Work order path planning methods, devices, electronic equipment, and media for photovoltaic power plants

Abstract

This invention discloses a method, apparatus, electronic device, and medium for work order path planning in a photovoltaic power station. The method includes: acquiring historical work order information of the photovoltaic power station and constructing a graph model based on the historical work order information; establishing a multi-head attention graph network prediction model to learn the features of each node in the graph model and the relationships between each node based on the constructed graph model; inputting the information of the work order to be planned and the information of the maintenance personnel in an idle state into the multi-head attention graph network prediction model, and outputting a first time consumption; determining a second time consumption based on the position of each work order to be planned; and processing the first time consumption and the second time consumption based on a particle swarm optimization algorithm to determine the work order path for each maintenance personnel. This method can determine a highly accurate planning path.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power plant maintenance technology, and in particular to a work order path planning method, device, electronic equipment and medium for photovoltaic power plants. Background Technology

[0002] In related technologies, the number of photovoltaic power plants is increasing rapidly, which has led to an increase in the demand for photovoltaic power plant maintenance. During the operation of photovoltaic power plants, it is necessary to efficiently arrange maintenance personnel to handle multiple work orders. However, due to the different abilities of maintenance personnel, the different urgency of different work orders, and the unknown maintenance time of work orders, the complexity of large-scale maintenance work orders and multi-maintenance personnel scheduling has increased dramatically, which makes work order path planning limited and unable to take into account multiple influencing factors to obtain a reasonable and accurate work order path plan. Summary of the Invention

[0003] Therefore, it is necessary to provide a method, device, electronic equipment, and medium for work order path planning in photovoltaic power plants that can solve the problem of inaccurate work order planning, in response to the above-mentioned technical issues.

[0004] A work order path planning method for a photovoltaic power plant includes the following steps:

[0005] The historical work order information of the photovoltaic power station is obtained, and a graph model is constructed based on the historical work order information; wherein, the nodes in the graph model represent historical work orders, and the edges in the graph model represent the movement time between the historical work orders;

[0006] Based on the constructed graph model, a multi-head attention graph network prediction model is established to learn the features of each node in the graph model and the relationships between each node;

[0007] The information of the work order to be planned and the information of the maintenance personnel in an idle state are input into the multi-head attention graph network prediction model, and the first time is output; the first time represents the predicted time for the maintenance personnel to complete the work order to be planned;

[0008] The second time consumption is determined based on the location of each work order to be planned; the second time consumption represents the movement time between each work order to be planned.

[0009] The first and second time consumption times are processed based on the particle swarm optimization algorithm to determine the work order path for each maintenance worker; the work order path is used to record the work orders and their order that the maintenance worker needs to process.

[0010] In the above scheme, the step of establishing a multi-head attention graph network prediction model based on the constructed graph model to learn the features of each node in the graph model and the relationships between each node includes:

[0011] Determine the node characteristics of node i and the node characteristics of its neighboring node j in the graph model;

[0012] Based on multi-head attention, the node features of node i and the node features of neighboring node j are processed to calculate the attention weight between node i and neighboring node j.

[0013] The attention weights are normalized to obtain attention coefficients;

[0014] The attention coefficient is used to perform weighted aggregation of the node features of the neighboring node j to obtain the weighted sum of the neighboring node features of node i;

[0015] The node features of node i are updated based on the weighted sum of the features of its neighboring nodes.

[0016] The updated node features of node i are input into the head of the fully connected neural network to obtain the third time consumption of the historical work order corresponding to node i.

[0017] If the actual maintenance time of the historical work order corresponding to the third time consumption meets the set conditions, the training of the multi-head attention graph network prediction model is completed.

[0018] In the above scheme, the node features include at least the cause of the work order failure, the number of failures, the maintenance personnel of the work order, and the actual maintenance time of the work order.

[0019] In the above scheme, the step of processing the first time consumption and the second time consumption based on the particle swarm optimization algorithm to determine the work order path for each maintenance worker includes:

[0020] Initialize the position and velocity of each particle; the particle represents a work order path for the maintenance worker.

[0021] Based on the position and fitness function of the particles, determine the fitness function value of each particle, and determine the global optimal solution based on the fitness function value of each particle;

[0022] Determine whether the global optimal solution satisfies the termination condition;

[0023] If the global optimal solution does not satisfy the termination condition, update the velocity and position of each particle, and update the global optimal solution until the updated global optimal solution satisfies the termination condition.

[0024] If the global optimal solution satisfies the termination condition, the global optimal solution is determined as the work order path.

[0025] In the above scheme, before initializing the position and velocity of each particle, the method further includes:

[0026] Construct the objective function and constraints;

[0027] The fitness function is defined based on the objective function and constraints; wherein,

[0028] The objective function represents the minimum total time consumption; the total time consumption includes the maintenance time of the work order and the time consumption for movement between work orders;

[0029] The constraints are used to limit the conditions that the work order path must meet.

[0030] In the above scheme, the work orders to be planned include a first work order and a second work order, with the first work order having a higher priority than the second work order. The step of processing the first and second time consumptions using the particle swarm optimization algorithm to determine the work order path for each maintenance worker includes:

[0031] Based on the particle swarm optimization algorithm, the first time consumption and the second time consumption of the first work order are processed to determine the first work order path for each of the maintenance personnel to process the first work order;

[0032] Based on the particle swarm optimization algorithm, the first time consumption and the second time consumption of the second work order are processed to determine the second work order path for each of the maintenance personnel to process the second work order; wherein, the determination time of the first work order path is earlier than the determination time of the second work order path.

[0033] In the above scheme, after determining the work order path for each maintenance worker, the method further includes:

[0034] Check for any temporarily added work orders awaiting planning;

[0035] If it is determined that there are temporarily added work orders to be planned, the work order path of each maintenance worker is re-determined based on all work orders to be planned; wherein, all work orders to be planned include temporarily added work orders to be planned and work orders that existed before the temporarily added work orders to be planned.

[0036] A work order path planning device for a photovoltaic power station, comprising:

[0037] A construction module is used to obtain historical work order information of the photovoltaic power station and construct a graph model based on the historical work order information; wherein, the nodes in the graph model represent historical work orders, and the edges in the graph model represent the movement time between the historical work orders;

[0038] The training module is used to build a multi-head attention graph network prediction model based on the constructed graph model, and to learn the features of each node in the graph model and the relationships between each node.

[0039] The prediction module is used to input the information of the work order to be planned and the information of the maintenance personnel in an idle state into the multi-head attention graph network prediction model, and output the first time consumption; the first time consumption represents the predicted time consumption for the maintenance personnel to complete the work order to be planned;

[0040] The determination module is used to determine a second time consumption based on the location of each work order to be planned; the second time consumption represents the movement time between each work order to be planned.

[0041] The planning module is used to process the first time consumption and the second time consumption based on the particle swarm optimization algorithm to determine the work order path for each of the maintenance personnel; the work order path is used to record the work orders and their order that the maintenance personnel need to process.

[0042] An electronic device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the above-mentioned work order path planning method for a photovoltaic power station.

[0043] A computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-mentioned work order path planning method for a photovoltaic power station.

[0044] The aforementioned work order planning method, device, electronic equipment, and medium for photovoltaic power plants utilize a graph model constructed from historical work order information of the photovoltaic power plant. A multi-head attention graph network prediction model is then established to learn the characteristics of each node in the graph model and the relationships between them. This allows the input of information about the work order to be planned and the information of maintenance personnel in an idle state into the multi-head attention graph network prediction model to obtain the first time consumption. Combined with the second time consumption determined by the location of the work order to be planned, the work order path for each maintenance personnel can be obtained through a particle swarm optimization algorithm. This fully utilizes the relationship between location and work order, as well as the differences in maintenance personnel experience, to achieve more accurate and efficient path planning. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating a work order path planning method for a photovoltaic power station in one embodiment.

[0046] Figure 2 This is a flowchart illustrating the work order path planning method for a photovoltaic power station in yet another embodiment;

[0047] Figure 3 This is a flowchart illustrating the work order path planning method for a photovoltaic power station in yet another embodiment;

[0048] Figure 4 This is a flowchart illustrating the work order path planning method for a photovoltaic power station in yet another embodiment;

[0049] Figure 5This is a flowchart illustrating the work order path planning process for a photovoltaic power station.

[0050] Figure 6 This is a structural block diagram of a work order path planning device for a photovoltaic power station in one embodiment. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the application and are not intended to limit the scope of this application.

[0052] The following is a detailed description of the visual details of the technical solutions in the embodiments of this application.

[0053] In one embodiment, such as Figure 1 As shown, a work order path planning method for a photovoltaic power plant is provided, which may include the following steps:

[0054] Step S101: Obtain historical work order information of the photovoltaic power station and construct a graph model based on the historical work order information.

[0055] Here, historical work order information for photovoltaic power plants is obtained, including those located in different areas. This historical work order information is then analyzed to construct a corresponding graph model.

[0056] A graph model is a mathematical model used to identify and analyze complex relationships. It describes the relationships between objects through a combination of nodes and edges. Nodes represent objects, and edges represent connections or associations between objects. In practical applications, graph models can also enrich the representation of relationships between objects through attributes on nodes and edges. Node attributes can represent the characteristics of objects, and edge attributes can represent the weights or other association information between objects.

[0057] In this embodiment, each historical work order can be represented as a node, and the relationships between historical work orders can be represented by edges, thus constructing a corresponding graph model. Assume that the graph model contains nodes A and B, connected by an edge. Node A corresponds to historical work order a, and node B corresponds to historical work order b. The edge connecting nodes A and B represents the movement of maintenance personnel between the positions corresponding to historical work order a and historical work order b. The edge attribute represents the time taken to move between the two work orders. Node attributes in the graph model can represent node characteristics.

[0058] It should be noted that in the process of constructing a graph model using historical work order information, the historical work order information also needs to be processed accordingly. Data with continuous attributes in the historical work order information (such as the number of photovoltaic modules in the photovoltaic power station, the number of work order failures, etc.) needs to be normalized, while data with discrete attributes needs to be encoded using one-hot encoding.

[0059] In one embodiment, node characteristics may include the cause of failure, number of failures, scale of the photovoltaic power station to which the failure belongs, repair location, repair personnel responsible for the repair, and repair duration of historical work orders.

[0060] Step S102: Based on the constructed graph model, establish a multi-head attention graph network prediction model to learn the features of each node in the graph model and the relationships between each node.

[0061] Here, the multi-head attention graph network prediction model is a graph neural network model that combines the multi-head attention mechanism with the ideas of graph neural networks, enabling it to effectively handle graph structure data with complex relationships. In the multi-head attention mechanism, each attention head focuses on a different subspace, allowing each node to adaptively focus on important information from its neighbors. This allows for the gradual fusion of relationship information between nodes during the learning process, thus better representing the graph structure.

[0062] In this embodiment, the multi-head attention graph network prediction model can update the feature vector of a node by using the neighbor information and the node's own features. Specifically, the multi-head attention graph network prediction model aggregates the neighbor information of a node through a multi-head attention mechanism and combines it with a graph neural network to update the feature vector of the node, thereby completing the learning of the node's features and the relationship between nodes.

[0063] Step S103: Input the information of the work order to be planned and the information of the maintenance personnel who are in an idle state into the multi-head attention graph network prediction model, and output the first time consumption.

[0064] Here, an idle repairman refers to a repairman who can handle work orders to be planned. The information of the work orders to be planned and the information of the repairmen in an idle state are input into the multi-head attention graph network prediction model. The information of the repairmen in an idle state can include the repairman's name, the total number of historical repair work orders, the repairman's experience level, etc., which can reflect the repairman's repair ability and experience.

[0065] Here, the information of the work orders to be planned and the information of the maintenance personnel need to be preprocessed in the same way as the model training. Continuous attributes in the data, such as the number of photovoltaic modules in the photovoltaic power station and the number of work order failures, need to be normalized. Discrete data in the data needs to be one-hot encoded (the encoding method here is the same as the encoding method used during model training).

[0066] In this embodiment, it is necessary to predict the time consumption for each idle repairman. For example, assuming that the idle repairmen are repairman A and repairman B, the multi-head attention graph network prediction model needs to predict the time consumption for repairman A to repair work order a and the time consumption for repairman B to repair work order a. In practical applications, by adding the corresponding repairman features to the input features of each work order to be planned, the multi-head attention graph network prediction model can output the corresponding first time consumption. It should be noted that the first time consumption here refers to the time required for a repairman to complete the repair of a work order to be planned.

[0067] Step S104: Determine the second time consumption based on the location of each work order to be planned.

[0068] Here, the information for each work order to be planned also records its location. Based on the location of each work order, the second time required to move between any two work orders can be determined. In practical applications, an interface providing map-related functions and services can be called to calculate the estimated time required to drive between the locations of each work order.

[0069] Step S105: Process the first and second time consumption based on the particle swarm optimization algorithm to determine the work order path for each maintenance worker.

[0070] Here, Particle Swarm Optimization (PSO) is a swarm intelligence-based optimization algorithm used to solve optimization problems. Each particle represents a solution, and the algorithm searches for the optimal solution by continuously adjusting its speed and position. The core idea of ​​PSO is to simulate the information sharing and cooperative behavior of birds foraging, enabling particles to adaptively search for the optimal solution in the solution space. Each particle updates its speed and position based on its historical best position (local optimal solution) and global best position (swarm optimal solution) to find a better solution.

[0071] In this embodiment, each repairman's path planning is represented as a particle, and the location point is used as the solution space for the example. The particle swarm optimization (PSO) algorithm can determine the total time corresponding to each solution based on the first and second time consumptions. By searching for different solutions, the PSO algorithm can determine the optimal solution from multiple solutions, that is, determine the work order path for each repairman, where this work order path is the path with the least time consumption. In practical applications, this work order path records the work orders that the repairman needs to process, as well as the processing order of the work orders.

[0072] It is understood that in this embodiment, the total time for the work order path includes the first time and the second time. In the process of using the particle swarm optimization algorithm to find the optimal solution, the first time brought about by the maintenance work order and the second time generated by the movement between work orders are considered at the same time. The determination of the first time is related to the ability and experience of the maintenance personnel. Based on this, by using the first time and the second time as the overall optimization goal, the work order path can be planned while considering multiple factors at the same time, so as to complete the dispatch and processing of all maintenance work orders in the shortest time, thereby improving the overall efficiency and service quality.

[0073] The following details the learning process of the multi-head attention graph network prediction model on the graph model. For example... Figure 2 As shown, Figure 2 This is a flowchart illustrating the work order path planning method for photovoltaic power plants.

[0074] Step S201: Determine the node characteristics of node i and the node characteristics of its neighboring node j in the graph model.

[0075] Step S202: Process the node features of node i and the node features of neighboring node j based on multi-head attention, and calculate the attention weight between node i and neighboring node j.

[0076] Step S203: Normalize the attention weights to obtain the attention coefficients.

[0077] Step S204: Use the attention coefficient to perform weighted aggregation of the node features of neighboring node j to obtain the weighted sum of the node features of neighboring node j.

[0078] Step S205: Update the node features of node i based on the weighted sum of the node features of neighboring nodes.

[0079] Step S206: Input the updated node features of node i into the head of the fully connected neural network to obtain the third time consumption of the historical work order corresponding to node i.

[0080] Step S207: If the actual maintenance time of the historical work order corresponding to the third time consumption meets the set conditions, the training of the multi-head attention graph network prediction model is completed.

[0081] Here, for the graph model G = (V, E), a multi-layer attention graph network prediction model is constructed, where V represents the set of nodes and E represents the set of edges. Each node V i Each node has a corresponding feature, denoted as h. i .

[0082] Assuming a graph model has N nodes, the node features of node i can be represented as h. i ∈R F , where F is the dimension of the node feature.

[0083] First, determine the node features h of one of the nodes i from the graph model. i And the node features h of node i's neighboring node j j It should be noted that, here, adjacent nodes refer to nodes that are connected to node i by an edge.

[0084] Multi-head attention is introduced to learn the relationships between different nodes. For each attention head 'a', the attention weights between node 'i' and its neighboring node 'j' can be calculated according to the following formula.

[0085]

[0086] In the above formula, α a ∈R 2D h is the parameter vector of the attention head a. i h represents the feature vector of node i. j W represents the feature vector of node j. a ∈R D×F is the weight matrix of attention head a, || represents the concatenation operation of vectors, the activation function is LeakyReLU, and D is the hidden feature dimension of each attention head.

[0087] In the above equation, the features of node i and node j are linearly transformed and nonlinearly mapped using the LeakyReLU activation function, thereby representing the relationship between node i and node j.

[0088] After determining the attention weights, the attention coefficients are obtained by normalizing the attention weights according to the following formula.

[0089]

[0090] in, Let i represent the set of neighboring nodes j of node i.

[0091] Attention coefficient in the above formula By exponentializing and normalizing the attention weights, we can reflect the degree of attention that node i pays to its neighboring node j. A larger attention coefficient indicates that node i pays more attention to the node features of its neighboring node j.

[0092] After determining the attention coefficients, the node features of neighboring node j are weighted and aggregated using the attention coefficients to obtain the weighted sum of the features of neighboring nodes of node i. The weighted sum of the features of neighboring nodes of node i can be obtained using the following formula.

[0093]

[0094] Where K is the number of attention heads, W a This is the weight matrix of the attention head 'a'. In the above formula, the attention coefficients are... Features W of neighboring nodes a h j Perform weighted aggregation to obtain the weighted sum h′ of the features of the neighboring nodes of node i. i This allows for the adaptive fusion of information from neighboring nodes using the attention coefficient.

[0095] After determining the weighted sum of the features of the neighboring nodes of node i, the node features of node i are updated using the following formula based on the weighted sum of the features of the neighboring nodes of node i.

[0096] h i =ReLU(Wh′) i +b)

[0097] Where W is the weight matrix used to update node features, and b is the bias vector. The above equation performs a nonlinear mapping on the weighted sum of neighboring node features by applying a linear transformation of the weighted sum and the ReLU activation function, resulting in the updated node features h. i This allows for the capture of information about the node itself and its global neighbors.

[0098] The updated node features of node i are input into the head of the fully connected neural network to obtain the predicted time consumption value of the corresponding work order for node i. Specifically,

[0099]

[0100] In this context, `head(·)` represents the output layer or the last layer of a multi-layer perceptron (MLP) constructed from a fully connected neural network. It is responsible for mapping the feature representations of the intermediate layers to the final prediction result. A multi-layer perceptron is a feedforward neural network model consisting of one or more fully connected hidden layers and an output layer. Each hidden layer comprises multiple neurons, each connected to all neurons in the previous layer. Each neuron receives the output of the previous layer, performs a non-linear transformation on the input through activation, and then passes the result to the next layer.

[0101] In practical applications, supervised learning is used to train the multi-head attention graph network prediction model. The third time consumption of historical work orders output by the multi-head attention graph network prediction model is compared with the actual time consumption of historical work orders. If the set conditions are met, the training of the multi-head attention graph network prediction model is terminated.

[0102] In practical applications, setting conditions can limit convergence and performance. Model training convergence refers to whether the model's parameters have converged to a stable state. Typically, model convergence can be assessed by monitoring changes in the model's loss function or accuracy on the training data. When the model's loss function or accuracy no longer changes significantly or tends to stabilize, the model can be considered converged, and training can be terminated. Besides model convergence, the model's performance on test data also needs to be considered. During model training, a portion of the data is usually reserved as a validation set to evaluate the model's performance on untrained data. By monitoring the model's performance on the validation set, it's possible to determine whether the model is overfitting or underfitting. If the model's performance on the validation set no longer improves or begins to decline, the model can be considered to have reached its optimal performance, and training can be terminated.

[0103] In one embodiment, Figure 3 A schematic flowchart of the work order path planning method for photovoltaic power plants is shown.

[0104] Step S301: Initialize the position and velocity of each particle.

[0105] Step S302: Determine the fitness function value of each particle based on its position and fitness function, and determine the global optimal solution based on the fitness function value of each particle.

[0106] Step S303: Determine whether the global optimal solution satisfies the termination condition.

[0107] Step S304: If the global optimal solution does not meet the termination condition, update the velocity and position of each particle, and update the global optimal solution until the updated global optimal solution meets the termination condition.

[0108] Step S305: If the global optimal solution satisfies the termination condition, the global optimal solution is determined as the work order path.

[0109] First, a group of particles is randomly generated, and the position and velocity of each particle are initialized. The position and velocity of each particle are random.

[0110] Next, based on the fitness function, the fitness function value corresponding to each particle is calculated. The fitness function value can measure the quality of the particle's solution, thus determining the global optimal solution based on the fitness function value of each particle.

[0111] After determining the global optimal solution, it is checked whether the global optimal solution meets the termination condition. The termination condition can be reaching the maximum number of iterations or reaching a certain fitness function threshold. If the termination condition is met, the global optimal solution is output.

[0112] If the termination condition is not met, the velocity and position of each particle are updated based on its current position, velocity, and the optimal solution. The velocity update is influenced by both the individual optimal solution and the global optimal solution. The changes in particle velocity and position can be expressed as follows:

[0113]

[0114] Among them, V i (t+1) ω is the velocity of particle i at time step t+1, ω is the inertia weight, controlling the influence of the previous velocity, c1 and c2 are acceleration constants, controlling the influence of the individual's historical optimal solution and the group's global optimal solution, and pBest. i gBest is the individual historical optimal solution, gBest is the group's global optimal solution, and rand() is a random number between 0 and 1. It is the position of particle i at time step t+1.

[0115] After updating the particles and velocities, the individual optimal solution and the global optimal solution are updated based on the fitness function value of each particle. Then, it is checked again whether the termination condition is met. If the termination condition is met, the corresponding work order path is obtained based on the final global optimal solution.

[0116] In one embodiment, the path planning process using the particle swarm optimization algorithm requires constructing an objective function and time constraints. Specifically, for the scheduling problem involving multiple maintenance work orders and multiple maintenance personnel, the objective function for the work order path planning problem in a photovoltaic power plant can be expressed as:

[0117]

[0118] The objective function represents minimizing the total time, including the time spent moving between locations and the time spent repairing each work order, where N represents the number of work orders to be planned, and c ij x represents the travel time from work order i to work order j. ij This indicates whether the movement has started from work order i and moved to work order j. The value is 1 if the movement is from work order i to work order j, and 0 otherwise. i y represents the repair time for work order i. i This indicates whether to execute the repair for work order i. If the repair for work order i is executed, the value is 1; otherwise, it is 0.

[0119] The constraints can be expressed as:

[0120]

[0121]

[0122]

[0123]

[0124] Specifically, (1a) it ensures that each location point has exactly one incoming edge, meaning each location point can only move from one adjacent location point; (1b) it ensures that M work orders are selected for repair, where M is the maximum number of work orders that a repairman can handle simultaneously; and (1c) it ensures that the repair time does not exceed the total available time T, where t i (1d) Ensure that maintenance work can only be performed after work order i has been selected.

[0125] After constructing the objective function and constraints, a fitness function is defined. The value of the objective function can be used as part of the fitness function, and solutions that violate the constraints can have their fitness set to a lower value so that these non-constrained values ​​cannot be considered the global optimum.

[0126] In one embodiment, Figure 4 This is a flowchart illustrating the work order path planning method for photovoltaic power plants.

[0127] Step S401: Based on the particle swarm optimization algorithm, process the first and second time consumption of the first work order to determine the first work order path for each maintenance worker to process the first work order.

[0128] Step S402: Based on the particle swarm optimization algorithm, process the first and second time consumption of the second work order to determine the second work order path for each maintenance worker to process the second work order.

[0129] Here, work orders to be planned can be divided into first work orders and second work orders according to their urgency. The first work order has a higher priority than the second work order, meaning that the first work order is the one that needs to be repaired first, while the second work order does not have any special requirements on the repair order.

[0130] Based on this, in this embodiment, the maintenance path for the first work order is planned first, and then the maintenance path for the second work order is planned.

[0131] In this embodiment, based on the first and second processing times of the first work order, a particle swarm optimization algorithm is used to determine the first work order path for each maintenance worker to handle the first work order. After determining the first work order path, based on the first and second processing times of the second work order, a particle swarm optimization algorithm is used to determine the second work order path for each maintenance worker to handle the second work order. This allows for prioritizing the processing of the first work order by maintenance workers and further planning the work order paths according to the urgency of the work orders.

[0132] In one embodiment, unplanned work orders at time t1 are processed to generate work order paths for those unplanned work orders. During this process, at time t2 after time t1, there may be temporarily added work orders to be planned. These temporarily added work orders at time t2 are not yet included in the determined work order paths. Therefore, after determining the work order paths based on the particle swarm optimization algorithm, it is necessary to further detect whether there are any temporarily added work orders to be planned. If there are no temporarily added work orders to be planned, the determined work order paths can be output, and maintenance personnel will work according to the corresponding work order paths.

[0133] When there are temporarily added work orders to be planned, the maintenance path for each repairman needs to be redefined based on all work orders to be planned, that is, based on the unplanned work orders at time t1 and the temporarily added work orders to be planned at time t2. Specifically, since the first and second time consumption of the unplanned work orders at time t1 has already been determined, in the process of redefining the work order path for the repairman, only the temporarily added work orders to be planned at time t2 need to be input into the multi-head attention graph network prediction model to obtain the first and second time consumption of the temporarily added work orders to be planned at time t2. Then, the particle swarm optimization algorithm is used to determine the repair path for the repairman for the work orders to be planned at time t1 and the temporarily added work orders to be planned at time t2.

[0134] like Figure 5 As shown, Figure 5 A flowchart illustrating the work order path planning process for a photovoltaic power station is shown.

[0135] Step 1: Input all the information of the work orders to be planned and the information of all the maintenance personnel that can be arranged at the maintenance site into the multi-head attention graph network prediction model.

[0136] Step 2: Predict the time required for each repairman to complete each work order.

[0137] Step 3: Determine if there are any work orders requiring priority repair. If yes, proceed to Step 4; otherwise, proceed to Step 5.

[0138] Step 4: Use the particle swarm optimization algorithm to plan the maintenance paths for work orders that require priority maintenance. Proceed to Step 6.

[0139] Step 5: Use the particle swarm optimization algorithm to plan the maintenance path for the work order to be repaired.

[0140] Step 6: Determine if there are any temporarily added repair work orders. If yes, proceed to Step 1; otherwise, end the process.

[0141] This application also provides an application embodiment. The process of the work order path planning method for photovoltaic power plants is described in detail below through the application embodiment.

[0142] 1. Historical data collection

[0143] We used maintenance work order data and all maintenance personnel information from all photovoltaic power stations in a certain region over the past year. Historical work order information and maintenance personnel information are shown in Tables 1 and 2. Due to data confidentiality, key features in Tables 1 and 2 have been processed. It should be noted that only partial data is shown in Tables 1 and 2.

[0144] Table 1

[0145]

[0146] Table 2

[0147]

[0148]

[0149] 2. Construction of a Multi-Head Attention Graph Network Prediction Model

[0150] Based on historical work orders and maintenance personnel information for a certain area, and simultaneously collecting address information from historical work orders, a graph model is constructed, and a multi-head attention graph network prediction model is trained.

[0151] 3. Data Input

[0152] After the multi-head attention graph network prediction model is trained, the work orders to be planned and the information of all available maintenance personnel are input into the multi-head attention graph network prediction model. The input data needs to undergo the same preprocessing as during training: normalization is performed on continuous attribute data (such as the number of photovoltaic modules in the photovoltaic power station, the number of work order failures, etc.), and one-hot encoding is performed on discrete attribute data. For processing the maintenance personnel information, time prediction needs to be performed separately for each maintenance personnel; that is, the input features of each work order need to be added to the features of the corresponding maintenance personnel to obtain the predicted time for each maintenance personnel on the work order.

[0153] 4. Work order time prediction

[0154] The 20 work orders to be planned and the information of 2 available maintenance personnel were input into a multi-head attention graph network prediction model to obtain the estimated time for each work order. Table 3 shows the estimated time for each maintenance personnel to complete each work order. Due to the large amount of data, Figure 3 Only a portion of the data is shown.

[0155] Table 3

[0156]

[0157] Based on the location of each work order, the estimated time for driving between each work order location is determined, and a distance matrix is ​​output. As shown in Table 4, the table represents different work order locations and the time for driving between work order locations (in hours).

[0158] Table 4

[0159]

[0160]

[0161] 5. Maintenance route planning

[0162] The Particle Swarm Optimization (PSO) algorithm was used to find the optimal work order path under different maintenance worker completion times. Traditional PSO and branch-and-bound methods were also employed. Since PSO and branch-and-bound methods cannot predict work order maintenance times, the time for all work orders was set to 0 during the solution process. After obtaining the optimal path, the work order time predicted by the multi-head attention graph network prediction model was added to the result. Table 5 records the optimal work order path solutions under different algorithms. It can be seen from Table 5 that the work order path determined by the multi-head attention graph network prediction model + PSO algorithm has the minimum total time and the most reasonable work order path planning.

[0163] Table 5

[0164] Multi-head attention graph network prediction model + particle swarm optimization algorithm 17.8 Particle Swarm Optimization 22.4 Branch and bound method 22.4

[0165] In the above embodiments, by making full use of historical work order information and incorporating the feature information of historical work orders into the multi-head attention graph network prediction model, the relationship between location points and work orders can be modeled more accurately. By combining the multi-head attention graph network prediction model with the particle swarm optimization algorithm, the relationship between location points and work orders, as well as the experience differences of maintenance personnel, can be fully utilized. The movement time between location points and the maintenance time of each work order are used as optimization targets to achieve more accurate and efficient path planning. This enables the dispatch of all maintenance work orders to be completed in the shortest possible time, thereby improving overall efficiency and service quality.

[0166] In one embodiment, a work order path planning device for a photovoltaic power station is provided, with reference to... Figure 6 As shown, the work order planning device 600 of the photovoltaic power station may include: a construction module 601, a training module 602, a prediction module 603, a determination module 604, and a planning module 605.

[0167] The system comprises the following modules: a construction module 601, which acquires historical work order information of the photovoltaic power station and constructs a graph model based on this information; where nodes in the graph model represent historical work orders and edges represent the movement time between historical work orders; a training module 602, which, based on the constructed graph model, establishes a multi-head attention graph network prediction model to learn the features of each node and the relationships between them; a prediction module 603, which inputs information about the work orders to be planned and information about maintenance personnel in an idle state into the multi-head attention graph network prediction model and outputs a first time consumption; the first time consumption represents the predicted time for a maintenance personnel to complete the work order to be planned; a determination module 604, which determines a second time consumption based on the position of each work order to be planned; the second time consumption represents the movement time between each work order to be planned; and a planning module 605, which processes the first and second times using a particle swarm optimization algorithm to determine the work order path for each maintenance personnel; the work order path records the work orders and their order that the maintenance personnel need to process.

[0168] In one embodiment, the training module 602 is specifically used to determine the node features of node i and the node features of neighboring node j in the graph model.

[0169] Based on multi-head attention, the node features of node i and the node features of neighboring node j are processed to calculate the attention weight between node i and neighboring node j.

[0170] The attention weights are normalized to obtain attention coefficients;

[0171] The attention coefficient is used to perform weighted aggregation of the node features of the neighboring node j to obtain the weighted sum of the neighboring node features of node i;

[0172] The node features of node i are updated based on the weighted sum of the features of its neighboring nodes.

[0173] The updated node features of node i are input into the head of the fully connected neural network to obtain the third time consumption of the historical work order corresponding to node i.

[0174] If the actual maintenance time of the historical work order corresponding to the third time consumption meets the set conditions, the training of the multi-head attention graph network prediction model is completed.

[0175] In one embodiment, the node features include at least the cause of the work order failure, the number of failures, the maintenance personnel for the work order, and the actual maintenance time for the work order.

[0176] In one embodiment, the planning module 605 is specifically used to initialize the position and velocity of each particle; the particle represents a work order path for the maintenance worker.

[0177] Based on the position and fitness function of the particles, determine the fitness function value of each particle, and determine the global optimal solution based on the fitness function value of each particle;

[0178] Determine whether the global optimal solution satisfies the termination condition;

[0179] If the global optimal solution does not satisfy the termination condition, update the velocity and position of each particle, and update the global optimal solution until the updated global optimal solution satisfies the termination condition.

[0180] If the global optimal solution satisfies the termination condition, the global optimal solution is determined as the work order path.

[0181] In one embodiment, the planning module 605 is specifically used to construct an objective function and constraints before initializing the position and velocity of each particle;

[0182] The fitness function is defined based on the objective function and constraints; wherein,

[0183] The objective function represents the minimum total time consumption; the total time consumption includes the maintenance time of the work order and the time consumption for movement between work orders;

[0184] The constraints are used to limit the conditions that the work order path must meet.

[0185] In one embodiment, the work order to be planned includes a first work order and a second work order, wherein the first work order has a higher priority than the second work order. The planning module 605 is specifically used to process the first time consumption and the second time consumption of the first work order based on the particle swarm algorithm, and determine the first work order path for each maintenance worker to process the first work order.

[0186] Based on the particle swarm optimization algorithm, the first time consumption and the second time consumption of the second work order are processed to determine the second work order path for each of the maintenance personnel to process the second work order; wherein, the determination time of the first work order path is earlier than the determination time of the second work order path.

[0187] In one embodiment, the planning module 605 is specifically used to detect whether there are any temporarily added work orders to be planned;

[0188] If it is determined that there are temporarily added work orders to be planned, the work order path of each maintenance worker is re-determined based on all work orders to be planned; wherein, all work orders to be planned include temporarily added work orders to be planned and work orders that existed before the temporarily added work orders to be planned.

[0189] Specific limitations regarding the work order path planning device for photovoltaic power plants can be found in the limitations of the work order path planning method for photovoltaic power plants mentioned above, and will not be repeated here. Each module in the aforementioned work order path planning device for photovoltaic power plants can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0190] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0191] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0192] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0193] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0194] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A work order path planning method for a photovoltaic power station, characterized in that, include: The historical work order information of the photovoltaic power station is obtained, and a graph model is constructed based on the historical work order information; wherein, the nodes in the graph model represent historical work orders, and the edges in the graph model represent the movement time between the historical work orders; Based on the constructed graph model, a multi-head attention graph network prediction model is established to learn the features of each node in the graph model and the relationships between each node; The information of the work order to be planned and the information of the maintenance personnel in an idle state are input into the multi-head attention graph network prediction model, and the first time is output; the first time represents the predicted time for the maintenance personnel to complete the work order to be planned; The second time consumption is determined based on the location of each work order to be planned; the second time consumption represents the movement time between each work order to be planned. The first and second time consumption times are processed based on the particle swarm optimization algorithm to determine the work order path for each maintenance worker; the work order path is used to record the work orders and their order that the maintenance worker needs to process.

2. The method according to claim 1, characterized in that, The step of establishing a multi-head attention graph network prediction model based on the constructed graph model to learn the features of each node in the graph model and the relationships between each node includes: Determine the node characteristics of node i and the node characteristics of its neighboring node j in the graph model; Based on multi-head attention, the node features of node i and the node features of neighboring node j are processed to calculate the attention weight between node i and neighboring node j. The attention weights are normalized to obtain attention coefficients; The attention coefficient is used to perform weighted aggregation of the node features of the neighboring node j to obtain the weighted sum of the neighboring node features of node i; The node features of node i are updated based on the weighted sum of the features of its neighboring nodes. The updated node features of node i are input into the head of the fully connected neural network to obtain the third time consumption of the historical work order corresponding to node i. If the actual maintenance time of the historical work order corresponding to the third time consumption meets the set conditions, the training of the multi-head attention graph network prediction model is completed.

3. The method according to claim 2, characterized in that, The node features include at least the cause of the work order failure, the number of failures, the maintenance personnel for the work order, and the actual maintenance time for the work order.

4. The method according to claim 1, characterized in that, The process of processing the first and second time consumption times based on the particle swarm optimization algorithm to determine the work order path for each maintenance worker includes: Initialize the position and velocity of each particle; the particle represents a work order path for the maintenance worker. Based on the position and fitness function of the particles, determine the fitness function value of each particle, and determine the global optimal solution based on the fitness function value of each particle; Determine whether the global optimal solution satisfies the termination condition; If the global optimal solution does not satisfy the termination condition, update the velocity and position of each particle, and update the global optimal solution until the updated global optimal solution satisfies the termination condition. If the global optimal solution satisfies the termination condition, the global optimal solution is determined as the work order path.

5. The method according to claim 4, characterized in that, Before initializing the position and velocity of each particle, the method further includes: Construct the objective function and constraints; The fitness function is defined based on the objective function and constraints; wherein, The objective function represents the minimum total time consumption; the total time consumption includes the maintenance time of the work order and the time consumption for movement between work orders; The constraints are used to limit the conditions that the work order path must meet.

6. The method according to claim 1, characterized in that, The work orders to be planned include a first work order and a second work order, with the first work order having a higher priority than the second work order. The step of processing the first and second work times using a particle swarm optimization algorithm to determine the work order path for each maintenance worker includes: Based on the particle swarm optimization algorithm, the first time consumption and the second time consumption of the first work order are processed to determine the first work order path for each of the maintenance personnel to process the first work order; Based on the particle swarm optimization algorithm, the first time consumption and the second time consumption of the second work order are processed to determine the second work order path for each of the maintenance personnel to process the second work order; wherein, the determination time of the first work order path is earlier than the determination time of the second work order path.

7. The method according to claim 1 or 6, characterized in that, After determining the work order path for each of the maintenance personnel, the method further includes: Check for any temporarily added work orders awaiting planning; If it is determined that there are temporarily added work orders to be planned, the work order path of each maintenance worker is re-determined based on all work orders to be planned; wherein, all work orders to be planned include temporarily added work orders to be planned and work orders that existed before the temporarily added work orders to be planned.

8. A work order path planning device for a photovoltaic power station, characterized in that, include: A construction module is used to obtain historical work order information of the photovoltaic power station and construct a graph model based on the historical work order information; wherein, the nodes in the graph model represent historical work orders, and the edges in the graph model represent the movement time between the historical work orders; The training module is used to build a multi-head attention graph network prediction model based on the constructed graph model, and to learn the features of each node in the graph model and the relationships between each node. The prediction module is used to input the information of the work order to be planned and the information of the maintenance personnel in an idle state into the multi-head attention graph network prediction model, and output the first time consumption; the first time consumption represents the predicted time consumption for the maintenance personnel to complete the work order to be planned; The determination module is used to determine a second time consumption based on the location of each work order to be planned; the second time consumption represents the movement time between each work order to be planned. The planning module is used to process the first time consumption and the second time consumption based on the particle swarm optimization algorithm to determine the work order path for each of the maintenance personnel; the work order path is used to record the work orders and their order that the maintenance personnel need to process.

9. An electronic device comprising a memory and a processor, the memory having a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images