Transmission line construction equipment parallel control method and device based on bi-lstm and ddpg algorithm

Abstract

A power transmission line construction equipment parallel control method and device based on Bi-LSTM and DDPG algorithm, the method comprises the following steps: optimizing the weight of the time sequence attention mechanism based on the differential evolution algorithm of the phase mutation strategy, constructing a Bi-LSTM network model through the feature attention mechanism and the optimized time sequence attention mechanism; inputting the construction equipment state data into the Bi-LSTM network model to obtain the construction equipment state prediction result; improving the DDPG algorithm based on the priority of the after-experience playback, constructing a decision control model by using the improved DDPG algorithm, and inputting the construction equipment state prediction result into the decision control model to obtain the optimal control strategy of the power transmission line construction equipment. The improved Bi-LSTM network model provides more accurate construction equipment state prediction, and the decision control model constructed by using the improved DDPG algorithm uses the state prediction result, combines deep learning and policy gradient method, and formulates a more optimized control strategy, so that the optimal decision solution is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of digital control of power transmission lines, and particularly relates to a parallel control method and device for power transmission line construction equipment based on Bi-LSTM and DDPG algorithms. BACKGROUND

[0002] With the rapid development of power transmission lines, power transmission line construction equipment faces severe challenges in control efficiency, stability and flexibility. Specifically, the power transmission and distribution system is becoming increasingly complex, and higher requirements are placed on the real-time, accuracy and intelligence of control. However, traditional control methods based on static models are difficult to adapt to the frequent changes of such complex dynamic systems. In addition, the time series analysis and optimization control algorithms widely used at present also have certain limitations in processing large-scale power grid complex data. In response to the above challenges, it is urgent to introduce cutting-edge technologies such as digital twinning, deep learning and reinforcement learning to build a digital and intelligent power transmission network and achieve efficient, stable and flexible control of line construction equipment.

[0003] In the prior art, the document "Application Status and Development Prospect of Digital Twin Technology in Power Transmission and Transformation Equipment State Evaluation" proposes to deeply integrate digital twin and state evaluation technology, rely on big data analysis, data mining and other technologies to build a power transmission and transformation equipment digital twin, which can realize differentiated prediction and evaluation of equipment state and intelligent diagnosis of faults, and promote the operation and maintenance management of power transmission and transformation equipment to the intelligent era. The document "Discussion on Intelligent Inspection System of Power Transmission Line Based on Digital Twin Technology" proposes a digital twin model of the power transmission line inspection system. Through the digital twin management and control platform, the model realizes panoramic monitoring and state evaluation of power transmission line equipment, as well as online real-time monitoring and early warning of the visual state of line external damage scene; but the existing digital twin system mainly plays an auxiliary decision-making role, rather than directly generating control output, and needs to rely on manual intervention, which limits the actual automatic control capability. The document "Parallel Control of Thermal Power Generating Unit for Large-scale New Energy Consumption" proposes a parallel control system of thermal power generating unit. Based on digital twin and parallel intelligent theory, the system performs parallel execution and feedback correction of artificial control system and actual system to significantly improve the flexible intelligent adjustment capability of the unit. The document "Construction of Digital Twin of Excavation Working Face and Parallel Intelligent Control Method" proposes a parallel intelligent control architecture of excavation working face based on digital twin and physical system. Through the construction of digital twin of excavation working face, the architecture realizes the interactive fusion of digital and physical excavation processes, so as to make advanced digital decision assist and guide physical excavation control, and achieve the goal of cooperation between digital twin and physical entity. The document "Short-term Power Load Forecasting Combined with Improved Residual Network and Bi-LSTM" proposes a bidirectional LSTM (Bi-LSTM) network which can utilize forward and reverse context information at the same time, thereby enhancing the feature representation and correlation modeling capability of LSTM. However, when Bi-LSTM processes long sequences, the intermediate information may still be attenuated or lost, and the model has insufficient interpretability. In addition, the DDPG network parameters are obtained by random initialization, which limits the convergence capability of the initial agent parameters. SUMMARY

[0004] The purpose of the present application is to overcome the above-mentioned defects and problems in the prior art, and to provide a power transmission line construction equipment parallel control method and device based on Bi-LSTM and DDPG algorithm. The improved Bi-LSTM network model provides more accurate construction equipment state prediction, and the decision control model constructed by the improved DDPG algorithm utilizes the state prediction results, combines deep learning and policy gradient method, and formulates a more optimized control strategy, thereby realizing the solution of optimal decision.

[0005] To achieve the above purpose, the technical solution of the present application is:

[0006] A power transmission line construction equipment parallel control method based on a Bi-LSTM and DDPG algorithm, comprising:

[0007] A differential evolution algorithm based on a phase mutation strategy is used to optimize the weights of the time sequence attention mechanism, a Bi-LSTM network model including an input layer, a feature attention layer, a Bi-LSTM network layer, a time sequence attention layer, a full connection layer and an output layer is constructed through the feature attention mechanism and the optimized time sequence attention mechanism, and the collected construction equipment state data is input into the Bi-LSTM network model to obtain a construction equipment state prediction result.

[0008] A DDPG algorithm is improved based on a priority-based post-experience replay, a decision control model is constructed using the improved DDPG algorithm, and the construction equipment state prediction result is input into the decision control model to obtain an optimal control strategy.

[0009] The optimal control strategy is used to control the power transmission line construction equipment.

[0010] A digital twin model of the power transmission line construction equipment is constructed based on digital twin technology, and the digital twin model includes the Bi-LSTM network model and the decision control model.

[0011] Based on the parallel control theory, a parallel intelligent control model between the digital twin model and the power transmission line equipment is constructed using state data bidirectional synchronization technology and control instruction mapping technology.

[0012] The differential evolution algorithm based on the phase mutation strategy includes:

[0013] The basic parameters of the initialization algorithm are initialized, including the population size N P , variable dimension L, iteration number N I , mutation factor F and cross ratio CR initialization; the population is initialized using the following formula:

[0014]

[0015] In the formula, represents the jth dimension optimization variable of the ith individual in the initial population, i.e., the weight of the time sequence attention mechanism; rand(0, 1) represents a random number in the interval (0, 1) subject to uniform distribution;

[0016] The evolution stage of the algorithm is evaluated according to the average Euclidean distance of the population:

[0017]

[0018] In the formula, is the average Euclidean distance of the gth generation population; is the jth dimension optimization variable of the ith individual in the gth generation population; is the best individual in the gth generation population;

[0019] Calculate

[0020]

[0021] wherein, is the normalized average Euclidean distance of the gth generation population; d max is the maximum value of the population average Euclidean distance; d min is the minimum value of the population average Euclidean distance;

[0022] According to Estimate the stage in which the current population is located:

[0023]

[0024] wherein, ST represents the evolution stage of the population; S1, S2, S3 represent the first, second, and third stages, respectively; d c is the stage control factor;

[0025] Update the selection probability of the current stage mutation strategy using the following formula:

[0026]

[0027] wherein, p k is the selection probability of the kth mutation strategy; NS k is the number of times that the test individual generated by the kth mutation strategy in the current generation successfully enters the next generation;

[0028] Generate a test individual after mixing the mutation individual and the target individual, specifically:

[0029]

[0030] wherein, represents the jth dimension optimization variable of the ith mutation individual in the gth generation population; represents the jth dimension optimization variable of the ith target individual in the gth generation population; represents the jth dimension optimization variable of the ith test individual in the g+1th generation population after the crossover operation; j rand is a random integer, j rand ∈ [1, 2,..., L];

[0031] Select the next generation using a greedy strategy, as shown in the following formula:

[0032]

[0033] wherein, For the g+1 generation of the i-th individual; For the g generation of the i-th individual; Indicates the i-th individual in the g+1 generation after the cross operation.

[0034] The input feature extraction link of the Bi-LSTM network layer is improved based on the feature attention mechanism:

[0035]

[0036] In the formula, q t =[q 1，t , q 2，t ,..., q K，t ] is the feature attention weight coefficient vector at time t; W q is a trainable weight matrix; x t =[x 1，t , x 2，t ,..., x K，t ] is the input feature vector; v q is the bias vector of the feature attention weight; σ(·) is the Sigmoid activation function; A t =[A 1，t , A 2，t ... A K，t ] is the feature attention weight vector, K is the number of features; x′ t is the weighted input vector;

[0037] The output layer of the Bi-LSTM network layer is improved based on the optimized time sequence attention mechanism:

[0038]

[0039] In the formula, p t =[p 1，t , p 2，t ,..., p N，t ] is the time sequence attention weight coefficient vector at time t; W r is a trainable weight matrix; H t =[H 1，t , H 2，t ,..., H N，t ] is the hidden layer state of the LSTM network unit; b r is the bias vector of the time sequence attention weight; ReLU(·) is the activation function; B t =[B 1，t , B 2，t ,..., B N，t ] is the time sequence attention weight vector, N is the length of the input sequence time window; H′t is a weighted input vector; is a matrix product.

[0040] The priority-based after-experience replay includes:

[0041] calculating a mean value D of the trajectory target distance traj is:

[0042] D traj (τ) = D traj (ag0, ag1,..., ag T ) = arg max(d t );

[0043] wherein τ is a trajectory; d t is a spatial distance between the expected achieved target g and the achieved target ag at time t;

[0044] calculating a priority value p i of the trajectory as:

[0045] p i = |-k*D trai +b|; wherein k is a slope of an inverse proportional function; and b is an intercept of the inverse proportional function.

[0046] wherein k is a slope of an inverse proportional function; and b is an intercept of the inverse proportional function.

[0047] calculating a priority sampling probability p(τ i ) of the trajectory being replayed as:

[0048]

[0049] wherein N is a total number of trajectories in the experience pool.

[0050] recalculating the reward as:

[0051]

[0052] wherein r1 is a reward calculated according to a spatial distance between the expected achieved target g and the achieved target ag; β is a reward control factor; r2 is a reward obtained at time t+1; r ε is a constant; d t+1 is a spatial distance between the expected achieved target g and the achieved target ag at time t+1.

[0053] the reward of the after-experience is:

[0054]

[0055] wherein r' is a binary reward of the after-experience; r κ is a critical reward value.

[0056] The specific flow of the improved DDPG algorithm is as follows:

[0057] (1) Initialize the DDPG training network parameter θ and update the target network parameter Then initialize the experience replay pool D, the iteration round e and the maximum iteration number E;

[0058] (2) Sample the target state g and the initial state s0;

[0059] (3) Environment interaction:

[0060] a, collect the environment state s t , select the action a t based on the current training network and the ε-greedy strategy;

[0061]

[0062] In the formula, ε is the exploration rate; λ is a randomly generated number between 0 and 1; A is the action set;

[0063] b, execute the action a t , and collect the environment state s t+1 obtained after the digital twin executes the action;

[0064] (4) Calculation of priority sampling;

[0065] a, calculate the spatial distance between the expected target g and the realized target ag;

[0066] b, calculate the mean value of the trajectory target distance;

[0067] c, calculate the priority value of the trajectory;

[0068] d, calculate the priority sampling probability of the trajectory being replayed;

[0069] (5) Post-experience replay:

[0070] a, store the transition sample (s t ‖g, a t , r t , s t+1 ‖g, p, D traj into the experience pool;

[0071] b, sample the trajectory from the experience pool according to the priority sampling probability of the trajectory being replayed;

[0072] c, then sample the sample (s t , a t , s t+1 ) from the trajectory, and sample the virtual target state g' from the trajectory g' ∈ {s t+1..., s T-1};

[0073] d, recompute the reward of the transition sample;

[0074] e, store the post-transition sample (s t ||g′, a t , r′ t , s t+1 ||g′, p, D traj ) into the experience pool D;

[0075] (6) sample a sample from the experience pool D to train the network, calculate the training network Q value and the target network value, and update the training network parameter θ;

[0076]

[0077]

[0078] wherein, is the target value of the sample extracted from the experience replay pool, μ and θ are the learning rate and the training network parameter respectively; Q(·) and are the state-action functions of the training network and the target network respectively;

[0079] (7) copy the training network parameter θ to update the target network parameter

[0080] (8) if t≥T, continue the following steps, otherwise t=t+1, return to step (3);

[0081] (9) if e≥E, exit the loop, otherwise e=e+1, return to step (3).

[0082] The improved DDPG algorithm is used to construct a decision control model, comprising:

[0083] State space: the state of the power transmission line construction equipment, including current, voltage and load prediction value; the prediction value of the power transmission line construction equipment parameter and the equipment operable constraint;

[0084] Action space:

[0085] A t ={δ t , ΔP f};

[0086] wherein, δ t is the switching action of the circuit breaker and the disconnector; ΔP t is the change amount of the construction equipment power;

[0087] Reward function:

[0088]

[0089] In the formula, w i and w j are the weights of the index reward term and the penalty term respectively; r i and p j are the i-th reward term value and the j-th penalty term value respectively.

[0090] A parallel control device for power transmission line construction equipment based on Bi-LSTM and DDPG algorithm, the device is applied to the method, the device comprises:

[0091] A construction equipment state prediction result acquisition module is configured to optimize the weights of a time sequence attention mechanism based on a differential evolution algorithm of a phase mutation strategy, construct a Bi-LSTM network model including an input layer, a feature attention layer, a Bi-LSTM network layer, a time sequence attention layer, a full connection layer and an output layer through a feature attention mechanism and the optimized time sequence attention mechanism, and input the collected construction equipment state data into the Bi-LSTM network model to obtain a construction equipment state prediction result;

[0092] An optimal control strategy acquisition module is configured to improve a DDPG algorithm based on a priority after-experience replay, construct a decision control model by using the improved DDPG algorithm, and input the construction equipment state prediction result into the decision control model to obtain an optimal control strategy;

[0093] A construction equipment control module is configured to control the power transmission line construction equipment through the optimal control strategy.

[0094] A parallel control device for power transmission line construction equipment based on Bi-LSTM and DDPG algorithm, comprising a memory and a processor;

[0095] The memory is configured to store computer program codes and transmit the computer program codes to the processor;

[0096] The processor is configured to execute the method according to the instructions in the computer program codes.

[0097] A computer readable storage medium, the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method.

[0098] Compared with the prior art, the method has the following beneficial effects:

[0099] This invention discloses a parallel control method and apparatus for transmission line construction equipment based on Bi-LSTM and DDPG algorithms. The method first inputs collected construction equipment state data into an improved Bi-LSTM network model to obtain construction equipment state prediction results. Then, the construction equipment state prediction results are input into a decision control model constructed using the improved DDPG algorithm to obtain an optimal control strategy. Finally, the transmission line construction equipment is controlled using the optimal control strategy. This invention provides more accurate construction equipment state prediction through the improved Bi-LSTM network model, while the decision control model constructed using the improved DDPG algorithm, combined with deep learning and policy gradient methods, formulates a more optimized control strategy based on the state prediction results, thereby achieving the solution of the optimal decision and accurately and efficiently solving the control problem of transmission line construction equipment. This invention can improve the control performance and disturbance rejection stability of transmission line construction equipment, promoting the construction of digital intelligent transmission networks. Attached Figure Description

[0100] Figure 1 This is a flowchart of the parallel control method for transmission line construction equipment based on Bi-LSTM and DDPG algorithms according to the present invention.

[0101] Figure 2 It is a digital twin virtual model of the power transmission line construction equipment proposed in this invention.

[0102] Figure 3 This is an architecture diagram of the improved Bi-LSTM network model proposed in this invention.

[0103] Figure 4 This is a schematic diagram of the voltage control results in an embodiment of the present invention.

[0104] Figure 5 This is a schematic diagram of the frequency control results in an embodiment of the present invention.

[0105] Figure 6 This is a schematic diagram of the indicator parameter values ​​set in the embodiments of the present invention.

[0106] Figure 7 This is a comparison chart of the convergence of the algorithm proposed in this invention with other algorithms.

[0107] Figure 8 This is a structural block diagram of the parallel control device for power transmission line construction equipment based on Bi-LSTM and DDPG algorithms according to the present invention.

[0108] Figure 9 This is a structural block diagram of the parallel control device for power transmission line construction equipment based on Bi-LSTM and DDPG algorithms according to the present invention. Detailed Implementation

[0109] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0110] See Figure 1 A parallel control method for transmission line construction equipment based on Bi-LSTM and DDPG algorithms, comprising:

[0111] S1. The differential evolution algorithm (DE) based on the stage mutation strategy optimizes the weights of the temporal attention mechanism. A Bi-LSTM network model including an input layer, a feature attention layer, a Bi-LSTM network layer, a temporal attention layer, a fully connected layer, and an output layer is constructed through the feature attention mechanism and the optimized temporal attention mechanism. The collected construction equipment status data is input into the Bi-LSTM network model to obtain the construction equipment status prediction results.

[0112] S2. The DDPG algorithm is improved based on priority-based post-experience playback (HPER). The improved DDPG algorithm is used to construct a decision control model, and the prediction results of the construction equipment status are input into the decision control model to obtain the optimal control strategy.

[0113] S3. Control the construction equipment of the transmission line through the optimal control strategy.

[0114] With the widespread application of new technologies such as wireless sensor networks, 5G, drone-based power line inspection, and visualization in the field of power transmission lines, the efficiency of power transmission line construction and operation has been significantly improved, laying the foundation for the intelligentization of power transmission lines. Facing the need for comprehensive perception and intelligent analysis of the construction status of power transmission lines, the rapid development of digital twin technology provides a new approach to solving these problems. Based on a general digital twin architecture, this invention proposes a specific digital twin architecture suitable for power transmission line construction equipment. See also Figure 2 This architecture is based on the cloud, pipe, edge, and terminal integrated structure of the power Internet of Things, and takes into account the unique characteristics of transmission line construction equipment. It mainly comprises three key parts: a perception layer, an edge computing layer, and a digital twin layer. Details are as follows:

[0115] a) Sensing Layer: This mainly includes two categories: construction equipment and data acquisition equipment. Construction equipment includes rotary drilling rigs, hydraulic tensioners, hydraulic traction machines, and winches, etc.; data acquisition equipment includes measurement and control equipment, video monitoring equipment, and environmental monitoring equipment, used for data acquisition and control of transmission line construction equipment.

[0116] b) Edge computing layer: The edge computing device contains power intelligent terminals and aggregation gateways with edge computing capabilities. The intelligent terminal is responsible for the access management, data analysis and collection, edge computing, and data forwarding to the cloud of the power line construction equipment. The intelligent terminal, together with the Internet of Things platform, realizes the connection and access of the terminal equipment, provides real-time data interaction, patrol strategy issuance, real-time control, and other open services of the Internet of Things. Edge computing realizes the cooperation of front-end and back-end data, ensuring the all-weather intelligent monitoring of the line. For complex environments with limited communication capabilities, the data is transmitted to the aggregation gateway through multi-hop aggregation, and then the gateway communicates with the cloud platform. This structure not only realizes efficient data transmission and storage, but also ensures real-time monitoring of the line in complex environments.

[0117] c) Digital twin layer: The digital twin layer is a virtual model that corresponds to the actual physical system through virtualization technology, and establishes a bidirectional interaction interface between the digital and physical systems to reflect and simulate the state, behavior, and performance parameters of the physical system in real time.

[0118] The specific functions of the three key parts on the power line construction equipment are as follows: A large amount of data is collected from physical entity equipment through various collection devices in the perception layer. The data is first aggregated and analyzed in the edge computing layer, and then the edge computing device transmits the data to the cloud platform through optical fiber and wireless network. The cloud platform receives the data and transmits it to the digital twin layer, and then performs data integration and simulation operation through modeling management, simulation services, and digital twin co-intelligence technology. Finally, through virtualization and visualization, human-computer interaction is realized, and users can issue control instructions through the visual interaction interface and finally send them to the physical power line construction equipment in the perception layer, realizing intelligent control of the equipment.

[0119] Based on the parallel control theory, and using state data bidirectional synchronization and control instruction mapping technology, a parallel intelligent control architecture of digital and physical fusion is constructed.

[0120] In the smart grid environment, to achieve the coordinated parallel control of the number and material fusion of power line construction equipment, state data bidirectional synchronization technology and control instruction mapping technology play a key role. First, deploy sensors and monitoring devices to collect real-time construction equipment state-related data, such as current, voltage, power, temperature, and other device operating parameters. Then, transmit the state data to the digital twin model to update the model's internal state and accurately reflect the real-time state of the actual system and environmental changes. Second, the digital twin generates control instructions, and through control instruction mapping technology, the control instructions generated by the digital twin model are accurately and effectively mapped to the actual physical system. Then, real-time monitoring of the actual physical system's response, i.e., the state changes of the actual power line construction equipment after implementation. These response data are used to verify the accuracy of the digital twin model and the effectiveness of the optimized control strategy. Finally, set up an online state detection and evaluation module to check the rationality of resource scheduling and load distribution of the physical system after receiving the control instructions of the digital twin. If the control efficiency is lower than expected, start a new round of digital simulation prediction and control update to complete the dynamic adjustment of the physical system. The response data are fed back to the digital twin model to adjust the model parameters and improve its fitting degree to the actual system, ensuring that the model is continuously optimized to improve adaptability.

[0121] Further, based on the digital twin technology, a digital twin model of the power line construction equipment is constructed, which includes a Bi-LSTM network model and a decision control model; based on the parallel control theory, and using state data bidirectional synchronization technology and control instruction mapping technology, a parallel intelligent control model between the digital twin model and the power line equipment is constructed.

[0122] In another embodiment of the present application, a differential evolution algorithm based on phase mutation strategy includes:

[0123] (1) Initialization

[0124] Initialize the basic parameters of the algorithm, including the population size N P , variable dimension L, the number of iterations N I , mutation factor F and the initialization of the cross ratio CR; the population is initialized using the following formula:

[0125]

[0126] In the formula, represents the j-th dimension of the i-th individual in the initial population, i.e., the weight of the timing attention mechanism; rand(0,1) represents a random number in the interval (0,1) that follows a uniform distribution;

[0127] (2) Mutation

[0128] Considering the simplicity and complexity of calculating the mean Euclidean distance (MED), the evolutionary stage of the algorithm is evaluated based on the mean Euclidean distance of the population:

[0129]

[0130] In the formula, is the average Euclidean distance of the g-th generation population; Let be the optimization variable of the j-th dimension for the i-th individual in the g-th generation population; It is the most outstanding individual in the g-th generation population;

[0131] calculate

[0132]

[0133] In the formula, d is the average Euclidean distance of the g-th generation population after normalization; max d represents the maximum value of the population's average Euclidean distance. min This represents the minimum value of the population's average Euclidean distance;

[0134] First, take the maximum value d of the population MED from the initial population MED. max Then, the MED and d of each evolutionary population are compared. max Compare, if MED is greater than d max Then replace d max .

[0135] Finally, according to Estimate the current stage of the population:

[0136]

[0137] In the formula, ST represents the evolutionary stage of the population; S1, S2, and S3 represent the first, second, and third levels, respectively; d c For stage control factors;

[0138] At each stage, the selection probability of the first-generation mutation strategy is initialized to 0.5, and then the selection probability of the mutation strategy in the current stage is updated using the following formula:

[0139]

[0140] In the formula, p k The selection probability of the k mutation strategy; NS k This represents the number of times that test individuals generated by the k-mutation strategy in the current generation successfully enter the next generation.

[0141] In order to avoid each mutation strategy from being lost in the evolutionary process due to poor performance, a constant ξ = 0.01 is added to the selection probability of each mutation strategy.

[0142] In the evolutionary process, first, the stage in which the current generation is located is determined, and then the mutation strategy is selected by using the roulette method according to the selection probability of each strategy in the corresponding stage, so as to realize adaptive selection of the mutation strategy.

[0143] (3) Crossover

[0144] After the mutation individual and the target individual are mixed in parameters, a test individual is generated, and the specific process is as follows:

[0145]

[0146] In the formula, xi g j represents the jth dimension optimization variable of the ith mutation individual in the gth generation population; xi g j represents the jth dimension optimization variable of the ith target individual in the gth generation population; xi g+1 j represents the jth dimension optimization variable of the ith test individual in the g+1th generation population after the crossover operation; j rand is a random integer, j rand ∈ [1, 2,..., L]. The mutation individual refers to a new vector obtained after the mutation operation on the individual vector, and the target individual is a vector without mutation. The target individual mostly contains relevant heuristic experience for making the target optimal, and is a high-quality parent individual left after the crossover selection. When the mixing condition is not met,

[0147] then it is directly equal to By mixing the mutation individual and the target individual, more different features can be introduced into the population, which to some extent guides the search direction, so that the newly generated test individual is more likely to move in a better direction. (4) Selection

[0148] The next generation is selected by using the greedy strategy, as shown in the following formula:

[0149]

[0150] In the formula, xi g+1 represents the ith individual in the g+1th generation;

[0151] xi g represents the ith individual in the gth generation; xi g+1 represents the ith individual in the g+1th generation after the crossover operation. In another embodiment of the present application, referring to

[0152] In another embodiment of the present application, referring to Figure 3 ​The improved Bi-LSTM network model comprises an input layer, a feature attention layer, a Bi-LSTM network layer, a time sequence attention layer and a full connection layer, and the LSTM network is improved based on the feature and time sequence attention mechanisms.

[0153] Firstly, a conventional Bi-LSTM model is constructed. The Bi-LSTM is composed of two stacked LSTM units, one of which processes the forward input sequence and the other of which processes the reverse input sequence, and then the outputs of the two are connected to capture the forward sequential information and the reverse context information. Each LSTM unit comprises an input gate (determining the amount of input information at the current time), a forgetting gate (determining the degree of forgetting the previous state at the current time) and an output gate (determining the amount of output information at the current time), and the three gating mechanisms are defined as follows:

[0154]

[0155] In the formula, i, f and o represent the input gate, the forgetting gate and the output gate respectively; is the Hadamard product; W and b represent the weight matrix and the bias vector of the network; x t , h t and c t are the input, output vector and memory unit of the hidden layer at t period.

[0156] Secondly, in order to solve the problem of feature correlation information loss in the traditional correlation analysis method, the input feature extraction link of the Bi-LSTM network layer is improved based on the feature attention mechanism. The mechanism dynamically allocates attention weights to mine the correlation between the input features and the target features, so as to focus on the relevant features with large information amount and reduce or ignore the influence of irrelevant features. The definition is as follows:

[0157]

[0158] In the formula, q t =[q 1，t , q 2，t ,..., q K，t ] is the feature attention weight coefficient vector at t time; W q is a trainable weight matrix; x t =[x 1，t , x 2，t ,..., x K，t ] is an input feature vector; v qis the bias vector of feature attention weight; and σ(·) is a Sigmoid activation function t = [A 1，t , A 2，t ,..., A K，t ] is a feature attention weight vector, K is the number of features; x' t is a weighted input vector;

[0159] Finally, in order to distinguish the different influences of each historical time on the current prediction output, the present application first improves the DE algorithm based on the phase mutation strategy. In this step, the weight of the time sequence attention mechanism is optimized by the improved DE algorithm. Finally, the output layer of the Bi-LSTM network layer is improved based on the optimized time sequence attention mechanism. This mechanism dynamically allocates weights to different time sequence information, autonomously learns the time sequence correlation between each time data, and thus enhances the information expression of key time. The definition is as follows:

[0160]

[0161] In the formula, p t = [p 1，t , p 2，t ,..., p N，t ] is the time sequence attention weight coefficient vector at t time; W r is a trainable weight matrix; H t = [H 1,t , H 2,t ,..., H N,t ] is the hidden layer state of the LSTM network unit; b r is the bias vector of the time sequence attention weight; ReLU(·) is an activation function; B t = [B 1，t , B 2，t ,..., B N，t ] is a time sequence attention weight vector, N is the length of the input sequence time window; H' t is a weighted input vector; is a matrix product.

[0162] In another embodiment of the present application, the DDPG algorithm realizes the ability to learn the optimal control strategy in the continuous action space by combining deep learning with the policy gradient method. However, the traditional DDPG network has many problems in training convergence speed and agent (control center) reward sparsity. In order to solve these problems, the priority idea is introduced into the post-experience replay mechanism, and the DDPG algorithm is improved based on the priority post-experience replay (PHER). In addition, in view of the possible overestimation problem in the post-experience reward re-estimation, the original reward is reshaped.

[0163] Firstly, a trajectory target distance function is defined, and the smaller the value of the function is, the more helpful the actions taken by the trajectory in achieving the target task to the agent approaching the target task. The trajectory target distance function mainly calculates the mean value between the virtual target and the target task within a round, and calculates the priority probability of the trajectory according to the mean value.

[0164] In a round, a complete trajectory τ can be represented by the tuple (S, A, P, R, γ). A trajectory contains a series of continuous states s t , where t represents the time step, t∈{0,1,...,T}. Each state s t ∈S also contains the state of the expected target, which represents the generalized subset of the expected state. The trajectory distance function D traj depends on the target state τ=s1,a1,s2,a2,...,s T-1 , a T-1 , s T . In the decision-making task, there are two targets, one is the expected target g, and the other is the target ag achieved after performing one step of operation. The agent expects to achieve the expected target after performing a number of steps of operation. In the simulation environment of the task, the spatial positions of the two targets are represented by the Cartesian coordinate system, which are (x g , y g , z g ) and (x ag , y ag , z ag ) respectively. Therefore, the spatial distance of the two targets is:

[0165]

[0166] The mean value D traj of the trajectory target distance is calculated as:

[0167] D traj (τ)=D traj (ag0,ag1,...,ag T )=arg max(d t ).

[0168] In the formula, τ is the trajectory; d t Let g be the spatial distance between the desired target g and the already achieved target ag at time t;

[0169] Secondly, the spatial distance between the desired target *g* and the already achieved target *ag* is calculated. Trajectory experiences with smaller distances are assigned higher priority for replay. At the beginning of each round, the agent begins exploring the environment using a random strategy. Sampled trajectories are stored in an experience pool. When the agent obtains a new trajectory, it calculates the distance using the spatial distance function of the target and stores this distance value along with the trajectory in the experience pool for subsequent priority ranking. Subsequently, this invention uses a function with a negative slope to calculate the priority value of the trajectory because this function is a monotonically decreasing function, and the trajectory distance value D... traj (τ i The smaller the value of ), the higher its priority. Let p be the priority value of each trajectory. i for:

[0170] p i =|-k*D traj +b|;

[0171] In the formula, k is the slope of the inverse proportional function; b is the intercept of the inverse proportional function.

[0172] When sampling from the experience pool, the agent directly uses the proportion of a trajectory's priority value to the sum of all trajectory priority values ​​as the priority sampling probability. This means that higher-priority trajectories are more likely to be sampled and replayed. Mathematically, the priority sampling probability p(τ) of a trajectory being replayed is... i )for:

[0173]

[0174] In the formula, N is the total number of trajectories in the experience pool;

[0175] Finally, the post-hoc sample reward is recalculated. In the goal-conditional DDPG algorithm, the reward for sampled post-hoc experiences is recalculated based on the distance between the achieved goal and the expected goal in a time step, thereby alleviating the overestimation of the post-hoc experience reward. This invention changes the way the reward for post-hoc experience samples is calculated, recalculating the reward based on the distance, as follows:

[0176]

[0177] In the formula, r1 is the reward calculated based on the spatial distance between the desired goal g and the achieved goal ag. In post-hoc experience replay, the experience is a quadruple (ag) t ||g,a t rt , ag t+1 ||g), wherein, ag t represents that the target has been achieved at t time, ag t+1 represents that the target has been achieved at t+1 time; β is a reward control factor; r2 is the reward obtained at t+1 time, which can guide the agent to approach the target more and more; r ε is a constant; d t is the spatial distance between the expected target g and the target ag t achieved at t time; d t+1 is the spatial distance between the expected target g and the target ag t+1 achieved at t+1 time. If d t+1 d t , it indicates that the agent is approaching the target, and r2=r ε ; if d t+1 d t , it indicates that the agent is moving away from the target, and r2=-r ε .

[0178] The reward of the post-experience is:

[0179]

[0180] In the formula, r' is the binary reward of the post-experience; r κ is a critical reward value.

[0181] The priority-based post-experience replay improved DDPG algorithm allows dynamic updating of experience data, and can more effectively utilize resources, improve the sampling frequency, the efficiency and stability of training, which means that the algorithm can converge to a higher reward level faster in the same time.

[0182] Further, the specific process of the improved DDPG algorithm is as follows:

[0183] (1) initialize the DDPN training network parameter θ and update the target network parameter Then, initialize the experience replay pool D, the iteration round e and the maximum iteration number E;

[0184] (2) sample the target state g and initialize the state S0;

[0185] (3) environmental interaction:

[0186] a, collect the environment state S t , select the action a t based on the current training network and the ε-greedy strategy;

[0187]

[0188] where ε is the exploration rate; λ is a randomly generated number between 0 and 1; A is the action set;

[0189] b. Perform action a t , and collect the environment state S obtained after the digital twin performs the action t+1 ;

[0190] (4) Priority sampling calculation;

[0191] a. Calculate the spatial distance between the expected goal g and the achieved goal ag;

[0192] b. Calculate the mean of the trajectory goal distance;

[0193] c. Calculate the priority value of the trajectory;

[0194] d. Calculate the priority sampling probability of the trajectory being replayed;

[0195] (5) Posterior experience replay:

[0196] a. Store the transition sample (s t ‖g, a t , r t , s t+1 ‖g, p, D traj ) into the experience pool;

[0197] b. Sample the trajectory from the experience pool according to the priority sampling probability of the trajectory being replayed;

[0198] c. Then sample the sample (s t , a t , s t+1 ) from the trajectory, and sample the virtual goal state g' ∈ {s t+1 ,..., s T-1} from the trajectory;

[0199] d. Recalculate the reward of the transition sample;

[0200] e. Store the posterior transition sample (s t ‖g', a t , r' t , s t+1 ‖g', p, D traj ) into the experience pool D;

[0201] (6) Sample the sample from the experience pool D to train the network, calculate the training network Q value and the target network value, and update the training network parameter θ;

[0202]

[0203] wherein, is the target value of the sample extracted from the experience replay pool, and μ and θ are the learning rate and the training network parameter, respectively; Q(·) and are the state-action functions of the training network and the target network, respectively;

[0204] (7) The target network parameter is updated by copying the training network parameter θ at a fixed frequency.

[0205] (8) If t≥T, the following steps are continued, otherwise t=t+1, and return to step (3);

[0206] (9) If e≥E, the loop is exited, otherwise e=e+1, and return to step (3).

[0207] In another embodiment of the present application, in the power transmission line construction equipment, through reasonable problem modeling, state representation, action definition and reward design, the improved DDPG algorithm can construct a decision control model to realize dynamic adjustment of the optimal control strategy in the actual control center.

[0208] State space: the state of the power transmission line construction equipment, including the current, voltage and load prediction value; the prediction value of the power transmission line construction equipment parameter and the device operable constraint;

[0209] Action space: adjustment of the device parameter, operation of the switch, etc.; the definition of the action is within the acceptable range of the physical system and consistent with the actual operation;

[0210] A t ={δ t , ΔP t};

[0211] wherein, δ t is the switch action of the circuit breaker and the disconnector; ΔP t is the change amount of the construction equipment power;

[0212] Reward function:

[0213]

[0214] wherein, w i and w j are the weights of the index reward item and the penalty item, respectively; r i and p j are the i-th reward item value and the j-th penalty item value, respectively.

[0215] The reward function is consistent with the optimization objective of the control center, which narrows down the fault range, reduces the loss, and improves the power supply reliability.

[0216] The application constructs a digital twin virtual model of a high-fidelity power transmission line construction equipment, then based on parallel control theory and using state data bidirectional synchronization and control instruction mapping technology, constructs a parallel intelligent control architecture of digital and physical fusion, so that digital intelligent optimization decision assists physical control, realizes intelligent cooperation and common intelligence mutual driving of digital and physical systems. A differential evolution algorithm (DE) based on phase mutation strategy is used to optimize the weights of the time sequence attention mechanism, and then a Bi-LSTM network model is improved through feature attention and the optimized time sequence attention mechanism, so as to dynamically capture the potential correlation between the time sequence characteristics of the construction equipment parameters and enhance the expression of key information; the Bi-LSTM network model is used to predict the state and performance of the construction equipment in the future. The DDPG algorithm is improved based on priority HPER, and the improved DDPG algorithm is used to construct a decision control model. The results predicted by the Bi-LSTM network model are used as one of the inputs of the decision control model, such as the predicted values of the line construction equipment parameters, current and voltage, which are used together with the current environmental state for decision making of the agent (control center). The control center formulates the optimal control decision according to the current state and the future state predicted by the Bi-LSTM network model, and the optimal strategy learned by the DDPG algorithm, such as adjusting the operating parameters or switch state of the construction equipment. In the actual construction process, the operation of the construction equipment is guided according to the decision of the control center, so as to realize the efficient completion of the power transmission line construction task. The improved Bi-LSTM network model provides more accurate state prediction, and the improved DDPG decision control model uses these prediction results to formulate more optimized control decisions by combining deep learning and policy gradient method, so as to realize the solution of the optimal decision. The application can improve the control performance and anti-disturbance stability of the power transmission line construction equipment, and promote the construction of digital intelligent power transmission network.

[0217] In order to verify the effectiveness of the proposed parallel control of digital and physical fusion, various comparison schemes are designed: 1. the proposed parallel control method; 2. parallel control based on improved DDPG decision only; 3. parallel control based on improved Bi-LSTM prediction only; 4. traditional control method. A case is set: construction equipment anomaly. Two important construction equipment are set, the load of which increases to 20%, and adjacent splittable lines are set. The voltage supply pressure of the node is evaluated, the control center formulates a priority power supply strategy for the important equipment, gives a controllable splitting scheme according to the voltage frequency stability analysis, and balances the load to improve the power supply reliability. The priority equipment recovery strategy based on state evaluation prediction is verified in the case scenario, and the voltage frequency anomaly caused by the abnormal increase of the equipment load is dealt with through the priority power supply strategy and effective splitting scheme. When the local line is insufficient due to the load increase of the equipment, part of the equipment load is split to the adjacent line for support in a controllable manner, so as to ensure the voltage frequency stability and power supply demand of the important equipment on the local side.Figure 4 and Figure 5 are the voltage and frequency control results of the case set by the present application. As can be seen from the figure, as the abnormal increase of the device load leads to the synchronous reduction of voltage and frequency, in order to ensure the reliability of power supply, the method of the present application restores the voltage and frequency to the normal level in the shortest possible time by reducing the energy consumption of the device and the complementary energy of the adjacent line.

[0218] Further, in order to intuitively compare the performance indicators of each scheme, the present application designs Figure 6 to represent the comparison results of the performance indicator parameters of the four schemes. As can be seen from the figure, in the frequency and voltage regulation dynamic quality control task, the digital twin driven closed-loop parallel control obtains the lowest voltage / frequency stabilization time of 0.019 / 0.508, and the minimum voltage / frequency dynamic deviation maximum value of 0.021 / 0.18. Compared with other schemes, the maximum deviation of the voltage and frequency dynamic indicators is reduced by more than 40% and 17%, respectively, and the stabilization overshoot time is shortened by nearly 3.5% and more than 34%, which greatly reduces the dependence and adjustment of the transmission line on stability during construction. In addition, the strategy proposed in the present application still obtains the optimal effect after the digital twin control instruction is sent to the physical end compared with the traditional static physical control. The voltage qualification rate of important equipment is improved by 11%, the power supply satisfaction rate of important equipment is also increased by 9.8%, and the load balancing degree is also improved by 6%, which will significantly reduce the possibility of line overload.

[0219] Figure 7 is the comparison of the convergence of the algorithm proposed in the present application and other algorithms. In order to verify the effectiveness of the improved algorithm proposed in the present application, the following comparison methods are set in the case scenario: 1. LSTM-DDPG; 2. LSTM-improved DDPG; 3. Bi-LSTM-improved DDPG; 4. The improved algorithm proposed in the present application. As can be seen from Figure 7 , the performance of the improved algorithm proposed in the present application is better than that of other methods. Specifically, the proposed improved algorithm converges after about 180 iterations, while the comparison method 2, the comparison method 3, and the comparison method 4 converge after about 380, 400, and 780 iterations, respectively, which indicates that the proposed method improves the speed of exploring the optimal strategy by 52.6%, 55%, and 76.9%, respectively. And compared with other methods, the proposed method obtains the highest reward value of -1.03. These significantly reflect the effectiveness of the method proposed in the present application.

[0220] Referring to Figure 8 , the present application also provides a power transmission line construction equipment parallel control device based on Bi-LSTM and DDPG algorithm. The device is applied to the power transmission line construction equipment parallel control method based on Bi-LSTM and DDPG algorithm. The device comprises:

[0221] The construction equipment state prediction result acquisition module is configured to optimize the weight of the time sequence attention mechanism based on a differential evolution algorithm of a stage mutation strategy, and to construct a Bi-LSTM network model including an input layer, a feature attention layer, a Bi-LSTM network layer, a time sequence attention layer, a full connection layer and an output layer through the feature attention mechanism and the optimized time sequence attention mechanism; and the construction equipment state prediction result is obtained by inputting the collected construction equipment state data into the Bi-LSTM network model.

[0222] The optimal control strategy acquisition module is configured to improve the DDPG algorithm based on a priority after-experience replay, to construct a decision control model by using the improved DDPG algorithm, and to obtain the optimal control strategy by inputting the construction equipment state prediction result into the decision control model.

[0223] The construction equipment control module is configured to control the power transmission line construction equipment through the optimal control strategy.

[0224] Referring to Figure 9 The application further provides a power transmission line construction equipment parallel control device based on a Bi-LSTM and DDPG algorithm.

[0225] The storage is configured to store computer program codes and transmit the computer program codes to the processor.

[0226] The processor is configured to execute the power transmission line construction equipment parallel control method based on the Bi-LSTM and DDPG algorithm according to instructions in the computer program codes.

[0227] The application further provides a computer readable storage medium, which stores computer programs, and the computer programs are executed by a processor to implement the power transmission line construction equipment parallel control method based on the Bi-LSTM and DDPG algorithm.

[0228] Generally, computer instructions used to implement the method of the application can be carried by any combination of one or more computer readable storage media. The non-transitory computer readable storage medium can include any computer readable medium except a signal propagating transitorily.

[0229] The computer readable storage medium may, for example, be tangible or intransitory and may include an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, the computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

[0230] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages, and specifically Python language and platform framework based on TensorFlow, PyTorch, etc. suitable for neural network calculation. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0231] The above device and non-transitory computer readable storage medium can refer to the specific description of the power transmission line construction equipment parallel control method based on Bi-LSTM and DDPG algorithm and its beneficial effects, which will not be repeated here.

[0232] Although the embodiments of the present application have been shown and described above, it should be understood that the above-described embodiments are exemplary, and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-described embodiments within the scope of the present application.

Claims

1. A power transmission line construction equipment parallel control method based on Bi-LSTM and DDPG algorithm, characterized in that, The differential evolution algorithm based on the phase mutation strategy is used to optimize the weight of the time sequence attention mechanism, a Bi-LSTM network model including an input layer, a feature attention layer, a Bi-LSTM network layer, a time sequence attention layer, a full connection layer and an output layer is constructed by using the feature attention mechanism and the optimized time sequence attention mechanism, and the collected construction equipment state data is input into the Bi-LSTM network model to obtain a construction equipment state prediction result; The DDPG algorithm is improved based on the priority-based post-experience replay, a decision control model is constructed by using the improved DDPG algorithm, and the construction equipment state prediction result is input into the decision control model to obtain an optimal control strategy; The optimal control strategy is used to control the power transmission line construction equipment.

2. The parallel control method for the power transmission line construction equipment based on the Bi-LSTM and DDPG algorithms according to claim 1, characterized in that: a digital twin model of the power transmission line construction equipment is constructed based on the digital twin technology, and the digital twin model includes the Bi-LSTM network model and the decision control model; The parallel intelligent control model between the digital twin model and the power transmission line equipment is constructed based on the parallel control theory and by using the state data bidirectional synchronization technology and the control instruction mapping technology. The differential evolution algorithm based on the phase mutation strategy comprises: The evolution stage of the algorithm is evaluated according to the average Euclidean distance of the population:

3. The parallel control method for power transmission line construction equipment based on Bi-LSTM and DDPG algorithm according to claim 1, characterized in that, The selection probability of the current stage mutation strategy is updated by using the following formula: Initialization of the basic parameters of the algorithm, including population size N P , variable dimension L, iteration number N I , mutation factor F and crossover ratio CR; the population is initialized by the following formula: In the formula, represents the optimization variable of the jth dimension of the ith individual in the initial population, that is, the weight of the timing attention mechanism; rand(0, 1) represents a random number subject to a uniform distribution in the interval (0, 1); The test individual is generated by mixing the parameters of the mutation individual and the target individual, and the specific process is as follows: wherein is the average Euclidean distance of the gth generation population; is the jth dimension of the optimization variable of the ith individual in the gth generation population; is the fittest individual in the gth generation population; Computing wherein is the average Euclidean distance of the gth generation population after normalization; d max is the maximum value of the population average Euclidean distance; d min is the minimum value of the population average Euclidean distance; According to Estimate the stage in which the current population is: In the formula, ST represents the evolution stage of the population; S1, S2, S3 represent the first, second, and third levels, respectively; d c is a stage control factor; The next generation is selected by using the greedy strategy, and the specific process is as follows: In the formula, p k is the selection probability of the k mutation strategy; NS k is the number of times that the test individual generated by the k mutation strategy in the current generation successfully enters the next generation; The input feature extraction link of the Bi-LSTM network layer is improved based on the feature attention mechanism: wherein denotes the jth dimension of the optimization variable of the ith mutant individual in the gth generation population; denotes the jth dimension of the optimization variable of the ith target individual in the gth generation population; denotes the jth dimension of the optimization variable of the ith test individual in the g+1th generation population after the crossover operation; j rnad is a random integer, j rand ∈ [1, 2,..., L]; The output layer of the Bi-LSTM network layer is improved based on the optimized time sequence attention mechanism: wherein is the g+1th individual of the ith generation; is the ith individual of the gth generation; denotes the ith individual of the g+1th population after the crossover operation.

4. The power transmission line construction equipment parallel control method based on Bi-LSTM and DDPG algorithm according to claim 1, characterized in that, The priority-based post-experience replay comprises: where q t 1，t 2，t K，t is the feature attention weight coefficient vector at time t; W q is a trainable weight matrix; x t 1，t 2，t K，t is the input feature vector; v q is the bias vector of the feature attention weight; σ(·) is the sigmoid activation function; A t 1，t 2，t K，t is the feature attention weight vector, K is the number of features.​​​​​​​​​ x′ t is the weighted input vector; In the formula, k is the slope of the inverse proportional function, and b is the intercept of the inverse proportional function; where p t 1：t 2，t N，t is the time sequence attention weight coefficient vector at time t; W r is a trainable weight matrix; H t 1，t 2，t N，t is the LSTM network unit hidden layer state; b r is the bias vector of the time sequence attention weight; ReLU(·) is the activation function; B t 1，t 2，t N，t is the time sequence attention weight vector, N is the input sequence time window length; H′ t is the weighted input vector; is the matrix product.​​​​​​​​​ 5. The parallel control method for power transmission line construction equipment based on Bi-LSTM and DDPG algorithm according to claim 1, characterized in that, In the formula, N is the total number of trajectories in the experience pool; The mean value D of the trajectory target distances is calculated traj is: D traj (τ) = D traj (ag0, ag1,..., ag T ) = arg max(d t ). In the formula, τ is a trajectory; d t is the spatial distance between the target g to be achieved at time t and the achieved target ag; The priority p of the trajectory is calculated i is: p i = | -k * D traj + b | ; The reward is recalculated as follows: Calculate the priority sampling probability p(τ) of the trajectory being replayed. i )for: The reward of the post-experience is as follows: The specific process of the improved DDPG algorithm is as follows: In the formula, r1 is a reward calculated according to a spatial distance between a desired achievement goal g and an achieved achievement goal ag; β is a reward control factor; r2 is a reward obtained at t+1; r ε is a constant; d t+1 is a spatial distance between the desired achievement goal g and the achieved achievement goal ag at t+1. (2) Sampling the target state g and the initial state S0; where r' is the post-hoc empirical binary reward; r k is the critical reward value.

6. The parallel control method for power transmission line construction equipment based on Bi-LSTM and DDPG algorithm according to claim 1, characterized in that, (3) Environmental interaction: (1) initialize DDQN training network parameters θ and update target network parameters Next, initialize experience replay pool D, iteration round e, and maximum iteration number E; In the formula, ε is the exploration rate, λ is a randomly generated number between 0 and 1, and A is the action set; (4) Calculation of priority sampling; a, collect the environment state S t , select action a based on the current training network and an e-greedy policy t ; a. Calculate the spatial distance between the expected target g and the achieved target ag; b. performing action a t and collecting the environment state S resulting from the digital twin performing the action t+1 ; b. Calculate the mean value of the trajectory target distance; c. Calculate the priority value of the trajectory; d. Calculate the priority sampling probability of the trajectory being replayed; (5) Post-experience replay: b. According to the priority sampling probability of the trajectory being replayed, sample the trajectory from the experience pool; d. Recalculate the reward of the transition sample; a. store the transferred samples (s t ||g, a t , r t , s t+1 ||g, p, D traj ) into the experience pool; (8) If t≥T, continue the following steps, otherwise t=t+1, and return to step (3); c. Then sample a sample (s t , a t , s t+1 ) from the trajectory t+1 ,..., s T-1} from the trajectory (9) If e≥E, exit the loop, otherwise e=e+1, and return to step (3). e. storing the post-transition sample (s t |g′, a t , r t ′, s t+1 |g′, p, D traj ) into the experience pool D; (6) Sample a sample pair from the experience pool D to train the network, calculate the training network q value and the target network value, and update the training network parameter θ; where, is the target value for experience replay pool sampling, μ and θ are the learning rate and the training network parameters, respectively; Q(·) and are the state-action functions of the training network and the target network, respectively. (7) updating the target network parameters by copying the training network parameters θ at a fixed frequency The decision control model is constructed by using the improved DDPG algorithm. ​ 7. The parallel control method for power transmission line construction equipment based on Bi-LSTM and DDPG algorithm according to claim 1, characterized in that, ​ State space: the state of the power transmission line construction equipment, including current, voltage and load prediction values; predicted values of power transmission line construction equipment parameters and equipment operable constraints; Action space: A t = {δ t , ΔP t}; In the formula, δ t is the switching action of the circuit breaker and disconnector; ΔP t is the change amount of the construction equipment power; Reward function: where w i and w j are the weights of the index reward term and the penalty term, respectively; r i and p j are the ith reward term value and the jth penalty term value, respectively.

8. A power transmission line construction equipment parallel control device based on Bi-LSTM and DDPG algorithm, characterized in that, The device is applied to the method of any one of claims 1-7, and the device comprises: a construction equipment state prediction result acquisition module, configured to optimize the weight of a time sequence attention mechanism based on a differential evolution algorithm of a phase mutation strategy, construct a Bi-LSTM network model comprising an input layer, a feature attention layer, a Bi-LSTM network layer, a time sequence attention layer, a full connection layer and an output layer through a feature attention mechanism and the optimized time sequence attention mechanism, and input the collected construction equipment state data into the Bi-LSTM network model to obtain a construction equipment state prediction result; an optimal control strategy acquisition module, configured to improve a DDPG algorithm based on a priority after-experience replay, construct a decision control model using the improved DDPG algorithm, and input the construction equipment state prediction result into the decision control model to obtain an optimal control strategy; a construction equipment control module, configured to control the power transmission line construction equipment through the optimal control strategy. 9.A parallel control device for power transmission line construction equipment based on a Bi-LSTM and DDPG algorithm, characterized in that, comprises a memory and a processor; the memory is configured to store computer program code and transmit the computer program code to the processor; the processor is configured to execute the method according to any one of claims 1-7 according to instructions in the computer program code.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer readable storage medium, and the computer program is executed by the processor to implement the method according to any one of claims 1-7.

Images