A data audit method and system for engineering cost projects

Abstract

The present invention relates to the field of artificial intelligence technology, and specifically relates to a data auditing method and system for engineering cost projects, including: obtaining all historical units and units to be audited based on the construction design drawings in the database and the construction design drawings to be audited, and using a graph convolutional neural network to obtain all historical modules and modules to be audited; obtaining the reference degree of each historical unit according to the structural similarity between historical modules and the cost characteristics between historical units; fusing the cost characteristics of historical units according to the reference degree of historical units and the structural similarity between the modules to be audited to which the units to be audited belong and the historical modules to which the historical units belong, so as to obtain the auditing characteristics of each unit to be audited; and further obtaining the auditing result according to the cost characteristics of each unit to be audited. That is, it realizes automatic auditing and ensures that the current auditing result is more reasonable, improves the effectiveness of the auditing work, and prevents situations such as under-estimation, omission, and over-reporting.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly relates to a data audit method and system for construction cost projects. Background Art

[0002] The industry competition in the construction engineering field is becoming increasingly fierce. How to reduce the construction cost on the basis of ensuring the construction quality has become the key for construction enterprises to improve their competitiveness. To reduce the construction cost, it is necessary to conduct a cost audit on the implementation process of the construction project, restrict its cost control, and keep the construction project in a refined management process for a long time, so as to reduce the construction cost of the construction project on the basis of ensuring the construction quality.

[0003] The quantity audit is the basic work of the construction project cost audit, which places high requirements on the auditors. On the one hand, the auditors must have strong reading and comprehension abilities. Only with this ability can they understand the connotation of each drawing. On the other hand, the auditors should be familiar with the calculation rules and methods. For example, for details such as how to divide the slab column, when to calculate according to the cornice, and when to calculate according to the canopy, they should be clearly grasped. Only by grasping the work details can situations such as underestimation, omission, and overreporting be prevented. This matter seems easy, but it is not easy to implement. It requires the auditors to continuously improve their professional level and accumulate experience to improve the effectiveness of the audit work.

[0004] However, there are few auditors with strong professionalism and rich experience, and the labor cost is high. Currently, the general auditors are not strong enough in professionalism or lack experience, and cannot maintain a high audit quality. In addition, the existing manual audit method is time-consuming and laborious, and the efficiency is not high. Summary of the Invention

[0005] To solve the above technical problems, an automated system with certain professionalism and audit experience is needed. The purpose of the present invention is to provide a data audit method and system for construction cost projects, and the specific technical solutions adopted are as follows:

[0006] The present invention provides a data audit method for construction cost projects, and the method includes the following steps:

[0007] Obtain all historical units on each construction design drawing in history, and obtain the cost characteristics of each historical unit. Construct each historical unit graph data according to the historical units on each construction design drawing, and input all the historical unit graph data into the graph convolutional neural network respectively to obtain all historical modules;

[0008] Calculate the structural similarity between all historical modules, and obtain the reference degree of each historical unit based on the cost difference between each historical unit and other historical units, as well as the structural similarity between the historical module to which each historical unit belongs and the historical modules to which other historical units belong;

[0009] Based on the units to be reviewed on the construction design drawing to be reviewed, construct the data of the units to be reviewed graph, and input the data of the drawing to be reviewed into the graph convolutional neural network to obtain all the modules to be reviewed;

[0010] Based on the structural similarity between each module to be reviewed and each historical module, as well as the reference degree of each historical unit, obtain the review features of each unit to be reviewed. Finally, obtain the cost review result based on the difference between the review features of the unit to be reviewed and the cost features of the unit to be reviewed.

[0011] Furthermore, the steps for obtaining the data of each historical unit graph include:

[0012] Read out the electronic construction design drawings of each historical project from the database, obtain all the historical units included in the construction design drawings, and assign an attribute vector to each historical unit;

[0013] Taking each historical unit as a node, the value of each node is the attribute vector of the corresponding historical unit. If any two historical units have direct physical contact on the construction design drawing, then the edge weight value between the nodes represented by the two historical units is 1, otherwise it is 0. Furthermore, construct a topological graph for all the historical units on each historical construction design drawing, which is called the data of each historical unit graph.

[0014] Furthermore, the steps for obtaining the attribute vector of the historical unit include:

[0015] The historical unit refers to the smallest building structure unit on the construction design drawing. Obtain the type of each historical unit, and the types include but are not limited to: beams, columns, walls, stairs, doors, windows, floors, and each type is represented by one-hot encoding; in addition, obtain all the vertex coordinates of the smallest circumscribed cube of each historical unit, and merge the type and the vertex coordinates of each historical unit into a one-dimensional vector, which is called the attribute vector.

[0016] Furthermore, the historical module refers to a set composed of multiple historical units. There are physical contacts and mutual forces between the historical units in the same historical module, forming a stable building structure with certain functions; the historical module is essentially a part of the building structure on the building.

[0017] Furthermore, the steps for obtaining the structural similarity between the historical modules include:

[0018] For any two historical modules, obtain any pair of historical units from the two historical modules respectively. If the pair of historical units is of the same type, the similarity degree of the pair of historical units is the cosine similarity of the attribute vectors of the pair of historical units. If the pair of historical units is not of the same type, the similarity degree of the pair of historical units is 0;

[0019] Use the KM algorithm to match the historical units in the two historical modules to obtain all pairs of matched historical units, so that all pairs of matched historical units have the maximum similarity degree;

[0020] Obtain the sum of the similarity degrees between all pairs of matched historical units, which is denoted as the overall similarity degree of the two historical modules; additionally, obtain the historical module with the largest number of historical units among the two historical modules, and call the number of historical units in this historical module the maximum capacity of the two historical modules. Call the ratio of the overall similarity degree to the maximum capacity the structural similarity of any two historical modules;

[0021] Similarly, obtain the structural similarity between any to-be-reviewed module and any historical module.

[0022] Furthermore, the cost of each historical unit refers to the high-dimensional vector composed of multiple charging indicators for each historical unit on a construction design drawing. The charging indicators include but are not limited to material prices, equipment usage costs during construction, and labor costs.

[0023] Furthermore, the steps for obtaining the reference degree of each historical unit include:

[0024] Obtain all historical units of the same type as each historical unit, which is called the same-type set of each historical unit. The elements in the same-type set of each historical unit refer to historical units;

[0025] Obtain the difference between the cost characteristics of each historical unit and the cost characteristics of each element in its same-type set, and then obtain the structural similarity between the historical module to which each historical unit belongs and the historical modules to which each element in its same-type set belongs. The product of the difference and the structural similarity is called the difference feature;

[0026] The L2 norm of the covariance matrix of the whitening processing results of all difference features of each historical unit is called the reference degree of each historical unit.

[0027] Furthermore, the steps for obtaining the review characteristics of the to-be-reviewed unit include:

[0028] Obtain the module to be audited where the unit to be audited is located, and all historical units of the same type as the unit to be audited. Use the structural similarity between the module to be audited and the historical module to which all the historical units belong as the first weight, use the reference degree of all the historical units as the second weight, and use the ratio of the first weight to the second weight as the final weight. Perform weighted summation on the cost characteristics of all the historical units to obtain the audit characteristics of the unit to be audited.

[0029] The present invention also provides a data audit system for engineering cost projects, including a processor and a memory. The processor is used to process the instructions stored in the memory to implement any one of the data audit methods for engineering cost projects described above.

[0030] The present invention has the following beneficial effects: By learning and extracting engineering cost characteristics from historical construction design drawings, and then assisting in the cost audit of the current construction design drawing, on the one hand, the function of automatic audit is realized, reducing the dependence on professionals and improving the audit efficiency; on the other hand, first extract the cost characteristics of each type of building structure unit at different building structures through data such as historical construction design drawings, and then combine the building structures in the design drawing to be audited to fuse the cost characteristics in history to obtain the audit result of the design drawing to be audited, ensuring that the rules of design and audit of different building units at different building structures can be considered during the audit, making full use of the audit experience of historical design drawings, ensuring that the current audit result is more reasonable, improving the effectiveness of the audit work, and preventing situations such as underestimation, omission, and overreporting. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0032] Figure 1 It is a flowchart of a data audit method for an engineering cost project provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] To further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following specifically describes, with reference to the accompanying drawings and preferred embodiments, a data review method and system for engineering cost projects according to the present invention, including its specific implementation manner, structure, features, and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0034] The following specifically describes the specific solution of a data review method and system for engineering cost projects provided by the present invention with reference to the accompanying drawings.

[0035] Please refer to Figure 1 , which shows a data review method for engineering cost projects according to the present invention. The method is characterized by including the following steps:

[0036] Step S001, obtaining a historical module and a module to be reviewed by using a graph convolutional neural network based on a historical construction design drawing and a design drawing to be reviewed.

[0037] Enterprises or institutions that review design drawings store the reviewed construction design drawings in a database. These design drawings are construction drawings designed by different construction units in different construction projects in history, and each construction design drawing not only contains the design data of the building but also the finally reviewed and approved project cost data. The construction design drawing described in the present invention refers to an electronic design drawing, and the design Figure 1 is generally divided into a front view, a top view, a side view, etc. That is, from the design drawing, the present invention can obtain the three-dimensional information of the building units on the drawing, and all the building structure data on the design drawing can be directly read out.

[0038] Read a construction design drawing from the database. The design drawing shows different types of building units, such as beams, columns, walls, stairs, doors, windows, floors, etc. The present invention refers to the building units on the design drawing in the database as historical units. The present invention uses one-hot numbers of types such as beams, columns, walls, stairs, doors, windows, floors, etc. to represent the type data of historical units, and obtains the minimum circumscribed cube of each historical unit according to the three views of the design drawing. The minimum circumscribed cube is equivalent to the minimum circumscribed rectangle in two-dimensional space and is used to describe the size and position information of the historical unit. The present invention combines the type of the historical unit and the coordinates of all vertices of the minimum circumscribed cube into a vector, which is called the attribute vector of the historical unit and is used to describe the type and position of the historical unit in the construction design drawing.

[0039] For a construction design drawing, all historical units thereon are used as nodes, and the attribute vectors of the historical units are used as the values of the nodes. If any two historical units have direct physical contact, the edge weight value between the nodes corresponding to these two historical units is 1, otherwise it is 0, thus generating a topological graph, which is called the historical unit graph data of the construction design drawing. The purpose of constructing the graph data in the present invention is to conveniently, concisely and completely represent the main features of the construction design drawing for convenient subsequent processing.

[0040] For a current construction design drawing to be audited, the building structure units on this drawing are simply referred to as units to be audited. Only for the purpose of distinguishing from the above-mentioned historical structure units, similarly, the unit graph data to be audited is obtained according to the construction design drawing to be audited.

[0041] Construct a graph convolutional neural network. After inputting a historical unit graph data into the graph convolutional neural network, all historical modules are obtained. Each of the historical modules refers to a set of some historical units. There are physical contact and interaction forces between the historical units in the same historical module, forming a stable building structure with a certain function. Substantially, the historical module is a part of the building structure on the building. For example, the building structure composed of columns, beams and floor slabs is a historical module. Another example is that certain types of brickwork and wooden structures also form a historical module, and so on. The reason for obtaining the historical modules is considered that the bill of quantities pricing rules or quota pricing rules of the same historical unit are different in different historical modules. For example, columns or beams, etc. in different historical modules, such as at the overlapping structures of beams, columns and slabs, cornice structures, canopy structures, etc., the measurement methods, concrete grades, quality of steel bars, etc. used are different, and thus the cost is also different. By introducing historical modules, the subsequent audit process can grasp the work details, so as to prevent situations such as underestimation, omission, over-reporting, etc.

[0042] It should be noted that the same historical unit may be included in different historical modules, and the same type of historical unit may be included in the same historical module. These situations do not affect the subsequent calculations of the present invention.

[0043] The training process of the graph convolutional neural network is supervised training. The method for obtaining the data set is as follows: all construction enterprises, construction drawing review enterprises, and construction drawing design enterprises share the construction design drawings. The historical unit graph data corresponding to these construction design drawings constitutes the data set. The label of each data in the data set is manually marked. Generally, when designing the construction drawing, the designer conveniently marks the historical units belonging to the same historical module, and thus the label of each data can be easily obtained. Then, the graph convolutional neural network is trained according to the data set.

[0044] On the other hand, in order to keep the data of the construction design drawings confidential, the graph convolutional neural network can be trained by using the method of federated learning; the construction design drawings can also be generated through computer simulators, such as game engines, architectural design engines, etc., so as to obtain a large number of data sets; it should be noted that the graph convolutional neural network used in the present invention has few layers, few parameters, and simple tasks, and does not require a large-scale data set. Even the construction design drawings in the database of the reviewing enterprise or institution can be used as a data set to train a relatively accurate graph convolutional neural network.

[0045] The present invention selects all the construction design drawings with reasonable review results from the database, constructs all the historical unit graph data according to the construction design drawings, and inputs the historical unit graph data into the graph convolutional neural network respectively to obtain all the historical unit modules; and obtains the to-be-reviewed unit graph data according to the to-be-reviewed construction design drawing, and inputs it into the graph convolutional neural network to obtain all the to-be-reviewed modules.

[0046] Step S002: Obtain the cost characteristics of each historical unit.

[0047] For all the historical modules obtained in step S001, each historical module is a set of some historical units, and each historical unit has cost data. The cost data of the historical unit described in the present invention refers to the cost result given when designing the construction drawing, which can be directly read from the database, and the cost data of each historical unit is reviewed by the reviewer or approved by the present invention; in order to more precisely characterize the cost data of each historical unit, the present invention uses a high-dimensional vector composed of cost indicators such as the material price of each historical unit, the equipment usage cost during the construction of each historical unit, and the labor cost. This high-dimensional vector is called the cost characteristic, and the present invention uses the cost characteristic to represent the cost data of each historical unit.

[0048] Furthermore, considering that the prices of each building material, labor costs, etc. are different in different periods, the average price of each material, the average equipment usage cost, and the average labor cost in each period are calculated, and then the cost characteristics of each historical unit are re-obtained by using the ratio of the material price of each historical unit to the average price of the corresponding material in the corresponding period, the ratio of the equipment usage cost during the construction of each historical unit to the average equipment usage cost in the corresponding period, the ratio of the labor cost to the average labor cost in the corresponding period, etc.

[0049] Step S003: Obtain the review characteristics of the to-be-reviewed unit according to the cost characteristics of the historical module and the historical unit.

[0050] For any two historical modules, obtain any two historical units from the two historical modules respectively and form a pair of historical units. If the two historical units are of the same type, the similarity degree of the two historical units is the cosine similarity of the attribute vectors of the pair of historical units. If the two historical units are not of the same type, the similarity degree of the pair of historical units is 0;

[0051] Use the KM algorithm to match the historical units in the two historical modules to obtain all the matched pairs of historical units, so that all the matched pairs of historical units have the maximum similarity degree;

[0052] Obtain the sum of the similarity degrees between all the matched pairs of historical units, which is denoted as the overall similarity degree of the two historical modules; additionally, obtain the historical module with the largest number of historical units among the two historical modules, and call the number of historical units in this historical module the maximum capacity of the two historical modules. Call the ratio of the overall similarity degree to the maximum capacity the structural similarity degree of any two historical modules.

[0053] The larger the structural similarity degree of any two historical modules, the more similar the building structures represented by the two historical modules are, and then the construction costs of the historical units in the two historical modules can be more mutually referable. Similarly, the structural similarity degree between any one to-be-reviewed module and any one historical module can be obtained.

[0054] For one of the to-be-reviewed units, obtain the to-be-reviewed module where the historical unit is located; and obtain all the historical units of the same type as the to-be-reviewed unit. Suppose there are N such historical units, and the cost feature of the nth historical unit is ;

[0055] Then the review feature of the to-be-reviewed unit is: .

[0056] Among them represents the structural similarity degree between the to-be-reviewed module A where the to-be-reviewed unit is located and the historical module where the nth historical unit is located. The larger this value is, the more the cost feature of the nth historical unit can be used to evaluate the cost of the to-be-reviewed unit, that is, the review feature of the to-be-reviewed unit is closer to ;

[0057] represents the reference degree of the nth historical unit. The specific calculation method is: obtain the structural similarity degree between the historical module where the nth historical unit is located and the historical module where the mth historical unit is located ; then obtain the cost feature of the mth historical unit and the cost feature of the nth historical unit ); When all constitute a set, which represents the distribution of the differences between other historical units and the nth historical unit; then use the ZCA whitening algorithm to whiten all the difference features in the set, and then calculate the covariance matrix of all the difference features in the set. The L2 norm of this covariance matrix is regarded as , representing the reference degree of the nth historical unit. The smaller this value is, the smaller the difference between the cost features of the nth historical unit and those of other historical units, indicating that the cost features of the nth historical unit can be used as a reference for the audit standard. Therefore, the audit features of the unit to be audited are closer to ; The larger it is, the greater the difference between the cost features of the nth historical unit and those of other historical units, and the stronger the uncertainty. Using the cost features of the nth historical unit as a reference in the audit process may introduce errors. Therefore, the present invention does not focus on the value. is the normalization coefficient.

[0058] The present invention introduces so that the obtained audit features can not only fully refer to and learn the cost experience in history, but also avoid the introduction of noise and incorrect data as much as possible, so as to prevent situations such as underestimation, omission, and overreporting in the subsequent audit results.

[0059] Thus, the audit features of the unit to be audited are obtained. Similarly, the audit features of all units to be audited on the construction design drawing to be audited are obtained.

[0060] Step S004: Obtain the audit result according to the audit features of the unit to be audited and the cost features of the unit to be audited.

[0061] The present invention believes that the cost features of all units to be audited have been given. Then, obtain the absolute value of the difference between the L2 norm of the cost features of each unit to be audited and the L2 norm of the audit features of the unit to be audited. When this value is greater than the threshold, it means that the cost of each unit to be audited has a problem, and each audit feature is used as the cost feature of each unit to be audited; when this value is less than the preset threshold, it means that the cost of each unit to be audited is okay and there is no need to modify the cost features.

[0062] The finally obtained cost features of each unit to be audited are the audit results.

[0063] The threshold described in the present invention can be set artificially. The present invention provides a method for obtaining the threshold: obtain the mean value of the L2 norms of the cost features of all units to be audited, denoted as a; obtain the mean value of the L2 norms of the audit features of all units to be audited, denoted as b; then the threshold is .

[0064] It should be noted that the above order of the embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. And the above specific embodiments of this specification have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0065] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments.

[0066] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A data review method for engineering cost projects, characterized in that, the method comprises the following steps: Obtain all historical units on each construction design drawing in history, and obtain the cost characteristics of each historical unit. Construct each historical unit graph data based on the historical units on each construction design drawing, and input all the historical unit graph data into a graph convolutional neural network respectively to obtain all historical modules; Calculate the structural similarity between all historical modules. Obtain the reference degree of each historical unit according to the cost difference between each historical unit and other historical units and the structural similarity between the historical module to which each historical unit belongs and the historical modules to which other historical units belong; Construct the to-be-reviewed unit graph data according to the to-be-reviewed units on the to-be-reviewed construction design drawing, and input the to-be-reviewed graph data into a graph convolutional neural network to obtain all to-be-reviewed modules; Obtain the review characteristics of each to-be-reviewed unit according to the structural similarity between each to-be-reviewed module and each historical module and the reference degree of each historical unit. Finally, obtain the cost review result according to the difference between the review characteristics of the to-be-reviewed unit and the cost characteristics of the to-be-reviewed unit; The historical module mentioned above refers to a set composed of multiple historical units. There is physical contact and interaction force between the historical units in the same historical module, forming a stable building structure with a certain function; the historical module is essentially a part of the building structure on the building; The steps for obtaining the reference degree of each historical unit include: Obtain all historical units of the same type as each historical unit, which is called the same-type set of each historical unit. The elements in the same-type set of each historical unit refer to historical units; Obtain the difference between the cost characteristics of each historical unit and the cost characteristics of each element in its same-type set, and then obtain the structural similarity between the historical module to which each historical unit belongs and the historical modules to which each element in its same-type set belongs. The product of the difference and the structural similarity is called the difference feature; The L2 norm of the covariance matrix of the whitening processing results of all difference features of each historical unit is called the reference degree of each historical unit; The steps for obtaining the review characteristics of the to-be-reviewed unit include: Obtain the to-be-reviewed module where the to-be-reviewed unit is located, and all historical units of the same type as the to-be-reviewed unit. Use the structural similarity between the to-be-reviewed module and the historical modules to which all the historical units belong as the first weight, use the reference degree of all the historical units as the second weight, and use the ratio of the first weight to the second weight as the final weight to perform weighted summation on the cost characteristics of all the historical units to obtain the review characteristics of the to-be-reviewed unit.

2. The data review method for engineering cost projects according to claim 1, characterized in that, the steps for obtaining each historical unit graph data include: Read out the electronic construction design drawings of each engineering project in history from the database, obtain all the historical units included in the construction design drawings, and assign an attribute vector to each historical unit; Taking each historical unit as a node, with the value of each node being the attribute vector of the corresponding historical unit. If any two historical units have direct physical contact on the construction design drawing, then the edge weight value between the nodes represented by the two historical units is 1; otherwise, it is 0. Furthermore, all historical units on each construction design drawing in history are constructed into a topological graph, which is called the data of each historical unit graph.

3. A data audit method for a project cost project according to claim 2, characterized in that, the steps for obtaining the attribute vector of the historical unit include: The historical unit refers to the smallest building structure unit on the construction design drawing. Obtain the type of each historical unit, and the type includes but is not limited to: beam, column, wall, staircase, door, window, floor, and each type is represented by one-hot encoding; in addition, obtain all vertex coordinates of the smallest circumscribed cube of each historical unit, and merge the type and the vertex coordinates of each historical unit into a one-dimensional vector, which is called the attribute vector.

4. A data audit method for a project cost project according to claim 1, characterized in that, the steps for obtaining the structural similarity between historical modules include: For any two historical modules, obtain any pair of historical units from the two historical modules respectively. If the pair of historical units is of the same type, then the similarity degree of the pair of historical units is the cosine similarity of the attribute vectors of the pair of historical units; if the pair of historical units is not of the same type, then the similarity degree of the pair of historical units is 0; Use the KM algorithm to match the historical units in the two historical modules to obtain all pairs of matched historical units, so that all pairs of matched historical units have the maximum similarity degree; Obtain the sum of the similarity degrees between all pairs of matched historical units, which is recorded as the overall similarity degree of the two historical modules; in addition, obtain the historical module with the largest number of historical units among the two historical modules, and call the number of historical units in this historical module the maximum capacity of the two historical modules, and call the ratio of the overall similarity degree to the maximum capacity the structural similarity between any two historical modules; Similarly, obtain the structural similarity between any to-be-audited module and any historical module.

5. A data audit method for a project cost project according to claim 1, characterized in that, The project cost of each historical unit refers to a high-dimensional vector composed of multiple charging indicators for each historical unit on a construction design drawing, and the charging indicators include but are not limited to material prices, equipment usage costs during construction, and labor costs.

6. A data audit system for a project cost project, characterized in that, it includes a processor and a memory, and the processor is used to process the instructions stored in the memory to implement a data audit method for a project cost project as described in any one of claims 1-5.

Images