Construction method of power distribution network project cost data analysis model

Abstract

The invention provides a construction method of a power distribution network project cost data analysis model. The method comprises the following steps: S1, mapping a physical structure of a power distribution network into a topological graph; constructing a graph neural network based on the power distribution network, and predicting the cost of the power distribution network according to the topological graph; identifying key external factors of the construction cost of the power distribution network according to the prediction result of the construction cost of the power distribution network by using a causal inference technology; constructing a cost optimization model according to a cost prediction result of the graph neural network based on the power distribution network and key external factors identified by using a causal inference technology, and optimizing the predicted project cost; and on the basis of actual data of a construction site, updating a graph structure and external influence factor characteristics in real time, and adjusting a model prediction value and an optimization strategy. According to the method, the limitation of the prior art on processing the complex topological structure of the power distribution network and identifying the cost influence factors is effectively overcome, the cost prediction precision can be remarkably improved, and more targeted optimization decision support is provided for project managers.

Description

Technical Field

[0001] The invention belongs to the field of model construction, and in particular relates to a method for constructing a distribution network engineering cost data analysis model. Background Art

[0002] In distribution network projects, cost control is one of the core contents of project management. However, distribution network projects involve complex equipment layout, line distribution, construction environment and other multi-dimensional factors. Accurate prediction and control of costs face many challenges in reality. Existing cost analysis methods for distribution network projects usually rely on historical data, empirical rules or simple linear regression models. These traditional methods often lack the ability to handle complex topological structures and cannot accurately characterize the nonlinear relationship between equipment nodes, lines and construction conditions. The equipment layout and topological relationship of distribution network projects have an important impact on the cost, especially in the complex relationship between multiple substations, lines and load nodes. Traditional linear analysis methods are difficult to capture this highly complex correlation. Therefore, existing technologies often lead to low cost prediction accuracy, and the actual cost may deviate from expectations, causing cost overruns or increased management difficulties for the project.

[0003] On the other hand, the cost of distribution network projects is also significantly affected by external environmental factors, such as fluctuations in material prices, changes in construction time, and uncontrollable weather conditions. However, existing cost analysis methods can usually only regard these external factors as fixed variables or noise data, and fail to deeply explore the specific impact of these factors on the cost and their causal relationship. Existing methods often use simple correlation analysis in data analysis, and fail to distinguish the actual contribution of different factors to the cost, making it difficult for managers to take effective optimization measures. Therefore, cost managers can usually only rely on past experience or simple statistical models, and cannot dynamically and accurately control costs during the construction process. In addition, due to the failure to effectively identify key factors, cost optimization measures are usually not targeted, and it is difficult to achieve accurate cost control in a complex construction environment. Summary of the invention

[0004] The purpose of the present invention is to propose a method for constructing a distribution network engineering cost data analysis model that can simultaneously handle the complex topological structure of the distribution network and deeply explore the causal relationship between various influencing factors to improve the accuracy of cost prediction and provide effective support for optimized decision-making.

[0005] In order to achieve the above object, the present invention provides a method for constructing a distribution network engineering cost data analysis model, the method comprising:

[0006] S1. Map the physical structure of the distribution network into a topology graph G = (V, E), where the structure of the power grid topology graph G is: each device node in the distribution network is represented as a node set V, each node v∈V contains its device attributes, each connection line is represented as an edge set E, each edge e∈E is the influence of the line, where the attribute of each node is represented by a vector h v Indicates that the attributes of the edge are represented by the vector h e Represented, and the equipment and line characteristics are represented as node feature matrices H v and edge feature matrix H e , and construct external features X ext , the external feature X ext Respectively with the node feature matrix H v and edge feature matrix H e Combine the corresponding nodes to form the final feature F v and the final feature F of the edge e ;

[0007] S2. Construct a graph neural network based on the distribution network to predict the cost of the distribution network according to the topological graph G=(V, E); wherein the graph neural network based on the distribution network is a multi-layer graph neural network, in which the characteristics of the nodes in each layer are updated by information propagation between their neighboring nodes, while considering the influence of the line. The update rule of the multi-layer graph neural network is expressed as follows:

[0008]

[0009] in, W represents the feature vector of node v at layer l; (l) Represents the weight matrix of the lth layer in the graph neural network, learning the linear transformation of node features; Represents the characteristics of edge e(v,u), connecting node v and its neighbor node u;

[0010] represents the weight matrix of the lth layer for edge features; b v represents the bias term of the node, taking into account the impact of the characteristics of the node itself on the cost; σ represents the nonlinear activation function;

[0011] S3. Use causal inference techniques to identify key external factors of distribution network project cost based on the results of distribution network cost forecast;

[0012] S4. Build a cost optimization model based on the cost prediction results of the distribution network-based graph neural network and the key external factors identified using causal inference technology to optimize the predicted project cost;

[0013] S5. Update the graph structure G and external influencing factor features X in real time based on the actual data of the construction site ext , adjust model prediction values ​​and optimization strategies.

[0014] Furthermore, the attribute of each node is represented by a vector h v Including the type, life, installation and maintenance cost of the equipment; the attributes of the edge are represented by the vector h e Including construction cost, line length; the external features X ext Including material prices, construction weather and estimated construction duration.

[0015] Furthermore, the S2 also includes adding a regularization term in the optimization process of the graph neural network based on the distribution network. Encourage the model to learn the sensitivity of external factors to cost prediction. The regularization term It is expressed as follows:

[0016]

[0017] in, represents the total project cost predicted by the graph neural network based on the distribution network; X ext,v represents the external factor of node v; λ represents the weight of the regularization term, which controls the degree of regularization.

[0018] Furthermore, when predicting the cost of the distribution network according to the topology graph G = (V, E), a weighted aggregation mechanism is designed according to the different contributions of different device nodes to the cost to generate an overall cost prediction. It is expressed as follows:

[0019]

[0020] Among them, α v represents the weight of node v, which is determined by its device type and importance; The final feature representation of node v reflects the contribution of the node to the overall cost;

[0021] The loss function of the graph neural network based on the distribution network is expressed as follows:

[0022]

[0023] in, represents the predicted cost of the i-th sample; C actual,i represents the actual cost of the i-th sample; N represents the number of samples.

[0024] Furthermore, the use of causal inference technology to identify key external factors of the distribution network project cost based on the results of the distribution network cost forecast specifically includes:

[0025] The causal relationship between external factors and predicted construction cost is quantified by the structural equation model, and then a causal diagram is constructed; wherein the structural equation model is expressed as follows:

[0026]

[0027] Among them, P material Indicates the material price, representing the impact of material cost on construction cost; W weather Indicates weather conditions during construction, which affect construction progress and cost; D duration Indicates the construction time and the cost caused by the construction period; F v and F e represents the characteristics of nodes and edges, reflecting the cost characteristics of equipment and lines respectively; β1, β2, β3 represent the influence coefficients of external factors; β v ,β e represents the weight of the impact of equipment and lines on the overall cost; ∈ represents the error term caused by unobserved factors;

[0028] Set the external factor sensitivity regularization term in the causal graph Reduce the error fluctuation caused by over-reliance on certain external factors, the sensitivity regularization term of the external factors It is expressed as follows:

[0029]

[0030] Among them, λ and μ represent the sensitivity control parameters to material price fluctuations and weather factors, respectively, and are usually adjusted according to the specific needs of the actual project; Indicates the impact of material price fluctuations on construction costs sensitivity; Indicates the impact of changing weather conditions on construction costs sensitivity.

[0031] Furthermore, the nodes of the causal graph include the total project cost predicted by the graph neural network based on the distribution network. External influencing factors characteristics X ext and the final feature F of the node v , the final feature F of the edge e ;

[0032] At the same time, causal intervention items are added to the causal diagram to simulate the adjustment strategy of key external factors. If the material price P is identified material The contribution to the overall cost is the largest, and intervention is carried out by adjusting the material procurement strategy, as shown below:

[0033]

[0034] Among them, δ represents the intervention intensity, which means reducing the impact of material price fluctuations on construction costs by adjusting material procurement strategies.

[0035] Furthermore, the cost optimization model is constructed as follows:

[0036] Construct cost optimization objective function to minimize predicted cost And by adjusting the external factor X ext The optimization is implemented as follows:

[0037]

[0038] Among them, δ i represents the intervention coefficient used to adjust external factors; β i X ext,i Represents the influence weight of each external factor; regularization term Used to control the sensitivity of external factors to improve the robustness of the model;

[0039] Adding regularization terms to the cost optimization model Limiting the impact of drastic fluctuations in external factors on construction costs is expressed as follows:

[0040]

[0041] Where μ represents the regularization strength parameter; Represents the historical average of external factors; Represents the historical standard deviation of external factors and indicates the normal fluctuation range.

[0042] Furthermore, the S5 specifically includes:

[0043] Design dynamic feedback equations to gradually integrate real-time data into the optimization process;

[0044] Design data fusion algorithms to determine the real-time impact of external factors and construction progress on costs by comparing actual data with predicted data;

[0045] Use sensors to collect real-time feedback, adjust the cost optimization model based on the real-time feedback, and dynamically iterate to adapt to the new external environment and construction progress. At the same time, recalculate the optimization objective function after each feedback update.

[0046] Further, the dynamic feedback equation is expressed as follows:

[0047]

[0048] in, Indicates the updated cost forecast result; represents the initial cost prediction result obtained through optimization; Xreal and X pred They are external factor data collected in real time and external factor data during prediction; real and P pred They are the feedback data of construction progress and actual cost and the data in the forecast respectively; α and β represent the feedback control coefficients, which are used to adjust the weight of different data on the final result;

[0049] The design data fusion algorithm determines the real-time impact of external factors and construction progress on costs by comparing actual data with predicted data. Specifically:

[0050] Real-time update of external factor data through sensors, manual records or market data X real and construction cost feedback P real ;

[0051] Calculate the difference between actual data and model prediction data externally: including external factor differences and construction cost differences. The external factor difference ΔX is expressed as follows:

[0052] ΔX=X real -X pred

[0053] Among them, X real represents the actual external factors, X pred It means predicting external factors;

[0054] The construction cost difference ΔP is expressed as follows:

[0055] ΔP=P real -P pred

[0056] Among them, P real represents the actual construction cost, P pred Represents the predicted construction cost

[0057] Integrate the differences ΔX and ΔP into the feedback loop of the graph neural network based on the distribution network to adjust the predicted cost Thereby updating the graph neural network based on distribution network;

[0058] The sensors are used to collect real-time feedback, and the cost optimization model is dynamically iterated based on the adjustment of the real-time feedback to adapt to the new external environment and construction progress. At the same time, after each feedback update, the optimization objective function is recalculated, which is expressed as follows:

[0059]

[0060] Among them, X real,i Indicates the current actual external factor data.

[0061] Furthermore, a convergence condition is introduced into the dynamic feedback equation to determine whether the real-time feedback is close enough to the actual data, which is expressed as follows:

[0062]

[0063] Where ∈ represents the error convergence threshold.

[0064] The beneficial technical effects of the present invention are at least as follows:

[0065] (1) Traditional cost analysis methods are difficult to handle the complex topological structure of distribution network projects. The present invention introduces graph neural network technology, which can capture the nonlinear relationship between device nodes and lines by modeling the topological structure of the distribution network. Through the multi-layer information propagation mechanism of GNN, the present invention can accurately identify the contribution of different devices and lines to the cost, and generate more accurate cost prediction results on this basis. This technology breaks through the limitations of traditional linear models, can handle more complex spatial layouts and construction conditions, and significantly improves the accuracy of predictions.

[0066] (2) In view of the limitations of the prior art in external factor analysis, the present invention innovatively introduces causal inference technology, which can effectively identify the actual impact of various external factors (such as material prices, construction time, weather conditions, etc.) on project costs. By constructing a causal graph, the present invention can not only identify the correlation between various factors, but also clarify the causal relationship between them, helping managers to better understand which factors play a key role in cost fluctuations. Different from the simple correlation analysis in the prior art, causal inference can provide a more scientific basis for cost optimization and ensure the effectiveness of optimization measures.

[0067] (3) The present invention combines graph neural networks and causal inference technology to adjust and update the cost prediction model in real time during the construction process, and provide targeted optimization strategies based on the actual project progress and changes in the external environment. For example, when the system recognizes that material prices may fluctuate significantly, the system can suggest project managers to pre-purchase materials or choose alternative solutions in advance, thereby effectively reducing cost risks. Through this dynamic adjustment and decision support function, the present invention can significantly improve the flexibility and effectiveness of cost control, helping project managers to better control costs in complex construction environments.

[0068] (4) The present invention solves the shortcomings of the prior art in processing complex topological structures, identifying key influencing factors, and dynamically optimizing decisions by combining graph neural networks and causal inference, and provides a more accurate and intelligent distribution network engineering cost analysis and optimization method. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0070] Figure 1 The present invention is a flow chart of a method for constructing a distribution network engineering cost data analysis model. DETAILED DESCRIPTION

[0071] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0072] like Figure 1 As shown, an embodiment of the present invention provides a method for constructing a distribution network engineering cost data analysis model, the method comprising the following steps S1-S5:

[0073] S1. Map the physical structure of the distribution network into a topology graph G = (V, E), where the structure of the power grid topology graph G is: each device node in the distribution network is represented as a node set V, each node v∈V contains its device attributes, each connection line is represented as an edge set E, each edge e∈E is the influence of the line, where the attribute of each node is represented by a vector h v Indicates that the attributes of the edge are represented by the vector h e Represented, and the equipment and line characteristics are represented as node feature matrices H v and edge feature matrix H e , and construct external features X ext , the external feature X ext Respectively with the node feature matrix H v and edge feature matrix H e Combine the corresponding nodes to form the final feature F v and the final feature F of the edge e .

[0074] S2. Construct a graph neural network based on the distribution network to predict the cost of the distribution network according to the topology graph G = (V, E).

[0075] S3. Use causal inference technology to identify the key external factors of distribution network project cost based on the results of distribution network cost forecast.

[0076] S4. Based on the cost prediction results of the distribution network-based graph neural network and the key external factors identified using causal inference technology, a cost optimization model is built to optimize the predicted project cost.

[0077] S5. Update the graph structure G and external influencing factor features X in real time based on the actual data of the construction site ext , adjust model prediction values ​​and optimization strategies.

[0078] The S1 specifically includes:

[0079] Construct a graph model from two aspects: distribution network topology and cost-related features:

[0080] Assuming that each device node in the distribution network (such as substations, transmission lines, etc.) is represented as a node set V, and each connection line is represented as an edge set E, the topological structure of the distribution network can be represented by a graph G = (V, E). Each node v∈V contains its device attributes, such as device type, installation cost, maintenance cost, etc., and each edge e∈E contains information such as line construction cost and line length. The attributes of each node are represented by a vector h v Indicates that the attributes of the edge are represented by the vector h e express.

[0081] The characteristics of each device node h v It is necessary to consider factors such as the type, lifespan, installation and maintenance costs of the equipment. The specific calculation is:

[0082]

[0083] Among them, T v Indicates the type of equipment, such as substation, transmission line, etc. Represents the installation cost of the equipment. Indicates the maintenance cost of the equipment. v Indicates the life of the equipment, indicating the expected service life of the equipment.

[0084] Furthermore, based on each edge e∈E, i.e., the line connecting different devices in the distribution network, considering key factors such as its construction cost and line length, the specific construction is as follows:

[0085]

[0086] in, Indicates the construction cost of the line and the cost of laying the line. e Indicates the length of the line, which is used to measure the spatial distance of the line.

[0087] Through this step, the present invention maps the physical structure of the distribution network into a graph G = (V, E), and represents the device and line characteristics as node feature matrices H v and edge feature matrix H e .

[0088] Furthermore, in order to enable the model to capture the cost fluctuations in the distribution network project affected by the external environment (such as material prices, construction weather, etc.), the present invention additionally introduces a set of external influencing factor features X ext , which represents the time series data or external market factors related to the project. The present invention considers the following key external factors:

[0089] X ext =[P material ,W weather ,D duration ]

[0090] Among them, P material Indicates the current market price of materials. weather Indicates the weather index during the construction period, which affects the construction progress. duration It indicates the estimated duration of construction and represents the overall time span of the project.

[0091] Furthermore, the present invention combines the external factor X ext With node feature H v and edge feature H e Combined to form a complete input feature set F. The final feature representation of the node is:

[0092] F v =[h v ,X ext ]

[0093] The final feature representation of the edge is:

[0094] F e =[h e ,X ext ]

[0095] In this way, each device node and line edge combines its own attributes and external environmental factors.

[0096] Furthermore, the final graph G = (V, E) can be defined by two parts: one is its topological structure (defined by nodes V and edges E); the other is the features F attached to each node and edge v and F e . Where V represents the equipment and load nodes. E represents the connection relationship between these nodes. v and F e The feature matrix representing each node and edge consists of device attributes and external factors, respectively.

[0097] Through this complete graph modeling process, we obtain the graph structure G and feature matrix F v and F e , which will become the input for subsequent graph neural network training.

[0098] The S2 specifically includes:

[0099] The topological structure of the distribution network has been modeled as a graph G = (V, E), with the characteristics of nodes and edges being F v and F e These input features include equipment attributes (such as equipment type, installation and maintenance costs) and external factors (such as material prices, weather, construction period, etc.). Specifically, the content input to the graph neural network (GNN) is:

[0100] Graph structure: G = (V, E), V represents the set of device nodes, and E represents the set of lines connecting devices.

[0101] Node feature matrix: F v =[h v ,X ext ], where h v Represents the physical characteristics of the device, X ext Represents external environmental factors.

[0102] Edge feature matrix: F e =[h e ,X ext ], where h e Indicates the attribute characteristics of the line, X ext Indicates external factors.

[0103] Furthermore, in order to adapt to the complexity of distribution network engineering, the present invention designs a multi-layer graph neural network (GNN), whose propagation mechanism can effectively capture the complex relationship between device nodes connected by lines. In each layer, the characteristics of node v are compared with those of its neighboring nodes. The information propagation between them is updated while considering the influence of lines (edges). The feature update of node v in layer l is:

[0104]

[0105] in, W represents the feature vector of node v at layer l. (l) Represents the weight matrix of the lth layer in the graph neural network, learning the linear transformation of node features. Represents the characteristics of the edge e(v,u), connecting the node v and its neighbor node u. Represents the weight matrix used for edge features in layer l. v It represents the bias term of the node, taking into account the impact of the characteristics of the node itself on the cost. σ represents the nonlinear activation function (usually ReLU is used).

[0106] In this mechanism, the edge features The interactive design with node characteristics is one of the core innovations. This is because in distribution network projects, the installation and maintenance costs of equipment are complexly related to the construction costs of connecting lines, line lengths, etc. Through the design of the propagation layer, the model can capture this specific influence relationship.

[0107] Furthermore, in order to make the model pay more attention to some key cost influencing factors in the project during the optimization process (such as the risk of over-budget and the high sensitivity of external factors to cost), the present invention introduces a customized regularization term to constrain the cost prediction. The regularization term is based on the following observations:

[0108] In distribution network projects, external factors such as material price fluctuations and construction weather often significantly affect the overall cost. In order to make the model better adapt to external environmental fluctuations, this paper designs a regularization term Encourage the model to learn the sensitivity of external factors to cost prediction:

[0109]

[0110] in, Represents the total project cost predicted by the model. ext,v Represents the external factors of node v. λ represents the weight of the regularization term, which controls the degree of regularization.

[0111] This regularization term limits the impact of changes in external factors on cost prediction, ensuring that the model can maintain a certain degree of prediction stability under different external environments, while capturing the differences in the impact of the external environment on the costs of different nodes.

[0112] Furthermore, after multi-layer GNN propagation update, the final feature representation of all nodes is obtained This paper designs an innovative weighted aggregation mechanism to generate an overall cost forecast This mechanism takes into account the different contributions of different equipment nodes to the cost. Weighted global cost prediction formula:

[0113]

[0114] Among them, α v Represents the weight of node v, which is determined by its device type and importance. It represents the final feature representation of node v, reflecting the contribution of the node to the overall cost.

[0115] In practical applications, α v It can be dynamically calculated based on the importance of the equipment or its connection with neighboring nodes. For example, the cost contribution of a substation to the entire project may be greater, while the contribution of a line node may be relatively smaller.

[0116] Furthermore, in order to effectively optimize the entire cost prediction, the present invention designs a loss function that combines the mean square error (MSE) and the aforementioned regularization term The goal is to minimize the prediction error of the model and control the impact of external factors on the prediction results. Loss function formula:

[0117]

[0118] in, represents the predicted cost of the i-th sample. actual,i Represents the actual cost of the i-th sample. N represents the number of samples. By optimizing the loss function, the model can gradually adjust its parameters so that the predicted cost is closer to the actual cost and maintains good robustness under different external conditions.

[0119] It can be understood that during the model training process, the present invention uses the gradient descent algorithm to optimize the model parameters, and continuously adjusts the parameters of the GNN (including W (l) and ). The training data includes the actual historical distribution network project cost data, as well as the related topological structure, node characteristics and external influencing factors. In each iteration, the model is trained through the loss function Calculate the error and update the parameters until the model converges and achieves good prediction accuracy.

[0120] In this step, the present invention designs a customized graph neural network (GNN) structure for the prediction of the construction cost of distribution network projects. The model generates an overall cost prediction value by combining the characteristics of device nodes and connection lines through a multi-layer message passing mechanism. At the same time, by introducing the external factor sensitivity regularization term The present invention ensures that the model can better adapt to changes in the external environment and optimize the accuracy of cost prediction. This step provides the core cost prediction capability for the entire patent and is a key part of achieving the patent goal.

[0121] The S3 specifically includes:

[0122] First, build a causal graph (CausalGraph), whose nodes include: (cost predicted by GNN), X ext (external factors, such as material prices, weather, construction period, etc.), F v and F e (Represents equipment and line characteristics, including equipment installation, maintenance costs, line length, etc.) The causal graph structure is used to describe the direct causal dependencies between these variables and to quantify the impact of each factor on the predicted cost. specific impact.

[0123] Furthermore, the present invention uses a structural equation model (SEM) to quantify the causal relationship between external factors and predicted construction costs. The SEM modeling method is expressed as follows:

[0124]

[0125] Among them, P material Indicates the material price, representing the impact of material cost on construction cost. weather Indicates weather conditions during construction, which affect construction progress and cost. duration Indicates the construction time and the cost caused by the construction period. v and F e Represents the characteristics of nodes and edges, reflecting the cost characteristics of equipment and lines respectively. β1, β2, β3 represent the influence coefficients of external factors. v ,β e Represents the weight of the impact of equipment and lines on the overall cost. ∈ represents the error term caused by unobserved factors. This formula is used to describe how external factors and project characteristics jointly affect the predicted cost. By estimating β1, β2, and β3, the present invention can clarify the contribution of external factors to the cost, especially the sensitivity of external factors to budget fluctuations.

[0126] Furthermore, since the cost of distribution network projects is easily affected by the external environment (for example, fluctuations in material prices and changes in weather), the present invention adds an innovative regularization term to the model to control the excessive impact of these key external factors on cost prediction. The present invention introduces an external factor sensitivity regularization term, which is expressed as Its purpose is to reduce the error fluctuation caused by over-reliance on certain external factors. The regularization term is designed as follows:

[0127]

[0128] Among them, λ and μ represent the sensitivity control parameters to material price fluctuations and weather factors, respectively, and are usually adjusted according to the specific needs of the actual project. Indicates the impact of material price fluctuations on construction costs sensitivity. Indicates the impact of changing weather conditions on construction costs The introduction of this regularization term helps to make the model more stable in the face of changes in external factors, ensuring that there will be no significant deviations in the cost forecast due to slight fluctuations in the external environment.

[0129] Furthermore, in order to further optimize cost control and perform simulation prediction, the present invention introduces causal intervention items to simulate the adjustment strategy of key external factors. For example, if the material price P is identified materialThe contribution to the overall cost is too large, and the present invention can intervene by adjusting the material procurement strategy. Intervention formula:

[0130]

[0131] δ represents the intervention intensity, which means reducing the impact of material price fluctuations on construction costs by adjusting material procurement strategies.

[0132] This intervention mechanism can effectively simulate the adjustment results of different management decisions on construction costs, so that the model can evaluate the effects of different intervention measures in advance in cost optimization. Through causal intervention, project managers can actively control the risks brought by key factors, thereby better controlling project costs.

[0133] It can be understood that through quantitative analysis of causal relationships and control of sensitivity, the model can clearly identify the factors that have the greatest impact on cost fluctuations. The present invention determines the external factors that have the most significant impact on cost based on the size of each causal coefficient (such as β1, β2, β3) in the model, and focuses on optimizing these key factors in the cost optimization stage.

[0134] At the same time, combined with the regularization term in step 2 This can further ensure that the model maintains good robustness in different external environments in actual projects. This analysis result provides a clear guiding direction for subsequent cost optimization and helps managers effectively control project costs.

[0135] This step shows in detail how to identify and quantify the key factors affecting the cost of distribution network projects through causal inference. The impact of external factors on the predicted cost is quantified through the structural equation model (SEM), and combined with innovative sensitivity regularization terms and causal intervention terms, the model is more stable and robust when dealing with fluctuations in the external environment. Ultimately, the causal analysis results provide clear guidance for project cost optimization, thereby ensuring that the project proceeds smoothly within the budget and can effectively respond to changes in the external environment. This step is the key to the entire patent process. Through causal analysis, we can deeply explore the external factors that affect the cost and serve the subsequent optimization decisions.

[0136] The S4 specifically includes:

[0137] Constructing an objective function to minimize the cost And by adjusting the external factor X ext To achieve optimization, the design of the objective function includes cost prediction and adjustment and control of external factors.

[0138]

[0139] in, Represents the predicted overall cost, output by the GNN model.i Represents the intervention coefficient used to adjust external factors. i X ext,i Represents the influence weight of each external factor (such as material price, weather, construction period). The regularization term of the second term Used to control the sensitivity of external factors to improve the robustness of the model.

[0140] Furthermore, in order to improve the model's ability to cope with external environmental fluctuations, a regularization term is added It is used to limit the impact of drastic fluctuations in external factors on construction costs. The regularization term is based on historical data to ensure that external factors remain within a reasonable range.

[0141]

[0142] Where μ represents the regularization strength parameter. Represents the historical average of external factors. Represents the historical standard deviation of external factors and the normal fluctuation range. This regularization term controls the model to not rely on extreme changes in external factors and improves the prediction stability.

[0143] Furthermore, based on the causal inference results, adjustment strategies are designed for key external factors. For example, the fluctuation of material prices can reduce its impact on the cost by adjusting the intervention coefficient δ1. The optimized cost prediction model can be expressed as:

[0144]

[0145] Through management strategies, such as early procurement, δ1 can reduce the impact of material cost fluctuations by locking in prices. For construction period adjustment, by speeding up the progress to reduce the impact of extended construction period, construction period optimization can be achieved by regulating δ3.

[0146] Further, optimize the process:

[0147] Input initial data: The external factors and their influence weights β obtained in step 3 i and the initial cost predicted by GNN Enter the optimization model.

[0148] Calculate the objective function: Use the cost optimization objective function Combined with regularization Calculate the impact of external factors on the total cost.

[0149] Apply intervention strategy: adjust intervention coefficient δ according to different external factors i , such as reducing the impact of material fluctuations on construction costs by locking in prices, or reducing the increase in costs caused by construction period by accelerating construction.

[0150] Iterative optimization: Use optimization algorithms such as gradient descent to perform multiple rounds of iterations on the cost, gradually reducing the overall cost until it converges to the optimal solution.

[0151] Output optimization results: Output the optimized total cost And provide corresponding management strategies, such as suggesting to lock in price or adjust construction plan.

[0152] This optimization model combines the causal analysis and cost forecasting in the previous steps, and proposes a customized optimization regularization term and intervention strategy design, especially for the uncertainty of external factors in distribution network projects. Through sensitivity control and regularization, the model ensures that external factors will not have too much impact on the forecast. At the same time, managers can use the intervention coefficient δ i Flexibly adjust the optimization strategy to ensure that the cost is effectively controlled in practical applications.

[0153] The S5 specifically includes:

[0154] The core goal of the dynamic feedback mechanism is to update the previous optimization model and forecast the cost through real-time data collection and feedback. The present invention gradually integrates real-time data into the optimization process by designing a dynamic feedback equation. The feedback mechanism achieves the adjustment of prediction and actual progress by introducing an update equation:

[0155]

[0156] in, Indicates the updated cost forecast result. Represents the initial cost forecast result obtained through optimization. real and X pred represents the external factor data collected in real time and the external factor data during prediction. real and P pred Denotes the construction progress and actual cost feedback data and the data in the forecast respectively. α and β denote feedback control coefficients, which are used to adjust the weight of different data on the final result.

[0157] The equation dynamically adjusts cost forecasts to keep them consistent with actual changes during construction and ensures that the system can respond promptly to external factors and fluctuations in construction progress.

[0158] Furthermore, in the dynamic feedback mechanism, the fusion of real-time data is the key. To this end, a data fusion algorithm is designed to judge the real-time impact of external factors and construction progress on costs by comparing actual data with predicted data. External Factors X real and P real Collected from construction sites and markets respectively, they represent the actual situation in the project.

[0159] The steps of data fusion are as follows:

[0160] Collect real-time data: Update external factors such as material prices, weather conditions, project progress, etc. in real time through sensors, manual records, or market data real and construction cost feedback P real .

[0161] Calculate the difference: Calculate the difference between the actual data and the model's predicted data. External factor difference:

[0162] ΔX=X real -X pred

[0163] Construction cost differences:

[0164] ΔP=P real -P pred

[0165] Update model: Integrate the differences ΔX and ΔP into the feedback loop of the original model to adjust the predicted cost The prediction model is then updated. Through continuous feedback and iteration of real-time data, the prediction results are ensured to be synchronized with the actual engineering dynamics.

[0166] Furthermore, based on the real-time feedback adjustment, the entire optimization model needs to be dynamically iterated to adapt to the new external environment and construction progress. Therefore, after each feedback update, the system will recalculate the optimization objective function:

[0167]

[0168] in, Indicates the cost result after adjustment through the dynamic feedback mechanism. real,i Represents the actual external factor data, replacing the external factors in the forecast. The design of the regularization term remains unchanged, and the impact of external factors on cost fluctuations continues to be controlled. The optimization function after dynamic update It can be updated in real time as external factors and construction progress change, and ensure the accuracy of cost control.

[0169] Furthermore, the dynamic feedback mechanism requires multiple iterations until the prediction model and the actual construction process reach a certain stable state. In order to determine the stopping condition of the feedback mechanism, the present invention introduces a convergence condition to determine whether the real-time feedback is close enough to the actual data. The convergence condition can be defined as:

[0170]

[0171] Among them, ∈ represents the error convergence threshold. Only when the cost forecast after feedback adjustment Prediction before optimization The system considers the feedback converged only when the difference between them is small enough.

[0172] It is understandable that the unique role of this dynamic feedback mechanism in patents is reflected in the following points:

[0173] Combination of sensitivity and real-time performance: Through real-time monitoring of the external environment and construction data, the feedback mechanism can promptly reflect various problems arising in the project and quickly adjust the cost forecast.

[0174] Dynamic adjustment and iterative optimization: After each feedback, the model is optimized through multiple iterations to make the prediction results gradually converge to the actual cost, ensuring that the model is synchronized with the actual project at any time.

[0175] Adaptive mechanism: Through the feedback mechanism, the model can adapt to changes in different construction stages and dynamically respond to external factors such as material prices and construction progress to ensure that cost control is more robust.

[0176] This dynamic feedback and real-time update mechanism ensures that the entire system can adjust the optimization strategy in a timely manner according to the progress of the project, thereby maintaining the accuracy and reliability of the cost forecast.

[0177] Finally, the system will output the adjusted cost forecast results after each feedback iteration. It also generates detailed reports, including the impact of changes in external factors on costs, deviations of construction progress from budget, etc., to provide managers with a basis for real-time decision-making.

[0178] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present invention. In practical applications, technicians in this field can select part or all of them according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0179] In addition, for technical details that are not described in detail in this embodiment, reference can be made to the parameter operation method provided in any embodiment of the present invention, and will not be repeated here.

[0180] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0181] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0182] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory / random access memory, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0183] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for constructing a distribution network engineering cost data analysis model, characterized in that: The method comprises: S1. Map the physical structure of the distribution network into a topology graph G = (V, E), where the structure of the power grid topology graph G is: each device node in the distribution network is represented as a node set V, each node v∈V contains its device attributes, each connection line is represented as an edge set E, each edge e∈E is the influence of the line, where the attribute of each node is represented by a vector h v Indicates that the attributes of the edge are represented by the vector h e Represented, and the equipment and line characteristics are represented as node feature matrices H v and edge feature matrix H e , and construct external features X ext , the external feature X ext Respectively with the node feature matrix H v and edge feature matrix H e Combine the corresponding nodes to form the final feature F v and the final feature F of the edge e ; S2. Construct a graph neural network based on the distribution network to predict the distribution network cost according to the topological graph G=(V, E); wherein the graph neural network based on the distribution network is a multi-layer graph neural network, in which the characteristics of the nodes in each layer are updated by information propagation between their neighboring nodes, while considering the influence of the line, and the update rule of the multi-layer graph neural network is expressed as follows: in, W represents the feature vector of node v at layer l; (l) Represents the weight matrix of the lth layer in the graph neural network, learning the linear transformation of node features; Represents the characteristics of edge e(v,u), connecting node v and its neighbor node u; represents the weight matrix of the lth layer for edge features; b v represents the bias term of the node, taking into account the impact of the characteristics of the node itself on the cost; σ represents the nonlinear activation function; S3. Use causal inference techniques to identify key external factors of distribution network project cost based on the results of distribution network cost forecast; S4. Build a cost optimization model based on the cost prediction results of the distribution network-based graph neural network and the key external factors identified using causal inference technology to optimize the predicted project cost; S5. Update the graph structure G and external influencing factor features X in real time based on the actual data of the construction site ext , adjust model prediction values ​​and optimization strategies.

2. The method for constructing a distribution network engineering cost data analysis model according to claim 1, characterized in that: The attributes of each node are represented by a vector h v Including the type, life, installation and maintenance cost of the equipment; the attributes of the edge are represented by the vector h e Including construction cost, line length; the external features X ext Including material prices, construction weather and estimated construction duration.

3. The method for constructing a distribution network engineering cost data analysis model according to claim 1, characterized in that: The S2 also includes adding a regularization term in the optimization process of the graph neural network based on the distribution network Encourage the model to learn the sensitivity of external factors to cost prediction. The regularization term It is expressed as follows: in, represents the total construction cost predicted by the graph neural network based on the distribution network; X ext,v represents the external factor of node v; λ represents the weight of the regularization term, which controls the degree of regularization.

4. The method for constructing a distribution network engineering cost data analysis model according to claim 3, characterized in that: When predicting the cost of the distribution network according to the topology graph G = (V, E), a weighted aggregation mechanism is designed according to the different contributions of different device nodes to the cost to generate an overall cost prediction. It is expressed as follows: Among them, α v represents the weight of node v, which is determined by its device type and importance; The final feature representation of node v reflects the contribution of the node to the overall cost; The loss function of the graph neural network based on the distribution network is expressed as follows: in, represents the predicted cost of the i-th sample; C actual,i represents the actual cost of the i-th sample; N represents the number of samples.

5. The method for constructing a distribution network engineering cost data analysis model according to claim 1, characterized in that: The use of causal inference technology to identify key external factors of the distribution network project cost based on the results of the distribution network cost forecast specifically includes: The causal relationship between external factors and predicted construction cost is quantified by the structural equation model, and then a causal diagram is constructed; wherein the structural equation model is expressed as follows: Among them, P material Indicates the material price, representing the impact of material cost on construction cost; W weather Indicates weather conditions during construction, which affect construction progress and cost; D duration Indicates the construction time and the cost caused by the construction period; F v and F e represents the characteristics of nodes and edges, reflecting the cost characteristics of equipment and lines respectively; β1, β2, β3 represent the influence coefficients of external factors; β v ,β e represents the weight of the impact of equipment and lines on the overall cost; ∈ represents the error term caused by unobserved factors; Set the external factor sensitivity regularization term in the causal graph Reduce the error fluctuation caused by over-reliance on certain external factors, the sensitivity regularization term of the external factors It is expressed as follows: Among them, λ and μ represent the sensitivity control parameters to material price fluctuations and weather factors, respectively, and are usually adjusted according to the specific needs of the actual project; Indicates the impact of material price fluctuations on construction costs sensitivity; Indicates the impact of changing weather conditions on construction costs sensitivity.

6. The method for constructing a distribution network engineering cost data analysis model according to claim 5, characterized in that: The nodes of the causal graph include the total project cost predicted by the graph neural network based on the distribution network External influencing factors characteristics X ext and the final feature F of the node v , the final feature F of the edge e ; At the same time, causal intervention items are added to the causal diagram to simulate the adjustment strategy of key external factors. If the material price P is identified material The contribution to the overall cost is the largest, and intervention is carried out by adjusting the material procurement strategy, as shown below: Among them, δ represents the intervention intensity, which means reducing the impact of material price fluctuations on construction costs by adjusting material procurement strategies.

7. The method for constructing a distribution network engineering cost data analysis model according to claim 5, characterized in that: The cost optimization model is constructed as follows: Construct cost optimization objective function to minimize predicted cost And by adjusting the external factor X ext The optimization is implemented as follows: Among them, δ i represents the intervention coefficient used to adjust external factors; β i X ext,i Represents the influence weight of each external factor; regularization term Used to control the sensitivity of external factors to improve the robustness of the model; Adding regularization terms to the cost optimization model Limiting the impact of drastic fluctuations in external factors on construction costs is expressed as follows: Where μ represents the regularization strength parameter; Represents the historical average of external factors; Represents the historical standard deviation of external factors and indicates the normal fluctuation range.

8. The method for constructing a distribution network engineering cost data analysis model according to claim 1, characterized in that: The S5 specifically includes: Design dynamic feedback equations to gradually integrate real-time data into the optimization process; Design data fusion algorithms to determine the real-time impact of external factors and construction progress on costs by comparing actual data with predicted data; Use sensors to collect real-time feedback, adjust the cost optimization model based on the real-time feedback, and dynamically iterate to adapt to the new external environment and construction progress. At the same time, recalculate the optimization objective function after each feedback update.

9. The method for constructing a distribution network engineering cost data analysis model according to claim 8, characterized in that: The dynamic feedback equation is expressed as follows: in, Indicates the updated cost forecast result; represents the initial cost prediction result obtained through optimization; X real and X pred They are external factor data collected in real time and external factor data during prediction; real and P pred They are the feedback data of construction progress and actual cost and the data in the forecast respectively; α and β represent the feedback control coefficients, which are used to adjust the weight of different data on the final result; The design data fusion algorithm determines the real-time impact of external factors and construction progress on costs by comparing actual data with predicted data. Specifically: Real-time update of external factor data through sensors, manual records or market data X real and construction cost feedback P real ; Calculate the difference between actual data and model prediction data externally: including external factor differences and construction cost differences. The external factor difference ΔX is expressed as follows: ΔX=X real -X pred Among them, X real represents the actual external factors, X pred It means predicting external factors; The construction cost difference ΔP is expressed as follows: ΔP=P real -P pred Among them, P real represents the actual construction cost, P pred Represents the predicted construction cost Integrate the differences ΔX and ΔP into the feedback loop of the graph neural network based on the distribution network to adjust the predicted cost Thereby updating the graph neural network based on distribution network; The sensors are used to collect real-time feedback, and the cost optimization model is dynamically iterated based on the adjustment of the real-time feedback to adapt to the new external environment and construction progress. At the same time, after each feedback update, the optimization objective function is recalculated, which is expressed as follows: Among them, X real,i Indicates the current actual external factor data.

10. The method for constructing a distribution network engineering cost data analysis model according to claim 8, characterized in that: Introducing a convergence condition into the dynamic feedback equation to determine whether the real-time feedback is close enough to the actual data is as follows: Where ∈ represents the error convergence threshold.

Images