Adaptive control system for electromechanical equipment based on deep deterministic policy gradient algorithm

Abstract

The present invention relates to an adaptive control system for electromechanical equipment based on a deep deterministic policy gradient algorithm, which belongs to the field of intelligent control of electromechanical equipment and solves the problem that the electromechanical equipment control system cannot be adaptively adjusted, resulting in the equipment operating state deviating from the ideal operating target. The present invention collects the operating state, control parameters and operating target data of the electromechanical equipment through a data acquisition module; the state transition model training module trains the LSTM network to obtain a state transition model; the operating target data prediction model training module trains the deep neural network to obtain an operating target data prediction model; the control parameter adaptive model training module trains the control parameter adaptive model based on the DDPG algorithm; the data generation module obtains control parameter data based on the trained control parameter adaptive model, and the control parameter execution module outputs the control parameter data to the electromechanical equipment. The present invention can adjust the operating state of the electromechanical equipment in real time, dynamically and automatically, and improve the intelligence level of the electromechanical equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent control of electromechanical equipment, and specifically relates to an adaptive control system for electromechanical equipment based on a deep deterministic policy gradient algorithm. Background Art

[0002] During the operation of electromechanical equipment, the complex coupling relationships between subsystems often cause slight mutations in control system parameters. These mutations can lead to unstable operating conditions. For example, in testing equipment, abnormal operating conditions during material performance testing can directly affect the accuracy of test results. In production equipment, the current operating state of the equipment can affect the execution of control parameters, potentially leading to energy waste and reduced production efficiency. The former is caused by sudden changes in equipment state, while the latter is caused by the control parameters failing to adapt to the current state of the equipment during actual operation, resulting in failure to achieve ideal output or production.

[0003] Currently, control systems for electromechanical equipment mostly use traditional control strategies, such as PID control and fuzzy control. These control methods can meet routine operating requirements in most cases. However, with the increasing complexity of the equipment's operating environment and working conditions, traditional electromechanical equipment control systems have shown certain limitations when dealing with complex, nonlinear, and time-varying systems. In particular, when it is necessary to adjust control parameters in real time and respond to dynamically changing working conditions, traditional control systems have poor adaptability and are unable to effectively optimize the equipment's operating performance. Adaptive control technology has gradually become an important direction for solving electromechanical equipment control problems. Traditional adaptive control systems improve system performance by adjusting controller parameters. However, many adaptive control systems are prone to over-adjustment or instability in complex systems, and lack the ability to accurately predict the equipment's operating status and optimize it in real time. Summary of the Invention

[0004] In order to solve the problem that the control system of electromechanical equipment cannot be adaptively adjusted during operation, resulting in the operating state of the electromechanical equipment deviating from the ideal operating target, the present invention provides an electromechanical equipment adaptive control system based on a deep deterministic policy gradient algorithm. The system can adaptively adjust the parameters of the future control system under the influence of the current operating state and control parameters of the electromechanical equipment to get as close as possible to the ideal operating target of the electromechanical equipment, thereby improving the performance of the electromechanical equipment and realizing the intelligent upgrade of the electromechanical equipment.

[0005] The technical solutions adopted by the present invention to solve the above-mentioned technical problems are as follows:

[0006] An adaptive control system for electromechanical equipment based on a deep deterministic policy gradient algorithm, the system comprising:

[0007] The data acquisition module is used to collect the operating status data, control parameter data and operating target data of the electromechanical equipment and process the collected data;

[0008] A data storage module, used for storing operation status data, control parameter data and operation target data;

[0009] State transfer model training module, used to The control parameter data and operating status data of a time unit are spliced together to form attribute data, the operating status data of the next time unit is used as label data, the attribute data and the corresponding label data are divided into a first training set and a first test set, and the long short-term memory neural network is trained using the first training set and the first test set. After training, a state transition model is obtained, and the state transition model is used to predict and output the operating status data of the electromechanical equipment in the next time unit;

[0010] An operation target data prediction model training module is used to form attribute data using the control parameter data and operation status data of one time unit, and use the operation target data of the same time unit as label data, divide the attribute data and the label data into a second training set and a second test set, and then use the second training set and the second test set to train the deep neural network. After training, an operation target data prediction model is obtained, and the operation target data prediction model is used to predict and output the operation target data of the electromechanical equipment in the time unit;

[0011] The control parameter adaptive model training module is used to train the control parameter adaptive model based on the deep deterministic policy gradient algorithm with the control parameter adaptive model as the intelligent agent and the electromechanical equipment operation system as the environment. During training, the control parameter data can be adjusted to the control parameter data of the previous time unit. times, first The control parameter data size range of each time unit conforms to formula (1):

[0012] (1)

[0013] in, Indicates the Control parameter data of electromechanical equipment per time unit;

[0014] The learning rate functions of the policy network and value network are set as shown in formula (2):

[0015] (2)

[0016] in, represents the learning rate baseline coefficient, represents the total number of iterations of the deep deterministic policy gradient algorithm, represents a positive integer greater than 0, and Hyperparameters set for the deep deterministic policy gradient algorithm runtime;

[0017] state For the The operating status data of the electromechanical equipment in time units, For the first The relationship between the control parameter data of a time unit and the action of the electromechanical equipment is shown in formula (3):

[0018] (3)

[0019] award As shown in formula (4):

[0020] (4)

[0021] in, For the The actual operation target data of electromechanical equipment under each time unit, For the The target data prediction model outputs the running target data under the time unit, is the relational function;

[0022] The interaction process between the agent and the environment when training the control parameter adaptive model is as follows:

[0023] collect The operating status data and control parameter data of the electromechanical equipment for each time unit are spliced and then standardized and input into the state transfer model and the control parameter adaptive model respectively. The state transfer model outputs the standardized operating status data of the next time unit, and the control parameter adaptive model outputs the adjusted control parameter data. The standardized operating status data is inversely standardized to form the real operating status data. The adjusted control parameter data is added to the current control parameter data to form the control parameter data of the next time unit. The inversely standardized operating status data of the next time unit and the control parameter data are spliced and input into the operation target data prediction model. The operation target data prediction model outputs the predicted operation target data. The operation target data and the real operation target data jointly determine the reward value obtained from the electromechanical equipment operation system in this interaction. The number of interactions between the intelligent agent and the environment in each round is times, and finally obtain the trained control parameter adaptive model;

[0024] Data generation module, used to collect the most recent The operating status data and control parameter data of the electromechanical equipment for a time unit are input into the trained control parameter adaptive model after data standardization and data splicing. The control parameter adaptive model outputs the adjustable control parameter data for the next time unit. The adjustable control parameter data is added to the current control parameter data to obtain the control parameter data for the next time unit.

[0025] The control parameter execution module is used to output the control parameter data of the future time unit obtained by the data generation module to the electromechanical equipment, so that the electromechanical equipment operates according to the control parameter data.

[0026] The beneficial effects of the electromechanical equipment adaptive control system based on the deep deterministic policy gradient algorithm proposed in this invention are:

[0027] This invention addresses the limitations of existing control systems in complex environments by introducing a deep deterministic policy gradient algorithm. Compared to traditional control systems, this invention achieves intelligent optimization of electromechanical equipment performance by adaptively adjusting control parameters, combining historical data predictions with real-time feedback. By combining deep learning with reinforcement learning, this invention not only improves control accuracy but also enhances the system's robustness and adaptability in dynamic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is an architectural diagram of an adaptive control system for electromechanical equipment based on a deep deterministic policy gradient algorithm according to an embodiment of the present invention;

[0029] Figure 2 A training framework for adaptive models with control parameters;

[0030] Figure 3 Return convergence curve for training the adaptive model with control parameters. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0032] like Figure 1As shown, this paper proposes an adaptive control system for electromechanical equipment based on a deep deterministic policy gradient algorithm. This system aims to optimize the operating state of electromechanical equipment and improve equipment performance through intelligent control parameter adjustment. Based on a deep reinforcement learning algorithm, this control system automatically adjusts control parameters to ensure that the equipment approaches the preset ideal operating target under different operating conditions. The core modules of this control system mainly include:

[0033] The data acquisition module is used to collect the operating status data, control parameter data and operating target data of the electromechanical equipment in real time, and process the collected data, where the data processing includes standardization, outlier elimination and null value filling;

[0034] A data storage module, used for storing operation status data, control parameter data and operation target data;

[0035] The state transition model training module is used to train the long short-term memory neural network. After training, the state transition model is obtained to predict the operating state data of the electromechanical equipment in the next time unit.

[0036] The operating target data prediction model training module is used to train the deep neural network. After training, the operating target data prediction model is obtained to predict the operating target data of the electromechanical equipment for one time unit;

[0037] Control parameter adaptive model training module, used to train the control parameter adaptive model based on the deep deterministic policy gradient algorithm;

[0038] A data generation module is used to optimize the control parameters based on the trained control parameter adaptive model to obtain the control parameter data for the next time unit;

[0039] The control parameter execution module is used to output the control parameter data of the next time unit to the electromechanical equipment, so that the electromechanical equipment operates according to the control parameter data to achieve target optimization.

[0040] Specifically, the data acquisition module collects and processes data, where data processing includes data standardization, outlier removal, and null value filling.

[0041] The data acquisition module collects data reflecting the operating status of the electromechanical equipment, namely operating status data; parameter data controlling the operation of the electromechanical equipment, namely control parameter data; and operating target data of the electromechanical equipment. The operating status data is acquired through sensors, the control parameter data is acquired from the control system of the electromechanical equipment, and the operating target data is acquired through sensors.

[0042] Data processing refers to standardizing the above three types of data, eliminating outliers and filling in null values.

[0043] When removing outliers, the data collection module adopts the box plot outlier removal method, and the data above the upper whisker and below the lower whisker of the box plot of each attribute value are replaced by null values.

[0044] When filling in null values, the data collection module uses the K nearest neighbor algorithm, where the domain size of the K nearest neighbor algorithm is set to 5.

[0045] The processed operation status data, control parameter data and operation target data are stored in the data storage module.

[0046] The state transition model can dynamically reflect the state relationship of the electromechanical equipment during operation. The state transition model training module trains the transition model of the electromechanical equipment operation state, so that the state transition model can be used to dynamically reflect the state relationship of the electromechanical equipment during operation. The operating status data of a time unit can accurately predict the operating status data of the next time unit, and its training method is based on deep learning.

[0047] In the state transfer model training module, The control parameter data and operating status data for each time unit are concatenated to form attribute data, with the operating status data for one time unit in the future serving as the label data. This attribute data and the corresponding label data are divided into a first training set and a first test set in a ratio of 7:3. A Long Short Term Memory (LSTM) neural network is then trained using these two sets to generate a state transition model. This model is used to predict and output the operating status data of electromechanical equipment for one time unit in the future. The LSTM network architecture consists of a 16-layer input layer, a 40-layer hidden layer, and a stack of 4 layers. The network consists of an input layer with 240 neurons, four hidden layers with 512 neurons, and an output layer with 8 neurons. The input and hidden layers use the RELU activation function. The LSTM operating parameters include a total number of 400 iterations, a learning rate of 0.0001, an Adam optimizer, a mean squared error loss function, and a training batch size of 256.

[0048] The operation target data prediction model training module trains the operation target data prediction model so that the operation target data of the electromechanical equipment in one time unit can be obtained based on the operation status data and control parameter data of the electromechanical equipment. Its training method is also based on deep learning.

[0049] In the operating target data prediction model training module, attribute data is composed of control parameter data and operating status data for one time unit, and the operating target data for the same time unit is used as label data. This attribute data and label data are divided into a second training set and a second test set in a ratio of 7:3. A deep neural network (DNN) is then trained using the second training set and the second test set to obtain an operating target data prediction model. This model is used to predict the operating target data of the electromechanical equipment for that time unit. The DNN architecture consists of an input layer with 16 neurons, three hidden layers with 512 neurons, and an output layer with 1 neuron. The input and hidden layers use the RELU activation function. The DNN operating parameters include a total number of 200 iterations, a learning rate of 0.001, an Adam optimizer, a mean squared error loss function, and a training batch size of 256.

[0050] The function of the control parameter adaptive model training module is to use the control parameter adaptive model as the intelligent agent and the electromechanical equipment operation system as the environment, and to train the control parameter adaptive model based on the deep deterministic policy gradient algorithm. Finally, the trained control parameter adaptive model is obtained, and the trained control parameter adaptive model is used to adjust the control parameter data of the electromechanical equipment in the next time unit.

[0051] In the control parameter adaptive model training module, after building the neural network of the control parameter adaptive model, the control parameter adaptive model is used as the intelligent agent, the electromechanical equipment operation system is used as the environment, and the control parameter adaptive model is used as the interaction framework between the intelligent agent and the environment. Figure 2 As shown in the figure, the control parameter adaptive model is trained based on the deep deterministic policy gradient algorithm. During training, the adjustable range of the control parameter data is set to the control parameter data of the previous time unit. times, first The control parameter data size range of each time unit conforms to formula (1):

[0052] (1)

[0053] in, Indicates the Control parameter data of electromechanical equipment per time unit, Indicates the Control parameter data of electromechanical equipment per time unit, The value of is 0.06.

[0054] The learning rate functions of the policy network (representing the control parameter adaptive model) and the value network are set as shown in formula (2):

[0055] (2)

[0056] in, represents the learning rate, represents the learning rate baseline coefficient, Indicates the total number of DDPG iterations, represents a positive integer greater than 0, and These are the hyperparameters newly added by the present invention for the DDPG runtime settings. and The two hyperparameters can be adjusted by adjusting the learning rate The convergence speed of the control parameter adaptive model can be converged and converged to a better value.

[0057] The elements required for DDPG training are designed as follows: For the The operating status data of the electromechanical equipment in a time unit; action For the first The relationship between the control parameter data of a time unit and the action of the electromechanical equipment is shown in formula (3):

[0058] (3).

[0059] award In the historical data set, The actual operation target data of electromechanical equipment is recorded in a time unit The running target data output by the running target data prediction model As an independent variable through a functional relationship The output value is shown in formula (4):

[0060] (4)

[0061] in, For the The actual operation target data of electromechanical equipment under each time unit, For the The target data output by the target data prediction model under the time unit is the target data. is a known relationship function. The larger the value, the better.

[0062] The interaction between the agent and the environment when DDPG trains the control parameter adaptive model is as follows, where the agent is the control parameter adaptive model and the environment is the electromechanical equipment operating system:

[0063] collect The operating status data and control parameter data of the electromechanical equipment for each time unit are spliced and then standardized before being input into the state transfer model and the control parameter adaptive model respectively. The state transfer model and the control parameter adaptive model respectively output the standardized operating status data and the adjusted control parameter data for the next time unit. The standardized operating status data are inversely standardized to form the real operating status data. The adjusted control parameter data and the current control parameter data are added to form the control parameter data for the next time unit. The inversely standardized operating status data for the next time unit and the control parameter data for the next time unit are spliced and input into the operating target data prediction model to output the predicted operating target data. The predicted operating target data and the real target value, i.e., the real operating target data, jointly determine the reward value obtained from the electromechanical equipment operating system for this interaction. The reward obtained by the intelligent agent in each round of interaction with the environment As shown in formula (5):

[0064] (5)

[0065] in, Indicates the number of interactions between the agent and the environment in each round.

[0066] The policy network, representing the adaptive control parameter model, has an LSTM network with one input layer of 16 neurons, one hidden layer of 40 neurons, and four stacked layers. The network architecture includes an input layer of 240 neurons, four hidden layers of 256 neurons, and an output layer of 8 neurons. The input and hidden layers use the RELU activation function. The value network has an input layer of 24 neurons, two hidden layers of 256 neurons, and an output layer of 1 neuron. The input and hidden layers use the RELU activation function.

[0067] When training the control parameter adaptive model based on DDPG, the learning rate functions of the policy network and the value network are set as shown in formula (2). The parameters set during the DDPG runtime include: the total number of iterations is 260, the discount rate is 0.96, the soft update parameter is 0.05, the experience replay pool capacity is 1000, the minimum experience replay pool capacity when training the model is 256, the training data batch size is 128, and the learning rate baseline coefficient for the value network and the policy network is . 0.00001, positive integer 10, the number of times the agent interacts with the environment during iteration is 32.

[0068] The data generation module collects the most recent The operating status data and control parameter data of the electromechanical equipment for a time unit are input into the trained control parameter adaptive model after data standardization and data splicing. The control parameter adaptive model outputs the adjustable control parameter data for the next time unit. The control parameter data for the next time unit is added to the current control parameter data to obtain the control parameter data for the next time unit.

[0069] The control parameter execution module outputs the control parameter data for the next time unit, obtained by the data generation module, to the electromechanical device, causing it to operate according to this control parameter data. Adjustments to the control parameter data can be made using a compensator, simply by using the compensator to compensate for the current control parameter data set.

[0070] This invention addresses the limitations of existing control systems in complex environments by introducing a deep deterministic policy gradient algorithm. Compared to traditional control systems, this invention achieves intelligent optimization of electromechanical equipment performance by adaptively adjusting control parameters, combining historical data predictions, and providing real-time feedback. By combining deep learning with reinforcement learning, this invention not only improves control accuracy but also enhances the system's robustness and adaptability in dynamic environments. This invention can dynamically and automatically adjust the operating status of electromechanical equipment in real time, optimizing its operational objectives to ensure optimal performance at all times, thereby enhancing the intelligence level of these devices.

[0071] The following describes the technical solution of the present invention in detail using a rotary kiln for uniform cement production as an example. During the operation of the rotary kiln, the goal of adjusting its control parameter data is to minimize coal energy consumption during uniform cement production.

[0072] The data acquisition module collects and processes data reflecting the operating status of the rotary kiln equipment. The status data has 8 attributes. The data acquisition module collects and processes parameter data that controls the operation of the rotary kiln equipment. The parameter data has 8 attributes. The data acquisition module collects and processes data on the amount of coal consumed by the rotary kiln equipment per unit time. The coal consumption data has 1 feature.

[0073] The state transition model training module concatenates control parameter data and operating status data from seven time units to form attribute data, using the operating status data for one future time unit as label data. This data is divided into training and test sets in a 7:3 ratio, and the state transition model is trained using a long short-term memory (LSTM) neural network. The network architecture employed is an LSTM network with 16 input layers, 40 hidden layers, and four stacked layers. The network consists of an input layer with 240 neurons, four hidden layers with 512 neurons, and an output layer with 8 neurons. The RELU activation function is used in both the input and hidden layers. The LSTM operating parameters include a total number of 400 iterations, a learning rate of 0.0001, an Adam optimizer, a mean squared error loss function, and a training batch size of 256.

[0074] The training module for the target data prediction model uses control parameter data and operating status data for one time unit as attribute data, and coal consumption data for the same time unit as label data. This data is divided into training and test sets in a ratio of 7:3, and the target data prediction model is trained based on a deep neural network (DNN). The DNN network architecture consists of an input layer with 16 neurons, three hidden layers with 512 neurons, and an output layer with 1 neuron. The input and hidden layers use the RELU activation function. The DNN operating parameters include a total number of iterations of 200, a learning rate of 0.001, an Adam optimizer, a mean squared error loss function, and a training batch size of 256.

[0075] The control parameter adaptive model training module trains the control parameter adaptive model. The control parameter adaptive model serves as the interaction framework between the agent and the environment. Figure 2 The adjustable range of the control parameter is -0.06 to 0.06 times of the control parameter data of the previous time unit, that is, The specific implementation process is as follows, where the intelligent agent is the control parameter adaptive model and the environment is the rotary kiln equipment operation system.

[0076] collect The operating status data and control parameter data of the rotary kiln equipment for each time unit are spliced and standardized and then input into the trained state transfer model and control parameter adaptive model respectively. The state transfer model and control parameter adaptive model output the standardized operating status data and the adjusted control parameter data for the next time unit respectively. The standardized operating status data are inversely standardized to form the real operating status data. The adjusted control parameter data and the current control parameter data are added to form the control parameter data for the next time unit. The inversely standardized operating status data for the next time unit and the control parameter data for the next time unit are spliced and input into the trained operation target data prediction model to output the predicted operation target data. The reward value is the actual coal consumption minus the predicted coal consumption. The reward obtained by the intelligent agent in each round of interaction with the environment As shown in formula (5).

[0077] The policy network architecture representing the control parameter adaptive model is an LSTM network with an input layer of 16 neurons, a hidden layer of 40 neurons, and 4 stacked layers. It has an input layer with 240 neurons, four hidden layers with 256 neurons, and an output layer with 8 neurons. The input layer and hidden layers use the RELU activation function. The value network architecture is an input layer with 24 neurons, two hidden layers with 256 neurons, and an output layer with 1 neuron. The input layer and hidden layers use the RELU activation function.

[0078] When training the control parameter adaptive model based on DDPG, the learning rate functions of the policy network and the value network are set as shown in formula (2). The parameters set during the DDPG runtime include: the total number of iterations is 260, the discount rate is 0.96, the soft update parameter is 0.05, the experience replay pool capacity is 1000, the minimum experience replay pool capacity when training the model is 256, the training data batch size is 128, and the learning rate baseline coefficient for the value network and the policy network is . 0.00001, positive integer 10, the number of times the agent interacts with the environment during iteration is 32.

[0079] The return convergence curve of the DDPG training process is as follows Figure 3 As shown in FIG, since the curve converges to a reward value greater than 0, it is proved that the method proposed in the present invention can reduce the coal consumption during the operation of the rotary kiln equipment.

[0080] The data generation module collects the operating status data and control parameter data of the rotary kiln equipment in the last seven time units, and inputs them into the trained control parameter adaptive model after data standardization and data splicing. The control parameter adaptive model outputs the adjusted control parameter data and adds it to the current control parameter data to form the control parameter data for the next one time unit.

[0081] The control parameter execution module outputs the control parameter data of one future time unit obtained by the data generation module to the rotary kiln equipment, so that the rotary kiln equipment operates according to the control parameter data.

[0082] The present invention aims to optimize the operation of electromechanical equipment. It trains a control parameter adaptive model that can output control parameter data for the electromechanical equipment for one future time unit based on the operating status data and control parameter data of historical time units. It can not only adaptively adjust the control parameter data to make the electromechanical equipment as close as possible to the ideal operating target and thus improve the performance of the electromechanical equipment, but also assist in the intelligent upgrade of the electromechanical equipment.

[0083] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0084] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. An adaptive control system for electromechanical equipment based on a deep deterministic policy gradient algorithm, characterized in that: include: The data acquisition module is used to collect the operating status data, control parameter data and operating target data of the electromechanical equipment and process the collected data; A data storage module, used for storing operation status data, control parameter data and operation target data; State transfer model training module, used to The control parameter data and operating status data of a time unit are spliced together to form attribute data, the operating status data of the next time unit is used as label data, the attribute data and the corresponding label data are divided into a first training set and a first test set, and the long short-term memory neural network is trained using the first training set and the first test set. After training, a state transition model is obtained, and the state transition model is used to predict and output the operating status data of the electromechanical equipment in the next time unit; An operation target data prediction model training module is used to form attribute data using the control parameter data and operation status data of one time unit, and use the operation target data of the same time unit as label data, divide the attribute data and the label data into a second training set and a second test set, and then use the second training set and the second test set to train the deep neural network. After training, an operation target data prediction model is obtained, and the operation target data prediction model is used to predict and output the operation target data of the electromechanical equipment in the time unit; The control parameter adaptive model training module is used to train the control parameter adaptive model based on the deep deterministic policy gradient algorithm with the control parameter adaptive model as the intelligent agent and the electromechanical equipment operation system as the environment. During training, the control parameter data can be adjusted to the control parameter data of the previous time unit. times, first The control parameter data size range of each time unit conforms to formula (1): （1） in, Indicates the Control parameter data of electromechanical equipment per time unit; The learning rate functions of the policy network and value network are set as shown in formula (2): （2） in, represents the learning rate baseline coefficient, represents the total number of iterations of the deep deterministic policy gradient algorithm, represents a positive integer greater than 0, and Hyperparameters set for the deep deterministic policy gradient algorithm runtime; state For the The operating status data of the electromechanical equipment in time units, For the first The relationship between the control parameter data of a time unit and the action of the electromechanical equipment is shown in formula (3): （3） award As shown in formula (4): （4） in, For the The actual operation target data of electromechanical equipment under each time unit, For the The target data prediction model outputs the running target data under the time unit, is the relational function; The interaction process between the agent and the environment when training the control parameter adaptive model is as follows: collect The operating status data and control parameter data of the electromechanical equipment for each time unit are spliced and then standardized and input into the state transfer model and the control parameter adaptive model respectively. The state transfer model outputs the standardized operating status data of the next time unit, and the control parameter adaptive model outputs the adjusted control parameter data. The standardized operating status data is inversely standardized to form the real operating status data. The adjusted control parameter data is added to the current control parameter data to form the control parameter data of the next time unit. The inversely standardized operating status data of the next time unit and the control parameter data are spliced and input into the operation target data prediction model. The operation target data prediction model outputs the predicted operation target data. The operation target data and the real operation target data jointly determine the reward value obtained from the electromechanical equipment operation system in this interaction. The number of interactions between the intelligent agent and the environment in each round is times, and finally obtain the trained control parameter adaptive model; Data generation module, used to collect the most recent The operating status data and control parameter data of the electromechanical equipment for a time unit are input into the trained control parameter adaptive model after data standardization and data splicing. The control parameter adaptive model outputs the adjustable control parameter data for the next time unit. The adjustable control parameter data is added to the current control parameter data to obtain the control parameter data for the next time unit. The control parameter execution module is used to output the control parameter data of the future time unit obtained by the data generation module to the electromechanical equipment, so that the electromechanical equipment operates according to the control parameter data.

2. The electromechanical equipment adaptive control system based on the deep deterministic policy gradient algorithm according to claim 1 is characterized in that: The network architecture of the long short-term memory neural network is an LSTM network with an input layer of 16, a hidden layer of 40, and a stack of 4 layers. It has an input layer with 240 neurons, four hidden layers with 512 neurons, and an output layer with 8 neurons. The input layer and hidden layers use the RELU activation function. The operating parameters of the long short-term memory neural network include: the total number of iterations is 400, the learning rate is 0.0001, the optimizer is the Adam optimizer, the loss function is the mean square error, and the training batch size is 256.

3. The electromechanical equipment adaptive control system based on the deep deterministic policy gradient algorithm according to claim 1 or 2, characterized in that: The network architecture of the deep neural network consists of an input layer with 16 neurons, three hidden layers with 512 neurons, and an output layer with 1 neuron. The input layer and hidden layers use the RELU activation function. The operating parameters of the deep neural network include: the total number of iterations is 200, the learning rate is 0.001, the optimizer is the Adam optimizer, the loss function is the mean square error, and the training batch size is 256.

4. The electromechanical equipment adaptive control system based on the deep deterministic policy gradient algorithm according to claim 1 or 2, characterized in that: The network architecture of the policy network is a long short-term memory neural network with an input layer of 16, a hidden layer of 40, and a stack of 4 layers. It has an input layer with 240 neurons, four hidden layers with 256 neurons, and an output layer with 8 neurons. The input layer and hidden layer use the RELU activation function.

5. The electromechanical equipment adaptive control system based on the deep deterministic policy gradient algorithm according to claim 1 or 2, characterized in that: The network architecture of the value network consists of an input layer with 24 neurons, two hidden layers with 256 neurons, and an output layer with 1 neuron. The input layer and hidden layer use the RELU activation function.

6. The electromechanical equipment adaptive control system based on the deep deterministic policy gradient algorithm according to claim 1 or 2, characterized in that: The parameters set during the runtime of the deep deterministic policy gradient algorithm include: total number of iterations is 260, discount rate is 0.96, soft update parameter is 0.05, experience replay pool capacity is 1000, minimum experience replay pool capacity when training the model is 256, training data batch size is 128, learning rate baseline coefficient 0.00001, positive integer 10, the number of times the agent interacts with the environment during iteration is 32.

7. The electromechanical equipment adaptive control system based on the deep deterministic policy gradient algorithm according to claim 1 or 2, characterized in that: The reward obtained by the agent's interaction with the environment in each round is shown in formula (5): （5） in, Represents the reward obtained by the agent in each round of interaction with the environment.

8. The electromechanical equipment adaptive control system based on the deep deterministic policy gradient algorithm according to claim 1 or 2, characterized in that: The value of is 0.06, The value of is 6.

9. The electromechanical equipment adaptive control system based on the deep deterministic policy gradient algorithm according to claim 1 or 2, characterized in that: The data acquisition module uses the box plot outlier elimination method to eliminate outliers in the operating status data, control parameter data and operating target data.

10. The electromechanical equipment adaptive control system based on deep deterministic policy gradient algorithm according to claim 1 or 2, characterized in that: The data acquisition module uses the K-nearest neighbor algorithm to fill in the empty values in the operating status data, control parameter data and operating target data, and the domain size of the K-nearest neighbor algorithm is set to 5.

Images