Simulation analysis method and system for power engineering cost based on digital twinning

Abstract

The application discloses a simulation analysis method and system for power engineering cost based on digital twinning, and relates to the technical field of simulation analysis. The method comprises the following steps: obtaining multiple engineering task multi-modal data and task dependency relationships for a power engineering; processing the multiple engineering task multi-modal data and the task dependency relationships based on digital twinning technology to construct a power engineering virtual model; inputting the power engineering virtual model into a graph neural network to extract task node features corresponding to each engineering task virtual module through a message passing mechanism; and defining input states of a reinforcement learning model according to the extracted task node features to generate task resource allocation actions for each engineering task through reinforcement learning. Thus, the digital twinning technology, graph theory feature extraction technology and reinforcement learning technology are comprehensively applied to monitor and simulate the cost and resource consumption of different links in the power engineering in real time and intelligently optimize resource allocation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of simulation analysis, and particularly relates to a simulation analysis method and system for power engineering cost based on digital twinning. BACKGROUND

[0002] Due to the complexity of the technology involved and the long construction period, cost control has always been a key problem in engineering management. The traditional power engineering cost management method relies on static budgeting, manual estimation, historical data review and other means. It is usually based on the preliminary setting of the engineering plan to make simple cost prediction and control. These methods can provide certain reference for cost prediction, but they rely on a large number of assumptions and simplifications, often ignoring the complex interactions between various links or engineering tasks in power engineering.

[0003] In addition, the data of each link in the power engineering is usually managed by different departments, and there is a lack of effective data integration and sharing mechanism, resulting in the problem of "data island". The information of different links is difficult to interconnect, and even the effective integration and analysis of data within the same link is difficult. For example, the complex relationship between engineering construction and equipment procurement, civil engineering tasks and mechanical and electrical installation tasks often cannot be fully reflected in the model. Therefore, when facing complex and dynamic engineering environment, the accuracy and efficiency of the traditional cost optimization control strategy are greatly reduced.

[0004] In view of the above problems, the industry has not yet proposed a better technical solution. SUMMARY

[0005] The present application provides a simulation analysis method and system for power engineering cost based on digital twinning, a storage medium, a computer program product and an electronic device, which at least solve the problems of relying on static assumptions and data islands in traditional power engineering project management.

[0006] In a first aspect, the embodiments of the present application provide a simulation analysis method for power engineering cost based on digital twinning, comprising: obtaining a plurality of engineering task multi-modal data and task dependency relationships for a power engineering; the engineering task multi-modal data includes task progress data, resource consumption data and engineering area meteorological data of the corresponding engineering task; the resource consumption data includes building material cost consumption information, engineering equipment cost use information and engineering labor cost use information of the corresponding engineering task; processing the plurality of engineering task multi-modal data and task dependency relationships based on digital twinning technology to construct a power engineering virtual model; the power engineering virtual model includes a plurality of engineering task virtual modules, each of the engineering task virtual modules is modeled by the corresponding engineering task multi-modal data, and there is an edge connection between the associated engineering task virtual modules indicated by the task dependency relationship; inputting the power engineering virtual model into a graph neural network to extract task node features corresponding to each of the engineering task virtual modules through a message passing mechanism; defining an input state of a reinforcement learning model according to the extracted task node features of each of the engineering task virtual modules to generate a task resource allocation action for each of the engineering tasks through reinforcement learning, the task resource allocation action is used to indicate a resource allocation scheme for the corresponding engineering task, the resource allocation scheme includes a building material allocation strategy, an engineering equipment allocation strategy and a labor force allocation strategy; the reward function of the reinforcement learning model is defined according to the total engineering completion cost, the total engineering completion cycle and the average engineering task resource use efficiency of the power engineering.

[0007] In a second aspect, an embodiment of the present application provides a simulation analysis system for power engineering cost based on digital twinning, comprising: a data acquisition unit configured to acquire a plurality of engineering task multi-modal data and task dependency relationships for a power engineering; the engineering task multi-modal data comprises task progress data, resource consumption data and engineering area meteorological data of a corresponding engineering task; the resource consumption data comprises building material cost consumption information, engineering equipment cost use information and engineering labor cost use information of the corresponding engineering task; a twinning modeling unit configured to process the plurality of engineering task multi-modal data and the task dependency relationships based on a digital twinning technology to construct a power engineering virtual model; the power engineering virtual model comprises a plurality of engineering task virtual modules, each of the engineering task virtual modules is modeled by corresponding engineering task multi-modal data, and there is an edge connection between the associated engineering task virtual modules indicated by the task dependency relationship; a feature extraction unit configured to input the power engineering virtual model into a graph neural network to extract task node features corresponding to each of the engineering task virtual modules through a message passing mechanism; a reinforcement learning analysis unit configured to define an input state of a reinforcement learning model according to the extracted task node features to generate a task resource allocation action for each of the engineering tasks through reinforcement learning, the task resource allocation action is used to indicate a resource allocation scheme for the corresponding engineering task, the resource allocation scheme comprises a building material allocation strategy, an engineering equipment allocation strategy and a labor force allocation strategy; a reward function of the reinforcement learning model is defined according to a total engineering completion cost, a total engineering completion cycle and an average engineering task resource use efficiency of the power engineering.

[0008] In a third aspect, an electronic device is provided, comprising: at least one processor, and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the simulation analysis method for power engineering cost based on digital twinning of any embodiment of the present application.

[0009] In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the simulation analysis method for power engineering cost based on digital twinning of any embodiment of the present application.

[0010] In a fifth aspect, an embodiment of the present application provides a computer program product comprising computer programs / instructions, which, when executed by a processor, implement the steps of the simulation analysis method for power engineering cost based on digital twinning of any embodiment of the present application.

[0011] The simulation analysis method and system for the cost of a power engineering based on digital twinning provided by the application can at least produce the following technical effects:

[0012] (1) By integrating the multi-modal data of multiple engineering tasks (such as task progress, resource consumption, meteorological data, etc.) and establishing a virtual model of the power engineering, the interaction relationship between each task can be more accurately reflected. The static budget and manual estimation of the traditional method cannot cope with the dynamically changing engineering environment, while the virtual model established by the digital twinning technology can simulate and track the project progress in real time, implement dynamic cost monitoring and resource scheduling, and adjust the resource allocation scheme according to the actual engineering progress, effectively avoiding the errors caused by the assumption bias and data simplification in the traditional method.

[0013] (2) By digital twinning technology, the data of each link is effectively integrated and correlated in the unified virtual model, which can share information in real time during project management, ensure the cooperation and optimization of each link, improve the cooperation efficiency between links, and help to identify potential risks and make targeted adjustments in time.

[0014] (3) The graph neural network (GNN) is used to process the complex dependency relationship between different engineering task virtual modules (i.e., nodes) in the power engineering virtual model (i.e., graph structure), dynamically transfer information and optimize the resource allocation of each task virtual module, thereby improving the overall execution efficiency of the project, and avoiding the waste of resources and confusion of scheduling caused by the fuzzy or simplified relationship between tasks in the traditional method.

[0015] (4) Through the reinforcement learning algorithm, the task state is updated in real time by digital twinning, which can learn and adjust itself between multiple engineering tasks, dynamically adjust the optimal resource allocation strategy, and support the implementation of a dynamic optimization decision support system. In addition, the reward function of reinforcement learning is based on the actual completion cost, total cycle and resource utilization efficiency of the power engineering, which can adaptively adjust the resource allocation strategy to achieve optimal execution within the budget range of the power engineering project. Therefore, through intelligent optimization, the resource allocation strategy can be adjusted in time when facing changing engineering environment, minimizing resource waste and improving the economic efficiency and sustainability of the project.

[0016] Through the technical solution, the digital twinning technology, graph feature extraction technology and reinforcement learning technology are comprehensively applied to monitor and simulate the cost and resource consumption of different links in the power engineering in real time, intelligently optimize resource allocation, enhance the cooperation efficiency between departments and unified management of engineering resources, improve resource utilization efficiency, and effectively reduce the overall investment cost of the power engineering. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced as follows. Obviously, the accompanying drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those of ordinary skill in the art without creative labor on the basis of these drawings.

[0018] Figure 1 A flow chart of an example of a simulation analysis method for power engineering cost based on digital twinning according to an embodiment of the present application is shown;

[0019] Figure 2 A principle schematic diagram of an example of state transition action in a reinforcement learning model is shown;

[0020] Figure 3 An operation flow chart of an example according to step S110 in Figure 1

[0021] Figure 4 An operation flow chart of an example of constructing a virtual model of power engineering based on digital twinning technology according to an embodiment of the present application is shown;

[0022] Figure 5 A structure connection schematic diagram of an example of a reinforcement learning model according to an embodiment of the present application is shown;

[0023] Figure 6 A structure block diagram of an example of a simulation analysis system for power engineering cost based on digital twinning according to an embodiment of the present application is shown;

[0024] Figure 7 A structure schematic diagram of an embodiment of the electronic device of the present application. DETAILED DESCRIPTION

[0025] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0026] In the technical solutions of the present application, the collection, storage, use, processing, transmission, provision and disclosure of user personal information, etc. comply with relevant laws and regulations and do not violate public order and good customs.

[0027] ​It should be noted that digital twinning is a method of synchronizing the physical world with the virtual world through real-time data, which can reflect the state and behavior of physical entities through highly accurate virtual models in real time.

[0028] Figure 1 A flowchart of an example of a simulation analysis method for power engineering cost based on digital twinning according to an embodiment of the present application is shown.

[0029] Regarding the execution subject of the method of the present application, it can be any controller or processor with computing or processing capability. Specifically, it can be implemented by a power engineering cost management cloud platform. By integrating data from different departments and links, a unified digital twinning model is constructed, breaking down the information barriers between engineering tasks. Data between links can be shared and interchanged, eliminating the problem of "data silos", improving the efficiency of department collaboration, and ensuring timely updating and feedback of information. Further, through simulation analysis and multi-dimensional data fusion, a comprehensive and accurate decision support tool is provided for engineering managers, helping them better assess cost risks, adjust resource allocation, and make reasonable cost optimization decisions, thereby significantly improving the intelligent management level of engineering projects.

[0030] In some examples, it can be integrated in an electronic device or terminal in a software, hardware or combination of software and hardware manner, and the type of terminal or electronic device can be diversified, such as mobile phone, tablet computer or desktop computer, etc.

[0031] As shown in Figure 1 In step S110, multiple engineering task multi-modal data and task dependency relationships for the power engineering are obtained.

[0032] It should be understood that the collection method of engineering task multi-modal data and task dependency relationships can be diversified, such as through sensors, smart devices, monitoring systems, management terminal interactions, etc. without limitation. In addition, the type of engineering task under the power engineering can also be diversified, and can be set or adjusted according to business needs, such as civil engineering tasks, substation equipment installation tasks, communication system construction tasks, intelligent control system installation tasks, power equipment maintenance tasks and power grid optimization tasks, etc. In some cases, the engineering task can also be designed in a multi-layer task structure, i.e. the engineering task can also include multiple levels of subtasks, such as excavation of foundation pit, foundation construction, reinforced concrete pouring, etc.

[0033] It is noted that there can be some potential correlations between different tasks. As an example, task dependency relationships can be determined by project plans (e.g., Gantt charts, network diagrams) and construction process flows. In some embodiments, the system can extract the precedence relationships, parallel or serial execution constraints between tasks according to the input of the project design phase, so as to determine the dependency order and resource conflicts between tasks. For example, the substation equipment installation task needs to be started after the completion of the civil engineering task, the intelligent control system installation task needs to be started after the completion of the communication system construction task, and so on.

[0034] Here, the engineering task multi-modal data includes task progress data, resource consumption data, and engineering area weather data of the corresponding engineering task. In some embodiments, the task progress data can include the current state of the engineering task (e.g., to be executed, in execution, completed, etc.), and the time node of the task (e.g., planned completion time, actual completion time, estimated delay time, etc.). The engineering area weather data can be various non-limiting weather parameters that can affect the progress and resource consumption of the project, such as temperature, wind speed, and precipitation, etc.

[0035] The resource consumption data includes building material cost consumption information, engineering equipment cost use information, and engineering labor cost use information of the corresponding engineering task. The building material cost consumption information can be the consumption amount and cost data of the building materials used in the engineering task (e.g., steel, cement, cable, etc.). The engineering equipment cost use information is used to record the use time, rental fee, maintenance cost, fuel consumption, etc. of the equipment (e.g., crane, excavator, concrete mixer, etc.) during the implementation of the engineering task. The engineering labor cost use information can record the number of labor, types of work, salary, and work efficiency, etc. of the engineering task.

[0036] In step S120, the plurality of engineering task multi-modal data and the task dependency relationship are processed based on digital twinning technology to construct a power engineering virtual model.

[0037] Here, the power engineering virtual model includes a plurality of engineering task virtual modules, each of which is modeled by the corresponding engineering task multi-modal data, and there is an edge connection between the associated engineering task virtual modules indicated by the task dependency relationship.

[0038] In some embodiments, based on the multi-modal data, a digital twin technology is used to virtually model the entire power engineering project, each engineering task is abstracted into a virtual module, and the attributes of the virtual module include task progress, resource consumption, meteorological data, etc. Exemplarily, various third-party digital twin platforms can be used, such as Siemens MindSphere, PTC ThingWorx, etc., which can provide real-time task status feedback for users through the construction of a visual dashboard. In addition, when modeling, the dependency relationship between engineering tasks is represented by a graph structure, which is specifically a graph composed of nodes (engineering tasks) and edges (relationships between tasks). It should be noted that in power engineering, there are often complex dependency relationships between tasks, and the use of a graph structure can effectively express the mutual relationship between tasks, especially the sequential dependency, parallel execution, and resource sharing relationship between tasks. In addition, each virtual module not only contains initial task data, but also can be dynamically updated according to real-time collected task progress and resource consumption data, so as to keep real-time synchronization of the status of each task in the real power engineering project.

[0039] In step S130, the power engineering virtual model is input to the graph neural network to extract the task node features corresponding to each engineering task virtual module through a message passing mechanism.

[0040] It should be noted that the graph neural network (GNN) is a deep learning model specially used for processing graph structure data. A graph is a data structure composed of nodes (vertices) and edges (edges), where nodes represent entities and edges represent relationships between entities. Through the graph neural network, information propagation and aggregation can be performed on the nodes and edges of the graph to learn the feature representation of the graph, capturing the complex topological structure information and mutual dependency relationship between nodes.

[0041] In this embodiment, each task virtual module of the power engineering is regarded as a node in the graph, and the dependency relationship (such as the order, resource sharing, etc.) between tasks is connected to each node as an edge. GNN propagates information through a message passing mechanism to pass feature information between nodes, and then extracts the task features of each node (i.e., engineering task), which includes progress, resource consumption, meteorological task influence, etc. Through the iterative propagation of the graph neural network, each task node can more accurately pass the key information between tasks, update the status of the task, and accurately capture and extract the dynamic features of the engineering task.

[0042] In step S140, the input state of the reinforcement learning model is defined according to the extracted task node features of each task, so as to generate a task resource allocation action for each engineering task through reinforcement learning.

[0043] It should be noted that reinforcement learning is a field of machine learning that emphasizes how an agent takes a series of actions in an environment to maximize cumulative rewards. In this process, the agent learns the optimal behavior strategy by constantly interacting with the environment. The agent selects an action according to the current state, and after executing the action, the environment will feedback a reward signal, and the environment state will also be updated. The goal of the agent is to learn a policy that maximizes the cumulative reward obtained in the long-term interaction process.

[0044] Figure 2 A schematic diagram showing the principle of an example of state transition actions in a reinforcement learning model.

[0045] As shown in Figure 2 , which involves actions corresponding to state transitions in a state space of multiple states S1-Sn, for example, a1 represents the action of state transition from S1 to S2, a2 represents the action of state transition from S2 to S1, a3 represents the action of state transition from S1 to S3, and so on. Here, the corresponding state transition can occur based on a policy, and each state transition policy can be used to occur different transitions, respectively. Exemplarily, based on the transition policy for state S1, actions a2 or a3 can occur.

[0046] It should be noted that the range of states that can be transitioned to from a state (also referred to as transferable states) can be limited or conditional, for example, none of S1-S3 can be state transitioned to S4-Sn, and the state that can be transitioned to from state S1 is S2 and S3, and so on.

[0047] In some embodiments, each action has a corresponding action reward, and each action reward can be determined based on a preset reward function. Generally, if the transition reward is greater, the action of the transition can be considered more valuable, and the system will preferentially select to execute this action. Exemplarily, if the reward corresponding to action a1 is greater than the reward corresponding to a3, it indicates that the action of transition a1 is more valuable.

[0048] In the embodiments of the present application, the task resource allocation action is used to indicate the resource allocation scheme for the corresponding engineering task, and the resource allocation scheme includes a building material allocation strategy, an engineering equipment allocation strategy, and a labor force allocation strategy. In this way, the decision-making process of task resource allocation is regarded as an "action selection" problem, and the reinforcement learning model is used to learn the optimal resource allocation strategy through continuous trial and error. The input state of the reinforcement learning model is composed of the extracted features of each task node, including task progress, resource consumption, external meteorological factors, and other information, which collectively affect the allocation decision of task resources.

[0049] In addition, the reward function of the reinforcement learning model is defined according to the total cost of the power engineering, the total cycle of the power engineering and the average use efficiency of the resource of the power engineering task, multiple optimization objectives are comprehensively considered, the resource allocation strategy is optimized through the trial and error process to maximize the comprehensive reward, so as to guide the resource scheduling for different tasks and ensure the completion on schedule and cost control of the power engineering project. Therefore, through the adaptive adjustment capability of the reinforcement learning model, intelligent resource allocation can be performed according to the real-time task state and the actual situation of the project, the reasonable allocation of resources between tasks is ensured, and excessive or insufficient allocation is avoided.

[0050] Through the embodiments of the present application, the complex dependency relationship between engineering tasks is comprehensively captured by comprehensively applying digital twinning, graph neural network and reinforcement learning model, the correlation of each task is effectively modeled, intelligent decision is made through dynamic data updating and task feature extraction, and reinforcement learning is introduced, so that the resource allocation is more reasonable. Therefore, when facing complex engineering environment, timely and accurate intelligent analysis and solution scheme for cost control can be provided, and the management efficiency of the power engineering project is significantly improved.

[0051] In some examples of the embodiments of the present application, the reward function of the reinforcement learning model is:

[0052]

[0053] In the formula, R t represents the reward value obtained by the reinforcement learning model according to the action strategy; C total represents the total cost of the project, C max represents the maximum cost budget of the power engineering, T total represents the total cycle of the project, T max represents the maximum duration of the project plan. U avg represents the average use efficiency of the resource of the project task, U avg The greater the resource use rate is, the higher the resource scheduling allocation is, and the higher the reward is. U max represents the maximum value of the use efficiency of the resource of the project task; w1, w2 and w3 represent weighting coefficients for controlling the importance of cost, cycle and resource use efficiency in the reward function, respectively; is an indicator function, which takes the value of 1 when the condition {·} is met, otherwise 0; λ1 and λ2 represent the cost overrun penalty coefficient and the duration overrun penalty coefficient, respectively. For example, as the total cost of the project increases, the penalty coefficient λ1 also increases; in addition, as the completion cycle is prolonged, the penalty coefficient λ2 also increases.

[0054] It should be noted that the total project completion cycle, the total project completion cost and the average use efficiency of the project task resources are important feedback data in the training process of the reinforcement learning model. These data can be obtained through the action execution process of the reinforcement learning model and the dynamic simulation of the project task environment. Specifically, the environment estimates the total project completion cycle T according to the resource allocation plan of the reinforcement learning model and the current task state, updates the task progress dynamically, takes the maximum value of the completion time of all tasks, and estimates the total project completion cycle T total In addition, the environment dynamically calculates the resource consumption of each task according to the resource allocation plan generated by the reinforcement learning model and the current task state, thereby obtaining the corresponding total project completion cost C total In addition, the task resource use efficiency is calculated by the ratio of the task progress increment to the resource allocation amount, U avg The average value of the task resource use efficiency of all project tasks is used to represent the average value of the task resource use efficiency of all project tasks. The environment generates an allocation plan according to the action of the reinforcement learning model, and calculates the task resource use efficiency at each time step through the dynamic change of the task progress.

[0055] Therefore, by combining the interaction of the reinforcement learning model and the engineering simulation environment, the duration, cost and efficiency are accurately calculated through the simulation environment, the optimization direction is provided for the reinforcement learning model, the task state is updated in real time, and the resource allocation is dynamically adjusted to reduce the duration and cost. Further, through the guidance of the reward function, multi-objective optimization of duration, cost and resource efficiency is realized.

[0056] In the embodiments of the present application, the reward function encourages the reduction of the total completion cost C total When the total cost gradually approaches the upper limit of the budget C max , the reward will decrease, thereby guiding the reinforcement learning model to actively optimize the resource allocation strategy, which can discover the risk of cost overrun in real time and trigger optimization of the engineering resource allocation strategy. In addition, through the design of the penalty term, if the actual cost exceeds the upper limit of the budget, the indicator function will be triggered to generate a fixed penalty value λ1, thereby further suppressing the behavior of exceeding the budget.

[0057] In addition, the reward function encourages the project completion cycle T total to tend to the maximum project plan duration T max The shorter the cycle, the higher the reward. In this way, the reinforcement learning model can be effectively guided to preferentially select strategies that can accelerate project progress when optimizing resource allocation, such as concentrating resources to complete critical tasks. Through the design of the penalty term, when T total exceeds T max , the penalty is triggered, further strengthening the constraint on the duration, and avoiding the model from ignoring the time target when pursuing cost or resource use rate, ensuring that the power engineering can be completed within the planned time.

[0058] Further, in the reward function, the reward resource usage rate is close to the maximum value U max For example, 100%), so that the reinforcement learning model can dynamically allocate resources, avoid the situation of idle or waste of resources, and ensure efficient use of resources (such as people, materials, and machines), thereby improving the overall execution efficiency of the project and avoiding waste of resources due to unreasonable scheduling.

[0059] Through the embodiments of the present application, based on the adjustment of w1, w2, and w3, when optimizing one target, the reward function will not ignore other targets, so that the system can comprehensively consider the cost, construction period, and resource usage efficiency when making dynamic decisions, avoiding the problem of global imbalance caused by excessive optimization of a single target. In addition, it can also meet the personalized management needs of different power engineering projects, such as cost-sensitive, construction period-sensitive, and the like. For example, in a project with a tight construction period, the weight of w2 can be increased to prioritize the optimization of the construction period; in a budget-sensitive project, the weight of w1 can be increased to focus on cost control.

[0060] Figure 3 An operation flowchart is shown according to an example of step S110 in Figure 1

[0061] As shown in Figure 3 In step S310, task data collection instructions are periodically sent to each preset task management terminal, and each task management terminal is used to maintain data updates of corresponding engineering tasks.

[0062] In some embodiments, the task management terminal can refer to one or more specific terminal devices, or it can refer to one or more clients with specific user accounts. In some scenarios, a power engineering project involves the cooperation of multiple departments, such as the civil engineering group, the cable installation group, the equipment installation group, the communication debugging group, and the like. Each department uploads the task-related data of the corresponding department through the maintenance of the corresponding task management terminal, realizing the real-time synchronization and sharing of engineering task data in the global scope of the engineering project.

[0063] In some embodiments, the system periodically (for example, daily or weekly) sends task data collection instructions to each task management terminal, so that the administrators of each department can upload engineering task multi-modal data, such as task progress, resource usage, weather conditions, and the like, through the task management terminal after receiving the instructions.

[0064] In step S320, in response to the task data collection instructions, the corresponding engineering task multi-modal data is received from each task management terminal.

[0065] ​In some embodiments, the system stores the data received from the task management terminal, for example, integrates the data from different task management terminals according to the task ID, timestamp, etc., forms a complete multi-modal data package for each task, and then stores it in a cloud database or a distributed database.

[0066] In step S330, an engineering task tree structure of the power engineering is obtained, and the engineering task tree structure includes decomposition level information and task plan information of each task.

[0067] Here, the decomposition level information contains the parent-child node relationship of the task, and the task plan information contains the planned start time, planned end time and task resource requirement of the task.

[0068] It should be noted that the engineering task tree is a tree-like hierarchical structure, which embodies the decomposition relationship of the engineering task. The root node represents the entire power engineering project, and each child node represents a specific engineering task or task group. The construction of the task tree is usually designed by the project manager during the project planning stage through a project management tool (such as Microsoft Project, Primavera, etc.). Each task node contains subtask, parent task and execution order information, so that the system can effectively extract the task level relationship of the power engineering project and clearly determine the position of each task node in the entire project. In addition, each task node not only contains hierarchical relationship information, but also includes the planned start time, planned end time, required resources, etc. of the task, which can be pre-set according to the project plan arrangement.

[0069] In step S340, according to the engineering task tree structure, the task dependency relationship between each engineering task is determined by quantifying the dependency strength between tasks.

[0070] It should be noted that the dependency relationship between different engineering tasks can be diversified, such as sequential dependency, parallel dependency, resource sharing, etc. By combining the parent-child task relationship in the task tree structure, the task plan information and the constraint conditions in project management (such as project delivery date, resource limitation, etc.), the dependency strength between each two tasks is quantified.

[0071] Exemplarily, the dependency strength between tasks can be quantitatively represented by the following formula:

[0072]

[0073] Where: Weight ij represents the dependency strength or weight between tasks v i and v j , Level ij represents the dependency strength or weight between tasks v i and v jcorrelation strength on the decomposition level; TimeGap ij for task v i j Time interval of v ij and v i , used to measure the closeness of tasks in time scheduling.

[0074] For decomposition level correlation strength Level ij , which can be defined by preset rule matching. Exemplarily, if v i is the direct parent task of v j , then Level ij = 1; if v i and v j are child tasks of the same parent task, then Level ij = 0.5; if v i and v j are located in different decomposition paths and have no direct parent-child relationship, then Level ij = 0.

[0075] For TimeGap ij , exemplarily, if the start time of task v j is later than the end time of v i , the time interval is positive; if the start time of task v j is earlier than or equal to the end time of v i , the time interval is 0 (indicating no time delay or overlapping execution).

[0076] Through the embodiments of the present application, the dependency strength between tasks is quantified by combining the decomposition level and the task planning time information, ensuring that the model accurately reflects the actual logical relationship of engineering tasks, and providing key reference information for subsequent task scheduling optimization.

[0077] In some examples of the embodiments of the present application, when it is detected that the first engineering task multi-modal data corresponding to the first engineering task exists update, the updated first engineering task multi-modal data is re-modeled based on digital twin technology to trigger the task resource allocation action for each engineering task.

[0078] In combination with the business application scenarios, after the administrators of various departments upload or update the engineering task multi-modal data to the system platform every day or every week, the digital twin technology is used to re-model the updated engineering multi-modal data, so that each task and link of the project can be displayed in real time in the virtual model, and the manager can understand the current state and resource consumption of each task at any time, thereby improving the transparency of project management. In addition, based on the re-modeled power engineering virtual model, the graph neural network is triggered to re-extract features, and the dynamic decision adjustment of resource allocation is re-performed through the reinforcement learning model. In this way, after the new resource allocation scheme is generated, the system will feed back the optimized resource allocation to the task management terminal to guide the scheduling and allocation of various resources (such as manpower, equipment, and materials), so as to ensure that the task is completed on schedule, for example, if the system finds potential risks, it can avoid the expansion of project risks through real-time early warning and resource adjustment and allocation.

[0079] Figure 4 An operation flowchart of an example of constructing a power engineering virtual model based on digital twin technology according to an embodiment of the present application is shown.

[0080] As shown in Figure 4 In step S410, a task state vector corresponding to each engineering task multi-modal data is extracted, and the task state vector includes a task completion percentage, a resource consumption deviation, and a meteorological task influence degree.

[0081] Here, the resource consumption deviation defines the difference between the resource consumption data of the corresponding engineering task and the expected resource consumption data. If the actual resource consumption of the task is 10% higher than the expected value, the deviation is +10%. The meteorological task influence degree defines the degree of interference of the engineering area meteorological data on the corresponding engineering task. For example, excessive wind speed may affect the safety of high-altitude operations, and excessive precipitation may cause the progress of civil engineering to lag. Through the task completion percentage in the task state vector, the amount of work completed in the execution process of the task is reflected.

[0082] In step S420, according to each task state vector, the corresponding engineering task virtual module is rendered in real time.

[0083] It should be noted that the virtual module is the digital representation of the engineering task in the digital twin system, and each task is rendered as a virtual module through its corresponding task state vector, thereby dynamically displaying the execution of each task. For example, the progress of the task will be adjusted in real time with the update of the completion percentage, the deviation of the resource consumption will affect the consumption progress bar of the related resources in the virtual module, and the change of the meteorological task influence degree will affect the state identifier of the task. In addition, during the rendering process, the system will update the virtual module in real time according to each item of data in the task state vector.

[0084] In step S430, edge connections are constructed between the engineering task virtual modules of the indicated association according to the task dependency relationship, and task priorities are labeled for each engineering task virtual module to construct the power engineering virtual model.

[0085] Here, the edge weight corresponding to the edge connection is defined by the dependency strength between the engineering tasks indicated by the task dependency relationship. For the description of the dependency strength between tasks and the related calculation details, reference can be made to the description in the foregoing with respect to other examples, which will not be described herein.

[0086] In some embodiments, the system automatically assigns a priority to each task virtual module according to the criticality, resource consumption, and dependency relationship of the task. A task with a higher priority generally means that it has a greater impact on the overall progress of the project, or that its delay can cause other tasks to lag. Therefore, the task priority will directly affect resource allocation and task scheduling, and a task with a high priority will be given more resources by the system to ensure.

[0087] Through the embodiments of the present application, the power engineering virtual model constructed based on digital twinning technology can reflect the changes in the state of the task, resource consumption, and external environment in real time during the execution of the task. The real-time rendering of the task virtual module and the accurate modeling of the task dependency relationship enable the system to flexibly respond to changes in the progress of the task, optimize resource allocation, and provide clear decision support for project managers. Ultimately, the real-time feedback and automated optimization of the system will greatly improve the efficiency and completion of project management.

[0088] As for the details of the labeling of the task priority, in some examples of the embodiments of the present application, the path cumulative impact score and the task dependency path depth corresponding to each engineering task virtual module are obtained, and the task priority is labeled for the corresponding engineering task virtual module according to the path cumulative impact score and the task dependency path depth.

[0089] Exemplarily, the calculation formula of the task priority is:

[0090]

[0091] In the formula, pr i represents the task priority of the i-th engineering task virtual module, All_Dts(i) represents the set of all subsequent engineering task virtual modules of the i-th engineering task virtual module. Depth(j) is the task dependency path depth of the subsequent engineering task virtual module j relative to the i-th engineering task virtual module, which is represented by the virtual module hierarchical distance. The path impact is attenuated according to the hierarchical depth Depth(j) between tasks, and the contribution of a deep task to i is smaller. represents the edge weight between the k-th engineering task virtual module and the l-th engineering task virtual module. The path cumulative influence score corresponding to i represents the product of all edge weights on the path from i to j, and the path influence intensity from i to j is evaluated for all subsequent j of i. By using the product of path weights to calculate the transmitted influence, the chain effect of dependence can be reflected.

[0092] Through the path cumulative influence score, the chain transmission influence of task i on all its subsequent tasks j is captured, and the more complex the path and the lower the edge weight, the weaker the cumulative influence. Thus, the interference of deep-level tasks on priority ranking is avoided, and the priority score is more reasonable.

[0093] By task dependency path depth, a depth attenuation term is constructed, and the deeper the depth, the smaller the transmitted influence, which reflects that the dependence of deep tasks on upstream tasks gradually weakens. When the dependency path of some tasks becomes more complex or the risk of delayed propagation of some tasks increases, the priority calculation will be dynamically adjusted to reflect the changes in task importance.

[0094] Through the embodiments of the present application, combined with path cumulative weight and path depth, the complex dependency relationship between tasks can be dynamically evaluated, and high-priority tasks can be quickly identified due to larger cumulative influence scores and lower path depths. Therefore, by dynamically considering task dependency structure and depth influence for priority marking, task ranking is no longer only concerned with a single task, but considers its global influence in the entire task network, which helps to achieve global optimization of the project goal (e.g., cost optimization without project delay).

[0095] In some examples of the embodiments of the present application, the graph neural network adopts a message passing mechanism based on a time dynamic graph. Specifically, according to the task state vector of each task virtual module, the node features of each task virtual module are initialized as follows:

[0096]

[0097] In the formula, denotes the initialized node feature of the engineering task virtual module i corresponding to the time step t, denotes the task completion percentage of i corresponding to the time step t, denotes the resource consumption deviation of i corresponding to the time step t, denotes the meteorological task influence degree of i corresponding to the time step t, T i current denotes the current time progress of the task of i, T i expected denotes the expected end time of the task of i.

[0098] The edge weight of the edge connected in the following way is dynamically updated:

[0099]

[0100] wherein, denotes the initial edge weight between the virtual module i of the engineering task and the virtual module j of the subsequent engineering task, pr j denotes the priority of the engineering task of the virtual module j, and denote the updated edge weight, the resource sharing intensity factor and the delay propagation risk factor between i and j at the corresponding time step t, respectively; γ1 and γ2 denote the adjustment weight parameters of the resource sharing intensity and the delay propagation risk on the edge weight, respectively, and β denotes the priority sensitivity parameter; denotes the amount of shared resources provided by the engineering task corresponding to i to the engineering task corresponding to j at the time step t; denotes the total resource usage of the engineering task corresponding to i at the time step t; denotes the delay time of the engineering task corresponding to i at the time step t, Plan_Duration i denotes the planned duration of the engineering task corresponding to i.

[0101] It should be noted that the resource sharing intensity factor and the delay propagation risk factor are the most critical parts of the dynamic edge weight formula, which directly affect the dependency strength between task nodes and are dynamically changing.

[0102] Specifically, the resource sharing intensity factor denotes the degree of resource support provided by i to j at the time step t. When the amount of shared resources provided by i to j is close to its total resource usage, close to 1, indicating that the resource support intensity is very high; when i does not provide any resources to j Therefore, by reflecting the actual dependency between tasks in terms of resources through the resource sharing intensity factor, it can be ensured that in the case of limited resources, the dependency relationship can accurately reflect the actual resource support degree between tasks, thereby optimizing task scheduling and resource allocation.

[0103] The delay propagation risk factor denotes the degree of negative impact of the task delay of i on the task of j. In engineering tasks, the delay of the preceding task often leads to the inability of the subsequent task to start on time, and even causes global duration delay. When i has no task delay denotes that the delay risk is zero. When the task delay time of i is longer, the larger, indicating that the delay risk is higher. Thus, by quantifying the direct impact of task delay on subsequent tasks through the delay propagation risk factor, it can help the model accurately capture the duration risk, optimize the task dependency relationship, and reduce the risk of overall engineering delay.

[0104] In addition, by incorporating the task priority difference |Pr i -Pr j | into the dynamic edge weight formula, the model can explicitly consider the urgency difference between tasks. In tasks with large priority differences, the dependency strength is appropriately weakened, which helps to ensure the priority of high-priority tasks in resource scheduling.

[0105] The message received by each engineering task virtual module from the subsequent engineering task virtual module at time step t is defined as:

[0106]

[0107] In the formula, denotes the dynamic attention weight between i and j at time step t, and || denotes the vector concatenation operation, denotes the initialization node feature of j at time step t, and a denotes a learnable attention vector, denotes the message aggregated from the subsequent engineering task virtual module to i at the corresponding time step t; is a normalization term to ensure

[0108] Here, the message passing mechanism combines task features and edge weights to weight the messages of neighbor nodes through a dynamic attention mechanism, so that each task node can focus on the neighbor nodes that have the greatest impact on it when updating its features. In the dynamic attention mechanism, the numerical stability is ensured through the normalization operation, and the feature contribution strength of the neighbor nodes is explicitly reflected, improving the accuracy of message passing and the reliability of task node feature updating.

[0109] The historical time series features of the nodes are captured through the LSTM module, and the aggregated results of the message passing are fused to update the node features:

[0110]

[0111] In the formula, W s denotes a learnable weight matrix, and σ denotes an activation function, denotes the historical time series feature of i at the corresponding time step t, and △T denotes a preset time window length, denotes the time series feature analysis of each historical feature of i within the time window through the LSTM module; denotes the extracted task node feature of the i-th engineering task virtual module at the corresponding time step t.

[0112] It should be noted that in the embodiments of the present application, the dynamic graph network adopts a time series modeling structure, and the corresponding task dependency graph Gt = (V, E t ), and introduce a temporal connection between time steps to model the time dynamic evolution process of task node features. The dynamic state of the current task is calculated through the direct neighbor features of the task node to realize local dependence aggregation; at the same time, the historical features of the task node are captured through the time sequence modeling based on the LSTM module to realize time sequence feature fusion.

[0113] Through the embodiments of the present application, by introducing a dynamic edge weight updating mechanism, the initial static task dependence relationship is combined with dynamic adjustment factors (task priority, resource sharing intensity and delay propagation risk), the dependence relationship between task nodes is more accurately quantified, and the dynamic updating of the dependence relationship can help decision makers quickly identify key task links. By introducing an LSTM module to model the historical features of the nodes, the historical state of the task node and the current dependence relationship are fused, the connection strength between tasks is dynamically optimized, the state evolution law of the task node at multiple time steps is captured, and the limitation of the state at a single time point is made up.

[0114] It should be noted that, for each of the foregoing method embodiments, in order to simply describe, each is described as a series of actions, but those skilled in the art should know that the present application is not limited by the order of the described actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present application. In the above embodiments, the description of each embodiment is focused on, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0115] Figure 5 A structural connection diagram of an example of a reinforcement learning model according to an embodiment of the present application is shown.

[0116] As shown in Figure 5 , the reinforcement learning model 500 includes a resource multi-head allocation network 510, a policy constraint adjustment network 520, and a task multi-head output network 530. Each head (511, 513…51n) of the resource multi-head allocation network 510 corresponds to a unique resource type, and each head (531, 533…53m) of the task multi-head output network 530 corresponds to a unique engineering task.

[0117] Each head (511, 513…51n) of the resource multi-head allocation network 510 is used to determine the initial allocation policy of each engineering task in the corresponding resource type for the input state; the resource types include building material resources, engineering equipment resources, and labor resources.

[0118] Exemplarily, three independent distribution heads are designed for three main resource types (building materials, engineering equipment and labor manpower) in the power engineering task, and each distribution head generates an initial distribution strategy for the corresponding resource type. It should be understood that the number of resource types can also be increased or adjusted according to the needs of the engineering project. Through the multi-head design of resource types, the characteristics of different resource types can be independently modeled, avoiding interference between resource distribution strategies. Thus, for each task, the initial distribution proportion is generated according to its characteristics and global resource conditions, which can ensure the logical rationality of resource distribution, for example, high-priority tasks and tasks with high resource demand should obtain more distribution.

[0119] The strategy constraint adjustment network 520 is used to count the total amount of strategy resource distribution corresponding to the initial distribution strategy of each engineering task under the corresponding resource type, and in the case that the first strategy resource distribution total amount corresponding to the first resource type exceeds the threshold value of the total amount of strategy resources available for distribution, the first initial distribution strategy under the first resource type is adjusted according to the task priority corresponding to each engineering task.

[0120] It should be noted that the main function of the strategy constraint adjustment network is to detect whether the initial distribution strategy exceeds the global resource limit, and dynamically adjust the distribution strategy to meet the constraint condition when the limit is exceeded, for example, adjust the distribution proportion according to the task priority (for example, Pr i ) to make high-priority tasks obtain resource distribution first. Thus, it is ensured that the final resource distribution scheme meets the global resource constraint, avoiding resource overload, and in the case of resource limitation, key tasks are prioritized to improve the overall efficiency of the project.

[0121] Each head (531, 533…53m) of the task multi-head output network 530 is respectively used to fuse each distribution strategy of the corresponding engineering task to output a corresponding task resource distribution action, which is used to indicate a resource distribution scheme for the corresponding engineering task.

[0122] It should be noted that the main function of the task multi-head output network is to fuse the distribution strategies of each task under different resource types, for example, to fuse the distribution strategies of each task on three types of resources to generate a final resource distribution action to guide the actual engineering resource distribution.

[0123] Through the reinforcement learning model structure provided by the embodiments of the present application, the multi-head network design ensures the pertinence and flexibility of resource distribution, the strategy constraint adjustment network responds to resource limitations in real time, realizes global optimization, and the distribution logic driven by task priority improves the completion efficiency of key tasks, thereby ensuring overall project duration and cost optimization. Finally, the resource distribution plan output by the model can provide fine and intelligent support for engineering management.

[0124] It should be noted that for each of the foregoing method embodiments, for the sake of simplicity, the method is described as a series of acts for easy of description, but the skilled person should appreciate that the application is not limited by the order of the acts described, as some acts can be performed in other orders or concurrently with each other. Furthermore, the skilled person should appreciate that the embodiments described in the specification are preferred embodiments, and the acts and modules described are not necessarily all required by the application. In the above embodiments, each of the embodiments is described with a focus on different aspects, and the parts not described in detail in one embodiment can be found in the description of the other embodiments.

[0125] Figure 6 A structural block diagram of an example of a power engineering cost simulation analysis system based on digital twinning according to an embodiment of the application is shown.

[0126] As shown in Figure 6 The power engineering cost simulation analysis system 600 based on digital twinning includes a data acquisition unit 610, a twinning modeling unit 620, a feature extraction unit 630, and a reinforcement learning analysis unit 640.

[0127] The data acquisition unit 610 is configured to acquire a plurality of engineering task multi-modal data and task dependency relationships for a power engineering; the engineering task multi-modal data includes task progress data, resource consumption data, and engineering area meteorological data of the corresponding engineering task; the resource consumption data includes building material cost consumption information, engineering equipment cost use information, and engineering labor cost use information of the corresponding engineering task.

[0128] The twinning modeling unit 620 is configured to process the plurality of engineering task multi-modal data and task dependency relationships based on a digital twinning technology to construct a power engineering virtual model; the power engineering virtual model includes a plurality of engineering task virtual modules, each of the engineering task virtual modules is modeled by the corresponding engineering task multi-modal data, and there is an edge connection between the associated engineering task virtual modules indicated by the task dependency relationship.

[0129] The feature extraction unit 630 is configured to input the power engineering virtual model into a graph neural network to extract task node features corresponding to each of the engineering task virtual modules through a message passing mechanism.

[0130] The reinforcement learning analysis unit 640 is configured to define an input state of a reinforcement learning model according to the extracted task node features, so as to generate a task resource allocation action for each engineering task by reinforcement learning, the task resource allocation action being used for indicating a resource allocation scheme for the corresponding engineering task, the resource allocation scheme including a construction material allocation strategy, an engineering equipment allocation strategy, and a labor force allocation strategy; and a reward function of the reinforcement learning model is defined according to the total engineering completion cost, the total engineering completion cycle, and the average engineering task resource utilization efficiency of the power engineering.

[0131] In some embodiments, the embodiments of the present application provide a non-volatile computer readable storage medium, wherein one or more programs including execution instructions are stored in the storage medium, the execution instructions being readable and executable by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to perform the steps of any of the above simulation analysis methods of power engineering cost based on digital twinning.

[0132] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer readable storage medium, the computer program including program instructions, when the program instructions are executed by a computer, causing the computer to perform the steps of any of the above simulation analysis methods of power engineering cost based on digital twinning.

[0133] In some embodiments, the embodiments of the present application also provide an electronic device, which includes at least one processor, and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the simulation analysis method of power engineering cost based on digital twinning.

[0134] Figure 7 is a hardware structure schematic diagram of an electronic device for performing the simulation analysis method of power engineering cost based on digital twinning provided by another embodiment of the present application, as shown in Figure 7 The device includes:

[0135] one or more processors 710 and a memory 720, Figure 7 The processor 710 is taken as an example.

[0136] The device for performing the simulation analysis method of power engineering cost based on digital twinning can also include an input device 730 and an output device 740.

[0137] The processor 710, the memory 720, the input device 730, and the output device 740 can be connected through a bus or other means,Figure 7 The bus connection is taken as an example.

[0138] The memory 720 is a non-volatile computer readable storage medium, which can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the simulation analysis method of power engineering cost based on digital twinning in the embodiments of the present application. The processor 710 executes various functions of the server and data processing by running the non-volatile software programs, instructions and modules stored in the memory 720, that is, implements the simulation analysis method of power engineering cost based on digital twinning in the above method embodiments.

[0139] The memory 720 can include a program storage area and a data storage area. The program storage area can store an operating system and at least one application required by a function. The data storage area can store data created during use of the electronic device, etc. In addition, the memory 720 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state memory device. In some embodiments, the memory 720 can optionally include a memory disposed remotely with respect to the processor 710, which can be connected to the electronic device through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0140] The input device 730 can receive input digital or character information, and generate signals related to user settings and function control of the electronic device. The output device 740 can include a display device such as a display screen.

[0141] The one or more modules are stored in the memory 720, and when executed by the one or more processors 710, perform the simulation analysis method of power engineering cost based on digital twinning in any of the above method embodiments.

[0142] The above product can perform the method provided in the embodiments of the present application, and has the corresponding function modules and beneficial effects of performing the method. Technical details not described in detail in the embodiments can be referred to the method provided in the embodiments of the present application.

[0143] The electronic device of the embodiments of the present application exists in various forms, including but not limited to:

[0144] (1) Mobile communication device: This type of device is characterized by having mobile communication function, and the main target is to provide voice and data communication. This type of terminal includes: smart phone, multimedia phone, functional phone, and low-end phone, etc.

[0145] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally have mobile Internet access features. Such terminals include PDA, MID and UMPC devices, etc.

[0146] (3) Portable entertainment devices: These devices can display and play multimedia content. Such devices include audio and video players, handheld game consoles, electronic books, and smart toys and portable car navigation devices.

[0147] (4) Other onboard electronic devices with data interaction functions, such as car-mounted devices installed on vehicles.

[0148] The device embodiments described above are only illustrative, wherein the units illustrated as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place or distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.

[0149] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus a general hardware platform, and of course can also be implemented by hardware. Based on such understanding, the above technical solutions or the part that contributes to the related art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some part of the embodiment.

[0150] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A simulation analysis method for power engineering costs based on digital twins, comprising: Acquire multimodal data and task dependencies for multiple engineering tasks in power engineering; The multimodal data of the engineering task includes the task progress data, resource consumption data and meteorological data of the engineering area for the corresponding engineering task; the resource consumption data includes the cost consumption information of building materials, the cost of engineering equipment and the cost of engineering labor for the corresponding engineering task. The power engineering virtual model is constructed by processing the multimodal data and task dependencies of the multiple engineering tasks based on digital twin technology. The power engineering virtual model includes multiple engineering task virtual modules, each of which is modeled by the corresponding engineering task multimodal data, and there are edge connections between the associated engineering task virtual modules indicated by the task dependencies. The power engineering virtual model is input into a graph neural network to extract the task node features corresponding to each virtual module of the engineering task through a message passing mechanism. The input state of the reinforcement learning model is defined based on the extracted features of each task node, so as to generate task resource allocation actions for each of the engineering tasks through reinforcement learning. The task resource allocation actions are used to indicate the resource allocation scheme for the corresponding engineering task. The resource allocation scheme includes building material allocation strategy, engineering equipment allocation strategy and labor allocation strategy. The reward function of the reinforcement learning model is defined based on the total cost of completing the power project, the total completion period of the project, and the average resource utilization efficiency of the project tasks. The process of processing multimodal data and task dependencies of multiple engineering tasks based on digital twin technology to construct a virtual model of the power engineering includes: Extract the task state vector corresponding to the multimodal data of each engineering task. The task state vector includes the task completion percentage, resource consumption deviation, and meteorological task impact. The resource consumption deviation defines the difference between the resource consumption data of the corresponding engineering task and the expected resource consumption data. The meteorological task impact defines the degree of interference of meteorological data in the engineering area on the corresponding engineering task. Based on each of the aforementioned task state vectors, the corresponding engineering task virtual modules are rendered in real time. Based on the task dependencies, edge connections are constructed between the indicated associated engineering task virtual modules, and task priorities are marked for each engineering task virtual module to construct a power engineering virtual model; wherein, the edge weights corresponding to the edge connections are defined by the dependency strength between the engineering tasks indicated by the task dependencies.

2. The method according to claim 1, wherein, The reward function of the reinforcement learning model is: ， In the formula, Indicates the time step The input state, and the reward value obtained by the reinforcement learning model based on the action policy; This indicates the total cost of completing the project. This represents the maximum cost budget for a power engineering project. Indicates the total project completion period. Indicates the maximum planned construction period. This indicates the average resource utilization efficiency of the project task. This represents the maximum efficiency of resource utilization in engineering tasks. These represent weighting coefficients, used to control the importance of cost, cycle time, and resource utilization efficiency in the reward function, respectively. It is an indicator function, when the condition is met. When the condition is met, the indicator function takes the value 1; otherwise, it takes the value 0. and These represent the cost overrun penalty coefficient and the schedule overrun penalty coefficient, respectively.

3. The method according to claim 1, wherein, The acquisition of multimodal data and task dependencies for multiple engineering tasks in power engineering includes: Task data acquisition instructions are periodically sent to each preset task management terminal; each task management terminal is used to maintain the data update of the corresponding engineering task. In response to the task data acquisition command, corresponding engineering task multimodal data are received from each of the task management terminals; and Obtain the engineering task tree structure of the power project. The engineering task tree structure includes the decomposition level information and task plan information of each task. The decomposition level information includes the parent-child node relationship of the task, and the task plan information includes the planned start time, planned end time and task resource requirements of the task. Based on the project task tree structure, the task dependency relationships between each project task are determined by quantifying the dependency strength between tasks.

4. The method according to claim 3, further comprising: If an update is detected in the multimodal data of the first engineering task corresponding to the first engineering task, the updated multimodal data of the first engineering task is remodeled based on digital twin technology to trigger the regeneration of task resource allocation actions for each of the engineering tasks.

5. The method according to claim 4, wherein, The process of assigning task priorities to each engineering task virtual module includes: Obtain the cumulative path impact score and task dependency path depth corresponding to each engineering task virtual module, and mark the task priority of the corresponding engineering task virtual module according to the cumulative path impact score and the task dependency path depth; The formula for calculating task priority is as follows: ， In the formula, Indicates the first Task priority of each engineering task virtual module Virtual module representing engineering tasks A collection of all subsequent engineering task virtual modules; It is a virtual module for subsequent engineering tasks. Relative Engineering Task Virtual Module The task dependency path depth is represented by the distance between virtual module levels; Indicates the first The virtual module of the engineering task and the first Edge weights between virtual modules of each engineering task; for The corresponding path cumulative influence score indicates from arrive The product of the weights of all edges on the path.

6. The method according to claim 5, wherein, The graph neural network employs a message passing mechanism based on a time-dynamic graph. Based on the task state vector of each task virtual module, initialize the node features of each task virtual module: ， In the formula, Indicates the corresponding time step Engineering task virtual module Initialization node characteristics, Indicates the corresponding time step of Task completion percentage Indicates the corresponding time step of The deviation in resource consumption. Indicates the corresponding time step of The impact of meteorological tasks express The current time progress of the task. express Expected completion time of the task; The edge weights of edge connections can be dynamically updated in the following way: ， ， ， In the formula, Virtual module representing engineering tasks Virtual modules for subsequent engineering tasks Initial edge weights between them Virtual module representing engineering tasks Task priority, , and These represent the corresponding time steps. exist and The updated edge weights, resource sharing strength factors, and delayed propagation risk factors between them; and The adjusted weight parameters represent the impact of resource sharing intensity and delayed propagation risk on edge weights, respectively. This represents the priority sensitivity parameter; Indicates at time step , The corresponding engineering tasks are provided The amount of shared resources for the corresponding engineering tasks; Indicates at time step , The total resource usage of the corresponding engineering task; express The corresponding engineering task in time step The delay time, express The planned duration of the corresponding engineering task; Each engineering task virtual module is set at time step The message received from the virtual module of subsequent engineering tasks is defined as: ， ， In the formula, Indicates the corresponding time step exist and Dynamic attention weights between them This represents a vector concatenation operation. express At time step Initialization node characteristics, This represents a learnable attention vector. This indicates aggregation from virtual modules of subsequent engineering tasks to... Corresponding time step The message; For the normalization term, to ensure ; The node features are updated by capturing historical temporal features of nodes using an LSTM module and integrating the aggregation results from message passing. ， ， In the formula, This represents the learnable weight matrix. This represents the activation function. Indicates the corresponding time step of Historical chronological characteristics, Indicates the preset time window length. This indicates that the LSTM module is used for... Perform time series feature analysis on each historical feature within the time window; This indicates the extracted corresponding time step. The The task node characteristics of a virtual module for an engineering task.

7. The method according to claim 5, wherein, The reinforcement learning model includes a resource multi-head allocation network, a policy constraint adjustment network, and a task multi-head output network; each head of the resource multi-head allocation network corresponds to a unique resource type, and each head of the task multi-head output network corresponds to a unique engineering task. Each head of the resource multi-head allocation network is used to determine the initial allocation strategy for each engineering task in the corresponding resource type based on the input state; the resource types include building material resources, engineering equipment resources, and labor resources; The strategy constraint adjustment network is used to count the total amount of strategy resource allocation corresponding to the initial allocation strategy of each engineering task under the corresponding resource type, and when it is detected that the total amount of the first strategy resource allocation corresponding to the first resource type exceeds the threshold of the total amount of available strategy resources, the first initial allocation strategy under the first resource type is adjusted according to the task priority corresponding to each engineering task. Each head of the task multi-head output network is used to integrate the various allocation strategies of the corresponding engineering task to output the corresponding task resource allocation action; the task resource allocation action is used to indicate the resource allocation scheme for the corresponding engineering task.

8. A simulation analysis system for power engineering costs based on digital twins, comprising: The data acquisition unit is used to acquire multimodal data and task dependencies for multiple engineering tasks in power engineering. The multimodal data of the engineering task includes the task progress data, resource consumption data and meteorological data of the engineering area for the corresponding engineering task; the resource consumption data includes the cost consumption information of building materials, the cost of engineering equipment and the cost of engineering labor for the corresponding engineering task. A twin modeling unit is used to process the multimodal data and task dependencies of the multiple engineering tasks based on digital twin technology to construct a virtual model of power engineering. The virtual model of power engineering includes multiple virtual modules of engineering tasks, each of which is modeled by the corresponding multimodal data of engineering tasks, and there are edge connections between the associated virtual modules of engineering tasks indicated by the task dependencies. The feature extraction unit is used to input the power engineering virtual model into the graph neural network so as to extract the task node features corresponding to each of the engineering task virtual modules through the message passing mechanism. The reinforcement learning analysis unit is used to define the input state of the reinforcement learning model based on the extracted features of each of the task nodes, so as to generate task resource allocation actions for each of the engineering tasks through reinforcement learning. The task resource allocation actions are used to indicate the resource allocation scheme for the corresponding engineering task. The resource allocation scheme includes building material allocation strategy, engineering equipment allocation strategy and labor allocation strategy. The reward function of the reinforcement learning model is defined based on the total cost of completing the power project, the total completion period of the project, and the average resource utilization efficiency of the project tasks. The process of processing multimodal data and task dependencies of multiple engineering tasks based on digital twin technology to construct a virtual model of the power engineering includes: Extract the task state vector corresponding to the multimodal data of each engineering task. The task state vector includes the task completion percentage, resource consumption deviation, and meteorological task impact. The resource consumption deviation defines the difference between the resource consumption data of the corresponding engineering task and the expected resource consumption data. The meteorological task impact defines the degree of interference of meteorological data in the engineering area on the corresponding engineering task. Based on each of the aforementioned task state vectors, the corresponding engineering task virtual modules are rendered in real time. Based on the task dependencies, edge connections are constructed between the indicated associated engineering task virtual modules, and task priorities are marked for each engineering task virtual module to construct a power engineering virtual model; wherein, the edge weights corresponding to the edge connections are defined by the dependency strength between the engineering tasks indicated by the task dependencies.

Images