is as follows:

[0014] Configure the feed substrate, then measure the feed nutritional components in the feed substrate through orthogonal experiments, and then add nutritional components to prepare 25 kinds of artificial feed for economic insects, and raise the larvae of target economic insects respectively to obtain 25 groups of data on the average weight increment of larvae; the feed nutritional components of each kind of artificial feed for economic insects and the corresponding average weight increment of larvae are used as a set of samples for standardization processing by StandardScaler, and the standardized data is divided into a training set, a validation set, and a test set.

[0015] As a further improvement of the economic insect breeding method based on machine learning for optimizing feed nutrient ratio in the present invention:

[0016] The offline training process is as follows:

[0017] Input the training set and the validation set into the five machine learning models respectively. Using the grid search method, traverse and combine the key hyperparameters of each model according to the value set. For each hyperparameter combination, use the training set for training and use the validation set to verify the performance indicators of the model; select the hyperparameter combination with the best performance indicators on the validation set as the best parameter configuration of the model, so as to obtain five trained machine learning models;

[0018] Then, input the test set into the five trained machine learning models respectively. Compare the predicted weight increment of the insect larvae output by each model with the average weight increment of the larvae and count the performance indicators. The model with the highest performance indicator is used as the offline trained economic insect feed nutrition evaluation model.

[0019] As a further improvement of the economic insect breeding method based on machine learning for optimizing feed nutrient ratio in the present invention:

[0020] The nutrient components include non-fiber carbohydrates, crude protein, vitamin B mixture, vitamin C, fat and cellulose.

[0021] As a further improvement of the economic insect breeding method based on machine learning for optimizing feed nutrient ratio in the present invention:

[0022] The feed nutrient components in the feed formula are as follows: the content ranges of non-fiber carbohydrates, crude protein, fat and cellulose are all 0%-100% and the total content does not exceed 100%, the content range of vitamin B mixture is 0-5000 mg / kg, and the content of vitamin C is 0-50000 mg / kg.

[0023] As a further improvement of the economic insect breeding method based on machine learning for optimizing feed nutrient ratio in the present invention:

[0024] The feed substrate is a mixture of casein acid hydrolysate (without vitamins), sucrose, glucose, soybean oil, B vitamin mixture, vitamin C, sodium carboxymethyl cellulose and cellulose.

[0025] As a further improvement of the economic insect breeding method based on machine learning for optimizing feed nutrient ratio in the present invention:

[0026] The performance indicators are root mean square error, mean absolute error or coefficient of determination.

[0027] The beneficial effects of the present invention are mainly reflected in:

[0028] 1. During the data collection process of the present invention, an artificial feed for economic insects is prepared by using a feed substrate prepared in the laboratory. The preparation materials are convenient and easy to obtain, avoiding the interference of unknown nutrients, ensuring the feeding advantage of insects on the artificial feed, and being able to clarify the demand of insects for various nutrient elements.

[0029] 2. In the model construction link of the present invention, a machine learning algorithm is used to construct a nutritional evaluation model for economic insect feed. The training set consists of a large amount of actual breeding data and can continuously enrich the data set as the breeding industry progresses. The generalization ability of the model is relatively high. The machine learning model can more accurately learn the patterns in the data, thus showing higher performance in the process of targeting different species of insects and using different types of food materials for feeding.

[0030] 3. Different algorithms of the present invention have different assumptions and sensitivities to the distribution and characteristics of data. Each economic insect can screen the algorithm with the highest fitting degree from five machine learning models, which is more universal and accurate.

[0031] 4. In the precise breeding link of the present invention, more than 30,000 feed formulas are randomly generated by a computer and the nutritional evaluation model of economic insect feed is used to predict the larva yield. From a data perspective, the final production capacity of economic insects under different nutritional conditions is simulated. The best feed nutritional formula for insect feed is determined according to the feed nutritional components with the highest yield. Finally, it is verified through actual breeding data, which is more rigorous.

[0032] 5. The existing means of raising economic insects is to directly place the larvae in the organic waste feed for natural growth, and it is impossible to further optimize the breeding means and food nutrition level. It lacks repeatability and stability and is not suitable for large-scale breeding operations. The present invention adds a data collection and machine learning analysis calculation model for the best nutritional components of insects to the process of raising economic insects, which can continuously optimize the feed nutritional components during the breeding process and further improve the output and feed conversion efficiency of raising economic insects.

[0033] 6. Compared with the evaluation method of using the traditional Pearson correlation analysis method and response surface analysis method to evaluate the experimental variables (feed nutritional components) on the results (average weight increment of larvae), the precise breeding method provided by the present method based on multiple breeding data and machine learning models can conveniently and quickly evaluate the breeding production capacity of any food material, reduce the trial-and-error cost; is easy to accurately adjust one or more nutritional components, optimize the feed nutrition, facilitate the use of low-cost organic waste in the economic insect breeding industry, and achieve the important goals of cost reduction, efficiency increase and agricultural green circular development; is applicable to large-scale breeding of economic insects and further improves the recycling rate of organic waste. Description of the Drawings

[0034] The specific implementation manners of the present invention will be further described in detail below in conjunction with the accompanying drawings.

[0035] Figure 1 It is a schematic flowchart of a method for culturing economic insects based on optimizing feed nutrient ratio by machine learning according to the present invention.

[0036] Figure 2 It is a schematic diagram of the relationship among the feed formula, weight gain, and feed nutrient components of the present invention;

[0037] Figure 2 In it:

[0038] A is a scatter plot of the weight gain of economic insects fed with 25 kinds of artificial feeds for economic insects;

[0039] B is a three-dimensional surface plot of the crude protein, carbohydrate content, and larval weight gain;

[0040] C is a three-dimensional surface plot of the fat, carbohydrate, and larval weight gain;

[0041] D is a three-dimensional surface plot of the fat, crude protein, and larval weight gain.

[0042] Figure 3 It is a Pearson correlation analysis chart of the nutrient components and weight gain data.

[0043] Figure 4 It is a comparison chart of the average weight gain of larvae and the predicted weight gain of insect larvae output by five machine learning models;

[0044] Figure 4 In it:

[0045] A is the extreme gradient boosting method model;

[0046] B is the decision tree model;

[0047] C is the random forest model;

[0048] D is the support vector regression model;

[0049] E is the linear regression model.

[0050] Figure 5 It is a diagram for explaining the influence of each nutrient component on the weight change of larvae using SHAP parameters;

[0051] Figure 5 In it:

[0052] A is a comparison chart of the influence degree of each nutrient component on the larval weight;

[0053] Figure B is a stacked chart showing the influence of each nutrient component on the SHAP value;

[0054] Figure C is a line chart showing the influence of crude protein on the SHAP parameter;

[0055] Figure D is a line chart showing the influence of non - fiber carbohydrates on the SHAP parameter;

[0056] Figure E is a line chart showing the influence of fat on the SHAP parameter.

[0057] Figure 6 It is a simulation result chart of the nutritional ratio model for the food of economic insects in the present invention;

[0058] Figure 6 Among them:

[0059] The left figure is a predicted weight increment chart of 30,000 kinds of insect larvae output by the model; the right figure is a difference chart between the average weight increment of larvae and the predicted weight increment of insect larvae.

[0060] Figure 7 It is a verification result chart of the ratio experiment of four kinds of organic wastes;

[0061] Figure 7 Among them:

[0062] Figure A is a feed nutrient component table of four kinds of organic wastes and the best feed nutrition formula;

[0063] Figures B and C are comparison charts of the feed nutrient components in wheat bran and the best feed nutrition formula;

[0064] Figures G and H are comparison charts of the feed nutrient components in corn straw and the best nutritional ratio;

[0065] Figures L and M are comparison charts of the feed nutrient components in chicken manure and the best nutritional ratio;

[0066] Figures Q and R are comparison charts of the feed nutrient components in food waste and the relative best nutritional ratio;

[0067] Figures D, I, N, and S are respectively the average weight increment change charts of black soldier fly larvae fed with four kinds of organic wastes supplemented with nutrients;

[0068] Figures E, J, O, and T are respectively the feed intake change charts of black soldier fly larvae fed with four kinds of organic wastes supplemented with nutrients;

[0069] Figures F, K, P, and U are respectively the food conversion rate charts of black soldier fly larvae fed with four kinds of organic wastes supplemented with nutrients.

[0070] Figure 8 It is a verification result chart of the mixing ratio of FW + WB combined feed;

[0071] Figure 8 In:

[0072] A is the predicted weight increment graph of insect larvae with different proportions of FW+WB combined feed;

[0073] B is the graph of the average weight increment of black soldier fly larvae changing with the proportion of FW+WB combined feed;

[0074] C is the graph of the food conversion rate changing with the proportion of FW+WB combined feed;

[0075] D is the graph of the reduction amount of organic waste changing with the proportion of FW+WB combined feed.

[0076] Figure 9 It is the verification result graph of the mixing ratio of FW+CS combined feed;

[0077] Figure 9 In:

[0078] A is the predicted weight increment graph of insect larvae with different proportions of FW+CS combined feed;

[0079] B is the graph of the average weight increment of black soldier fly larvae changing with the proportion of FW+CS combined feed;

[0080] C is the graph of the food conversion rate changing with the proportion of FW+CS combined feed;

[0081] D is the graph of the reduction amount of organic waste changing with the proportion of FW+CS combined feed.

[0082] Figure 10 It is the verification result graph of the mixing ratio of WB+CM combined feed;

[0083] Figure 10 In:

[0084] A is the predicted weight increment graph of insect larvae with different proportions of WB+CM combined feed;

[0085] B is the graph of the average weight increment of black soldier fly larvae changing with the proportion of WB+CM combined feed;

[0086] C is the graph of the food conversion rate changing with the proportion of WB+CM combined feed;

[0087] D is the graph of the reduction amount of organic waste changing with the proportion of WB+CM combined feed.

[0088] Figure 11 It is the verification result graph of the mixing ratio of WB+CS combined feed;

[0089] Figure 11 In:

[0090] Figure of predicted weight gain of insect larvae with different proportions of WB+CS combined feed;

[0091] Figure of the change in the average weight gain of black soldier fly larvae with the proportion of WB+CS combined feed;

[0092] Figure of the change in feed conversion rate with the proportion of WB+CS combined feed;

[0093] Figure of the change in the reduction of organic waste with the proportion of WB+CS combined feed. Detailed implementation method

[0094] The present invention will be further described below in conjunction with specific embodiments, but the protection scope of the present invention is not limited thereto:

[0095] Embodiment 1. An economic insect breeding method based on optimizing the nutritional ratio of feed by machine learning, as Figure 1 shown, includes three key links: data collection, construction of an economic insect feed nutrition evaluation model, and precision breeding, and can quickly and accurately infer the impact on the growth and development of fed economic insects according to different nutritional components of the feed, and is used in the industrial production of economic insects.

[0096] 1. Data collection

[0097] In the data collection link of the present invention, 25 kinds of artificial feeds for economic insects are prepared through feed substrates in a laboratory environment, and are used to raise insects respectively, and numerical values such as the growth rate and larval production performance (larval weight gain) of the target economic insects are collected. Specifically:

[0098] This method first prepares a feed substrate in the laboratory. The feed substrate used is a mixture prepared from casein acid hydrolysate (without vitamins), sucrose, glucose, soybean oil, a mixture of B vitamins, vitamin C, sodium carboxymethyl cellulose, and cellulose in different proportions. Then, orthogonal experiments are applied to determine the feed nutrient components (i.e., the content of each nutrient) in the feed substrate, quantify each nutrient component in the feed substrate, and add corresponding amounts of nutrients to the feed substrate through feed additives according to the feed nutrient components shown in Feed Formulas 1 - 25, as shown in Table 1, so as to formulate 25 kinds of artificial diets for target economic insects with different nutrient components to rear economic insects. The nutrient components include non-fiber carbohydrates (NFCs), crude protein (CP), vitamin B mixture (Vb), vitamin C (Vc), fat, and cellulose. Then, record and collect the values such as the growth rate and larval production performance (larval weight gain) of the target economic insects reared with each feed formula as the data source for subsequent model training.

[0099] Each feed formula of the artificial diet for economic insects should contain at least 6 nutrient components, as shown in Table 1.

[0100] Table 1. Feed nutrient component table of 25 feed formulas of artificial diets for economic insects

[0101]

[0102] In the data collection process, the 25 kinds of artificial diets for economic insects of the present invention configured according to Table 1 are different from ordinary artificial diets. The feed substrate used in the configuration uses vitamin-free casein acid hydrolysate as the nitrogen source, which can effectively exclude the interference of nutrients such as vitamins, fats, and carbohydrates in the nitrogen source component of the feed on the final experimental results; the feed substrate only uses nitrogen source, carbon source (glucose, fructose), inert filler (cellulose), fat (beef tallow), B vitamins (vitamin B1, B2, B3, B5, B6, B7, B9, B12), and vitamin C as the main nutrients. While excluding the interference of other trace nutrients, it meets the minimum requirements for the normal growth and development of insects, facilitating the determination of their requirements for various nutrients.

[0103] 2. The link of constructing the economic insect feed nutrition evaluation model mainly constructs the economic insect feed nutrition evaluation model of the present invention by collecting the feed nutrition components and the corresponding growth characteristic data and using the machine learning algorithm with the highest fitting degree among the five algorithms of decision tree (DT), random forest (RF), extreme gradient boosting (XGBoost), linear regression (LR), and support vector regression (SVR).

[0104] 2.1. Construct the training data set

[0105] The larvae of the target economic insects are respectively fed with the 25 kinds of artificial feeds for economic insects obtained in step 1, and the average weight increment of the larvae (fed for 10 days) is collected. The feed nutrition components in the artificial feed for economic insects and the corresponding average weight increment of the larvae are used as a set of sample data. More than 1000 sets of data are required to construct the training data set to ensure the accuracy of the model. The average value, standard deviation, and distribution of each sample in the data set are calculated through a Python script to standardize the data set by StandardScaler, that is, each sample is centered by the mean of the features, and then scaled by the standard deviation to make the data set conform to the normal distribution. Finally, the standardized data set is divided into a training set, a validation set, and a test set in a ratio of 6:2:2. The training set and the validation set are used for the training and parameter tuning of the model, while the test set is used to evaluate the generalization performance of the final model.

[0106] 2.2. Offline training

[0107] Select five machine learning models of the prior art, including decision tree (DT), random forest (RF), extreme gradient boosting (XGBoost), support vector regression (SVR), and linear regression (LR), as candidate models for the economic insect feed nutrition evaluation model of the present invention to construct corresponding instances.

[0108] Taking the training set and validation set constructed in step 2.1 as inputs, train five pre-built machine learning models (Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), Linear Regression (LR)) respectively. Using the grid search method, traverse and combine the key hyperparameters of each model according to the predefined value sets. For each hyperparameter combination, use the training set to train the model, and use the validation set to calculate the performance metrics of the model on the target task (i.e., obtaining the predicted weight increment of insect larvae), such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), or Coefficient of Determination (R 2 ). By traversing all hyperparameter combinations, record the performance of each combination on the validation set. Select the hyperparameter combination with the best performance on the validation set (R 2 highest, lowest RMSE and MAE values) as the best parameter configuration for the model, thus obtaining five trained machine learning models.

[0109] Then, input the test set constructed in step 2.1 into the five trained machine learning models respectively to obtain the prediction results of the predicted weight increment of insect larvae. Compare the prediction results of each model with the data of the average weight increment of larvae in the test set, and calculate the corresponding performance metrics (MSE, MAE, and R2). For example, for black soldier fly, the comparison graphs of the predicted weight increment of insect larvae and the average weight increment of larvae output by the five machine learning models are as shown in Figure 4 A - E. Among them, the performance metrics of the models generated by different algorithms are: XGboost method, R 2 = 0.8093, RMSE = 0.0233, MAE = 0.0180; DT method, R 2 = 0.8083, RMSE = 0.0224, MAE = 0.0181; RF method, R 2 = 0.8087, RMSE = 0.0224, MAE = 0.0180; SVR method, R 2 = -0.153, RMSE = 0.056, MAE = 0.047; LR method, R 2 = 0.808, RMSE = 0.0222, MAE = 0.018. The performance metrics of the five trained machine learning models on the same test set show that XGBoost has better prediction performance and relatively stable prediction ability (R 2The highest, with the lowest RMSE and MAE values). Preferably, the Extreme Gradient Boosting (XGBoost) algorithm with the highest performance index is used as the economic insect feed nutrition evaluation model.

[0110] 2.3 Calculate the weights of each nutrient component and the optimal nutrient ratio

[0111] Furthermore, use the economic insect feed nutrition evaluation model (i.e., the offline-trained XGBoost algorithm) to evaluate the impact of different feed nutrient components (i.e., different contents of nutrient components) on the larval weight of economic insects. Use the SHapley Additive exPlanations of the XGBoost algorithm to calculate the importance parameters (or importance degrees) of different feed nutrient components in the feed on insect development. The results show that an appropriate proportion of crude protein in the feed is crucial for the effective conversion of organic waste, as Figure 5 shown.

[0112] 3. Precision farming link

[0113] Based on obtaining the optimal machine learning model, write Python scripts to generate more than 30,000 feed formulas of different economic insect artificial feeds with random distributions of each feed nutrient component (according to the marginal effects of each nutrient element, set the content ranges of non-fiber carbohydrates, crude protein, fat, and cellulose to be 0%-100%, and the total combined content of the above four nutrient components does not exceed 100%; the content range of vitamin B mixture is 0 - 5000 mg / kg; the content of vitamin C is 0 - 50000 mg / kg). Input the generated more than 30,000 different feed formulas into the offline-trained economic insect feed nutrition evaluation model in sequence. One feed formula obtains a predicted weight increment of insect larvae through the offline-trained economic insect feed nutrition evaluation model, and a total of more than 30,000 data of predicted weight increments of insect larvae are obtained. Subsequently, arrange the more than 30,000 predicted weight increments of insect larvae in descending order, and select the feed formulas of 5 target economic insects corresponding to the TOP5 predicted weight increments of insect larvae to be formulated into economic insect artificial feeds in the laboratory for actual breeding experiments of target economic insects. Then, take the feed formula with the highest average weight increment of the larvae of target economic insects after 10 days of actual breeding as the final optimal feed nutrition formula of target economic insects for large-scale breeding of economic insects.

[0114] The present invention can timely import economic insect breeding-related data (such as the nutrient ratio of different organic wastes, the feed nutrient components in the feed formulas of different economic insect artificial feeds, and the corresponding average weight data increments of larvae) into the economic insect feed nutrition evaluation model, and the accuracy of the model will gradually increase with the continuous enrichment of the data set, having very high potential application value.

[0115] 4. Verification Experiment

[0116] Experiment 1:

[0117] To verify the accuracy of the constructed nutritional evaluation model for economic insect feed, taking the black soldier fly as an example, 30,000 feed formulations containing different feed nutritional components were randomly generated by computer. The 30,000 different feed nutritional components were respectively input into the economic insect feed nutritional evaluation model (XGBoost model) trained offline in Step 2 to predict the predicted weight increment of the insect larvae of the black soldier fly. The prediction results are as Figure 6 shown in A. And the top 10 feed formulations (Random diet, RD) with the highest predicted weight increment of insect larvae were selected to conduct a feeding experiment on the black soldier fly. After 10 days of feeding, the average weight increment of the larvae of the black soldier fly and the predicted weight increment of the insect larvae are as Figure 6 shown in B. There is no significant difference between the actual feeding results and the model prediction results. Therefore, the 11,497th (RD - 11497) feed formulation, which is the best among the 10 feeds, was selected as the best feed nutrition formulation (Best model) for the black soldier fly. In summary, the economic insect feed nutritional evaluation model constructed by the machine learning model is sufficient to provide weight results similar to those of larvae reared under laboratory conditions.

[0118] Experiment 2. Verification of the Correlation between Nutritional Components and Weight Gain Data

[0119] According to the positive correlation between the content of each nutritional component and the weight increment, the traditional response surface analysis method was used to explore the relationship between the 6 nutritional components in Table 1 and the weight increment, and the most critical nutritional components in the food of economic insects were initially calculated. For the black soldier fly, a response surface was fitted by collecting 6 feed nutritional components and the corresponding larval production performance (larval weight) in 25 artificial feeds for economic insects to simulate the true limit state surface. By selecting the feed nutritional components corresponding to the highest point of the larval production performance on the response surface, the optimal ratio between different feed nutritional components was initially calculated. The feed formulation, weight increment results, and the interaction of feed nutritional components are as Figure 2 shown, where Figure 2 (A) shows the statistical results of the weight increase of the black soldier fly fed with 25 artificial feeds for economic insects with different feed nutritional components, Figure 2 (B - D) show the results of using response surface analysis to find the optimal ratio between different feed nutritional components. Among them, Figure 2 (B) is the calculation result of the optimal ratio of non - fiber carbohydrates and crude protein. When the content of non - fiber carbohydrates in the artificial feed for economic insects is 27% and the protein content is 40%, the weight increment of the larvae of the black soldier fly is 0.187 g, which is the best weight increment.Figure 2 (C) is the calculation result of the optimal ratio of fat and non-fiber carbohydrates. When the content of non-fiber carbohydrates in the artificial feed for economic insects is 30% and the fat content is 2%, the weight increase of the larvae of Hermetia illucens is 0.179 g, which is the optimal weight gain. Figure 2 (D) is the calculation result of the optimal ratio of fat and crude protein. When the content of crude protein in the artificial feed for economic insects is 54% and the fat content is 3.2%, the weight increase of the larvae of Hermetia illucens is 0.177 g, which is the optimal weight gain.

[0120] Then, the traditional Pearson correlation analysis method is used to perform a correlation analysis on the nutrient components and weight gain data to determine the nutrient components that have the strongest correlation with the growth efficiency of economic insects, such as Figure 3 shown. The Pearson correlation analysis calculates the correlation between the average weight increase of the larvae and each feed nutrient component. Among them, the correlation between the crude protein content and the weight increase is 0.53, which is the highest among all feed nutrient components. Therefore, it is the most critical nutrient component affecting the yield of Hermetia illucens larvae. The remaining several nutrient components have marginal effects (when the proportion of the feed nutrient component reaches a certain level, it will no longer continue to increase the weight of the larvae, and may even lead to a decrease in nutrient content). Figure 3 In

[0121] NFCs (Non-fiber carbohydrates) is non-fiber carbohydrates; CP (Crude protein) is crude protein; Vb (Vitamin B group) is vitamin B mixture; Vc (Vitamin C) is vitamin C; Fat is fat; Cellulose is cellulose; Weight gain is weight increase; The red square represents a positive correlation, and the blue square represents a negative correlation. Figure 2 As can be seen from

[0122] Experiment 3: Four Feed Ratio Experiments

[0123] Based on the verified nutritional evaluation model for economic insect feed, it can provide nutritional supplementation suggestions for the lacking nutrients in economic insect feed (which can be supplemented through feed additives); at the same time, it can specify several different food materials (or organic waste) for mixing, and obtain the best mixing ratio (i.e., the best feed nutrition formula) that meets the nutritional requirements for the growth and development of economic insects according to the economic insect feed nutritional evaluation model.

[0124] In this four-feed ratio experiment, gas chromatography-mass spectrometry (LC-MS) was used to determine the contents of various nutritional components (NFCs, non-fiber carbohydrates; CP, crude protein; Vb, vitamin B mixture; Vc, vitamin C; Fat, fat) in wheat bran, corn stover, chicken manure, and food waste (i.e., feed nutritional components), as Figure 7 shown in A. The best feed nutrition formula (Best model) was obtained through 30,000 random feed predictions and actual breeding experiments. Refer to the results of Experiment 1.

[0125] Through determination, it was found that wheat bran, corn stover, chicken manure, and food waste all showed varying degrees of nutritional deficiencies compared to the best feed nutrition formula (Best model) for black soldier fly. The proportions of non-fiber carbohydrates, crude protein, crude fat, vitamin B mixture, and vitamin C in wheat bran and the lack amounts relative to the best feed nutrition formula (Best model) are as Figure 7 shown in B and Figure 7 C; the proportions of non-fiber carbohydrates, crude protein, crude fat, vitamin B mixture, and vitamin C in corn stover and the lack amounts relative to the best nutritional ratio are as Figure 7 shown in G and Figure 7 H; the proportions of non-fiber carbohydrates, crude protein, crude fat, vitamin B mixture, and vitamin C in chicken manure and the lack amounts relative to the best nutritional ratio are as Figure 7 shown in L and Figure 7 M; the proportions of non-fiber carbohydrates, crude protein, crude fat, vitamin B mixture, and vitamin C in food waste and the lack amounts relative to the best nutritional ratio are as Figure 7 shown in Q and Figure 7 R.

[0126] Then, referring to the Best Model of optimal nutrition formula, additives were added to wheat bran, corn straw, chicken manure, and food waste respectively to make up for the insufficient nutrition. They were used to raise black soldier fly larvae in a laboratory environment respectively. The changes in the weight gain, feed intake, and food conversion rate of the black soldier fly larvae fed with the feed after supplementing the nutrients were calculated. Among them, the change in the weight gain of the black soldier fly larvae fed with the wheat bran after supplementing the nutrients is as shown in Figure 7 Figure D, the change in feed intake is as shown in Figure 7 Figure E, and the change in food conversion rate is as shown in Figure 7 Figure F. The change in the weight gain of the black soldier fly larvae fed with the corn straw after supplementing the nutrients is as shown in Figure 7 Figure I, the change in feed intake is as shown in Figure 7 Figure J, and the change in food conversion rate is as shown in Figure 7 Figure K. The change in the weight gain of the black soldier fly larvae fed with the chicken manure after supplementing the nutrients is as shown in Figure 7 Figure N, the change in feed intake is as shown in Figure 7 Figure O, and the change in food conversion rate is as shown in Figure 7 Figure P. The change in the weight of the black soldier fly larvae fed with the food waste after supplementing the nutrients is as shown in Figure 7 Figure S, the change in feed intake is as shown in Figure 7 Figure T, and the change in food conversion rate is as shown in Figure 7 Figure U.

[0127] The significance of the difference was calculated by the Mann-Whitney U test method. p < 0.05 indicates significant difference. The results show that Figure 7 the significance of the difference p of D, E, I, J, K, N, O, P, and S is less than 0.01, belonging to extremely significant difference; the significance of the difference p of Figures T and U is less than 0.05, belonging to significant difference.

[0128] It can be Figure 7 seen that after comparing with the optimal nutrition ratio, using feed additives (Additives) to make up for the lack of nutrition in each organic waste and using it to raise the larvae of economic insects, it can be observed that there are significant improvements in the weight gain, food consumption, and food conversion rate of the larvae. The weight gain of the larvae raised with different types of organic waste after supplementing the nutrients has been significantly improved, indicating that supplementing the feed nutrition components in the food materials based on the prediction results of the economic insect feed nutrition evaluation model can effectively improve the production indexes of the larvae, which is a very feasible breeding strategy.

[0129] Experiment 4. Two kinds of feed ratio experiments

[0130] Since the cost of using feed additives is relatively high, organic waste with different feed nutrient components can be mixed in a scientific and reasonable proportion, which can also improve the feed nutrition to a certain extent.

[0131] (1) Kitchen waste + wheat bran ratio experiment

[0132] Two different proportions of food materials (kitchen waste + wheat bran, FW + WB) were evenly mixed into 10 FW + WB combined feeds at gradient ratios of 0:10, 1:9, 2:8, 3:7, 4:6, 5:5, 6:4, 7:3, 8:2, 9:1, 10:0. The gas chromatography-mass spectrometry (LC-MS) method was used to determine the feed nutrient components in the FW + WB combined feeds at each ratio, and the economic insect feed nutrition evaluation model was used to obtain the predicted weight increment of insect larvae, as Figure 8 shown in A.

[0133] Meanwhile, in the laboratory environment, black soldier fly larvae were fed with 10 FW + WB combined feeds, and the average weight increment of the larvae was measured to further verify the accuracy of the prediction results. The FW + WB combined feeds with ratios of 0:10, 6:4, and 10:0 were used to feed black soldier fly larvae, and the weight increment ( Figure 8 B), food conversion rate ( Figure 8 C), and reduction of organic waste ( Figure 8 D) were statistically analyzed for the improvement of the three important indicators. As Figure 8 can be seen, the average weight increment of black soldier fly larvae fed with FW + WB combined feeds is higher than that fed with single feeds. Especially when the ratio is 6:4, it can effectively improve the average weight increment, food conversion efficiency, and consumption of organic waste of black soldier fly larvae. Feeding with the FW + WB combined feed with a ratio of 6:4 can increase by 3% - 5% compared with single feed.

[0134] (2) Kitchen waste + corn straw ratio experiment

[0135] Two different proportions of food materials (kitchen waste + corn straw, FW + CS) were evenly mixed into 10 FW + CS combined feeds at gradient ratios of 0:10, 1:9, 2:8, 3:7, 4:6, 5:5, 6:4, 7:3, 8:2, 9:1, 10:0. The gas chromatography-mass spectrometry (LC-MS) method was used to determine the feed nutrient components in the FW + CS combined feeds at each ratio, and the economic insect feed nutrition evaluation model was used to obtain the predicted weight increment of insect larvae, as Figure 9 shown in A.

[0136] Meanwhile, in the laboratory environment, black soldier fly larvae were fed with 10 combinations of FW + CS feed, and the average weight gain of the larvae was measured to further verify the accuracy of the prediction results. The comparison of the average weight gain of black soldier fly larvae fed with FW + CS feed mixtures in the ratios of 0:10, 8:2, and 10:0 and single feed is shown in Figure 9 Figure B, the comparison of food conversion rate is shown in Figure 9 Figure C, and the reduction of organic waste is shown in Figure 9 Figure D. The model output shows that the average weight gain of black soldier fly larvae fed with FW + CS feed mixtures is higher than that fed with single feed. The average weight gain of the larvae continuously increases with the increase in the proportion of food waste until it reaches the highest value at a ratio of 8:2. Subsequently, further increase in food waste cannot continue to increase the average weight gain of the larvae, but instead causes a certain degree of decrease. This result indicates that there are nutrients in corn straw that are lacking in food waste. When the above two materials, food waste and corn straw, are mixed in an appropriate ratio (such as food waste: corn straw = 8:2), it can achieve a better effect than feeding with single feed. The food conversion rate of the larvae fed with a mixture of 8:2 food materials can reach 17.98%, which is significantly higher than that of feeding with food waste alone (16.72%) and corn straw alone (0.37%). In addition, this mixing method has a significant improvement effect on both the average weight gain of the larvae and the consumption of organic waste.

[0137] (3) Wheat bran + chicken manure ratio experiment

[0138] Two different ratios of food materials (wheat bran + chicken manure, WB + CM) were evenly mixed into 10 combinations of WB + CM feed at gradient ratios of 0:10, 1:9, 2:8, 3:7, 4:6, 5:5, 6:4, 7:3, 8:2, 9:1, and 10:0. The composition of nutrients in the WB + CM feed combinations at each ratio was determined using gas chromatography - mass spectrometry (LC - MS), and the predicted weight gain of insect larvae was obtained using the economic insect feed nutrition assessment model, as shown in Figure 10 Figure A.

[0139] Meanwhile, in the laboratory environment, black soldier fly larvae were fed with 10 combinations of WB + CM feed, and the average weight gain of the larvae was measured to further verify the accuracy of the prediction results. The comparison of the average weight gain of black soldier fly larvae fed with WB + CM feed mixtures in the ratios of 0:10, 6:4, and 10:0 and single feed is shown in Figure 10 Figure B, the comparison of food conversion rate is shown in Figure 10 Figure C, and the reduction of organic waste is shown in Figure 10As shown in D, the model output indicates that the average weight gain of black soldier fly larvae fed with the WB+CM combined feed is higher than that of those fed with a single feed. The average weight gain of the larvae continuously increases with the increase in the proportion of chicken manure until it reaches the highest value when the ratio is 6:4. Subsequently, further increase in chicken manure cannot continue to increase the larval weight, but instead causes a certain degree of decrease and reaches a stable value. This result shows that there are nutrients relatively lacking in the other in both wheat bran and chicken manure. When the wheat bran and chicken manure materials are mixed in an appropriate ratio (such as wheat bran:chicken manure = 6:4), it can achieve a better effect than feeding with a single feed. From the results, when the above materials are mixed at a ratio of 6:4, the average weight gain of the larvae increases by 23.84% compared to feeding with wheat bran alone and by 12.14% compared to feeding with chicken manure alone; the feed conversion rate increases by 4.7% compared to wheat bran and by 9.7% compared to chicken manure; in terms of feed digestibility, the digestibility of black soldier fly larvae fed with the mixed feed is 70.14%, which is also significantly higher than 47.36% of pure chicken manure and 67.22% of pure wheat bran.

[0140] (4) Wheat bran + corn straw ratio experiment

[0141] Two different proportions of food materials (wheat bran + corn straw, WB+CS) were evenly mixed into 10 WB+CS combined feeds at gradient ratios of 0:10, 1:9, 2:8, 3:

[0142] 7, 4:6, 5:5, 6:4, 7:3, 8:2, 9:1, 10:0. The gas chromatography-mass spectrometry (LC-MS) method was used to determine the composition of nutrients in the WB+CS combined feeds at each ratio, and the economic insect feed nutrition evaluation model was used to obtain the predicted weight gain of insect larvae, as Figure 10 shown in A.

[0143] Meanwhile, in the laboratory environment, black soldier fly larvae were fed with 10 WB+CS combined feeds respectively to measure the average weight gain of the larvae, and further verify the accuracy of the prediction results. The average weight gain of black soldier fly larvae fed with the WB+CS combined feeds mixed at ratios of 0:10, 5:5, and 10:0 is as Figure 11 shown in B, the comparison of food conversion rates is as Figure 11 shown in C, and the reduction amount of organic waste is as Figure 11As shown in Figure D, the model output shows that the WB+CS combination feed did not effectively improve the average weight gain and food conversion efficiency of black soldier flies larvae. Unlike other combinations, when the proportion of wheat bran is relatively low, the mixture of wheat bran and corn straw cannot significantly increase the average weight gain and feed conversion rate of black soldier flies larvae. However, when the proportion of wheat bran is higher than 70%, the average weight gain and feed conversion-related parameters of black soldier flies larvae are significantly improved. However, compared with feeding with wheat bran alone, mixing corn straw does not improve the various production parameters of black soldier flies. The above results show that not all food materials can improve their nutritional value by mixing. Therefore, it is necessary to use the optimal food nutrient ratio model provided by this method for preliminary calculations and simulations, and select appropriate food materials for mixing to achieve the effect of 1+1>2.

[0144] In summary, for some food materials or organic waste, the performance of organisms fed in combination with other food materials is better than that of organisms fed with nutrient-deficient organic waste alone. Appropriate mixed feed can increase both larval production and the efficiency of waste bioconversion.

[0145] Finally, it should be noted that the above examples are only some specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and there are many variations. All variations that can be directly derived or associated with the content disclosed by a person skilled in the art should be considered as the protection scope of the present invention.

Claims

1. A method for optimizing feed nutritional ratio based on machine learning, characterized in that The steps include: Step S1, constructing a training data set by collecting feed nutritional components and average larval weight data of target economic insects; Step S2, using the training data set to perform offline training on five machine learning models respectively, and selecting the model with the highest performance index as the economic insect feed nutritional evaluation model; Machine learning models include binary decision trees, random forests, extreme gradient boosted trees, support vector regression, and linear regression; Step S3, randomly generating at least 30,000 feed formulas, and then sequentially inputting them into the offline trained economic insect feed nutrition evaluation model to obtain at least 30,000 predicted weight increments of insect larvae; Subsequently, 5 groups of feed formulas corresponding to the top 5 predicted weight gains of insect larvae were selected and configured as artificial feeds for economic insects for the breeding of target economic insects. The feed formula with the highest average weight gain of the larvae of the target economic insects after 10 days of breeding was taken as the optimal feed nutritional formula.

2. The method according to claim 1, characterized in that: The process of constructing the training data set is as follows: The feed substrate was prepared, and the feed nutritional components in the feed substrate were determined through orthogonal experiments. Then, the nutritional components were added to formulate 25 kinds of artificial feeds for economic insects, and the larvae of the target economic insects were raised respectively, and the data of the average weight gain of 25 groups of larvae were obtained accordingly; the feed nutritional components of each economic insect artificial feed and the corresponding average weight gain of the larvae were combined into a group of samples, which were standardized by StandardScaler, and the standardized data were divided into training set, validation set and test set.

3. The method according to claim 2, characterized in that: The offline training process is as follows: The training set and the validation set are respectively input into the five machine learning models, and the key hyperparameters of each model are traversed and combined according to the value set using the grid search method. For each hyperparameter combination, the training set is used for training, and the performance index of the model is verified using the validation set; the hyperparameter combination with the best performance index on the validation set is selected as the optimal parameter configuration of the model to obtain the five trained machine learning models; Then, the test set was input into five trained machine learning models respectively, and the predicted weight gain of insect larvae output by each model was compared with the average weight gain of larvae, and the performance index was statistically analyzed. The model with the highest performance index was used as the offline trained economic insect feed nutrition evaluation model.

4. The method according to claim 3, characterized in that: The nutritional ingredients include non-fiber carbohydrates, crude protein, vitamin B mixture, vitamin C, fat and cellulose.

5. The method according to claim 4, characterized in that: The feed nutritional components in the feed formula are as follows: non-fiber carbohydrates, crude protein, fat and cellulose content ranges from 0% to 100% and the total content does not exceed 100%, the vitamin B mixture content ranges from 0 to 5000 mg / kg, and the vitamin C content ranges from 0 to 50000 mg / kg.

6. The method according to claim 5, characterized in that: The feed substrate is a mixture of vitamin-free casein acid hydrolyzate, sucrose, glucose, soybean oil, a B vitamin mixture, vitamin C, sodium cellulose and cellulose.

7. The method according to claim 6, characterized in that: The performance indicator is root mean square error, mean absolute error or determination coefficient.

Images