An energy optimization approach for industrial wireless rechargeable sensor networks

Abstract

The present invention discloses an energy optimization method for an industrial wireless rechargeable sensor network. The method combines the requirements of production line monitoring tasks, the performance of sensors, the performance of mobile charging vehicles, etc., establishes a mathematical model of the industrial wireless rechargeable sensor network, and transforms the target optimization problem into two sub-problems of selecting the optimal sensor activation and optimizing the charging path of the mobile charging vehicle for solution; then the path planning problem of the mobile charging vehicle is modeled into a Markov decision process, and a deep reinforcement learning algorithm is used to train the network model to obtain a deep reinforcement learning model that can be used to solve the path planning problem; secondly, an approximate algorithm based on marginal products is used in combination with the deep reinforcement learning model, and the optimal sensor set and the corresponding optimal charging path of the mobile charging vehicle are obtained by continuously and iteratively selecting the sensors to be activated and calculating the corresponding charging path of the mobile charging vehicle.

Description

Technical Field

[0001] The invention belongs to an energy optimization method, and in particular relates to an energy optimization method for an industrial wireless rechargeable sensor network. Background Art

[0002] Industrial wireless sensor networks are widely used to monitor various parameters on production lines, which greatly helps production automation. However, the sensors in industrial wireless sensor networks are powered by batteries with limited capacity and cannot be recharged. When the sensors are exhausted, they can only be replaced to maintain the operation of the entire wireless sensor network. In industrial environments, it is difficult to replace sensors. This greatly limits the further development of production automation. In particular, with the rapid development of intelligent manufacturing in recent years, there is a higher demand for sensor processing capabilities, which leads to faster sensor power consumption. Therefore, traditional industrial wireless sensor networks are difficult to cope with such scenarios. To solve the above problems, the concept of industrial wireless rechargeable sensor networks came into being. However, the large-scale deployment of industrial wireless rechargeable sensors and the addition of mobile charging vehicles have increased the energy consumption of the entire industrial wireless rechargeable sensor network. In the context of green manufacturing, energy optimization of industrial wireless rechargeable sensor networks is of great significance. At present, energy optimization of industrial wireless rechargeable sensor networks often only considers improving the energy efficiency of mobile charging vehicles, such as using one charging vehicle to charge multiple sensors at the same time; multiple mobile charging vehicles work together to charge sensors, etc. Such unilateral considerations make it difficult to optimize the overall energy of the entire industrial wireless rechargeable sensor network.

[0003] Therefore, when optimizing the energy of industrial wireless rechargeable sensor networks, it is necessary to consider both the energy optimization of sensors and the energy optimization of mobile charging vehicles. This approach has important practical significance. Summary of the invention

[0004] The purpose of the present invention is to provide an energy optimization method for industrial wireless rechargeable sensor networks, so as to achieve overall energy optimization of industrial wireless sensor networks and thus reduce carbon emissions.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is as follows.

[0006] An energy optimization method for an industrial wireless rechargeable sensor network, the method comprising the following steps:

[0007] (1) Determine the energy requirements of the production line monitoring task of the industrial wireless rechargeable sensor network to be optimized, the energy and loss performance of the sensor, and the energy parameters of the mobile charging vehicle;

[0008] (2) Establishing a sensor model, a mobile charging vehicle model, and a production line monitoring task model. The sensor model includes an energy demand model, a charging deadline model, a sensing coverage model, and a sensor energy consumption model. The mobile charging vehicle model includes a mobile charging vehicle energy consumption model and a mobile charging vehicle arrival time model.

[0009] (3) Construct the energy optimization problem for industrial wireless rechargeable sensor networks. The function expression of the energy optimization problem for industrial wireless rechargeable sensor networks is as follows:

[0010]

[0011] Constraints:

[0012]

[0013]

[0014] π g ≠π g' ,g≠g';g=1,L,|H 1 |,g'=1,L,|H 1 |,(c)

[0015]

[0016]

[0017]

[0018] H=H 0 ∪H 1 ,(g)

[0019]

[0020] Where γ is the energy consumed by the mobile charging vehicle per unit distance traveled, dist(π g ,π g+1 ) is the last access target π g and will access the target π g+1 The distance between them, T is the duration of the production line monitoring task cycle, is the power consumption rate of sensor i, For sensor i to task The contribution of Z is the set of all tasks. For the task demand, For mobile charging vehicles to reach the transmission target π g time, is the target deadline (such as the sensor charging deadline), The target π g The power demand, E MCV is the battery capacity of the mobile charging vehicle, S is the set of all sensors, H is all activated sensors, which consists of the sensor set H without charging demand 0 And the sensor set H that needs charging 1 composition, It is the charging path for the mobile charging vehicle.

[0021] The above formula represents minimizing the energy consumption of the mobile charging vehicle and sensors, where constraint (a) indicates that the perception coverage probability contributed by the activated sensor set H needs to be greater than the requirement of each task; constraint (b) indicates that the mobile charging vehicle needs to reach the sensor and charge it before the charging deadline of the sensor; constraint (c) indicates that each sensor can only be visited once on the driving path of the mobile charging vehicle; constraint (d) indicates the energy consumed by the mobile charging vehicle; constraint (e) indicates that the mobile charging vehicle starts from the base and returns to the base after completing the charging task; constraint (f) indicates that the activated sensors are a subset of all sensors; constraint (g) indicates that the activated sensors include a sensor set with charging requirements and a sensor set without charging requirements; constraint (h) indicates the charging path of the mobile charging vehicle;

[0022] (4) The path planning problem of the mobile charging vehicle is modeled as a Markov decision process, and the network is trained using a deep reinforcement learning algorithm to obtain a deep reinforcement learning model for charging path planning of the mobile charging vehicle;

[0023] (5) The marginal product-based approximation algorithm is combined with the deep reinforcement learning model of the mobile charging vehicle charging path planning to jointly solve the optimal sensor set activation problem and the mobile charging vehicle path planning problem to obtain the optimal sensor set and the corresponding mobile charging vehicle charging path.

[0024] In the above method, the sensor model in step (2) includes the energy demand model, charging deadline model, sensing coverage probability model, and energy consumption model of the sensor, which are specifically as follows:

[0025] The mathematical expression of the energy demand model is as follows:

[0026]

[0027] Where: is the energy consumed by sensor i in performing the production line monitoring task per unit time, T is the duration of a production line monitoring task cycle, is the initial energy of sensor i, is the lower limit of energy for sensor i to remain activated. 1 , it means that its initial energy is not enough to continuously perform tasks within the production line monitoring task cycle, and a mobile charging vehicle is needed to replenish its energy, inherent energy demand; on the contrary, when sensor i belongs to H 0 When , its initial energy is sufficient, so there is no energy demand;

[0028] The mathematical expression of the charging cut-off time model is as follows:

[0029]

[0030] This formula means that only when sensor i belongs to H 1 When the charging end time is i , and the charging cut-off time is set when the remaining power of the sensor is consumed to the energy lower limit moment;

[0031] The mathematical expression of the perception coverage probability model is as follows:

[0032]

[0033] Where: represents the production line monitoring task of type j and number m, α i represents the signal attenuation coefficient of sensor i, Represents sensor i and task The Euclidean distance, r i is the sensing radius of sensor i, i∈S j Indicates the task type and task that sensor i can perform So this formula means that when the location of the production line monitoring task is within the sensor's perception radius, and the type of task that the sensor can perform corresponds to the type of the task, the sensor i's perception coverage probability for this task is Otherwise, 0;

[0034] The mathematical expression of the energy consumption model is as follows:

[0035]

[0036] This formula represents the energy consumed by the activated sensor set H in one task cycle;

[0037] Furthermore, the mobile charging vehicle model in step (2) includes an energy consumption model of the mobile charging vehicle and an arrival time model of the mobile charging vehicle, which are specifically as follows:

[0038] Energy consumption model of mobile charging vehicle:

[0039]

[0040] Where: γ represents the energy consumed by the mobile charging vehicle per unit distance traveled, dist(π g-1 ,π g ), represents the distance between two visited targets;

[0041] Arrival time model of mobile charging vehicle:

[0042]

[0043] Where: represents the time when the mobile charging vehicle reaches the last visited target, ε represents the charging efficiency of the mobile charging vehicle, υ represents the driving speed of the mobile charging vehicle, Indicates the driving route of the mobile charging vehicle.

[0044] Furthermore, the mathematical expression of the production line monitoring task model in step (2) is as follows:

[0045]

[0046] Where: Indicates production line monitoring tasks The demand, that is, the set of activated sensors H is used to perform tasks The perception coverage probability contributed by the task must meet the requirements of the task

[0047] In step (4), the steps of using deep reinforcement learning to solve the path planning problem of the mobile charging vehicle are as follows:

[0048] (4.1) Construct the mobile charging vehicle path planning problem:

[0049]

[0050] Constraints:

[0051]

[0052] π g ≠π g' ,g≠g';g=1,L,|H 1 |,g'=1,L,|H 1 |,(c)

[0053]

[0054]

[0055]

[0056] (4.2) The mobile charging vehicle path planning problem is modeled as a Markov decision process:

[0057] 1) State space: including the location of the sensor, the charging deadline of the sensor, the power requirement of the sensor, the location of the mobile charging vehicle, and the remaining power of the mobile charging vehicle;

[0058] The mathematical expression of the state space is:

[0059]

[0060] In the formula is an array containing the states of the above states (sensors and charging vehicles) at time step g;

[0061] 2) Action space: Sensors waiting to be charged that have not yet been charged by the mobile charging vehicle

[0062] 3) Return: In this problem, the sum of the energy consumed by the mobile charging vehicle and the arrival time of the mobile charging vehicle exceeding the cut-off time of the sensor is taken as a penalty (negative return), and its mathematical expression is:

[0063]

[0064] in, For mobile charging vehicles to reach sensor π g Exceed sensor π g The duration of the deadline is expressed as:

[0065]

[0066] 4) Strategy: The strategy for guiding the mobile charging vehicle to drive (initially set to random);

[0067] (4.3) Based on the Markov decision process constructed above, the actor-critic deep reinforcement learning algorithm is used to solve the charging path planning problem of the mobile charging vehicle. The steps are as follows:

[0068] First, an actor network with an attention mechanism is constructed. This actor network contains two recurrent neural networks: an encoding neural network and a decoding neural network. The encoding neural network is constructed using a one-dimensional convolutional layer. The encoding neural network encodes the input (the status information of the sensor and the mobile charging vehicle) and encodes it into a higher-dimensional vector; the decoding neural network is constructed using a gated neural unit (GRU). The decoding neural network combines the attention mechanism to calculate the conditional probability of each input being used as the output of this time step (as the next access target). Its mathematical expression is as follows:

[0069]

[0070]

[0071] Where: d g is the hidden state of the decoding neural network, is the input i (sensor or mobile charging vehicle) at time step g after being encoded by the encoding neural network, ω,W 1 ,W 2 is the network parameter to be learned, and softmax is used for normalization L,

[0072] Next, construct the critic network, which consists of two convolutional layers and one fully connected layer;

[0073] Finally, the two networks are trained using the actor-critic approach, resulting in a deep reinforcement learning model.

[0074] Furthermore, in step (5), the steps of jointly solving the problem of selecting the optimal sensor set activation problem and the path planning problem of the mobile charging vehicle are as follows, combining the marginal product-based approximation algorithm and the deep reinforcement learning model:

[0075] (5.1) Determine the candidate sensor set, whose mathematical expression is:

[0076]

[0077] Concurrently in Indicates that the energy consumption of the mobile charging vehicle is 0;

[0078] (5.2) Select the sensor with the largest marginal product value from the candidate sensor set:

[0079]

[0080] Where: is the corresponding charging path of the mobile charging vehicle when sensor i is added, calculated by the deep reinforcement learning model, where h indicates whether the added sensor has a charging demand. Its mathematical expression is:

[0081]

[0082] When the path obtained by the deep learning model exceeds the charging deadline of any sensor or the total energy consumption of the mobile charging vehicle exceeds its battery capacity, the sensor is deleted. All candidate sensors are continuously looped until the sensor with the largest marginal product value is selected;

[0083] (5.3) Update the activated sensor set H 1 =H 1 ∪{iselected}, the energy consumption of mobile charging vehicles Check every unfinished task, if there is a task in adding i selected After Satisfaction Then it is excluded from the unfinished task set and its corresponding candidate sensor set is deleted from the candidate sensor set. The mathematical expression is:

[0084]

[0085]

[0086] (5.4) Update the candidate sensor set:

[0087] S candidate =S candidate \{i selected}

[0088] (5.5) Repeat steps (5.2)-(5.4) until the requirements of all production line monitoring tasks are met, and the final optimal activated sensor set H and the corresponding charging path of the mobile charging vehicle are output.

[0089] Beneficial effects: Compared with the prior art, the substantial progress and significant effects of the energy optimization method for industrial wireless rechargeable sensor networks provided by the present invention include the following points:

[0090] (1) Reduce the energy consumption of industrial wireless rechargeable sensor networks;

[0091] (2) Charge the sensor in a timely manner so that it can continue to work stably during the mission cycle;

[0092] (3) Reduce environmental pollution and carbon emissions;

[0093] (4) Improve the energy utilization of wireless rechargeable sensor networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 This is a system model diagram of an industrial wireless rechargeable sensor network;

[0095] Figure 2 This is the experimental result diagram. DETAILED DESCRIPTION

[0096] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments of the specification, but the protection scope of the present invention is not limited to the embodiments.

[0097] The present invention provides an energy optimization method for an industrial wireless rechargeable sensor network. First, a sensor model, a mobile charging vehicle model, and a production line monitoring task model are established, and an energy optimization problem for an industrial wireless rechargeable sensor network is constructed based on the above models; then the charging path planning problem of the mobile charging vehicle is modeled into a Markov decision process, and the network is trained using the actor-critic method in deep reinforcement learning to obtain a deep reinforcement learning model that can solve the charging path planning; finally, an approximate algorithm based on marginal products is used, and combined with a deep reinforcement learning model, to jointly solve the problem of selecting the optimal sensor set activation and the path planning problem of the mobile charging vehicle, and finally the optimal sensor set and the corresponding charging path of the mobile charging vehicle are obtained. Compared with the method of only optimizing the charging path planning of the mobile charging vehicle, this joint optimization method more effectively reduces the overall energy consumption of the industrial wireless rechargeable sensor network.

[0098] Specifically, the energy optimization method for industrial wireless rechargeable sensor networks described in the present invention has a system model as follows: Figure 1 As shown, the following steps are included:

[0099] (1) Determine the performance of the production line monitoring task, the performance of the sensor, and the performance of the mobile charging vehicle of the industrial wireless rechargeable sensor network to be optimized.

[0100] (2) Based on step (1), the sensor model, mobile charging vehicle model, and production line monitoring task model are established respectively, and based on the above models, the industrial wireless charging sensor network problem is formed. The specific mathematical expressions of each model are as follows:

[0101] A. The mathematical expression of the energy demand model is as follows:

[0102]

[0103] Where: is the energy consumed by sensor i in performing the production line monitoring task per unit time, T is the duration of a production line monitoring task cycle, is the initial energy of sensor i, is the lower limit of energy for sensor i to remain activated. 1 , it means that its initial energy is not enough to continuously perform tasks within the production line monitoring task cycle, and a mobile charging vehicle is needed to replenish its energy, inherent energy demand; on the contrary, when sensor i belongs to H 0 When , its initial energy is sufficient, so there is no energy demand;

[0104] B. The mathematical expression of the charging cut-off time model is as follows:

[0105]

[0106] This formula means that only when sensor i belongs to H 1 When the charging end time is i , and the charging cut-off time is set when the remaining power of the sensor is consumed to the energy lower limit moment;

[0107] C. The mathematical expression of the perception coverage probability model is as follows:

[0108]

[0109] Where: represents the production line monitoring task of type j and number m, α i represents the signal attenuation coefficient of sensor i, Represents sensor i and task The Euclidean distance, r i is the sensing radius of sensor i, i∈S j Indicates the task type and task that sensor i can perform So this formula means that when the location of the production line monitoring task is within the sensor's perception radius, and the type of task that the sensor can perform corresponds to the type of the task, the sensor i's perception coverage probability for this task is Otherwise, 0;

[0110] D. The mathematical expression of the energy consumption model is as follows:

[0111]

[0112] This formula represents the energy consumed by the activated sensor set H in one task cycle;

[0113] In step (2), the mobile charging vehicle model includes an energy consumption model of the mobile charging vehicle and an arrival time model of the mobile charging vehicle:

[0114] E. Energy consumption model of mobile charging vehicle:

[0115]

[0116] Where: γ represents the energy consumed by the mobile charging vehicle per unit distance traveled, dist(π g-1 ,π g ), represents the distance between two visited targets;

[0117] F. Arrival time model of mobile charging vehicle:

[0118]

[0119] Where: represents the time when the mobile charging vehicle reaches the last visited target, ε represents the charging efficiency of the mobile charging vehicle, υ represents the driving speed of the mobile charging vehicle, Indicates the driving route of the mobile charging vehicle.

[0120] In step (2), the mathematical expression of the production line monitoring task model is as follows:

[0121]

[0122] Where: Indicates production line monitoring tasks The demand, that is, the set of activated sensors H is used to perform tasks The perception coverage probability contributed by the task must meet the requirements of the task

[0123] (3) According to the model established in step (2), the function expression of the energy optimization problem for industrial wireless rechargeable sensor networks is obtained as follows:

[0124]

[0125] Constraints:

[0126]

[0127]

[0128] π g ≠π g' ,g≠g';g=1,L,|H 1 |,g'=1,L,|H 1 |,(c)

[0129]

[0130]

[0131]

[0132] H=H 0 ∪H 1 ,(g)

[0133]

[0134] This formula represents minimizing the energy consumption of the mobile charging vehicle and sensors, where constraint (a) indicates that the perception coverage probability contributed by the activated sensor set H needs to be greater than the requirement of each task; constraint (b) indicates that the mobile charging vehicle needs to reach the sensor and charge it before the charging deadline of the sensor; constraint (c) indicates that each sensor (except the base) can only be visited once on the driving path of the mobile charging vehicle; constraint (d) indicates the energy consumed by the mobile charging vehicle (including the energy consumed for driving and charging the sensor); constraint (e) indicates that the mobile charging vehicle starts from the base and returns to the base after completing the charging task; constraint (f) indicates that the activated sensors are a subset of all sensors; constraint (g) indicates that the activated sensors include a sensor set with charging requirements and a sensor set without charging requirements; constraint (h) indicates the charging path of the mobile charging vehicle.

[0135] (4) Using deep reinforcement learning to solve the path planning problem of the mobile charging vehicle includes the following steps:

[0136] (4.1) Construct the mobile charging vehicle path planning problem:

[0137]

[0138] Constraints:

[0139]

[0140] π g ≠π g' ,g≠g';g=1,L,|H 1 |,g'=1,L,|H 1 |,(c)

[0141]

[0142]

[0143]

[0144] (4.2) The mobile charging vehicle path planning problem is modeled as a Markov decision process:

[0145] 1) State space: including the location of the sensor, the charging deadline of the sensor, the power requirement of the sensor, the location of the mobile charging vehicle, and the remaining power of the mobile charging vehicle. The mathematical expression of the state space is:

[0146]

[0147] In the formula is an array containing the states of the above states (sensors and charging vehicles) at time step g;

[0148] 2) Action space: Sensors waiting to be charged that have not yet been charged by the mobile charging vehicle

[0149] 3) Return: In this problem, the sum of the energy consumed by the mobile charging vehicle and the arrival time of the mobile charging vehicle exceeding the cut-off time of the sensor is taken as a penalty (negative return), and its mathematical expression is:

[0150]

[0151] in, For mobile charging vehicles to reach sensor π g Exceed sensor π g The duration of the deadline is expressed as:

[0152]

[0153] 4) Strategy: The strategy for guiding the driving of the mobile charging vehicle (initially set to random).

[0154] (4.3) Based on the Markov decision process constructed above, the actor-critic deep reinforcement learning algorithm is used to solve the charging path planning problem of the mobile charging vehicle. The steps are as follows:

[0155] First, an actor network with an attention mechanism is constructed. This actor network contains two recurrent neural networks: an encoding neural network and a decoding neural network. The encoding neural network is constructed using a one-dimensional convolutional layer. The encoding neural network encodes the input (the status information of the sensor and the mobile charging vehicle) and encodes it into a higher-dimensional vector; the decoding neural network is constructed using a gated neural unit (GRU). The decoding neural network combines the attention mechanism to calculate the conditional probability of each input being used as the output of this time step (as the next access target). Its mathematical expression is as follows:

[0156]

[0157]

[0158] Where: d g is the hidden state of the decoding neural network, is the input i (sensor or mobile charging vehicle) at time step g after being encoded by the encoding neural network, ω,W 1 ,W 2 is the network parameter to be learned, and softmax is used for normalization L,

[0159] Next, construct the critic network, which consists of two convolutional layers and one fully connected layer;

[0160] Finally, the two networks are trained using the actor-critic approach, resulting in a deep reinforcement learning model.

[0161] (5) The specific steps of combining the marginal product-based approximation algorithm and the deep reinforcement learning model to jointly solve the problem of selecting the optimal sensor set activation problem and the path planning problem of the mobile charging vehicle are as follows:

[0162] (5.1) Determine the candidate sensor set, whose mathematical expression is:

[0163]

[0164] Concurrently in Indicates that the energy consumption of the mobile charging vehicle is 0;

[0165] (5.2) Select the sensor with the largest marginal product value from the candidate sensor set:

[0166]

[0167] Where: is the corresponding charging path of the mobile charging vehicle when sensor i is added, calculated by the deep reinforcement learning model, where h indicates whether the added sensor has a charging demand. Its mathematical expression is:

[0168]

[0169] When the path obtained by the deep learning model exceeds the charging deadline of any sensor or the total energy consumption of the mobile charging vehicle exceeds its battery capacity, the sensor is deleted. All candidate sensors are continuously looped until the sensor with the largest marginal product value is selected;

[0170] (5.3) Update the activated sensor set H 1 =H 1 ∪{i selected}, the energy consumption of mobile charging vehicles Check every unfinished task, if there is a task in adding i selected After Satisfaction Then it is excluded from the unfinished task set and its corresponding candidate sensor set is deleted from the candidate sensor set. The mathematical expression is:

[0171]

[0172]

[0173] (5.4) Update the candidate sensor set:

[0174] S candidate =S candidate \{i selected}

[0175] (5.5) Repeat steps (42)-(44) until the requirements of all production line monitoring tasks are met and the final optimal activated sensor set H and the corresponding charging path of the mobile charging vehicle are output.

[0176] Based on the above methods and steps, an application example is specifically given below to fully illustrate the implementation process of the present invention and the significant technical effects produced thereby.

[0177] A. Parameter settings

[0178] In this embodiment, the size of the industrial area is set to 80 meters x 80 meters; 800 sensors are randomly deployed in this area, and there are 4 types of these sensors, with 200 sensors of each type; accordingly, in this industrial area, there are 4 types of production line monitoring tasks, and the task requirements (coverage probability) are randomly selected within 0.5 to 0.7, and the duration of each task cycle is 1 hour. Therefore, before the start of each task cycle, different types of tasks and task requirements will be randomly generated; there are 4 types of sensor perception radius, which are 10 meters, 15 meters, 20 meters, and 25 meters, the battery capacity is 10.8kJ, the power consumption rate is 0.5J / s, the energy lower limit is 540J, the initial power is randomly selected from 1080J to 3240J, and the attenuation coefficient is randomly selected from 0.1 to 0.3; the battery capacity of the mobile charging vehicle is 128kJ, the driving energy consumption is 20J / m, the driving speed is 2m / s, and the charging rate is 15W.

[0179] B. Result Analysis

[0180] The embodiment of the present invention writes the model and algorithm program based on the Pycharm2020 software platform, and compares it with the traditional method greedy method and the existing method reward-penalty ratio method to reflect the superiority of this method.

[0181] like Figure 2 The experimental results show that with the increase in the number of tasks, the overall energy consumption of the industrial wireless rechargeable sensor network continues to increase. However, compared with the other two methods, it is found that the method proposed in the present invention is superior to the other two methods.

[0182] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An energy optimization method for industrial wireless rechargeable sensor networks, characterized in that: The method comprises the following steps: (1) Determine the energy requirements of the production line monitoring task of the industrial wireless rechargeable sensor network to be optimized, the energy and loss performance of the sensor, and the energy parameters of the mobile charging vehicle; (2) Establishing a sensor model, a mobile charging vehicle model, and a production line monitoring task model. The sensor model includes an energy demand model, a charging deadline model, a sensing coverage model, and a sensor energy consumption model. The mobile charging vehicle model includes a mobile charging vehicle energy consumption model and a mobile charging vehicle arrival time model. (3) Construct the energy optimization problem for industrial wireless rechargeable sensor networks. The function expression of the energy optimization problem for industrial wireless rechargeable sensor networks is as follows: Constraints: p g ≠π g' ,g≠g';g=1,…,|H1|,g'=1,…,|H1|,(c) H=H0∪H1,(g) Where γ is the energy consumed by the mobile charging vehicle per unit distance traveled, dist(π g ,π g+1 ) is the last access target π g and will access the target π g+1 The distance between them, T is the duration of the production line monitoring task cycle, is the power consumption rate of sensor i, For sensor i to task The contribution of Z is the set of all tasks. For the task demand, For mobile charging vehicles to reach the transmission target π g time, The deadline for the target, The target π g The power demand, E MCV is the battery capacity of the mobile charging vehicle, S is the set of all sensors, H is all activated sensors, which consists of the sensor set H0 without charging demand and the sensor set H1 with charging demand. A charging path for mobile charging vehicles; The above formula represents minimizing the energy consumption of the mobile charging vehicle and sensors, where constraint (a) indicates that the perception coverage probability contributed by the activated sensor set H needs to be greater than the requirement of each task; constraint (b) indicates that the mobile charging vehicle needs to reach the sensor and charge it before the charging deadline of the sensor; constraint (c) indicates that each sensor can only be visited once on the driving path of the mobile charging vehicle; constraint (d) indicates the energy consumed by the mobile charging vehicle; constraint (e) indicates that the mobile charging vehicle starts from the base and returns to the base after completing the charging task; constraint (f) indicates that the activated sensors are a subset of all sensors; constraint (g) indicates that the activated sensors include a sensor set with charging requirements and a sensor set without charging requirements; constraint (h) indicates the charging path of the mobile charging vehicle; (4) The path planning problem of the mobile charging vehicle is modeled as a Markov decision process, and the network is trained using a deep reinforcement learning algorithm to obtain a deep reinforcement learning model for charging path planning of the mobile charging vehicle; The mobile charging vehicle path planning problem is modeled as a Markov decision process as follows: 1) State space: including the location of the sensor, the charging deadline of the sensor, the power requirement of the sensor, the location of the mobile charging vehicle, and the remaining power of the mobile charging vehicle; The mathematical expression of the state space is: In the formula is an array containing the states of the upper sensor and the mobile charging vehicle at time step g; 2) Action space: sensors to be charged that have not yet been charged by the mobile charging vehicle; 3) Return: The sum of the energy consumed by the mobile charging vehicle and the arrival time of the mobile charging vehicle exceeding the cut-off time of the sensor is taken as the penalty. The mathematical expression is: in, For mobile charging vehicles to reach sensor π g Exceed sensor π g The duration of the deadline is expressed as: 4) Strategy: Strategy to guide the driving of mobile charging vehicles; Based on the Markov decision process constructed above, the actor-critic deep reinforcement learning algorithm is used to solve the charging path planning problem of the mobile charging vehicle. The steps are as follows: First, an actor network with an attention mechanism is constructed. The actor network includes two recurrent neural networks, an encoding neural network and a decoding neural network. The encoding neural network is constructed using a one-dimensional convolutional layer. The encoding neural network encodes the input and encodes it into a vector with a higher dimension. The decoding neural network is constructed using a gated neural unit. The decoding neural network combines the attention mechanism to calculate the conditional probability of each input being used as the output of this time step. Its mathematical expression is as follows: Where: d g is the hidden state of the decoding neural network, is the input i, obtained by encoding the neural network at time step g, ω, W1, W2 are the network parameters to be learned, and softmax is used to normalize Next, construct the critic network, which consists of two convolutional layers and one fully connected layer; Finally, the actor-critic method is used to train these two networks, resulting in a deep reinforcement learning model; (5) The marginal product-based approximation algorithm is combined with the deep reinforcement learning model of the mobile charging vehicle charging path planning to jointly solve the optimal sensor set activation problem and the mobile charging vehicle path planning problem to obtain the optimal sensor set and the corresponding mobile charging vehicle charging path.

2. The energy optimization method for industrial wireless rechargeable sensor networks according to claim 1 is characterized in that: The sensor model in step (2) is as follows: The mathematical expression of the energy demand model is as follows: Where: is the energy consumed by sensor i in performing the production line monitoring task per unit time, T represents the duration of the production line monitoring task cycle, is the initial energy of sensor i, It is the lower limit of energy for sensor i to remain activated. When the selected activated sensor i belongs to H1, it means that its initial energy is not enough to continuously perform tasks within the production line monitoring task cycle, and a mobile charging vehicle is needed to replenish its energy. On the contrary, when sensor i belongs to H0, its initial energy is sufficient and there is no energy demand. The mathematical expression of the charging cut-off time model is as follows: This formula indicates that only when sensor i belongs to H1, it has a charging cut-off time ddl i , and the charging cut-off time is set when the remaining power of the sensor is consumed to the energy lower limit moment; The mathematical expression of the perception coverage probability model is as follows: Where: represents the production line monitoring task of type j and number m, α i represents the signal attenuation coefficient of sensor i, Represents sensor i and task The Euclidean distance, r i is the sensing radius of sensor i, i∈S j Indicates the task type and task that sensor i can perform This formula indicates that when the location of the production line monitoring task is within the sensor's perception radius and the type of task that the sensor can perform corresponds to the type of the task, the sensor i's perception coverage probability for this task is Otherwise, 0; The mathematical expression of the sensor energy consumption model is as follows: This formula represents the energy consumed by the activated sensor set H in one task cycle.

3. The energy optimization method for industrial wireless rechargeable sensor networks according to claim 1 is characterized in that: The mobile charging vehicle model in step (2) is as follows: Energy consumption model of mobile charging vehicle: Where: γ represents the energy consumed by the mobile charging vehicle per unit distance traveled, dist(π g-1 ,π g ), represents the distance between two visited targets; Arrival time model of mobile charging vehicle: Where: represents the time when the mobile charging vehicle reaches the last visited target, ε represents the charging efficiency of the mobile charging vehicle, υ represents the driving speed of the mobile charging vehicle, Indicates the driving route of the mobile charging vehicle.

4. The energy optimization method for industrial wireless rechargeable sensor networks according to claim 1, characterized in that: In step (2), the mathematical expression of the production line monitoring task model is as follows: Where: Indicates production line monitoring tasks The demand, that is, the set of activated sensors H is used to perform tasks The perception coverage probability contributed by the task must meet the requirements of the task 5. The energy optimization method for industrial wireless rechargeable sensor networks according to claim 1, characterized in that: Step (4) constructs the mobile charging vehicle path planning problem as follows: Constraints: p g ≠π g' ,g≠g';g=1,…,|H1|,g'=1,…,|H1|,(c) 6. The energy optimization method for industrial wireless rechargeable sensor networks according to claim 1, characterized in that: In step (5), the steps of jointly solving the problem of selecting the optimal sensor set activation problem and the path planning problem of the mobile charging vehicle are as follows, combining the marginal product-based approximation algorithm and the deep reinforcement learning model: (5.1) Determine the candidate sensor set, whose mathematical expression is: set up in Indicates that the energy consumption of the mobile charging vehicle is 0; (5.2) Select the sensor with the largest marginal product value from the candidate sensor set. The mathematical expression for the selection is as follows: Where: is the corresponding charging path of the mobile charging vehicle when sensor i is added, calculated by the deep reinforcement learning model, where h indicates whether the added sensor has a charging demand. Its mathematical expression is: When the path obtained by the deep learning model exceeds the charging deadline of any sensor or the total energy consumption of the mobile charging vehicle exceeds its battery capacity, the sensor is deleted; all candidate sensors are continuously cycled until the sensor with the largest marginal product value is selected; (5.3) Update the activated sensor set H1 = H1∪{i selected }, the energy consumption of mobile charging vehicles Check every unfinished task, if there is a task in adding i selected After Satisfaction Then it is excluded from the unfinished task set and its corresponding candidate sensor set is deleted from the candidate sensor set. The mathematical expression is: (5.4) Update the candidate sensor set: S candidate =S candidate \{i selected } (5.5) Repeat steps (5.2)-(5.4) until the requirements of all production line monitoring tasks are met, and the final optimal activated sensor set H and the corresponding charging path of the mobile charging vehicle are output.

Images