A method for controlling energy consumption of a vehicle oven

Abstract

The application discloses a kind of vehicle oven energy consumption control methods, it includes the following steps:100: collection oven data sample, at least including energy consumption data and oven process data, the oven process data includes oven discharge temperature and the frequency of discharge fan;200: select the key features of oven data sample;300: to key features are normalized;400: construct reinforcement learning model, set the action space of reinforcement learning model based on the key features, create the energy consumption reward function of reinforcement learning model;500: the key features and the number of car body in oven are input into the reinforcement learning model and train it, so that: the value of the energy consumption reward function is constantly optimized by the oven discharge temperature and the frequency of discharge fan output by the reinforcement learning model;600: based on trained reinforcement learning model, real-time adjustment is carried out to the control parameters of oven equipment.

Description

Technical Field

[0001] The present invention relates to a control method, and in particular to an energy consumption control method. Background Art

[0002] Automobile manufacturing involves four major processes: stamping, welding, painting, and final assembly. The paint shop is a major energy consumer within the automaker. The paint drying room is a key piece of equipment in the paint shop, used to dry the car bodies. Using four energy sources—natural gas, electricity, chilled water, and compressed air—it ensures that the bodies meet the baking process requirements. Furthermore, the paint drying room is a key energy consumer within the paint shop.

[0003] To ensure the car body baking process meets the requirements, the oven equipment parameters are set based on maximum load. This leaves a significant margin under load conditions, resulting in excessively high exhaust temperatures. Furthermore, the oven equipment's long startup and shutdown times mean that during short shift changes, the equipment remains in production, and energy consumption is directly discharged. This results in significant energy waste, which is then discharged as high-temperature exhaust gases. Real-time control of equipment parameters can address this significant energy waste in ovens.

[0004] The proportional-integral-derivative (PID) control method is a relatively mature technique in industrial control, but it is generally suitable for controlling low-dimensional targets. Drying room equipment has high parameter dimensionality, with characteristic dimensions exceeding 10, which represents a high-dimensional target. Using the proportional-integral-derivative (PID) control method can easily lead to oscillation, resulting in poor control effectiveness and limited energy savings, less than 1% of total energy consumption. Summary of the Invention

[0005] The purpose of the present invention is to provide a vehicle drying room energy consumption control method, which can dynamically adjust the control parameters of the equipment according to changes in the production environment, reduce the temperature of exhaust gas emissions, and significantly reduce energy waste.

[0006] In order to achieve the above object, the present invention proposes a vehicle drying room energy consumption control method, which comprises the steps of:

[0007] 100: Collecting drying room data samples, where the drying room data samples include at least energy consumption data and drying room process data, where the drying room process data includes drying room exhaust temperature and exhaust fan frequency;

[0008] 200: Select key features of drying room data samples;

[0009] 300: Normalize key features;

[0010] 400: Build a reinforcement learning model, set an action space of the reinforcement learning model based on the key features, and create an energy consumption reward function of the reinforcement learning model;

[0011] 500: Input the key features and the number of car bodies in the drying room into the reinforcement learning model to train the model, so that the drying room exhaust temperature and exhaust fan frequency output by the reinforcement learning model continuously optimize the value of the energy consumption reward function;

[0012] 600: Based on the trained reinforcement learning model, the control parameters of the drying room equipment are adjusted in real time.

[0013] In this technical solution, a reinforcement learning model is used to control the control parameters of the drying room equipment. Here, the model is defined as a standard reinforcement learning problem, and the environment E is modeled as a Markov decision process, with the state space S, the action space A, the initial state p(s1), and the transition probability p(s t+1 |s t , a t ), the reward function is r t =r(s t , a t ). The agent selects different actions a at time t. t ∈R N Interact with the environment. After each interaction, the environment releases a new state s t+1 , the behavior of the agent can be defined by a random policy π, which refers to the action mapping π taken by the agent according to the state of the environment: t ->a t This technical solution approaches the upper limit of the control effect by constructing a reasonable reinforcement learning network structure, and improves the accuracy by training the model by continuously adjusting the model hyperparameters.

[0014] In this technical solution, the inputs to the reinforcement learning model are the selected key features (normalized) and the number of car bodies in the drying room. The outputs of the reinforcement learning model are the drying room exhaust temperature and the frequency of the exhaust fan, both of which are drying room process data.

[0015] After obtaining the trained reinforcement learning model, the reinforcement learning model can be deployed online to achieve real-time parameter adjustment of the drying room equipment according to the status of the drying room equipment and the number of car bodies.

[0016] Furthermore, in the vehicle drying room energy consumption control method of the present invention, the drying room data sample further includes at least one of drying room equipment data, conveying equipment data, and drying room production data.

[0017] To achieve the purpose of controlling the energy consumption of the drying room in real time, in some embodiments, the data sampling frequency may be 1 / 60 Hz, that is, one sample is collected every minute.

[0018] Furthermore, in the vehicle drying room energy consumption control method of the present invention, step 100 also includes data cleaning of the collected drying room data samples.

[0019] In some implementations, data can be normalized using custom data structures. Furthermore, drying room equipment experiences power on / off, power outages, and sensor calibration, which can introduce a significant amount of abnormal data. Due to this characteristic of drying room data, a data cleansing step is preferably required to address this abnormal data.

[0020] The abnormal data processing process can include locating abnormal data, matching the cause of the abnormality, and prescribing the appropriate treatment. Abnormal data is located by drawing visual graphics such as histograms and scatter plots. The main causes of abnormalities include power on / off, device power off, and sensor calibration. The cause of the abnormality needs to be matched; after matching, the appropriate treatment can be prescribed. For example, abnormal data related to power on / off can be replaced with the previous value; abnormal data related to device power off can be processed through linear interpolation or data deletion; abnormal data related to sensor calibration can be processed through linear interpolation.

[0021] Furthermore, in the vehicle drying room energy consumption control method of the present invention, step 200 includes: selecting key features of the drying room data sample based on an integrated feature selection method, which includes the steps of:

[0022] For n-dimensional drying room data sample (or feature sample) X={x1,x2,...,x n} (including drying room process data, drying room equipment data, conveying equipment data, drying room production data) are respectively subjected to similarity-based feature selection to obtain the similarity-based feature selection score score1(f k ).

[0023] Among them, in the method based on similarity feature selection, the data set is X∈R nxd , that is, n samples, d features. The similarity between different samples can be expressed in an association matrix X∈R nxn The correlation matrix is ​​symmetric. The utility maximization of the feature subset S in the similarity-based feature selection method can be further decomposed into formula (1):

[0024]

[0025] Among them, U(S) represents the utility of feature subset S, U(f) is the utility function of feature f, is the transformation of the initial eigenvector f, express The derivative of It is the transformation of its affinity matrix S (that is, the feature subset). In general, the more similar the samples are, the closer their eigenvalues ​​are, and the higher the score of the feature is.

[0026] For n-dimensional drying room data sample X={x1,x2,...,x n} respectively perform feature selection based on mutual information to obtain the feature selection score score2(f k ).

[0027] The feature selection algorithm based on mutual information can solve the problem that the feature selection algorithm based on similarity cannot handle feature redundancy.

[0028] The feature selection algorithm based on mutual information requires sample categories and measures the importance of features by the correlation between sample categories and sample features. The theoretical basis of mutual information is entropy. The entropy of a discrete random variable X is defined as follows:

[0029]

[0030] Among them, x i is a specific value of the random variable X, p(x i ) is the value of x i probability.

[0031] The conditional probability of X given the discrete variable Y is formula (3):

[0032]

[0033] Among them, p(y j ) is y j The prior probability, p(x i |y j ) is a given y j Next x i The conditional probability of .

[0034] The information gain or mutual information between X and Y is used to measure the mutual information between X and Y as formula (4):

[0035] I(X;Y)=H(X)-H(X|Y) (4)

[0036] The feature selection algorithm based on mutual information can be unified in formula (5):

[0037]

[0038] Among them, the first term I(X k ; Y) is X k The mutual information with Y, I represents the mutual information, β represents I(X j ;X k) can be adjusted according to actual conditions, X j Indicates other features, X k The second item represents the feature to be scored. Consider X k The smaller the mutual information with other features, the better. In this formula, λ represents I(X j ;X k The weight coefficient of |Y) can be adjusted according to the actual situation. Given Y, in the same Y, X k The greater the mutual information with other features, the better.

[0039] For n-dimensional drying room data sample X={x1,x2,...,x n} respectively perform sparse learning-based feature selection to obtain the sparse learning-based feature selection score score3(f k ).

[0040] Feature selection based on sparse learning involves solving a non-smooth optimization problem. The feature selection algorithm based on sparse learning aims to minimize the fitting error and the sparse regularization term. The sparse regularization term makes many feature coefficients very small or completely zero, and these features can be eliminated. Its unified calculation formula is shown in Equation (6), where the first term is the loss term and the second term is the penalty term:

[0041] min W loss(W;X,y)+α||W|| (6)

[0042] Where W represents the weight coefficient of the model, X represents all features, y represents the output, and α represents the penalty coefficient.

[0043] The feature selection score based on similarity, the feature selection score based on mutual information, and the feature selection score based on sparse learning are summed to obtain score(f k ):

[0044] where f k represents the kth feature.

[0045] Then, the top m features with the highest scores are selected as key features, where m ≤ n.

[0046] In addition, in certain embodiments of the present technical solution, before selecting key features of drying room data samples based on the integrated feature selection method, the sample data may be classified by cluster analysis to reduce the amount of subsequent calculations.

[0047] Furthermore, in the vehicle drying room energy consumption control method described in the present invention, step 200 further includes: sorting features related to (i.e., affecting) equipment energy consumption based on the physical properties of the drying room equipment, and integrating at least the first three features related to equipment energy consumption with key features extracted using the integrated feature selection method to form key features, before proceeding to step 300. For example, natural gas consumption is related to valve opening, or electricity consumption is related to fan frequency. The present technical solution can also select key features based on these physical properties of the equipment, and then integrate them with the key features selected using the integrated feature selection method to form the final selected key features.

[0048] This technical solution performs normalization processing on the selected key features in step 300 to resolve the influence of different dimensions and improve the speed of model convergence. Through normalization processing, all key features are concentrated between 0 and 1.

[0049] Furthermore, in step 400 of the vehicle drying room energy consumption control method of the present invention, the reinforcement learning model adopts a deterministic policy gradient model, which includes an actor and a critic.

[0050] Furthermore, in the vehicle drying room energy consumption control method of the present invention, the actor includes three hidden layers and one fully connected layer, and / or the commentator includes three hidden layers and one fully connected layer.

[0051] In this technical solution, to ensure the process requirements for the car bodies, after selecting the key characteristics of the drying room, it is necessary to set an action space (or control range) for these key characteristics. In this technical solution, based on the process requirements, two action spaces are defined: one for the car-in-the-box state and one for the car-out-of-the-box state. The action space for the car-in-the-box state can be set based on the process parameter range, while the action space for the car-out-of-the-box state can be set based on the equipment requirements. Ultimately, the action space is dynamically selected based on the number of car bodies.

[0052] Furthermore, in step 400 of the vehicle drying room energy consumption control method of the present invention, the energy consumption reward function Reward is:

[0053] Reward=Reward1+Reward2+Reward3;

[0054]

[0055]

[0056]

[0057] Among them, Reward1 represents the reward function of the drying room exhaust temperature, Reward2 represents the reward function of the drying room exhaust fan frequency, Reward3 represents the energy consumption penalty function, mean(temperature) and σ(temperature) represent the average value and standard deviation of the drying room exhaust temperature obtained by standardizing the collected drying room exhaust temperature, respectively; T actual represents the actual exhaust temperature obtained after adjusting the drying room equipment parameters in real time; λ represents the number of car bodies; mean(frequency) and σ(frequency) represent the mean value and standard deviation of the exhaust fan frequency obtained by standardizing the collected exhaust fan frequency of the drying room, respectively; F actual Indicates the actual exhaust fan frequency obtained after adjusting the drying room equipment parameters in real time; max temperature and max frequency Represent the maximum drying room exhaust temperature and the maximum drying room exhaust fan frequency before adjusting the drying room equipment parameters, respectively, which serve as energy consumption thresholds. The Reward3 energy consumption penalty function is introduced as an energy consumption penalty term in this technical solution. For values ​​less than the threshold, no penalty is imposed; for values ​​greater than or equal to the threshold, a negative penalty is imposed.

[0058] The energy consumption reward function is used to evaluate the effect of dynamic output actions, and can make the reinforcement learning model gradually lean towards actions with higher output reward values ​​during training. In order to better evaluate the effect, the present invention constructs a reward function based on energy consumption. The goal of the present invention is to save energy in the drying room. The energy consumption of the drying room is equal to the process energy consumption plus the emission energy consumption. The process energy consumption must be met, so the goal of the present invention can be converted to minimize the emission energy consumption. The emission energy consumption includes air volume and temperature, and the trade-off between the two needs to be considered during design. In addition, the reward function needs to consider the positive evaluation item of energy consumption and the energy consumption penalty item. Based on this, the present invention proposes the above-mentioned energy consumption reward function.

[0059] The vehicle drying room energy consumption control method of the present invention has the following advantages and beneficial effects:

[0060] This invention uses an integrated feature selection method to reduce the dimensionality of drying room data. Normalization is performed to normalize key drying room features, improving model convergence speed. The action space of key features is set to ensure the process requirements of drying room equipment. An energy consumption-based drying room reward function is created to establish an energy consumption evaluation mechanism. The model is trained to improve accuracy by adjusting the reinforcement learning network structure and hyperparameters. Deployment of the trained model enables real-time parameter adjustment of drying room equipment.

[0061] In a preferred embodiment of the present invention, data structure is normalized and noise is removed through data cleaning.

[0062] The method of the present invention can solve the problem of large energy waste in the drying room during the automobile manufacturing process. It can dynamically adjust the key parameters of the equipment according to the number of produced car bodies and changes in the equipment status to achieve energy consumption control, with significant energy-saving effects, and can reduce energy consumption by about 20%. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 The figure is a schematic flow chart of the steps of the vehicle drying room energy consumption control method according to one embodiment of the present invention.

[0064] Figure 2 The present invention is a flowchart of the steps of selecting key features of drying room data samples based on an integrated feature selection method in one embodiment of the vehicle drying room energy consumption control method of the present invention.

[0065] Figure 3 The training process of the reinforcement learning model of the vehicle drying room energy consumption control method described in the present invention in one embodiment is shown.

[0066] Figure 4 The energy-saving effect of the vehicle drying room energy consumption control method described in the present invention in one embodiment is shown. DETAILED DESCRIPTION

[0067] The vehicle drying room energy consumption control method of the present invention will be further explained and illustrated below in conjunction with the accompanying drawings and specific embodiments. However, such explanation and illustration do not constitute an undue limitation to the technical solution of the present invention.

[0068] Figure 1 The figure is a schematic flow chart of the steps of the vehicle drying room energy consumption control method according to one embodiment of the present invention.

[0069] like Figure 1 As shown, in this embodiment, the vehicle drying room energy consumption control method includes the steps of:

[0070] S1: Drying room data samples are collected using IoT technology. These samples include energy consumption data, drying room process data, drying room equipment data, conveyor equipment data, and drying room production data. Drying room process data includes the drying room exhaust temperature and exhaust fan frequency. In one embodiment, a total of 32 drying room energy consumption features are collected, with one sample collected every minute, for a total of approximately 100,000 samples collected over a month.

[0071] S2: Data Cleaning to Handle Abnormal Data: The abnormal data handling process can include locating abnormal data, matching the cause of the abnormality, and prescribing the appropriate treatment. Abnormal data is located by drawing visual graphics such as histograms and scatter plots. Abnormal causes mainly include power on / off, device power off, and sensor calibration. The causes need to be matched; after matching, the appropriate treatment can be prescribed. For example, abnormal data from power on / off can be replaced with the previous value; abnormal data from device power off can be handled through linear interpolation or data deletion; abnormal data from sensor calibration can be handled through linear interpolation.

[0072] S3: Select key features of drying room data samples based on an integrated feature selection method:

[0073] For n-dimensional drying room data sample (or feature sample) X={x1,x2,...,x n} (including drying room process data, drying room equipment data, conveying equipment data, drying room production data) are respectively subjected to similarity-based feature selection to obtain the similarity-based feature selection score score1(f k );

[0074] For n-dimensional drying room data sample X={x1,x2,...,x n} respectively perform feature selection based on mutual information to obtain the feature selection score score2(f k );

[0075] For n-dimensional drying room data sample X={x1,x2,...,x n} respectively perform sparse learning-based feature selection to obtain the sparse learning-based feature selection score score3(f k );

[0076] The feature selection score based on similarity, the feature selection score based on mutual information, and the feature selection score based on sparse learning are summed to obtain score(f k ):

[0077] where f k represents the kth feature.

[0078] Then, the top 11 features with the highest scores are selected as key features. Table 1 lists these 11 key features:

[0079] Table 1. Ranking of key features in one embodiment

[0080]

[0081]

[0082] S4: According to the physical properties of the oven equipment, the features related to the energy consumption of the equipment (i.e. affecting the energy consumption of the equipment) are sorted out, and the top three features related to the energy consumption of the equipment are selected, in this embodiment, the incinerator exhaust fan frequency, the incinerator hearth temperature and the heat preservation 1 zone heating box circulating fan frequency are selected as the three key features, wherein the incinerator hearth temperature is not selected in the 11 key features selected in S3, therefore, the incinerator hearth temperature is introduced as a key feature in this example, and finally 12 key features are selected.

[0083] S5: Normalizing the key features to improve the convergence speed of the model.

[0084] S6: Setting the action space of the reinforcement learning model based on the key features:

[0085] In order to ensure the process requirements of the vehicle body, after the key features of the oven are selected, the action space of the key features needs to be set. In view of the process requirements, the action space is divided into two kinds: vehicle state action space and no vehicle state action space, which can be specifically shown in Table 2 in this embodiment; the oven action space in the vehicle state is set according to the process parameter range; the oven action space in the no vehicle state is set according to the equipment requirements; finally, according to the number of vehicle bodies, the dynamic space is dynamically selected;

[0086] Table 2. Action space

[0087]

[0088]

[0089] S7: Creating the energy consumption reward function Reward of the reinforcement learning model:

[0090] Reward = Reward1 + Reward2 + Reward3;

[0091]

[0092]

[0093]

[0094] Wherein, Reward1 represents the oven discharge temperature reward function, Reward2 represents the oven discharge fan frequency reward function, and Reward3 represents the energy consumption penalty function, mean(temperature) and σ(temperature) represent the mean value and standard deviation of the oven discharge temperature obtained by standardizing the collected oven discharge temperature; T actualrepresents the actual exhaust temperature obtained after adjusting the drying room equipment parameters in real time; λ represents the number of car bodies; mean(frequency) and σ(frequency) represent the mean value and standard deviation of the exhaust fan frequency obtained by standardizing the collected exhaust fan frequency of the drying room, respectively; F actual Indicates the actual exhaust fan frequency obtained after adjusting the drying room equipment parameters in real time; max temperature and max frequency Represent the maximum drying room exhaust temperature and the maximum drying room exhaust fan frequency before adjusting the drying room equipment parameters, respectively, and serve as energy consumption thresholds. The Reward3 energy consumption penalty function is introduced in this technical solution as an energy consumption penalty term. For values ​​below the threshold, no penalty is applied; for values ​​greater than or equal to the threshold, a negative penalty is applied. A higher value for the energy consumption reward function, the better.

[0095] S8: Input the key features and the number of car bodies in the drying room into the constructed reinforcement learning model for training, so that: the drying room exhaust temperature and exhaust fan frequency output by the reinforcement learning model continuously optimize the value of the energy consumption reward function:

[0096] In one embodiment, a deterministic policy gradient model (DDPG model) can be used, which includes an actor and a critic. The actor consists of three hidden layers and one fully connected layer, and the network structure is 574X32X64X32X12, where 574 is the input dimension, that is, the number of environment variables in the state space; 12 is the output dimension, that is, the number of action variables in the action space; the forward propagation excitation function is the rectified linear unit (ReLU). The critic also consists of three hidden layers and one fully connected layer, and the network structure is 586X32X64X32X12, where 586 is the input dimension, which is the sum of the number of environment variables in the state space and the number of action variables in the action space; the forward propagation excitation function is also ReLU. The target network structures of the actor and critic are the same as their respective network structures.

[0097] By continuously adjusting the hyperparameters of the reinforcement learning model, the reinforcement learning model is trained to improve accuracy. In one embodiment, after optimization, the values ​​of the hyperparameters of the reinforcement learning model are as shown in Table 3.

[0098] Table 3. Hyperparameter values

[0099]

[0100] In one embodiment, training is performed according to the above network and hyperparameters, and the energy consumption changes during training are as follows: Figure 3 As shown in the figure, energy consumption fluctuates greatly before the 118th training run, as the reinforcement learning model continuously explores and seeks greater rewards. Afterward, energy consumption stabilizes at around the set value of 160. The figure shows that parameters are constantly exploring to achieve the lowest energy consumption throughout the training process.

[0101] S9: Based on the trained reinforcement learning model, the control parameters of the drying room equipment are adjusted in real time.

[0102] By adopting the above technical solution, the present invention realizes the energy consumption control of the drying room while ensuring the process requirements, with significant energy saving effect, reducing energy consumption by about 25%. Figure 4 The effect of this implementation is shown.

[0103] It should be noted that the prior art within the scope of protection of the present invention is not limited to the embodiments given in this application document. All prior art that does not contradict the solutions of the present invention, including but not limited to prior patent documents, prior publications, prior public uses, etc., can be included in the scope of protection of the present invention.

[0104] In addition, the combination of the various technical features in this case is not limited to the combination described in the claims of this case or the combination described in the specific embodiments. All technical features recorded in this case can be freely combined or combined in any way unless there is a contradiction between them.

[0105] It should also be noted that the above-listed embodiments are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above-listed embodiments. Similar variations or modifications that can be directly derived from or easily conceived by those skilled in the art based on the disclosure of the present invention should fall within the scope of protection of the present invention.

Claims

1. A method for controlling energy consumption in a vehicle drying room, characterized in that: Including steps: 100: Collecting drying room data samples, where the drying room data samples include at least energy consumption data and drying room process data, where the drying room process data includes drying room exhaust temperature and exhaust fan frequency; 200: Select key features of drying room data samples; 300: Normalize key features; 400: Constructing a reinforcement learning model, setting an action space of the reinforcement learning model based on the key features, and creating an energy consumption reward function for the reinforcement learning model; the action space includes an action space for a state where a car is in the drying room and an action space for a state where no car is in the drying room; 500: Input the key features and the number of car bodies in the drying room into the reinforcement learning model to train the model, so that the temperature of the drying room exhaust and the frequency of the exhaust fan output by the reinforcement learning model continuously optimize the value of the energy consumption reward function; 600: Based on the trained reinforcement learning model, the control parameters of the drying room equipment are adjusted in real time.

2. The vehicle drying room energy consumption control method according to claim 1, characterized in that: The drying room data sample further includes at least one of drying room equipment data, conveying equipment data, and drying room production data.

3. The vehicle drying room energy consumption control method according to claim 1, characterized in that: The step 100 further includes performing data cleaning on the collected drying room data samples.

4. The vehicle drying room energy consumption control method according to claim 1, characterized in that: The step 200 includes: selecting key features of the drying room data sample based on an integrated feature selection method, which includes the steps of: For n-dimensional drying room data sample X={x1,x2,...,x n } respectively perform similarity-based feature selection to obtain the similarity-based feature selection score score1(f k ); For n-dimensional drying room data sample X={x1,x2,...,x n } respectively perform feature selection based on mutual information to obtain the feature selection score score2(f k ); For n-dimensional drying room data sample X={x1,x2,...,x n } respectively perform sparse learning-based feature selection to obtain the sparse learning-based feature selection score score3(f k ); The feature selection score based on similarity, the feature selection score based on mutual information, and the feature selection score based on sparse learning are summed to obtain score(f k ): Where f represents the kth feature; The top m features with the highest scores are selected as key features, where m ≤ n.

5. The vehicle drying room energy consumption control method according to claim 4, characterized in that: The step 200 further includes: sorting the features related to the energy consumption of the equipment according to the physical properties of the drying room equipment, and fusing at least the first three features related to the energy consumption of the equipment with the key features selected based on the integrated feature selection method as the key features, and then proceeding to step 300.

6. The vehicle drying room energy consumption control method according to claim 1, characterized in that: In step 400 , the reinforcement learning model adopts a deterministic policy gradient model, which includes an actor and a critic.

7. The vehicle drying room energy consumption control method according to claim 6, characterized in that: The actor includes three hidden layers and one fully connected layer, and the critic includes three hidden layers and one fully connected layer.

8. The vehicle drying room energy consumption control method according to any one of claims 1 to 7, characterized in that: In step 400, the energy consumption reward function Reward is: Reward=Reward1+Reward2+Reward3; Among them, Reward1 represents the reward function of the drying room exhaust temperature, Reward2 represents the reward function of the drying room exhaust fan frequency, Reward3 represents the energy consumption penalty function, mean(temperature) and σ(temperature) represent the average value and standard deviation of the drying room exhaust temperature obtained by standardizing the collected drying room exhaust temperature, respectively; T actual represents the actual exhaust temperature obtained after adjusting the drying room equipment parameters in real time; λ represents the number of car bodies; mean(frequency) and σ(frequency) represent the mean value and standard deviation of the exhaust fan frequency obtained by standardizing the collected exhaust fan frequency of the drying room, respectively; F actual Indicates the actual exhaust fan frequency obtained after adjusting the drying room equipment parameters in real time; max temprature and max frequency They respectively represent the maximum value of the drying room exhaust temperature and the maximum value of the drying room exhaust fan frequency before the drying room equipment parameters are adjusted.

Images