Auxiliary construction project decision-making method based on post-evaluation-pre-planning feed-forward deduction

By converting post-evaluation data into graph structures and using the transfer closure and shortest path algorithm, combining Bayesian network for construction project decision reasoning, the problem of post-evaluation data in architectural design is solved, and intelligent decision-making assistance and The processing of data scarcity improves the scientificity and efficiency of the design.

CN118941148APending Publication Date: 2024-11-12SICHUAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411007148.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The lack of effective post-evaluation data utilization in architectural design leads to experience reliance on early planning and prone to repeated misdesigns. In addition, deep learning technology is difficult to apply due to scarce data, so it is impossible to effectively perform feed-forward reasoning.

Method used

The post-evaluation data is converted into a graph structure, and directed acyclic graph is constructed using the passing closure and shortest path algorithm, and decision-making reasoning is performed through Bayesian networks to avoid deep learning's dependence on the data set.

Benefits of technology

It realizes intelligent decision-making assistance based on post-evaluation data, avoids the cumbersomeness and errors of manual reasoning, reduces the cost of data collection, and improves the scientificity and efficiency of construction projects.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941148A_ABST
    Figure CN118941148A_ABST
Patent Text Reader

Abstract

The invention provides an auxiliary building project decision-making method based on post-evaluation-pre-planning feed-forward deduction, and belongs to the technical field of building projects, and the method comprises the steps: obtaining building project table data of a post-evaluation-pre-planning index system, and constructing an initial directed acyclic network diagram; finding out a mapping relationship between all building pre-planning and post-evaluation indexes in the initial directed acyclic network diagram by using a transitive closure; obtaining initial directed acyclic weight data; and according to the constructed directed acyclic network graph, performing path deduction based on the Bayesian network to obtain an optimal decision. Decision suggestions are provided for pre-planning of buildings of the same type by using post-evaluation data, an architect is assisted to make an intelligent decision, and waste of social resources caused by repeated mistakes of the same problem is avoided; meanwhile, from the perspective of a computer, a new research thought is provided for improvement and path deduction of an incomplete graph network structure under small sample data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of construction projects, and in particular relates to an auxiliary construction project decision-making method based on post-evaluation-forward planning and feedforward deduction. Background Art

[0002] Architects often rely on design briefs for architectural design. However, there are unscientific design briefs in the industry. Some owners determine the positioning and needs of the project based on subjective will, experience and habits. If architects only design according to such positioning, there will be risks in the long run. In recent years, people have gradually realized the importance of post-evaluation theory, attached great importance to the early planning and review of buildings, ensured the correctness of the construction direction, verified the concept and found problems through the accumulation of post-evaluation data, and worked hard for carbon emission reduction from the root. The significance of using post-evaluation is not only to provide feedback and corrections to the problems of the evaluation body building, but more importantly, it provides analysis basis and reference for the planning and design of similar buildings. The development of pre-building planning and post-evaluation system is still in the theoretical stage. At present, there is no relevant mature technology to assist architects in using post-evaluation data for early building planning and design.

[0003] At present, the early planning and design of buildings rely more on the experience accumulation of architects, without considering the post-evaluation data, which has the following shortcomings:

[0004] 1. The previous buildings may have had unreasonable designs, but there was no post-evaluation system for feedback and review, so architects were unaware of problems and the experience they had learned was wrong. Relying solely on experience for design will lead to repeated mistakes and a huge waste of social resources.

[0005] 2. With the development of the post-evaluation system, more and more architects have noticed the importance of post-evaluation data and want to use the post-evaluation data of the same type of buildings to assist in the preliminary planning of new buildings. Knowing that there is a problem with a certain post-evaluation indicator and wanting to improve it, but due to the complexity of the indicators, it is difficult to manually find all the pre-planning indicators that affect the post-evaluation indicator, and it is impossible to quantify the weight of the pre-planning indicator on the post-evaluation indicator, thus failing to assist architects in the preliminary planning of construction projects.

[0006] From a computer perspective, implementing feedforward reasoning from post-evaluation indicators to forward planning indicators is essentially a graph reasoning problem. The indicators of the forward planning-backward evaluation system are structured data based on tables, but it is difficult for tables to intuitively and efficiently represent the complex relationship between indicators, and it is even more difficult to reason. Therefore, the table data is converted into graph structured data, with the forward planning and post-evaluation indicators as nodes of the graph, and the mapping relationship between the indicators as directed edges. The feedforward reasoning from the back-evaluation to the forward planning is abstracted into a graph reasoning problem. Deep learning technology has made rapid progress in the field of graph reasoning, but it has the following disadvantages:

[0007] 1. Deep learning relies heavily on a complete data set. However, due to the high cost of data collection in the construction field, privacy and security issues, and low willingness to share data within the industry, data sets related to the construction field are scarce and data collection is difficult. Graph reasoning technology based on deep learning is difficult to apply in this scenario.

[0008] 2. Deep learning uses data from existing buildings, but some existing buildings have unreasonable designs. The results obtained from learning from incorrect data are also wrong and no improvements are made. Summary of the invention

[0009] In view of the above-mentioned deficiencies in the prior art, the present invention provides an auxiliary construction project decision-making method based on post-evaluation-forward planning deduction. The present invention transforms the theoretical system of fore-planning and post-evaluation in the field of construction from a theoretical level into technical practice, and uses post-evaluation data to provide decision-making suggestions for the fore-planning of buildings of the same type, assisting architects in making intelligent decisions, avoiding repeated mistakes in the same problem and thus wasting social resources, and solving the problem that manual fore-planning decision-making is difficult and ineffective; at the same time, from a computer perspective, it provides a new research idea for the improvement of incomplete graph network structures and path deduction under small sample data, and uses traditional graph theory algorithms to avoid the requirements of deep learning algorithms for data sets.

[0010] In order to achieve the above purpose, the technical solution adopted by the present invention is: an auxiliary construction project decision-making method based on post-evaluation-front planning and feedforward deduction, comprising the following steps:

[0011] S1. Obtain the tabular data of the construction project of the post-evaluation-pre-planning indicator system and construct an initial directed acyclic network graph;

[0012] S2. Using transitive closure, find the mapping relationship between all pre-construction planning and post-evaluation indicators in the initial directed acyclic network graph to improve the adjacency matrix;

[0013] S3, according to the processing result of step S2, using the weight learning algorithm based on the shortest path, obtaining the initial directed acyclic weight data to complete the construction of all data of the directed acyclic graph;

[0014] S4. According to the constructed directed acyclic network diagram, the optimal decision is obtained based on the path deduction of the Bayesian network to complete the decision on the construction project.

[0015] The beneficial effects of the present invention are as follows: in the stage of building planning, the post-evaluation data of existing similar buildings are considered. The present invention utilizes the post-evaluation-forward planning feedforward deduction technology to assist architects in scientific design and decision-making, which can avoid repeated mistakes in the same design problem and is beneficial to the development of urban construction. The present invention utilizes computer software to assist architects in feedforward reasoning from post-evaluation to forward planning, thus avoiding the problems of manual mapping reasoning, cumbersome work and poor results, and realizes the leap from the theoretical stage to the technical implementation of the building pre-planning and post-evaluation system. At the same time, the use of traditional graph theory technology overcomes the problem of data scarcity in the field of architecture, avoids the problem of deep learning technology's dependence on data sets, does not require the collection of data sets, and greatly reduces labor costs and the cost of consumables for collecting data.

[0016] Furthermore, the S1 is specifically: obtaining the construction project table data of the post-evaluation-pre-planning indicator system, and constructing a directed acyclic network diagram using the following formula:

[0017] G=(V,A,W)

[0018] Among them, G represents a directed acyclic graph, V represents the node set in the initial directed acyclic network graph, A represents the adjacency matrix of the initial directed acyclic network graph, and W represents the edge weight matrix.

[0019] The beneficial effect of the above further scheme is that the initial data of the post-evaluation-fore-planning indicator system is structured data based on tables, but it is difficult for tables to intuitively and efficiently represent the complex relationships between nodes, and it is even more difficult to make reasoning decisions. Therefore, the tabular data is converted into graph structured data, and the fore-planning and post-evaluation indicators are used as nodes of the graph, and the mapping relationship between the indicators is used as directed edges. The post-evaluation-fore-planning feedforward deduction is abstracted into a graph-based reasoning problem.

[0020] Furthermore, the S2 comprises the following steps:

[0021] S201, initializing the transitive closure matrix as an adjacency matrix;

[0022] S202, calculating the square of the transitive closure matrix;

[0023] S203, updating the transitive closure matrix to the operation result of the non-zero elements in the square calculation result and the elements in the adjacency matrix;

[0024] S204, determining whether the value of the transitive closure matrix no longer changes, if so, the finally obtained transitive closure matrix is ​​the transitive closure of the initial directed acyclic network graph, and proceeding to S205, otherwise, returning to step S201;

[0025] S205, assigning the transitive closure matrix to the adjacency matrix, obtaining the mapping relationship between all pre-construction planning and post-evaluation indicators in the initial directed acyclic network graph, so as to improve the adjacency matrix.

[0026] The beneficial effect of the above further scheme is that the mapping relationship between pre-construction planning and post-evaluation indicators mainly depends on literature and expert evaluation, and the relationship mapping is annotated manually. It is difficult for manual annotation to fully capture all mapping relationships between indicators. Step S2 uses transitive closure to find all direct or indirect mapping relationships in the graph, improves the adjacency matrix of the directed acyclic graph, and lays the foundation for the next step of using the shortest path algorithm to learn edge weights.

[0027] Furthermore, the S3 comprises the following steps:

[0028] S301, assuming that the directed acyclic graph is a complete graph, assign the adjacency matrix to the weight matrix W, and randomly initialize the weight matrix W' to a two-dimensional matrix in the interval [1,9], where there exists a directed acyclic graph G, whose weight matrix is ​​W, and there exists a directed acyclic graph G', whose weight matrix is ​​W';

[0029] S302, bringing the weight matrix W into the Floyd-Warshall algorithm for calculation to obtain the shortest path P between each node pair;

[0030] S303, bringing the weight matrix W' into the Floyd-Warshall algorithm to calculate and obtain the path P', and comparing the shortest paths between all node pairs in the path P' with the actual shortest paths in the shortest path P;

[0031] S304, according to the comparison result, cyclically adjust the size of the weight matrix W', and according to the adjustment result, determine whether the comparison result of the path P' is consistent with the shortest path P, if so, proceed to step S305, otherwise, return to step S303;

[0032] S305 , taking the inverse of the weight matrix W′ and assigning it to the weight matrix W, so as to obtain the weight data of the initial network graph, and complete the construction of all the data of the directed acyclic graph.

[0033] The beneficial effect of the above further scheme is that the mapping relationship between indicators in the pre-planning and post-evaluation system can only be manually marked to show whether there is a relationship, that is, whether there is a directed edge between two nodes, but it is difficult to quantify the strength of their relationship, that is, the directed edge has no weight value. Through the above scheme, using the weight learning algorithm based on the shortest path proposed in the present invention, the weight learning of the directed edge can be performed by learning the graph structure, thereby obtaining a reasonable weight matrix, which provides conditions for the next step of reasoning using the Bayesian network.

[0034] Furthermore, the S4 comprises the following steps:

[0035] S401, normalizing the weight data of the constructed directed acyclic network graph;

[0036] S402, based on the normalization processing result, taking any post-evaluation node as the starting point in the Bayesian network, and deducing the pre-planning nodes and paths related to the starting point along the Bayesian network path;

[0037] S403, performing probability calculation on the pre-planning nodes and paths, and obtaining the optimal decision in the pre-construction planning stage according to the probability calculation results, thereby completing the decision on the construction project.

[0038] The beneficial effect of the above further scheme is that through this method, in the Bayesian network, taking any post-evaluation node as the starting point, the related pre-planning nodes and paths can be deduced along the network path, and the probability of these pre-planning nodes and paths can be calculated, so as to assist decision makers in making the best decision in the early planning stage of the building and provide important guidance.

[0039] Furthermore, the weight data of the initial network graph is normalized, which is specifically:

[0040] Normalize the weight data of the initial network graph to obtain the matrix N, and assign the matrix N to the weight matrix W:

[0041]

[0042] Among them, N[i][j] means that the matrix N is the i-th row and j-th column, n represents the size of the weight matrix W, W[i][j] represents the element of the i-th row and j-th column in the weight matrix, and 0≤i<n, 0≤j<n.

[0043] The beneficial effect of the above further scheme is that the Bayesian network requires a conditional probability table, so the weight matrix needs to be row-normalized to ensure that the sum of the probabilities starting from each node is 1, that is, to form a probability distribution. Traverse each row of the weight matrix, calculate the sum of all elements in the row, record it as the norm of the row, divide each element of the row by the norm of the row to achieve row normalization, and adjust the edge weights in the graph to a probability distribution.

[0044] Furthermore, the expression for probability calculation is as follows:

[0045] P(X i )=∑ p∈Q P(p)=∑ p∈Q ∏ e∈Ep P(e)

[0046] P(p)=∏ e∈Ep P(e)

[0047] Among them, P(X i ) represents the target node X i The probability of p representing the path and Q representing the starting node R to the target node X i The set of all paths, P(p) represents the probability of path p, Ep represents the set of edges passed by path p, and P(e) represents the conditional probability of edge e on path p.

[0048] The beneficial effect of the above further scheme is: the above formula calculates the probability of each path and the front planning node passed through starting from a selected post-evaluation node. The larger the calculated value, the stronger the correlation. If you want to improve the value of the second post-evaluation node, you can start from the front planning node with a large calculated value, thereby helping architects to design the front planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 The figure is a flow chart of the method of the present invention.

[0050] Figure 2 Schematic diagram of the hierarchical structure of the pre-planning-post-evaluation indicator system.

[0051] Figure 3 Schematic diagram of partial mapping relationship between pre-planning and post-evaluation nodes.

[0052] Figure 4 Schematic diagram of Bayesian network reasoning. DETAILED DESCRIPTION

[0053] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0054] Example

[0055] Before describing the present invention, the following terms are explained:

[0056] Pre-planning: refers specifically to the research work in the field of architecture in which architects set the goals of the overall plan from the perspective of the discipline of architecture. It not only relies on experience and specifications, but also on actual investigations. It uses modern scientific and technological means such as computers to objectively analyze the research goals, and finally quantitatively derives the methods and procedures that should be followed to achieve the established goals.

[0057] Post-evaluation: A set of systematic procedures and methods for evaluating buildings and their environment after they have been built and used for a period of time. The principle is to collect feedback information by comparing the expected purpose of the building design with the actual use, so as to provide a reliable and objective basis for the design and decision-making of similar buildings and environments in the future.

[0058] Overview of the post-evaluation-pre-planning index system: The pre-planning-post-evaluation index system is constructed on the basis of the existing architectural index research and theoretical method framework, while fully considering the opinions and suggestions of experts in multiple fields such as architectural planning, planning, low-carbon green research, and evaluation. The index system takes the three dimensions of spatial characteristics, low-carbon performance, and users as the starting point, divides the indicators of large public buildings into three levels, and further refines the specific content of each dimension. The structure diagram is as follows Figure 2 shown.

[0059] The development of the pre-construction planning and post-evaluation system is still in the theoretical stage. It is urgent to transform the theoretical system of pre-construction planning and post-evaluation from the theoretical level into technical practice. However, the manual mapping of post-evaluation indicators and pre-planning indicators is labor-intensive, the mapping relationship is not comprehensive, and it is difficult to quantify the influence weights between indicators. There is an urgent need for a tool to assist architects in completing this work. At present, deep learning technology is highly dependent on data sets. However, data in the construction field is scarce and the collection cost is high. A new technical solution needs to be designed to solve the problem of data scarcity in the construction field, such as Figure 1 As shown, the present invention provides an auxiliary construction project decision-making method based on post-evaluation-front planning and feedforward deduction, and its implementation method is as follows:

[0060] S1. Obtain the tabular data of the construction project of the post-evaluation-pre-planning indicator system and construct an initial directed acyclic network diagram, which is specifically:

[0061] Obtain the tabular data of the construction project of the post-evaluation-pre-planning indicator system, and use the following formula to construct a directed acyclic network graph:

[0062] G=(V,A,W)

[0063] In this embodiment, the index system of post-evaluation-pre-planning can be abstracted as a directed acyclic graph (DAG), where G represents a directed acyclic graph, V represents a set of nodes in the initial directed acyclic network graph, i.e., the various indicators of the pre-planning and post-evaluation index systems (post-evaluation: 20 at the first level, 54 at the second level, and 170 at the third level; pre-planning: 16 at the first level, 38 at the second level, and 117 at the third level; a total of 415 nodes); A represents the adjacency matrix of the initial directed acyclic network graph, i.e., the mapping relationship between the pre-planning and post-evaluation nodes. Figure 3 As shown in the figure (the mapping relationship within the separate indicator systems of pre-planning and post-evaluation is mapped according to the hierarchical relationship, and the mapping relationship between post-evaluation and pre-planning is manually preliminarily labeled by experts, and the post-evaluation indicators are mapped and connected with the pre-planning indicators that affect them); W represents the edge weight matrix, that is, the strength of the mapping relationship between the pre-planning and post-evaluation nodes (this data was not available initially and was calculated later by the algorithm).

[0064] S2. Use transitive closure to find the mapping relationship between all pre-construction planning and post-evaluation indicators in the initial directed acyclic network graph to improve the adjacency matrix. The implementation method is as follows:

[0065] S201, initializing the transitive closure matrix as an adjacency matrix;

[0066] S202, calculating the square of the transitive closure matrix;

[0067] S203, updating the transitive closure matrix to the operation result of the non-zero elements in the square calculation result and the elements in the adjacency matrix;

[0068] S204, determining whether the value of the transitive closure matrix no longer changes, if so, the finally obtained transitive closure matrix is ​​the transitive closure of the initial directed acyclic network graph, and proceeding to S205, otherwise, returning to step S201;

[0069] S205, assigning the transitive closure matrix to the adjacency matrix, obtaining the mapping relationship between all pre-construction planning and post-evaluation indicators in the initial directed acyclic network graph, so as to improve the adjacency matrix.

[0070] In this embodiment, the mapping relationship between pre-construction planning and post-evaluation indicators mainly relies on literature and expert evaluation, and preliminary annotation of relationship mapping is performed manually. It is difficult to fully capture all mapping relationships between indicators through manual annotation. The present invention uses transitive closure to find all direct or indirect mapping relationships in the graph and improve the adjacency matrix. In the directed acyclic graph G, the adjacency matrix is ​​represented by A and the transitive closure matrix is ​​represented by T. The calculation of transitive closure can be achieved by continuously iterating the following steps:

[0071] (1) Initialize the transitive closure matrix T to the adjacency matrix A;

[0072] (2) Calculate the square T^2 of the transitive closure matrix T;

[0073] (3) Update the transitive closure matrix T to the logical OR result of the non-zero elements in the square T^2 and the elements of the adjacency matrix A;

[0074] (4) Repeat steps (2) and (3) until the value of T no longer changes. The final transitive closure matrix T is the transitive closure of the directed acyclic graph G. The transitive closure matrix T is assigned to the adjacency matrix A to achieve the improvement of the mapping relationship between nodes.

[0075] S3. Using the shortest path-based weight learning algorithm, the initial directed acyclic weight data is obtained to complete the construction of all data of the directed acyclic graph. The implementation method is as follows:

[0076] S301, assuming that the directed acyclic graph is a complete graph, assign the adjacency matrix to the weight matrix W, and randomly initialize the weight matrix W' to a two-dimensional matrix in the interval [1,9], where there exists a directed acyclic graph G, whose weight matrix is ​​W, and there exists a directed acyclic graph G', whose weight matrix is ​​W';

[0077] S302, bringing the weight matrix W into the Floyd-Warshall algorithm for calculation to obtain the shortest path P between each node pair;

[0078] S303, bringing the weight matrix W' into the Floyd-Warshall algorithm to calculate and obtain the path P', and comparing the shortest paths between all node pairs in the path P' with the actual shortest paths in the shortest path P;

[0079] S304, according to the comparison result, cyclically adjust the size of the weight matrix W', and according to the adjustment result, determine whether the comparison result of the path P' is consistent with the shortest path P, if so, proceed to step S305, otherwise, return to step S303;

[0080] S305 , taking the inverse of the weight matrix W′ and assigning it to the weight matrix W, so as to obtain the weight data of the initial network graph, and complete the construction of all the data of the directed acyclic graph.

[0081] In this embodiment, the shortest path algorithm is used to learn the weight information of the graph, and the multi-source Floyd algorithm Floyd-Warshall is selected to learn the weight matrix of the directed acyclic graph. During initialization, it is assumed that the graph is a complete graph, and the weight matrix is ​​randomly initialized; this randomly initialized weight matrix is ​​brought into the Floyd-Warshall algorithm for calculation to obtain the shortest path between each node pair, and the actual shortest path in the graph is used as a benchmark for comparison, and the weight value is continuously adjusted so that the shortest path found is consistent with the actual path. At this time, the weight matrix edge learns the topological structure information of the graph. The specific algorithm steps are as follows:

[0082] There is a directed graph G, where the vertex set is V, the adjacency matrix is ​​A, and the weight matrix is ​​W; suppose there is a directed graph G', where the vertex set is V' which is the same as V, the adjacency matrix A' is a complete graph, and the weight matrix is ​​W'.

[0083] (1) Initialization: Assign the adjacency matrix A to the weight matrix W, that is, assume that the weight value of each edge is equal to 1; the weight matrix W' is randomly initialized to a two-dimensional matrix in the interval [1-9].

[0084] (2) Benchmark shortest path: Substitute the weight matrix W into the Floyd-Warshall algorithm to obtain the shortest path P, which is set as the benchmark shortest path.

[0085] (3) Comparative update: Compare the shortest paths between all node pairs in P' with the actual shortest paths in the shortest path P, and adjust the size of the weight matrix W'.

[0086] The weight value obtained based on the shortest path is essentially the cost value of each edge, which needs to be further processed and converted into weights. The inverse of the final weight matrix W' is taken and then assigned to the weight matrix W. At this point, all the data of the directed acyclic graph is constructed.

[0087] In this embodiment, the weight learning method based on the shortest path uses the Floyd-Warshall algorithm, and the Dijkstra shortest path algorithm can also be used. It can be calculated several times, and the shortest paths between all node pairs can be calculated in each round. The weight learning method based on the shortest path can be replaced by a weight learning algorithm based on the minimum spanning tree.

[0088] S4. According to the constructed directed acyclic network diagram, the optimal decision is obtained by path deduction based on the Bayesian network, and the decision on the construction project is completed. The implementation method is as follows:

[0089] S401, normalizing the weight data of the constructed directed acyclic network graph;

[0090] S402, based on the normalization processing result, taking any post-evaluation node as a starting point in the Bayesian network, and deducing the pre-planning nodes and paths related to the starting point along the Bayesian network path;

[0091] S403, performing probability calculation on the pre-planning nodes and paths, and obtaining the optimal decision in the pre-construction planning stage according to the probability calculation results, thereby completing the decision on the construction project.

[0092] In this embodiment, the Bayesian network requires a conditional probability table, so the weight matrix W needs to be normalized. This ensures that the sum of the probabilities starting from each node is 1, that is, a probability distribution is formed. Traverse each row of the weight matrix W, calculate the sum of all elements in the row, record it as the norm of the row, and divide each element of the row by the norm of the row to achieve row normalization. The specific formula is as follows:

[0093] There is a weight matrix W, where the element in the i-th row and j-th column is W[i][j], 0≤i<n, 0≤j<n, and n is 415. After row normalization, the resulting matrix is ​​N, and N is assigned to W. The calculation formula is:

[0094]

[0095] Among them, N[i][j] means that the matrix N is the i-th row and j-th column, n represents the size of the weight matrix W, W[i][j] represents the element of the i-th row and j-th column in the weight matrix, and 0≤i<n, 0≤j<n.

[0096] The joint probability of variables in a Bayesian network can be obtained by multiplying the conditional probability distribution of the nodes. Based on the multiplication rule of joint probability and Bayes' theorem, the present invention determines the starting node through probability propagation, and can calculate the probability value of the leaf node and the joint probability of each path. Specifically, it can be summarized as follows:

[0097] A Bayesian network with n nodes, where the node variables are represented by X1,X2,...,X n , where X n Represents the nth node variable, given the starting node R and the target leaf node X i Define path p as starting from the starting node R, passing through a series of nodes and finally reaching the target leaf node X i A path, the set of edges on the path p is represented as Ep, from the starting node R to the target leaf node X i The set of all paths is represented by Q, P represents the probability, then the probability of path p can be expressed as:

[0098] P(p)=∏ e∈Ep P(e)

[0099] Then, the target leaf node Xi The probability of can be calculated by adding the probabilities of all paths:

[0100] P(X i )=∑ p∈Q P(p)=∑ p∈Q ∏ e∈Ep P(e)

[0101] Among them, P(X i ) represents the target node X i The probability of p representing the path and Q representing the starting node R to the target node X i The set of all paths, P(p) represents the probability of path p, Ep represents the set of edges passed by path p, and P(e) represents the conditional probability of edge e on path p.

[0102] Through this method, in the Bayesian network, any post-evaluation node can be used as the starting point to deduce the related pre-planning nodes and paths along the network path, and the probability of these pre-planning nodes and paths can be calculated to assist decision makers in making the best decision in the pre-planning stage of the building and provide important guidance. Figure 4 As shown, 1-5 are post-evaluation nodes, 6-8 are pre-planning nodes. Referring to the same type of buildings, assuming that the data of post-evaluation node 1 is not ideal and it is hoped to be improved, then the probability of the pre-planning nodes related to it can be calculated. The higher the probability, the stronger the correlation. In the subsequent planning of the same type of buildings, these pre-planning indicators should be considered. The probabilities of all paths and nodes can be calculated according to the formula. Path 1-3-7: 0.2*0.8; Node 7: P(1-3-7)+P(1-4-7)+P(1-5-7). Similarly, the values ​​of nodes 6 and 8 can be calculated. The larger the value, the stronger the correlation, and the design of the pre-planning indicator can be considered.

Claims

1. A construction project decision-making assistance method based on post-evaluation-pre-planning feedforward deduction, characterized in that: The following steps are involved: S1. Obtain the tabular data of the construction project of the post-evaluation-pre-planning indicator system and construct an initial directed acyclic network graph; S2. Using transitive closure, find the mapping relationship between all pre-construction planning and post-evaluation indicators in the initial directed acyclic network graph to improve the adjacency matrix; S3, according to the processing result of step S2, using the weight learning algorithm based on the shortest path, obtaining the initial directed acyclic weight data to complete the construction of all data of the directed acyclic graph; S4. According to the constructed directed acyclic network diagram, the optimal decision is obtained based on the path deduction of the Bayesian network to complete the decision on the construction project.

2. The auxiliary construction project decision-making method based on post-evaluation-pre-planning feedforward deduction according to claim 1 is characterized in that: The S1 is specifically: obtaining the construction project table data of the post-evaluation-pre-planning indicator system, and constructing a directed acyclic network diagram using the following formula: G=(V,A,W) Among them, G represents a directed acyclic graph, V represents the node set in the initial directed acyclic network graph, A represents the adjacency matrix of the initial directed acyclic network graph, and W represents the edge weight matrix.

3. The auxiliary construction project decision-making method based on post-evaluation-front planning and feedforward deduction according to claim 1 is characterized in that: The S2 comprises the following steps: S201, initializing the transitive closure matrix as an adjacency matrix; S202, calculating the square of the transitive closure matrix; S203, updating the transitive closure matrix to the operation result of the non-zero elements in the square calculation result and the elements in the adjacency matrix; S204, determining whether the value of the transitive closure matrix no longer changes, if so, the finally obtained transitive closure matrix is ​​the transitive closure of the initial directed acyclic network graph, and proceeding to S205, otherwise, returning to step S201; S205, assigning the transitive closure matrix to the adjacency matrix, obtaining the mapping relationship between all pre-construction planning and post-evaluation indicators in the initial directed acyclic network graph, so as to improve the adjacency matrix.

4. The auxiliary construction project decision-making method based on post-evaluation-front planning and feedforward deduction according to claim 1 is characterized in that: The S3 comprises the following steps: S301, assuming that the directed acyclic graph is a complete graph, assign the adjacency matrix to the weight matrix W, and randomly initialize the weight matrix W' to a two-dimensional matrix in the interval [1,9], where there exists a directed acyclic graph G, whose weight matrix is ​​W, and there exists a directed acyclic graph G', whose weight matrix is ​​W'; S302, bringing the weight matrix W into the Floyd-Warshall algorithm for calculation to obtain the shortest path P between each node pair; S303, bringing the weight matrix W' into the Floyd-Warshall algorithm to calculate and obtain the path P', and comparing the shortest paths between all node pairs in the path P' with the actual shortest paths in the shortest path P; S304, according to the comparison result, cyclically adjust the size of the weight matrix W', and according to the adjustment result, determine whether the comparison result of the path P' is consistent with the shortest path P, if so, proceed to step S305, otherwise, return to step S303; S305 , taking the inverse of the weight matrix W′ and assigning it to the weight matrix W, so as to obtain the weight data of the initial network graph, and complete the construction of all the data of the directed acyclic graph.

5. The auxiliary construction project decision-making method based on post-evaluation-pre-planning feedforward deduction according to claim 1 is characterized in that: The S4 comprises the following steps: S401, normalizing the weight data of the constructed directed acyclic network graph; S402, based on the normalization processing result, taking any post-evaluation node as the starting point in the Bayesian network, and deducing the pre-planning nodes and paths related to the starting point along the Bayesian network path; S403, performing probability calculation on the pre-planning nodes and paths, and obtaining the optimal decision in the pre-construction planning stage according to the probability calculation results, thereby completing the decision on the construction project.

6. The auxiliary construction project decision-making method based on post-evaluation-pre-planning feedforward deduction according to claim 5 is characterized in that: The weight data of the initial network graph is normalized, which is specifically: Normalize the weight data of the initial network graph to obtain the matrix N, and assign the matrix N to the weight matrix W: Among them, N[i][j] means that the matrix N is the i-th row and j-th column, n represents the size of the weight matrix W, W[i][j] represents the element of the i-th row and j-th column in the weight matrix, and 0≤i<n, 0≤j<n.

7. The auxiliary construction project decision-making method based on post-evaluation-front planning and feedforward deduction according to claim 5 is characterized in that: The expression for probability calculation is as follows: P(X i )=∑ p∈Q P(p)=∑ p∈Q ∏ e∈Ep P(e) P(p)=∏ e∈Ep P(e) Among them, P(X i ) represents the target node X i The probability of p representing the path and Q representing the starting node R to the target node X i The set of all paths, P(p) represents the probability of path p, Ep represents the set of edges passed by path p, and P(e) represents the conditional probability of edge e on path p.