Fruit and vegetable drying control method based on combination of GA-SVM and fuzzy PID

Abstract

The invention discloses a fruit and vegetable drying control method based on combination of GA-SVM and fuzzy PID. The method is characterized by comprising the following steps that drying data of fruits and vegetables needing to be dried are obtained and screened, an SVM model and a kernel function are selected, optimal hyper-parameters are obtained through a GA genetic algorithm, an SVM prediction model is trained and verified, and drying temperature and wind speed parameters are actively adjusted through fuzzy PID according to the predicted water content. A control system can respond to the temperature and the air speed required by the next stage after responding to the change of the water content along with time in advance, so that over-regulation or under-regulation of the drying temperature is avoided; the quality of dried crops is improved, and energy waste is reduced; and the control requirement that different crops have different drying processes can be met.

Description

Technical Field

[0001] The invention relates to a fruit and vegetable drying parameter control technology, in particular to a fruit and vegetable drying control method based on the combination of GA-SVM (genetic algorithm support vector machine) and fuzzy PID. Background Art

[0002] With the improvement of energy conservation and emission reduction requirements and the advancement of technology, heat pump drying technology has gradually been promoted and applied. This technology uses a low-temperature heat source to heat the air through a heat pump system to achieve low-temperature drying of fruits and vegetables. It has the advantages of energy saving, high efficiency and environmental friendliness. The heat pump system can convert heat energy in a low-temperature environment into usable heat through compression, condensation, expansion and other processes, thereby achieving the temperature and humidity regulation required during the drying process.

[0003] At present, the industry mostly uses traditional PID and fuzzy PID control. The adjustment of traditional PID and fuzzy PID control is based on the current error and error change, and mainly relies on real-time feedback. For example, the patent publication number CN115540525A provides an air energy heat pump dryer controller and its control method, which presets the drying temperature and drying temperature difference, and detects the indoor temperature; if the indoor temperature satisfies the indoor temperature ≤ drying temperature - drying temperature difference, the dryer is turned on and runs the drying mode; if the indoor temperature satisfies the indoor temperature ≥ drying temperature; the dryer is shut down. And so on. However, when faced with complex nonlinear relationships, these existing technologies cannot adapt, especially when the environment changes drastically, large errors may occur, and they cannot respond to future temperature changes in advance, resulting in inaccurate temperature control, over-adjustment or under-adjustment of the drying temperature, and the quality of the dried crops and energy waste.

[0004] Therefore, it is particularly important to be able to better adjust key parameters such as temperature, humidity, air flow, and automatically optimize the drying process according to the characteristics of different fruits and vegetables. This can not only improve the drying efficiency, but also ensure the drying quality of fruits and vegetables and the preservation of nutrients. Summary of the invention

[0005] The purpose of the present invention is to solve the above problems and provide a fruit and vegetable drying control method based on GA-SVM and fuzzy PID, which combines the control method of genetic algorithm optimized support vector machine (GA-SVM) prediction model and fuzzy PID, integrates predictive control and real-time feedback control, and has a more accurate, flexible and stable control effect in the control of fruit and vegetable drying.

[0006] The above technical problems of the present invention are mainly solved by the following technical solutions: a fruit and vegetable drying control method based on the combination of GA-SVM and fuzzy PID, characterized by comprising the following contents: obtaining the drying data of the fruits and vegetables to be dried and screening the data, selecting the SVM model and the kernel function, obtaining the best hyperparameters by using the GA genetic algorithm, training and verifying the SVM prediction model, and actively adjusting the drying temperature and wind speed parameters by the fuzzy PID according to the predicted moisture content; wherein:

[0007] S1. Obtain drying data of fruits and vegetables to be dried, including initial moisture content, temperature and humidity of the external environment, internal drying temperature and wind speed, and moisture content in different time periods.

[0008] S2. First clean the acquired data, remove duplicate data and process missing values, then normalize the data, and finally divide the data set.

[0009] S3. Select the SVM regression model. In the regression model, the output target moisture content of the dried material is obtained according to the input features and expressed as a function f(x) to make it close to the target value y. Assume that the input feature vector is x∈R d , y∈R, then the linear regression model is expressed as:

[0010] f(x)=ω T x+b

[0011] Where: ω is the weight vector of the regression model, b is the bias term, x is the input feature, and T is the transpose of the vector.

[0012] S4, control error range ε, the function to be optimized is as follows:

[0013]

[0014] Where εi, They represent the upper deviation and lower deviation of the ith data point respectively; C is the penalty factor, which is used to control the balance between error and model complexity; ε is the error tolerance, which indicates the tolerance range of the model to prediction error.

[0015] S5. The SVM regression model uses a kernel function to map the input features to a higher-dimensional space in order to find a suitable regression plane in the linear space.

[0016] S6. The final linear regression model function optimization includes constraints, so that the error is within the range of ε and the complexity of the regression model is minimized. Through Lagrangian dual optimization, the final optimization goal can be expressed as:

[0017]

[0018] Where ai ,a j ≥0, which is the Lagrange multiplier.

[0019] S7. Generate multiple candidate hyperparameter combinations through genetic algorithm (GA), train different SVM models, and evaluate their fitness based on the prediction error of the model.

[0020] S8. Train the SVM regression model using the training set and validation set data classified in step S2 by optimizing the hyperparameters.

[0021] S9. After the SVM regression model training is completed, it is tested with the test group data prepared in the previous step S2. The initial moisture content of the material, the temperature and humidity of the external environment, the internal drying temperature and wind speed and time are input. The SVM model will output the predicted moisture content of the material based on the learned mapping relationship.

[0022] S10. According to the prediction signal of SVM, the fuzzy PID controller actively adjusts the temperature, humidity and wind speed of the dryer to ensure that the moisture content of fruits and vegetables gradually decreases to the predetermined target.

[0023] In the aforementioned fruit and vegetable drying control method based on the combination of GA-SVM and fuzzy PID, preferably, the normalization in step S2 is to scale the data to a specified range: [0, 1] or [-1, 1]; the formula is:

[0024]

[0025] Where: x′ is the normalized data, x is the input feature, min(x) is the minimum value of the feature, and max(x) is the maximum value of the feature.

[0026] In the aforementioned fruit and vegetable drying control method based on the combination of GA-SVM and fuzzy PID, preferably, in step S2, the data set is divided into: a training set for training SVM model weights and biases, a validation set for selecting model hyperparameters (C, σ) and tuning the model, and a performance test set for evaluating the final model.

[0027] In the aforementioned fruit and vegetable drying control method based on the combination of GA-SVM and fuzzy PID, preferably, in step S3, in the regression model, the input features based on which the SVM regression model is based are: initial moisture content, temperature and humidity of the external environment, internal drying temperature and wind speed and time.

[0028] In the aforementioned fruit and vegetable drying control method based on the combination of GA-SVM and fuzzy PID, preferably, in step S4, the constraint condition of the function to be optimized is:

[0029]

[0030] In the formula, ε i , They represent the upper deviation and lower deviation of the i-th data point respectively; C is the penalty factor, which is used to control the balance between error and model complexity.

[0031] In the aforementioned fruit and vegetable drying control method based on the combination of GA-SVM and fuzzy PID, as a preferred embodiment, in the kernel function of step S5, the radial basis kernel function is used to process the complex nonlinear relationship between the data, and the formula is:

[0032]

[0033] Where: σ is the parameter of the kernel function, which controls the similarity between data points.

[0034] In the aforementioned fruit and vegetable drying control method based on the combination of GA-SVM and fuzzy PID, as a preferred embodiment, in step S7, the genetic algorithm (GA) algorithm simulates the selection and biological genetic process in nature, starting from the initial population, through selection, crossover, and mutation operations, to produce a group that is more conducive to and suitable for environmental survival, and then obtains the optimal solution.

[0035] In the aforementioned fruit and vegetable drying control method based on the combination of GA-SVM and fuzzy PID, as a preferred embodiment, in step S8, during the training process of the SVM regression model, the SVM will learn the mapping relationship between the initial moisture content of the material, the temperature and humidity of the external environment, the internal drying temperature and wind speed and time, and the moisture content of the drying material, and optimize the model by minimizing errors and intervals.

[0036] In the aforementioned fruit and vegetable drying control method based on the combination of GA-SVM and fuzzy PID, as a preferred embodiment, in step S9, the mean absolute percentage error, root mean square error, mean absolute error and determination coefficient are used as criteria for evaluating the quality of the moisture content prediction model. If the criteria are not met, data and hyperparameter optimization are continued. If the criteria are met, the moisture content prediction under different input parameters can be provided for the material drying.

[0037] This technical solution obtains a sufficient set of drying data of crops to be dried, cleans them, removes duplicate data and processes missing values, and then normalizes the data. Since the training of the SVM model is very sensitive to the scale of the features, if the numerical ranges of different features vary greatly, the model may tend to use features with a larger range, resulting in poor model performance. The data set is then divided so that the data meets the statistical significance characteristics.

[0038] This scheme effectively divides the acquired drying data set, obtains the optimization function through the linear regression model, and then focuses on the characteristics that the initial moisture content of the material, the external environment temperature and humidity, the internal drying temperature and wind speed, the drying time, etc. affect the drying rate of fruits and vegetables in a complex way during the fruit and vegetable drying operation, and the corresponding moisture content relationship is usually nonlinear. A kernel function is designed to map the input features to a higher-dimensional space, so as to find a suitable regression plane in the linear space to deal with its nonlinear problem.

[0039] This technical solution has a constrained and optimized linear regression model function, which makes the error within the set range and minimizes the complexity of the regression model. Furthermore, this solution generates multiple candidate hyperparameter combinations through genetic algorithms, selects appropriate hyperparameters that affect the model training effect, trains different SVM models, and evaluates the fitness based on the prediction error of the model. Finally, GA helps find the most suitable parameter configuration and obtains the optimal solution.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] (1) Using GA-SVM (genetic algorithm support vector machine) to provide prediction signals for the fuzzy PID controller allows the control system to respond in advance to the temperature and wind speed required for the next stage after the moisture content changes over time, thereby avoiding over- or under-adjustment of the drying temperature.

[0042] (2) GA-SVM (genetic algorithm support vector machine) is used to provide prediction signals for the fuzzy PID controller, which improves the control accuracy, improves the quality of dried crops, and reduces energy waste.

[0043] (3) The use of GA-SVM (genetic algorithm support vector machine) to provide prediction signals for the fuzzy PID controller is tailored to a certain crop, so the control system can meet the control requirements of different crops with different drying processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is an overall implementation process diagram of the present invention.

[0045] Figure 2 It is a basic flow chart of the present invention.

[0046] Figure 3 It is a specific flow chart of the present invention. DETAILED DESCRIPTION

[0047] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0048] This embodiment is a fruit and vegetable drying control method based on GA-SVM combined with fuzzy PID, such as Figure 1 As shown, the method includes obtaining the drying data of fruits and vegetables to be dried and screening the data; selecting a suitable SVM model and kernel function; using the GA genetic algorithm to obtain the best hyperparameters to improve the accuracy of the SVM prediction model; training and verifying the SVM prediction model until the accuracy requirement is met; and actively adjusting the drying temperature, wind speed and other parameters by the fuzzy PID according to the predicted moisture content.

[0049] The specific process is as follows:

[0050] S1. Obtain 500-1000 sets of drying data of a certain type of fruit and vegetable to be dried under multiple drying processes, including initial moisture content, temperature (room temperature) and humidity of the external environment, drying temperature and wind speed inside the dryer, and moisture content in different time periods.

[0051] S2. Clean the acquired data, remove duplicate data, process missing values, and then normalize the data. The training of the SVM model is very sensitive to the scale of features. If the numerical ranges of different features vary greatly, the model may tend to use features with a larger range, resulting in poor model performance. Normalization: Scale the data to a certain range (usually [0,1] or [-1,1]). The formula is:

[0052]

[0053] Where: x′ is the normalized data, x is the input feature, min(x) is the minimum value of the feature, and max(x) is the maximum value of the feature.

[0054] The dataset is then divided into a training set for training the SVM model weights and biases, a validation set for selecting the model’s hyperparameters (such as C, σ) and tuning the model, and a test set for evaluating the performance of the final model to ensure that the model is not overfitted.

[0055] S3. Select the SVM regression model. Support vector machines are generally divided into two major problems: classification and regression. Since the goal of this embodiment is to predict moisture content (a continuous variable), the SVM regression model is used. In the regression model, SVM will learn how to obtain the output target: moisture content of the dried material based on the input features: initial moisture content, temperature and humidity of the external environment, internal drying temperature and wind speed and time. Usually, the goal of the regression model is to find a function f(x) that can be as close to the target value y as possible. Assume that the input feature vector is x∈R d , y∈R, then the linear regression model can be expressed as:

[0056] f(x)=ω T x+b

[0057] Where: ω is the weight vector of the regression model, b is the bias term, x is the input feature, and T is the transpose of the vector.

[0058] S4. Usually the linear regression model function needs to consider a certain error, and the error is controlled within the range ε, so that the error of most data points does not exceed ε, and the complexity of the model is minimized. The function to be optimized is as follows:

[0059]

[0060] Where εi, They represent the upper deviation and lower deviation of the ith data point respectively; C is the penalty factor, which is used to control the balance between error and model complexity; ε is the error tolerance, which indicates the tolerance range of the model to prediction error.

[0061] The constraints are:

[0062]

[0063] In the formula, ε i , They represent the upper and lower deviations of the ith data point respectively; C is the penalty factor, which is used to control the balance between error and model complexity. ε is the error tolerance, which indicates the tolerance range of the model to the prediction error.

[0064] S5. Since the initial moisture content of the material in the fruit and vegetable dryer, the temperature and humidity of the external environment, the internal drying temperature and wind speed, and the drying time affect the drying rate of fruits and vegetables in a complex way, the corresponding moisture content relationship is usually nonlinear. Therefore, the SVM regression model needs to use a kernel function to map the input features to a higher-dimensional space in order to find a suitable regression plane in the linear space. The radial basis kernel function is suitable for when there are complex nonlinear relationships between data, and it can effectively handle nonlinear problems. The formula is as follows:

[0065]

[0066] Where: σ is the parameter of the kernel function, which controls the similarity between data points.

[0067] S6. The final linear regression model function optimization includes constraints, so that the error is within the range of ε and the complexity of the regression model is minimized. Through Lagrangian dual optimization, the final optimization goal can be expressed as:

[0068]

[0069] Where a i ,a j ≥0, which is the Lagrange multiplier.

[0070] S7. After determining the SVM linear regression model function, it is necessary to select appropriate hyperparameters, such as penalty parameter C and kernel function σ parameter. The selection of these hyperparameters will affect the training effect of the model. In this embodiment, a plurality of candidate hyperparameter combinations are generated by genetic algorithm (GA), different SVM models are trained, and their fitness is evaluated according to the prediction error of the model. Thus, GA can help find the most suitable parameter configuration.

[0071] The GA algorithm mainly simulates the selection and biological genetic process in nature. Starting from the initial population, through selection, crossover, mutation and other operations, it produces a group that is more conducive to and suitable for environmental survival. This group will continue to evolve and eventually evolve into the group that is most adapted to the environment, and then obtain the optimal solution. Its basic process is as follows: Figure 2 shown.

[0072] S8. Train the SVM regression model using the previously classified training and validation set data with optimized hyperparameters.

[0073] During the training process, SVM will learn the mapping relationship between the initial moisture content of the material, the temperature and humidity of the external environment, the internal drying temperature and wind speed and time, and the moisture content of the drying material, and optimize the model by minimizing the error and interval.

[0074] The specific process is as follows Figure 3 shown.

[0075] S9. After the SVM regression model training is completed, it can be tested with the prepared test group data. Input the initial moisture content of the material, the temperature and humidity of the external environment, the internal drying temperature and wind speed and time. The SVM model will output the predicted moisture content of the material based on the learned mapping relationship.

[0076] This embodiment uses the mean absolute percentage error, root mean square error, mean absolute error and determination coefficient as the criteria for evaluating the quality of the moisture content prediction model. If the criteria are not met, data and hyperparameter optimization will continue. If the criteria are met, the moisture content prediction under different input parameters can be provided for the material drying.

[0077] S10. According to the prediction signal of SVM, the fuzzy PID controller can actively adjust the temperature, humidity and wind speed of the dryer, so as to ensure that the moisture content of fruits and vegetables is gradually reduced to the predetermined target.

[0078] The above embodiments are for explanation of the present invention, not for limitation of the present invention, and the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without creative work inspired by the present invention shall fall within the scope of protection of the present invention.

Claims

1. A fruit and vegetable drying control method based on GA-SVM combined with fuzzy PID, characterized in that The following contents are included: obtaining the drying data of fruits and vegetables to be dried and screening the data, selecting the SVM model and kernel function, using the GA genetic algorithm to obtain the best hyperparameters, training and verifying the SVM prediction model, and actively adjusting the drying temperature and wind speed parameters by the fuzzy PID according to the predicted moisture content; among which: S1. Obtain drying data of fruits and vegetables to be dried, including initial moisture content, temperature and humidity of the external environment, internal drying temperature and wind speed, and moisture content in different time periods; S2. Clean the acquired data, remove duplicate data and process missing values, normalize the data, and finally divide the data set; S3. Select the SVM regression model. In the regression model, the output target moisture content of the dried material is obtained according to the input features and expressed as a function f(x) to make it close to the target value y. Assume that the input feature vector is x∈R d , y∈R, then the linear regression model is expressed as: f(x)=ω T x+b Where: ω is the weight vector of the regression model, b is the bias term, x is the input feature, and T is the transpose of the vector; S4, control error range ε, the function to be optimized is as follows: Where εi, They represent the upper deviation and lower deviation of the i-th data point respectively; C is the penalty factor, which is used to control the balance between error and model complexity; ε is the error tolerance, which indicates the tolerance range of the model to the prediction error; S5, SVM regression model uses kernel function to map input features to a higher dimensional space in order to find a suitable regression plane in linear space; S6. The final linear regression model function optimization includes constraints, so that the error is within the range of ε and the complexity of the regression model is minimized. Through Lagrangian dual optimization, the final optimization goal can be expressed as: Where a i ,a j ≥0, Lagrange multiplier; S7, Generate multiple candidate hyperparameter combinations through genetic algorithm (GA), train different SVM models, and evaluate their fitness based on the prediction error of the model; S8, using the training set and validation set data classified in step S2 to train the SVM regression model through the optimized hyperparameters; S9, SVM regression model training is completed, and the test group data prepared in the previous step S2 is used for testing. The initial moisture content of the material, the temperature and humidity of the external environment, the internal drying temperature and wind speed and time are input. The SVM model will output the predicted moisture content of the material based on the learned mapping relationship; S10. According to the prediction signal of SVM, the fuzzy PID controller actively adjusts the temperature, humidity and wind speed of the dryer to ensure that the moisture content of fruits and vegetables gradually decreases to the predetermined target.

2. The fruit and vegetable drying control method based on GA-SVM combined with fuzzy PID according to claim 1 is characterized in that: The normalization in step S2 is to scale the data to a specified range: [0, 1] or [-1, 1]; the formula is: Where: x′ is the normalized data, x is the input feature, min(x) is the minimum value of the feature, and max(x) is the maximum value of the feature.

3. Fruit and vegetable drying control based on GA-SVM combined with fuzzy PID according to claim 1 or 2 The method is characterized in that In step S2, the data set is divided into: a training set for training the SVM model weights and biases, a validation set for selecting the model's hyperparameters (C, σ) and tuning the model, and a test set for evaluating the performance of the final model.

4. The fruit and vegetable drying control method based on GA-SVM combined with fuzzy PID according to claim 1 is characterized in that: In step S3, in the regression model, the input features of the SVM regression model are: initial moisture content, temperature and humidity of the external environment, internal drying temperature and wind speed and time.

5. The fruit and vegetable drying control method based on GA-SVM combined with fuzzy PID according to claim 1 is characterized in that: In step S4, the constraints of the function to be optimized are: In the formula, ε i , They represent the upper deviation and lower deviation of the i-th data point respectively; C is the penalty factor, which is used to control the balance between error and model complexity.

6. The fruit and vegetable drying control method based on GA-SVM combined with fuzzy PID according to claim 1 is characterized in that: In step S5, the radial basis kernel function is used to process the complex nonlinear relationship between the data. The formula is: Where: σ is the parameter of the kernel function, which controls the similarity between data points.

7. The fruit and vegetable drying control method based on GA-SVM combined with fuzzy PID according to claim 1 is characterized in that: In step S7, the genetic algorithm (GA) algorithm simulates the selection and biological genetic process in nature, starting from the initial population, through selection, crossover, and mutation operations, to produce a more favorable and suitable The group that survives in the environment can then come up with the optimal solution.

8. The fruit and vegetable drying control method based on GA-SVM combined with fuzzy PID according to claim 1 is characterized in that: In step S8, during the training process of the SVM regression model, the SVM will learn the mapping relationship between the initial moisture content of the material, the temperature and humidity of the external environment, the internal drying temperature and wind speed and time, and the moisture content of the drying material, and optimize the model by minimizing the error and interval.

9. The fruit and vegetable drying control method based on GA-SVM combined with fuzzy PID according to claim 1 is characterized in that: In step S9, the mean absolute percentage error, root mean square error, mean absolute error and determination coefficient are used as the criteria for evaluating the quality of the moisture content prediction model. If the criteria are not met, data and hyperparameter optimization will continue. If the criteria are met, the moisture content prediction under different input parameters can be provided for the material drying.

Images