Engineering cost dynamic data labeling method and system based on BIM (Building Information Modeling) technology

Abstract

The invention provides a BIM technology-based engineering cost dynamic data labeling method and system, and the method comprises the steps: carrying out the efficient construction cost component element extraction of target BIM engineering cost dynamic data through combining a standardized deep learning network and a customized deep learning network, generating a confidence coefficient sequence related to a preset construction cost mapping knowledge set, and carrying out the construction cost component element extraction of the target BIM engineering cost dynamic data; therefore, the target cost component elements can be accurately identified. Furthermore, a standardized deep learning network and an annotation network are utilized to carry out detailed annotation on the target cost component element, and a confidence coefficient sequence associated with another preset cost mapping knowledge set is generated, so that a target cost description label of the target cost component element is accurately obtained. According to the method, the marking accuracy and efficiency of the dynamic data of the engineering cost are remarkably improved, deep mining and fine management of the BIM engineering cost data are also realized, and accurate estimation, cost control and project management of the engineering cost are facilitated.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a method and system for annotating dynamic data of engineering cost based on BIM technology. Background Art

[0002] In the field of construction engineering, dynamic data management of construction cost plays a vital role in project cost control, budget planning and decision support. With the rapid development of Building Information Modeling (BIM) technology, BIM has become an important tool in construction cost management. It provides rich and visual construction data, which provides strong support for the accurate calculation of construction cost. However, in the process of processing BIM construction cost data, how to efficiently and accurately mark the cost component elements in dynamic data has always been a problem to be solved.

[0003] Traditional methods of engineering cost data annotation often rely on manual operations, which are not only time-consuming and labor-intensive, but also easily affected by human factors, making it difficult to ensure the accuracy and consistency of the annotation results. In recent years, although some automated annotation methods based on machine learning or deep learning technology have been proposed, these methods usually rely on a single deep learning network, and their annotation effect and generalization ability are still limited for complex and changeable BIM engineering cost dynamic data.

[0004] In addition, the existing automatic annotation methods often lack the detailed distinction and accurate description of different cost component elements, which makes it difficult for the annotation results to meet the refined needs of engineering cost management in practical applications. Therefore, how to combine the characteristics of BIM technology to develop a method that can efficiently and accurately annotate cost component elements in dynamic engineering cost data has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a method for dynamic data annotation of construction cost based on BIM technology, the method comprising:

[0006] Acquire target BIM engineering cost dynamic data to be annotated, extract cost component elements from the target BIM engineering cost dynamic data through a standardized deep learning network, generate a first confidence sequence associated with a first preset cost mapping knowledge set, extract cost component elements from the target BIM engineering cost dynamic data through a customized deep learning network, generate a second confidence sequence associated with the first preset cost mapping knowledge set, wherein the first preset cost mapping knowledge set includes a plurality of reference cost component elements;

[0007] According to the first confidence sequence and the second confidence sequence, a target cost component element in the target BIM engineering cost dynamic data is obtained;

[0008] According to the target BIM engineering cost dynamic data and the target cost component element, the target cost component element is annotated by the standardized deep learning network to generate a third confidence sequence associated with a second preset cost mapping knowledge set, and the target cost component element is annotated by the annotation network to generate a fourth confidence sequence associated with the second preset cost mapping knowledge set, wherein the second preset cost mapping knowledge set includes a plurality of reference cost description tags;

[0009] Obtaining a target cost description label of the target cost component element according to the third confidence sequence and the fourth confidence sequence;

[0010] Based on the target cost component element and the target cost description tag of the target cost component element, the annotation result of the target BIM engineering cost dynamic data is generated.

[0011] On the other hand, an embodiment of the present invention also provides a dynamic data labeling system for engineering cost based on BIM technology, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0012] Based on the above aspects, the embodiment of the present application combines a standardized deep learning network with a customized deep learning network to efficiently extract cost component elements from the target BIM engineering cost dynamic data, and generates a confidence sequence associated with a preset cost mapping knowledge set, thereby accurately identifying the target cost component elements. Furthermore, the target cost component elements are carefully annotated using a standardized deep learning network and an annotation network, and a confidence sequence associated with another preset cost mapping knowledge set is generated to accurately obtain the target cost description label of the target cost component element. This method not only significantly improves the annotation accuracy and efficiency of engineering cost dynamic data, but also realizes deep mining and refined management of BIM engineering cost data, which is helpful for accurate estimation of engineering costs, cost control and project management. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the execution flow of the method for dynamic data labeling of engineering cost based on BIM technology provided by an embodiment of the present invention.

[0014] Figure 2It is a schematic diagram of the hardware architecture of a dynamic data annotation system for engineering cost based on BIM technology provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 It is a flow chart of a method for dynamic data labeling of engineering cost based on BIM technology provided by an embodiment of the present invention. The method for dynamic data labeling of engineering cost based on BIM technology is introduced in detail below.

[0016] Step S110, obtain the target BIM engineering cost dynamic data to be annotated, extract cost component elements from the target BIM engineering cost dynamic data through a standardized deep learning network, and generate a first confidence sequence associated with a first preset cost mapping knowledge set, extract cost component elements from the target BIM engineering cost dynamic data through a customized deep learning network, and generate a second confidence sequence associated with the first preset cost mapping knowledge set, wherein the first preset cost mapping knowledge set includes multiple reference cost component elements.

[0017] In this embodiment, in a construction project, there is target BIM engineering cost dynamic data to be annotated. Assume that this construction project is a comprehensive commercial building construction project, which includes various building structure parts, such as foundation engineering, main structure, decoration, water supply and drainage system, electrical system, etc., each part contains many cost component elements.

[0018] First, obtain the target BIM engineering cost dynamic data of this commercial building project. These target BIM engineering cost dynamic data can come from the cost estimates, budgets and actual cost records of each stage of the project, and are constantly updated as the project progresses. For example, the cost data of the foundation engineering part may include the cost of different types of foundation piles, the amount and price of concrete, the cost of earth excavation, etc. These target BIM engineering cost dynamic data are dynamic and may change with the progress of the project due to factors such as material price fluctuations and changes in engineering quantities.

[0019] Next, the target BIM engineering cost dynamic data is extracted from the cost component elements using a standardized deep learning network. The standardized deep learning network is constructed and trained based on the general cost knowledge of multiple projects. For example, this network has learned a large number of basic cost component element patterns in different types of construction projects. When the target BIM engineering cost dynamic data of the commercial building project is processed, possible cost component elements can be identified, and a first confidence sequence associated with the first preset cost mapping knowledge set is generated for each identified cost component element. The first preset cost mapping knowledge set contains multiple reference cost component elements, such as various types of foundation components (such as prefabricated piles, cast-in-place piles, etc.), main structure components (such as beams, columns, plates, etc.), etc. Assuming that the standardized deep learning network identifies the cast-in-place pile cost component element in the foundation project, a confidence value for this cast-in-place pile cost component element in the first preset cost mapping knowledge set can be given to form an element in the first confidence sequence. If this confidence value is high, it means that the network is relatively certain of this recognition result.

[0020] At the same time, the customized deep learning network also extracts cost component elements from the target BIM engineering cost dynamic data. The customized deep learning network is generated by learning network parameters based on the characteristic distance between the first estimated cost component element and the actual cost component element of each first template BIM engineering cost dynamic data. For this commercial building project, the customized deep learning network may be specially customized for commercial building projects of similar scale and function. It also analyzes the target BIM engineering cost dynamic data of the commercial building. For example, when analyzing the foundation engineering part, since it is customized for commercial buildings, it may be more accurate to identify cost component elements suitable for the characteristics of commercial buildings. For example, there may be special anti-floating design component elements in the foundation engineering of large commercial buildings. The customized deep learning network can better identify these elements and generate a second confidence sequence associated with the first preset cost mapping knowledge set. For example, for the special anti-floating component elements in the foundation engineering, the customized deep learning network gives a confidence value in the first preset cost mapping knowledge set to form an element in the second confidence sequence.

[0021] Step S120: obtaining a target cost component element in the target BIM engineering cost dynamic data according to the first confidence sequence and the second confidence sequence.

[0022] Continuing with the above commercial building project as an example, the first confidence sequence and the second confidence sequence contain information about the degree of possibility of each reference cost component element in the target BIM engineering cost dynamic data.

[0023] Assume that the confidence of the cast-in-place pile cost component element in the foundation engineering part in the first confidence sequence is 0.8, while the confidence of the cast-in-place pile cost component element in the second confidence sequence generated by the customized deep learning network is 0.9. At the same time, in the first confidence sequence, the confidence of another component element in the foundation engineering, the independent foundation, is 0.3, and in the second confidence sequence, the confidence of the independent foundation is 0.2.

[0024] Based on the characteristic distance between the first confidence sequence and the second confidence sequence, the first extraction loss of the second confidence sequence is determined. For example, if the difference between the corresponding elements in the two sequences is large, such as the confidence difference of the cast-in-place pile is 0.1 and the confidence difference of the independent foundation is 0.1, these differences are comprehensively considered to calculate the first extraction loss. Then, based on the deviation between the second confidence sequence and the first extraction loss, the first revised confidence of the first confidence sequence is determined. If the deviation of the second confidence sequence from the first extraction loss as a whole indicates that it is more reliable, then the confidence of the same cost component elements in the first confidence sequence and the second confidence sequence may be increased. For example, the confidence of the cast-in-place pile in the first confidence sequence is revised to 0.85.

[0025] Next, the second target confidence sequence is determined based on the deviation value between the corrected first confidence sequence and other related confidence sequences (such as confidence sequences related to other network processing results that may be involved later). For example, assuming that there are other confidence sequences of the results of the extraction of cost component elements in the target BIM engineering cost dynamic data by other networks, the final second target confidence sequence is obtained after comprehensive consideration of these factors. Based on this second target confidence sequence, the target cost component elements in the target BIM engineering cost dynamic data are obtained. For example, cast-in-place piles and special anti-floating component elements are finally determined as target cost component elements, because these target cost component elements have higher confidence in the final confidence sequence, while independent foundations are excluded due to their lower confidence.

[0026] Step S130: Based on the target BIM project cost dynamic data and the target cost component elements, the target cost component elements are labeled through the standardized deep learning network to generate a third confidence sequence associated with a second preset cost mapping knowledge set; the target cost component elements are labeled through the labeling network to generate a fourth confidence sequence associated with the second preset cost mapping knowledge set, and the second preset cost mapping knowledge set includes multiple reference cost description tags.

[0027] In the commercial building project, target cost component elements have been identified, such as cast-in-place piles and special anti-floating component elements.

[0028] According to the target BIM engineering cost dynamic data and these target cost component elements, the target cost component elements are annotated through a standardized deep learning network. The standardized deep learning network is generated by learning network parameters based on the third annotation results of each third template BIM engineering cost dynamic data and the characteristic distance between the template annotation data. This network will generate a third confidence sequence associated with the second preset cost mapping knowledge set for each target cost component element. The second preset cost mapping knowledge set includes multiple reference cost description tags, such as "cast-in-place pile-high-strength concrete cast-in-place pile (diameter 1 meter, depth 30 meters), C30 concrete, cost per meter is XXX yuan" and other similar description tags. Assume that in the third confidence sequence given by the standardized deep learning network to the cast-in-place pile target cost component element, the confidence of the description tag "high-strength concrete cast-in-place pile (diameter 1 meter, depth 30 meters), C30 concrete, cost per meter is XXX yuan" is 0.7.

[0029] At the same time, the target cost component elements are annotated through the annotation network. The annotation network is generated by learning network parameters based on the characteristic distance between the second estimated cost description label and the actual cost description label of each second template BIM engineering cost dynamic data. The annotation network will also generate a fourth confidence sequence associated with the second preset cost mapping knowledge set for the target cost component elements. For example, in the fourth confidence sequence given by the annotation network to the cast-in-place pile target cost component element, the confidence of the description label "high-strength concrete cast-in-place pile (diameter 1 meter, depth 30 meters), C30 concrete, cost XXX yuan per meter" is 0.8.

[0030] Step S140: obtaining a target cost description label of the target cost component element according to the third confidence sequence and the fourth confidence sequence.

[0031] Still taking the cast-in-place pile target cost component element in the commercial building project as an example, the target cost description label of the target cost component element is obtained based on the third confidence sequence generated by the standardized deep learning network and the fourth confidence sequence generated by the labeling network.

[0032] Assume that the confidence of the description label "high-strength concrete cast-in-place piles (1 meter in diameter, 30 meters in depth), C30 concrete, cost per meter is XXX yuan" in the third confidence sequence is 0.7, and the confidence of the description label in the fourth confidence sequence is 0.8. At the same time, there may be other networks (such as the confidence sequence corresponding to the results of the customized deep learning network in labeling mentioned in step S120, which is assumed to be the sixth confidence sequence here) that affect the confidence of this description label. For example, the confidence of the description label in the sixth confidence sequence is 0.75. Taking into account the information of these confidence sequences, it is determined that the target cost description label corresponding to the cast-in-place pile target cost component element is "high-strength concrete cast-in-place piles (1 meter in diameter, 30 meters in depth), C30 concrete, cost per meter is XXX yuan", because this description label has the highest comprehensive confidence among multiple confidence sequences.

[0033] Step S150, generating a labeling result of the target BIM engineering cost dynamic data according to the target cost component element and the target cost description label of the target cost component element.

[0034] In the commercial building project, target cost component elements (such as bored piles and special anti-floating component elements) and their respective target cost description labels (such as "high-strength concrete bored piles (1 meter in diameter, 30 meters in depth), C30 concrete, cost of XXX yuan per meter" for bored piles, etc.) have been determined.

[0035] Based on these target cost component elements and the target cost description tags of the target cost component elements, the annotation results of the target BIM engineering cost dynamic data are generated. This annotation result will contain all target cost component elements and their corresponding target cost description tags. For example, the annotation result will clearly list the cost component element information of the cast-in-place pile and its detailed cost description tag, the cost component element information of the special anti-floating component element and its corresponding cost description tag, etc. This annotation result can be used in many aspects of the project, such as cost analysis, cost control, and budget review. For example, in terms of cost control, by comparing the actual cost with the information such as the cost per meter in the cost description tag in the annotation result, cost deviations can be discovered in time and measures can be taken to adjust them. At the same time, during the budget review, the cost component elements and their cost description tags in the annotation results can be used to check whether the budget preparation is reasonable, whether there are omissions or overestimation of the cost, etc.

[0036] Based on the above steps, the embodiment of the present application combines a standardized deep learning network with a customized deep learning network to efficiently extract cost component elements from the target BIM engineering cost dynamic data, and generates a confidence sequence associated with a preset cost mapping knowledge set, thereby accurately identifying the target cost component elements. Furthermore, the standardized deep learning network and the annotation network are used to carefully annotate the target cost component elements, and a confidence sequence associated with another preset cost mapping knowledge set is generated to accurately obtain the target cost description label of the target cost component element. This method not only significantly improves the annotation accuracy and efficiency of engineering cost dynamic data, but also realizes the deep mining and refined management of BIM engineering cost data, which is helpful for accurate estimation of engineering cost, cost control and project management.

[0037] In a possible implementation, the customized deep learning network is generated by learning network parameters based on the characteristic distance between the first estimated cost component element and the actual cost component element of each first template BIM engineering cost dynamic data. The annotation network is generated by learning network parameters based on the characteristic distance between the second estimated cost description label and the actual cost description label of each second template BIM engineering cost dynamic data.

[0038] The standardized deep learning network is generated by learning network parameters based on the characteristic distance between the third annotation results of each third template BIM engineering cost dynamic data and the template annotation data, the third annotation results include the third estimated cost component elements and the third cost description labels, and the template annotation data include the actual cost component elements and the actual cost description labels of the third template BIM engineering cost dynamic data.

[0039] In a possible implementation, the method further includes:

[0040] Step A110: extract cost component elements from the target BIM engineering cost dynamic data through a labeling network according to the target BIM engineering cost dynamic data, and generate a fifth confidence sequence associated with the first preset cost mapping knowledge set.

[0041] Step A120, based on the target BIM project cost dynamic data and the target cost component elements, the target cost component elements are labeled through the customized deep learning network to generate a sixth confidence sequence associated with the second preset cost mapping knowledge set.

[0042] Wherein, step S120 includes:

[0043] Step S121: Determine a first extraction loss of the second confidence sequence based on the feature distance between the first confidence sequence and the second confidence sequence, and determine a second extraction loss of the fifth confidence sequence based on the feature distance between the first confidence sequence and the fifth confidence sequence.

[0044] Step S122: determining a first revised confidence for the first confidence sequence according to a deviation value between the second confidence sequence and the first extraction loss, and revising the first confidence sequence according to the first revised confidence.

[0045] Step S123, determining a second target confidence sequence according to the deviation value between the modified first confidence sequence and the second extraction loss, and obtaining the target cost component element according to the second target confidence sequence.

[0046] Step S140 includes: obtaining a target cost description label corresponding to the target cost component element according to the third confidence sequence, the fourth confidence sequence and the sixth confidence sequence.

[0047] In this embodiment, first, the customized deep learning network is generated by learning network parameters based on the characteristic distance between the first estimated cost component element and the actual cost component element of each first template BIM engineering cost dynamic data. A plurality of previously completed large-scale commercial building projects are used as the source of the first template BIM engineering cost dynamic data. In these projects, for each first template BIM engineering cost dynamic data, the customized deep learning network with initialized parameters is used to extract the cost component element to obtain the first estimated cost component element. For example, in a completed large-scale commercial shopping center project, the foundation engineering part actually contains a variety of cost component elements, such as large-diameter cast-in-place piles, raft foundations, and underground continuous walls, which are actual cost component elements. However, the first estimated cost component element that may be initially identified by the customized deep learning network may have partial deviations, such as misidentifying some deeper large-diameter cast-in-place piles as small-diameter cast-in-place piles or misestimating the thickness of the raft foundation. By calculating the characteristic distances between these first estimated cost component elements and the actual cost component elements, such as calculating the differences between the two in terms of structural characteristics, material usage, cost characteristics related to construction technology, etc., the network parameters of the customized deep learning network are adjusted and learned according to these characteristic distances, so that the network can more accurately identify cost component elements when processing similar commercial building projects.

[0048] The annotation network is generated by learning network parameters based on the characteristic distance between the second estimated cost description label and the actual cost description label of each second template BIM engineering cost dynamic data. For example, for a series of commercial building projects as the second template BIM engineering cost dynamic data, in a commercial office building project, for the beam component element, the actual cost description label is "rectangular cross-section reinforced concrete beam, C30 concrete, beam height 1 meter, width 0.5 meters, cost per meter is XXX yuan", while the second estimated cost description label initially obtained through the annotation network may be "reinforced concrete beam, C30 concrete, the size of the beam is not accurately described, and the cost per meter estimate has a large deviation". By calculating the characteristic distance between this second estimated cost description label and the actual cost description label in terms of description accuracy, cost numerical accuracy, etc., the network parameters of the annotation network are adjusted, thereby improving the annotation accuracy of the annotation network in subsequent projects.

[0049] The standardized deep learning network is generated by learning network parameters based on the characteristic distance between the third annotation results of each third template BIM engineering cost dynamic data and the template annotation data. Among the many different types of construction projects as the third template BIM engineering cost dynamic data, for a large shopping mall renovation project, the third estimated cost component elements and third cost description labels in the third annotation results may be different from the actual cost component elements and actual cost description labels in the template annotation data. For example, for the facade renovation of the shopping mall, the estimated cost component elements may have omitted some special decorative components, and the cost value in the cost description label is also different from the actual one. By calculating the characteristic distance of these differences, such as considering the integrity of component elements and the accuracy of cost values, the parameters of the standardized deep learning network are optimized so that it can operate more accurately when processing the target commercial building project.

[0050] In the target commercial building project, based on the target BIM engineering cost dynamic data, the target BIM engineering cost dynamic data is extracted through the annotation network to generate a fifth confidence sequence associated with the first preset cost mapping knowledge set. For example, in the electrical system part of the target commercial building, the annotation network will identify various cost component elements such as distribution boxes, cable trays, etc. and generate a fifth confidence sequence. For the distribution box, the annotation network gives a confidence value in the first preset cost mapping knowledge set based on its own learning results and the characteristics of the electrical system cost data of the target commercial building. This value reflects the degree of certainty of the annotation network in identifying the distribution box as a cost component element. The same is true for other cost component elements, thereby forming a fifth confidence sequence.

[0051] According to the target BIM engineering cost dynamic data and target cost component elements, the target cost component elements are annotated through a customized deep learning network to generate a sixth confidence sequence associated with the second preset cost mapping knowledge set. For example, for the target cost component elements that have been determined in the target commercial building, such as large-diameter cast-in-place piles in the foundation project, the customized deep learning network will annotate them and generate the sixth confidence sequence based on the relevant information in the target BIM engineering cost dynamic data, such as the depth, diameter, concrete strength, etc. of the cast-in-place piles. For the possible annotation of "large-diameter cast-in-place piles, C30 concrete, depth of 30 meters, cost of XXX yuan per meter", the customized deep learning network will give a confidence value in the second preset cost mapping knowledge set, and the annotation of other target cost component elements is similar, thereby forming the sixth confidence sequence.

[0052] In the process of obtaining the target cost component elements in the target BIM engineering cost dynamic data according to the first confidence sequence and the second confidence sequence, the structural part of the target commercial building is taken as an example. According to the characteristic distance between the first confidence sequence and the second confidence sequence, the first extraction loss of the second confidence sequence is determined. Assuming that the confidence of a certain structural component in the first confidence sequence is 0.8 and that in the second confidence sequence is 0.7, the first extraction loss is calculated by comparing the differences between the two in each cost component element, such as the confidence difference corresponding to the relevant features such as the type and scale of the structural component. At the same time, according to the characteristic distance between the first confidence sequence and the fifth confidence sequence, the second extraction loss of the fifth confidence sequence is determined. For example, there is a difference between the confidence of a certain component in the first confidence sequence and the confidence of the same component in the fifth confidence sequence, and the second extraction loss is determined by comprehensively considering the difference of all components. According to the deviation value between the second confidence sequence and the first extraction loss, the first modified confidence of the first confidence sequence is determined. If the deviation between the second confidence sequence and the first extraction loss shows that the second confidence sequence has a unique advantage in certain component elements, then the confidence of the corresponding component element in the first confidence sequence is adjusted to obtain the first revised confidence. For example, for a certain key structural component, the first revised confidence will be appropriately increased or decreased according to the deviation. The first confidence sequence is revised according to the first revised confidence, and then the second target confidence sequence is determined according to the deviation between the revised first confidence sequence and the second extraction loss. The relationship between the revised first confidence sequence and the second extraction loss on each component element is comprehensively considered to obtain the second target confidence sequence, and finally the target cost component element is obtained according to the second target confidence sequence. For example, among the structural components of a commercial building, those components with higher confidence in the second target confidence sequence are determined as target cost component elements.

[0053] When obtaining the target cost description label of the target cost component element according to the third confidence sequence and the fourth confidence sequence, take the decoration and renovation part of the target commercial building as an example. According to the third confidence sequence, the fourth confidence sequence and the sixth confidence sequence, the target cost description label corresponding to the target cost component element is obtained. Assuming that the target cost component element is a marble floor, the confidence of the description label "natural marble floor, thickness 2 cm, cost per square meter is XXX yuan" in the third confidence sequence is 0.7, the confidence of the description label in the fourth confidence sequence is 0.8, and the confidence of the sixth confidence sequence is 0.75. Considering the confidence of this description label and other possible description labels in these three confidence sequences, it is finally determined that the target cost description label of the marble floor is "natural marble floor, thickness 2 cm, cost per square meter is XXX yuan", because this description label has the highest overall confidence after comprehensively considering the three confidence sequences.

[0054] In a possible implementation, the method further includes: obtaining auxiliary guidance data for the engineering cost dynamic data annotation request.

[0055] Among them, the auxiliary guidance data includes component extraction guidance data and component labeling guidance data. The component extraction guidance data is used to guide the extraction of cost component elements of the target cost knowledge domain in the target BIM engineering cost dynamic data, and the component labeling guidance data is used to guide the labeling of the extracted cost component elements.

[0056] The first confidence sequence is obtained by extracting cost component elements from the target BIM engineering cost dynamic data and component extraction guidance data through a standardized deep learning network. The second confidence sequence is obtained by extracting cost component elements from the target BIM engineering cost dynamic data and component extraction guidance data through a customized deep learning network.

[0057] The third confidence sequence is obtained by annotating the target cost component elements through a standardized deep learning network based on the target BIM engineering cost dynamic data, the target cost component elements and the component annotation guidance data. The fourth confidence sequence is obtained by annotating the target cost component elements through an annotation network based on the target BIM engineering cost dynamic data, the target cost component elements and the component annotation guidance data.

[0058] In a possible implementation, the step of obtaining auxiliary guidance data of the engineering cost dynamic data annotation request includes:

[0059] Step B110, obtaining target data from the source data node corresponding to the engineering cost dynamic data annotation request, wherein the target data includes the type of engineering project involved, the usage scenario of the cost data, and the final application purpose of the annotation data.

[0060] Step B120, determining a target data source according to the target data, wherein the target data source includes an existing engineering cost database, a project document library, an industry standard specification library, and a related expert experience knowledge base.

[0061] Step B130: establishing data screening rules according to the annotation target of the engineering cost dynamic data annotation request, and extracting an initial engineering cost related data set from the target data source based on the data screening rules.

[0062] Step B140, performing feature analysis on the initial engineering cost related data set to generate data feature analysis results, the data feature analysis results including the structural features, semantic features, time series features and numerical features of the initial engineering cost related data set, as well as the representation and association of cost component elements in different data sources, the semantic changes and consistency requirements of cost description labels in different contexts, and the differences in description labels of the same cost components in different engineering projects. The time series features include the dynamic change characteristics of the cost with the progress of the project, and the impact of the dynamic change characteristics on the extraction and labeling of cost component elements.

[0063] Step B150, determining the type of auxiliary guidance data for the engineering cost dynamic data annotation request based on the data feature analysis result, the type including: rule-based guidance data, example-based guidance data, and association-based guidance data.

[0064] Step B160, constructing a structural framework of the guidance data based on the type of the auxiliary guidance data, wherein the rule-based guidance data is classified and organized according to the cost knowledge domain, the example-based guidance data is stored in layers according to component type and project type, and the association-based guidance data is constructed in the form of a relationship map.

[0065] Step B170, determining the hierarchical structure of the auxiliary boot data to establish dependency relationships and calling sequences between data at different levels.

[0066] Step B180, according to the determined boot data type and the hierarchical structure of the auxiliary boot data, extract the corresponding target data content from the initial engineering cost related data set, and integrate the extracted target data content of different types according to the constructed structural framework to construct the auxiliary boot data.

[0067] In this embodiment, first, obtaining the auxiliary guidance data of the engineering cost dynamic data annotation request is an important prerequisite for the accurate annotation of the entire engineering cost dynamic data. The auxiliary guidance data includes component extraction guidance data and component annotation guidance data, which play a key guiding role in the extraction and annotation process of the cost component elements of the target BIM engineering cost dynamic data. Among them, the component extraction guidance data is used to guide the extraction of the cost component elements of the target cost knowledge domain in the target BIM engineering cost dynamic data, and the component annotation guidance data is used to guide the annotation of the extracted cost component elements. For example, in a commercial building project, for the target BIM engineering cost dynamic data, it covers the cost information of various parts from basic engineering to decoration and decoration, and the target cost knowledge domain may be a specific structural part or a specific system (such as an electrical system). The component extraction guidance data can help accurately extract the relevant cost component elements from these complex data, and the component annotation guidance data can ensure that the extracted cost component elements are accurately annotated.

[0068] The first confidence sequence is obtained by extracting cost component elements from the target BIM engineering cost dynamic data based on the target BIM engineering cost dynamic data and component extraction guide data through a standardized deep learning network. Taking the electrical system of a commercial building as an example, the target BIM engineering cost dynamic data contains cost-related information such as various electrical equipment and line laying. The component extraction guide data may contain feature information about the cost component elements of the electrical system, such as typical features of cost component elements such as different types of distribution boxes and cable specifications. The standardized deep learning network analyzes and processes the target BIM engineering cost dynamic data based on this information. When identifying the cost component element of the distribution box, the confidence of the distribution box in the entire cost component element system is calculated based on the relevant data in the target BIM engineering cost dynamic data (such as equipment model, power, etc.) combined with the feature information in the component extraction guide data, thereby forming an element in the first confidence sequence. The same is true for other cost component elements, and finally a complete first confidence sequence is obtained.

[0069] The second confidence sequence is obtained by extracting cost component elements from the target BIM engineering cost dynamic data and component extraction guidance data through a customized deep learning network. In commercial building projects, the customized deep learning network also extracts cost component elements based on the target BIM engineering cost dynamic data and component extraction guidance data. For example, for special structural parts of commercial buildings, there may be some customized structural components. The customized deep learning network uses the special structural component feature information in the component extraction guidance data (such as special shapes, special materials, and other cost component element related information) and the relevant data in the target BIM engineering cost dynamic data (such as structural dimensions, material usage, etc.) to identify these special structural components and give confidence, forming a second confidence sequence.

[0070] The third confidence sequence is obtained by annotating the target cost component elements through the standardized deep learning network based on the target BIM engineering cost dynamic data, target cost component elements and component annotation guidance data. In the commercial building project, when the target cost component elements (such as cast-in-place piles in the foundation project) have been determined, the target BIM engineering cost dynamic data contains various relevant data of the cast-in-place piles (such as depth, diameter, concrete strength, etc.), and the component annotation guidance data may contain information such as specifications and requirements for cast-in-place pile cost annotation. The standardized deep learning network annotates the cast-in-place piles based on this information. For example, for an annotation such as "cast-in-place piles-high-strength concrete cast-in-place piles (diameter 1 meter, depth 30 meters), C30 concrete, cost per meter is XXX yuan", the confidence of this annotation in the entire annotation system is calculated according to the specific data in the target BIM engineering cost dynamic data and the annotation specifications in the component annotation guidance data, thereby forming an element in the third confidence sequence. The annotations of other target cost component elements are similar, and finally the third confidence sequence is obtained.

[0071] The fourth confidence sequence is obtained by annotating the target cost component elements through the annotation network based on the target BIM engineering cost dynamic data, the target cost component elements and the component annotation guidance data. In the commercial building project, the annotation network is annotated based on the target BIM engineering cost dynamic data, the target cost component elements and the component annotation guidance data. For example, for the target cost component elements (such as marble floor) of the decoration and decoration part of the commercial building, the target BIM engineering cost dynamic data contains information such as the area and material grade of the marble floor, and the component annotation guidance data contains the relevant specifications and requirements for the marble floor cost annotation (such as the price range of different grades of marble, the cost adjustment corresponding to the laying process, etc.). The annotation network annotates the marble floor based on this information. For example, for the annotation of "natural marble floor, thickness 2 cm, cost per square meter is XXX yuan", according to the specific data in the target BIM engineering cost dynamic data and the annotation specifications in the component annotation guidance data, the confidence of this annotation in the entire annotation system is calculated, thereby forming an element in the fourth confidence sequence. The annotations of other target cost component elements are similar, and finally the fourth confidence sequence is obtained.

[0072] In the specific steps of obtaining auxiliary guidance data for the engineering cost dynamic data annotation request, the target data is obtained from the source data node corresponding to the engineering cost dynamic data annotation request. The target data includes the type of engineering project involved, the use scenario of the cost data, and the final application purpose of the annotation data. In the commercial building project, the engineering project type is a large commercial building. The use scenarios of the cost data may include project budget preparation, cost control, settlement review, etc. The final application purpose of the annotation data may be to accurately evaluate the project cost and ensure that the project cost is within a reasonable range.

[0073] The target data source is determined based on these target data, and the target data source includes the existing engineering cost database, project document library, industry standard specification library, and relevant expert experience knowledge base. In the commercial building project, the existing engineering cost database may contain the cost data of similar commercial building projects in the past. The project document library contains the design documents, construction plans and other documents of the commercial building project, which may involve cost-related information. The industry standard specification library contains the standard specifications of the construction industry on cost calculation, cost component definition, etc., and the relevant expert experience knowledge base contains the experience and knowledge of experts in commercial building cost management.

[0074] According to the annotation target of the dynamic data annotation request for the engineering cost, data screening rules are established, and the initial engineering cost-related data set is extracted from the target data source based on these data screening rules. For example, if the annotation target is to accurately mark the cost component elements of the structural part of a commercial building, the data screening rules may be to screen out data related to the cost of the structural part, screen out the cost data of the structural part of previous commercial buildings from the engineering cost database, screen out the cost information in the structural design-related documents from the project document library, screen out the structural cost calculation specifications from the industry standard specification library, etc., and screen out the expert's experience knowledge on the cost management of commercial building structures from the expert experience knowledge base, thereby forming an initial engineering cost-related data set.

[0075] The data set related to the initial engineering cost is analyzed to generate data feature analysis results. In the commercial building project, the data feature analysis results include the structural features, semantic features, time series features and numerical features of the initial engineering cost related data set, as well as the representation and association of cost component elements in different data sources, the semantic changes and consistency requirements of cost description labels in different contexts, and the differences in description labels of the same cost components in different engineering projects. In terms of structural features, for example, the data structure in the engineering cost database may be stored according to different building parts and cost classifications, and the cost information in the project document library may be a structure associated with a specific design structure and construction process. In terms of semantic features, there may be differences in the semantic representation of the same cost component element in different data sources. For example, in the engineering cost database, professional cost terms may be used to describe beam components, while in the project document library, they may be described from a design perspective. Time series features include the dynamic change characteristics of cost with project progress, and the impact of dynamic change characteristics on the extraction and annotation of cost component elements. For example, in a commercial building project, as the project progresses from foundation engineering to main structure and then to decoration and renovation, the importance and value of cost component elements at different stages will change. In the foundation engineering stage, the cost component elements of foundation piles are the focus of attention. The dynamic changes in their cost (such as the increase in pile length due to changes in geological conditions) will affect the extraction and labeling of cost component elements.

[0076] According to the results of data feature analysis, the type of auxiliary guidance data for the dynamic data annotation request of the engineering cost is determined, and the types include rule-based guidance data, example-based guidance data, and association-based guidance data. In a commercial building project, rule-based guidance data may be standard specifications for cost calculation and definition of cost component elements, such as cost calculation rules for structural components; example-based guidance data may be specific cost component elements and annotation examples in previous similar commercial building projects, such as the cost component elements and annotation instances of the foundation piles of a completed commercial building; association-based guidance data may be the association relationship between different cost component elements (such as the impact of the association relationship between beams and columns in structural stress on the cost) constructed in the form of a relationship map.

[0077] According to the type of auxiliary guidance data, the structural framework of the guidance data is constructed. Rule-based guidance data is classified and organized according to the cost knowledge domain. For example, the rules related to structural cost and the rules related to electrical cost are classified separately. Example-based guidance data is stored in layers according to component type and project type. For example, examples of cost component elements of foundation engineering and examples of cost component elements of main structure are stored in layers according to different project types (such as commercial buildings, commercial office buildings, etc.). Association-based guidance data is constructed in the form of a relationship map. For example, the force correlation relationship between structural components such as beams, columns, and plates and the impact of this correlation relationship on the cost are constructed into a relationship map.

[0078] Determine the hierarchical structure of auxiliary guidance data to establish the dependency relationship and calling order between data at different levels. For example, in a commercial building project, rule-based guidance data may be at a higher level, providing basic rule basis for cost component element extraction and annotation; example-based guidance data is at an intermediate level, providing example references for specific operations; and association-based guidance data is at a deeper level, used to handle complex associations between cost component elements.

[0079] According to the determined guiding data type and the hierarchical structure of auxiliary guiding data, the corresponding target data content is extracted from the initial engineering cost-related data set, and the extracted target data content of different types is integrated according to the constructed structural framework to construct auxiliary guiding data. For example, rule-based guiding data content such as structural cost calculation rules is extracted from the initial engineering cost-related data set, and organized according to the cost knowledge domain classification; example-based guiding data content such as foundation pile cost component elements and annotation examples in different commercial building projects are extracted, and stored in layers according to component type and project type; association-based guiding data content such as the association relationship between structural components such as beams and columns is extracted to construct a relationship map, and finally these different types of target data content are integrated to construct a complete auxiliary guiding data, which is used to provide accurate guidance in the extraction and annotation process of cost component elements in commercial building projects.

[0080] In a possible implementation, step S120 may further include:

[0081] A first extraction loss of the second confidence sequence is determined according to a feature distance between the first confidence sequence and the second confidence sequence.

[0082] A first modified confidence for the first confidence sequence is determined according to a deviation value between the second confidence sequence and the first extraction loss.

[0083] The first confidence sequence is modified according to the first modified confidence to generate a first target confidence sequence.

[0084] The target cost component element is obtained according to the first target confidence sequence.

[0085] In this embodiment, in the cost analysis of commercial buildings, the first confidence sequence and the second confidence sequence contain confidence information about various cost component elements. For example, in the structural part of the commercial building, the first confidence sequence is obtained by a standardized deep learning network based on the target BIM engineering cost dynamic data and component extraction guidance data. There may be a confidence value for the beam component, and there is also a corresponding confidence value for the column component, etc.; the second confidence sequence is obtained by a customized deep learning network based on the same target BIM engineering cost dynamic data and component extraction guidance data, and also includes the confidence values ​​of components such as beams and columns. The calculation of the feature distance is based on the confidence difference of these components in the two sequences and the properties of the components themselves. For example, for beam components, the confidence in the first confidence sequence is 0.8, and the confidence in the second confidence sequence is 0.7. At the same time, the influence of the relevant properties such as the size and material of the beam on this difference is considered to comprehensively calculate the feature distance, and then determine the first extraction loss of the second confidence sequence. The first extraction loss reflects the degree of deviation of the second confidence sequence relative to the first confidence sequence in terms of cost component element extraction.

[0086] Next, the first modified confidence for the first confidence sequence is determined based on the deviation between the second confidence sequence and the first extraction loss. In the commercial building project, there is a certain relationship between the second confidence sequence and the first extraction loss. For example, if the deviation between the confidence of some cost component elements in the second confidence sequence and the first extraction loss is large, it means that the performance of these component elements in the second confidence sequence is inconsistent with the overall extraction loss. Taking the column component as an example, if the confidence of the column component in the second confidence sequence is 0.6, and the deviation value related to the column component calculated according to the first extraction loss is large, this may mean that there are special circumstances in the identification of the column component by the customized deep learning network. The confidence of the column component in the first confidence sequence is adjusted according to this deviation value, so as to determine the first modified confidence for the first confidence sequence. Similarly, for other cost component elements, by comprehensively considering the deviation value of all component elements, the confidence values ​​in the first confidence sequence are adjusted accordingly.

[0087] Then, the first confidence sequence is corrected according to the first corrected confidence to generate the first target confidence sequence. In the commercial building project, each confidence value in the first confidence sequence is replaced with the first corrected confidence to obtain the first target confidence sequence. For example, the original confidence of the beam component in the first confidence sequence is 0.8, which becomes 0.85 after correction according to the first corrected confidence; the original confidence of the column component is 0.7, which becomes 0.72 after correction, etc. This first target confidence sequence combines the information of the first confidence sequence and the second confidence sequence, and obtains a more optimized confidence sequence by analyzing and processing the feature distance, first extraction loss, deviation value, etc. between the two.

[0088] Finally, according to the first target confidence sequence, the target cost component element is obtained. In the cost management of commercial buildings, the confidence value in the first target confidence sequence reflects the possibility of each cost component element in the target BIM engineering cost dynamic data. For example, in the electrical system of a commercial building, cost component elements such as distribution boxes and cable trays are screened according to the confidence values ​​in the first target confidence sequence. If the confidence value of the distribution box in the first target confidence sequence is higher than the set threshold, the distribution box is determined as the target cost component element; if the confidence value of the cable tray is lower than the threshold, it is excluded. In this way, the cost component elements of various parts of the commercial building (such as the structural part, the decoration part, the electrical system, etc.) are screened, and the target cost component elements in the target BIM engineering cost dynamic data are finally determined. These target cost component elements will serve as the basis for subsequent operations such as annotation, which is of great significance for accurately evaluating the engineering cost of commercial buildings.

[0089] In a possible implementation, the step of determining the first confidence sequence and the second confidence sequence includes:

[0090] Step C110, based on the target BIM engineering cost dynamic data, cost component elements are extracted from the target BIM engineering cost dynamic data through a standardized deep learning network to generate a first basic confidence sequence associated with a first preset cost mapping knowledge set, and cost component elements are extracted from the target BIM engineering cost dynamic data through a customized deep learning network to generate a second basic confidence sequence associated with the first preset cost mapping knowledge set.

[0091] The first basic confidence sequence includes the first basic confidence of each reference cost component element, and the second basic confidence includes the second basic confidence of each reference cost component element.

[0092] Step C120: extracting, from the reference cost component elements, target reference component elements whose first basic confidence and second basic confidence are both greater than or equal to a set confidence, according to the first basic confidence sequence and the second basic confidence sequence.

[0093] Step C130: remove the confidences corresponding to the non-target reference component elements in the first basic confidence sequence and the second basic confidence sequence, respectively, to generate the first confidence sequence and the second confidence sequence.

[0094] In this embodiment, taking the foundation engineering of a commercial building as an example, the target BIM engineering cost dynamic data includes cost information of various aspects of the foundation engineering, such as cost data of different types of foundation piles (cast-in-place piles, prefabricated piles, etc.), cost information of earth excavation and backfilling, cost of foundation waterproofing engineering, etc. The standardized deep learning network analyzes and processes these data based on its existing learning model and algorithm, and matches and calculates confidence for the reference cost component elements in the first preset cost mapping knowledge set. The first preset cost mapping knowledge set includes many cost component elements that may appear in the foundation engineering, such as various types of pile foundations, different grades of concrete, various types of steel bars, etc. For the reference cost component element of cast-in-place piles, the standardized deep learning network calculates a confidence value based on the relevant data in the target BIM engineering cost dynamic data (such as the diameter, depth, concrete strength, etc. of the cast-in-place piles), and this value is the first basic confidence of the cast-in-place piles in the first basic confidence sequence. Similarly, the customized deep learning network extracts cost component elements for the same target BIM engineering cost dynamic data. For the reference cost component element of cast-in-place piles, the second basic confidence in the second basic confidence sequence is calculated based on its own customized learning results and algorithms. Other reference cost component elements such as prefabricated piles and foundation waterproofing membranes also obtain their respective first basic confidence and second basic confidence under the processing of the two networks, thus forming a complete first basic confidence sequence and second basic confidence sequence.

[0095] Then, according to the first basic confidence sequence and the second basic confidence sequence, the target reference component element whose first basic confidence and second basic confidence are both greater than or equal to the set confidence is extracted from each reference cost component element. In the commercial building project, the set confidence is a threshold determined according to project requirements and past experience, for example, set to 0.6. In the foundation engineering part, for the reference cost component element of cast-in-place pile, if its first basic confidence in the first basic confidence sequence is 0.7 and its second basic confidence in the second basic confidence sequence is 0.8, both of which are greater than the set confidence of 0.6, then the cast-in-place pile is determined as the target reference component element. For other reference cost component elements, such as precast piles, the first basic confidence is 0.5 and the second basic confidence is 0.65. Since the first basic confidence is less than the set confidence, the precast pile cannot be determined as the target reference component element. Through such comparison and screening, the target reference component element that meets the conditions is determined from the numerous reference cost component elements.

[0096] Finally, the confidences corresponding to the non-target reference component elements in the first basic confidence sequence and the second basic confidence sequence are eliminated respectively to generate the first confidence sequence and the second confidence sequence. In the foundation engineering of a commercial building, for the first basic confidence sequence, after the target reference component elements (such as cast-in-place piles) are determined, the confidences corresponding to other non-target reference component elements (such as precast piles, etc.) are eliminated from the first basic confidence sequence. For example, the first basic confidence sequence originally contains the confidences of reference cost component elements such as cast-in-place piles, precast piles, and foundation waterproofing membranes. After eliminating the confidences of non-target reference component elements such as precast piles, the confidences of the remaining target reference component elements such as cast-in-place piles constitute the first confidence sequence. Similarly, for the second basic confidence sequence, after eliminating the confidences corresponding to the non-target reference component elements, the second confidence sequence is obtained. The first confidence sequence and the second confidence sequence obtained in this way are more focused on the target reference component elements, providing a more targeted data basis for the subsequent accurate acquisition of the target cost component elements in the target BIM engineering cost dynamic data, which helps to improve the accuracy and efficiency of the entire engineering cost analysis and management.

[0097] In a possible implementation, the target BIM engineering cost dynamic data includes a plurality of target cost component elements, and the annotation result includes a plurality of target cost component elements and target cost description tags of the target cost component elements.

[0098] The labeling result is generated by enabling x rounds of deep learning process, each round of deep learning process is a round of deep learning component extraction process or a round of deep learning component labeling process, and the deep learning result of each round of deep learning process is a target cost component element or a target cost description label.

[0099] Among them, the specific steps of the a-th round of deep learning process include:

[0100] Step D110, determining the output category of the output result to be output according to the deep learning result of the a-1th round of deep learning process, wherein the output category is the component extraction category or the component annotation category, and the output category corresponding to the first round of deep learning process is the component extraction category.

[0101] Step D120, based on the target BIM project cost dynamic data and the deep learning results of the previous a-1 round of deep learning process, obtain a standardized confidence sequence through a standardized deep learning network, obtain a customized confidence sequence through a customized deep learning network, and obtain a labeled confidence sequence through a labeled network.

[0102] Among them, the standardized confidence sequence is the first confidence sequence or the third confidence sequence, the customized confidence sequence is the second confidence sequence or the sixth confidence sequence, and the labeled confidence sequence is the fourth confidence sequence or the fifth confidence sequence.

[0103] Step D130, generating a deep learning result of the a-th round of deep learning process based on the standardized confidence sequence, the customized confidence sequence and the labeled confidence sequence.

[0104] In this embodiment, taking the electrical system of a commercial building as an example, the target cost component elements may include distribution boxes, cable trays, wires and cables, etc. Each round of deep learning process is either a deep learning component extraction process or a deep learning component labeling process, and the result of each round of deep learning is a target cost component element or a target cost description label.

[0105] For the a-th round of deep learning process, first determine the output category of the output result based on the deep learning results of the a-1-th round of deep learning process. If it is the first round of deep learning process, the corresponding output category is the component extraction category. This is because in the initial stage, it is necessary to first identify the target cost component elements from the target BIM engineering cost dynamic data. For example, in the first round of deep learning process, for the target BIM engineering cost dynamic data of the electrical system of the commercial building, the first thing to do is to find out the cost component elements such as distribution boxes and cable trays. This process is based on the algorithm of the deep learning network and the electrical system-related cost information in the target BIM engineering cost dynamic data, such as equipment model, quantity, specifications and other information.

[0106] Next, based on the target BIM project cost dynamic data and the deep learning results of the previous a-1 round of deep learning process, a standardized confidence sequence is obtained through a standardized deep learning network, a customized confidence sequence is obtained through a customized deep learning network, and annotated confidence sequence is obtained through an annotation network. Assume that in the second round of deep learning (at this time a=2), the first round has identified the target cost component element of the distribution box. For the standardized deep learning network, the cost description labels related to the distribution box can be analyzed based on the electrical system cost information in the target BIM project cost dynamic data (such as the cost corresponding to different functional modules of the distribution box, the price difference between different brands, etc.) and the first round of deep learning results (the existence of the distribution box component has been determined), and the confidence corresponding to each possible cost description label is calculated, thereby obtaining a standardized confidence sequence (here may be the third confidence sequence, because it is in the round related to component annotation). For example, for the possible cost description label "distribution box-10 circuits, brand A, each cost is XXX yuan", the standardized deep learning network calculates a confidence value based on the data.

[0107] Similarly, the customized deep learning network will also perform analysis and processing based on the same target BIM engineering cost dynamic data and the first round of deep learning results. Since the customized deep learning network is generated by learning network parameters based on the characteristic distance between the first estimated cost component element and the actual cost component element of each first template BIM engineering cost dynamic data, in the scenario of the electrical system of a commercial building, the cost description label of the distribution box can be analyzed based on the previous learning experience of similar commercial building electrical systems to obtain a customized confidence sequence (here may be the sixth confidence sequence). For example, for the label "distribution box-10 circuits, brand A, each cost is XXX yuan", the customized deep learning network gives a confidence value based on its own learning results.

[0108] The annotation network will also operate based on the target BIM engineering cost dynamic data and the first round of deep learning results. The annotation network is generated by learning network parameters based on the feature distance between the second estimated cost description label and the actual cost description label of each second template BIM engineering cost dynamic data. It will analyze the cost description label of the distribution box to obtain the annotation confidence sequence (which may be the fourth confidence sequence). For example, for the label "Distribution box-10 circuits, brand A, each cost is XXX yuan", the annotation network gives a confidence value based on its own learning and calculation.

[0109] Finally, the deep learning results of the ath round of deep learning process are generated based on the standardized confidence sequence, customized confidence sequence and labeled confidence sequence. In the above second round of deep learning process, the confidence of the label "distribution box-10 circuits, brand A, each cost is XXX yuan" and other possible labels in the three confidence sequences are comprehensively considered. If the confidence of this label in the standardized confidence sequence is 0.7, the confidence in the customized confidence sequence is 0.8, and the confidence in the labeled confidence sequence is 0.75, a specific algorithm (such as weighted average or other comprehensive evaluation algorithms) is used to determine whether this label is the final target cost description label or to further adjust the subsequent rounds of deep learning operations based on this result. If this label is finally determined as the target cost description label, then this is the deep learning result of the second round of deep learning process; if other possible labels need to be further analyzed or the next round of component extraction and other operations need to be performed, then corresponding adjustments will be made based on this result to provide a basis for the subsequent deep learning process until a complete labeling result is generated, including all target cost component elements and their target cost description labels in the electrical system of the commercial building. This process is repeated in each round of deep learning process, and the operation category of the next round is determined based on the results of the previous round. The corresponding confidence sequence is obtained through different deep learning networks, and then the deep learning results of each round are generated based on these confidence sequences, and the complete labeling results are gradually constructed.

[0110] In a possible implementation, the training step of the customized deep learning network includes:

[0111] Step S101, obtain a first sample learning data sequence, wherein the first sample learning data sequence includes multiple first template BIM engineering cost dynamic data carrying template annotation data, and the template annotation data of each first template BIM engineering cost dynamic data includes the actual cost component element of the first template BIM engineering cost dynamic data.

[0112] In this embodiment, a series of completed large-scale commercial construction projects are taken as examples, and these projects serve as the source of the dynamic data of the first template BIM engineering cost. Each project has its own unique engineering cost composition. For example, in the foundation engineering part of one of the commercial construction projects, the actual cost component elements in its template annotation data include large-diameter cast-in-place piles, raft foundations, underground continuous walls, etc., and also include the relevant cost information of these component elements in the project, such as the depth, diameter, concrete strength grade and corresponding cost value of the cast-in-place piles. The data of these projects together constitute the first example learning data sequence, which provides basic data for the training of customized deep learning networks.

[0113] Step S102: Based on each first template BIM engineering cost dynamic data, cost component elements are extracted from each first template BIM engineering cost dynamic data through a customized deep learning network with initialized parameters to generate a first estimated cost component element of each first template BIM engineering cost dynamic data.

[0114] For example, when processing data from one of the commercial construction projects, the customized deep learning network analyzed the project’s engineering cost data based on its initialized parameters and algorithms. When analyzing the foundation engineering portion, since the network parameters may be inaccurate at the beginning, some features of the large-diameter cast-in-place piles may be misidentified, and the generated first estimated cost component elements may be biased. For example, the diameter of the large-diameter cast-in-place piles may be miscalculated, or part of the concrete usage of the raft foundation may be mistakenly calculated for other components, resulting in a less accurate first estimated cost component element. This situation is common in the early stages of training because the network is still in the learning process.

[0115] Step S103: Based on each first template BIM engineering cost dynamic data and the first estimated cost component element, the first estimated cost component element is labeled through a customized deep learning network with initialized parameters to generate a first estimated cost description label for the first estimated cost component element; based on the first estimated cost component element and the first estimated cost description label of each first template BIM engineering cost dynamic data, a labeling result of each first template BIM engineering cost dynamic data is generated.

[0116] Taking the large-diameter cast-in-place piles in foundation engineering as an example, the customized deep learning network attempts to label them based on the first estimated cost component elements extracted previously (even if there are some errors). If the diameter of the cast-in-place piles in the first estimated cost component elements is estimated incorrectly, then when labeling its cost description label, an inaccurate first estimated cost description label may be generated based on the incorrect diameter information, such as "large-diameter cast-in-place piles (wrong diameter), C30 concrete, cost per meter is XXX yuan (wrong value)". Then, based on the first estimated cost component elements and first estimated cost description labels of each first template BIM engineering cost dynamic data, the labeling results of each first template BIM engineering cost dynamic data are generated. This labeling result contains inaccurate cost component elements and cost description label information, but it is an intermediate result in the network training process.

[0117] Step S104, determining the first network learning error based on the characteristic distance between the first estimated cost component element and the actual cost component element of each first template BIM engineering cost dynamic data, and training the customized deep learning network of the initialization parameters based on the first network learning error.

[0118] For example, for the large-diameter cast-in-place pile component element, the erroneous diameter, concrete usage, etc. in the first estimated cost component element are compared with the correct values ​​in the actual cost component element, and the differences in structural characteristics, material usage, etc. are calculated to obtain the characteristic distance. These characteristic distances reflect the first network learning error in combination. Based on this first network learning error, the customized deep learning network with initialization parameters is trained. During the training process, the network adjusts its internal parameters according to the error, such as adjusting the neuron connection weights related to the identification and labeling of cast-in-place pile cost component elements, to reduce the error. With the processing and learning of more first template BIM engineering cost dynamic data, the customized deep learning network gradually improves its accuracy in extracting and labeling cost component elements.

[0119] The training steps of the labeling network include:

[0120] Step S201, obtain a second sample learning data sequence, wherein the second sample learning data sequence includes multiple second template BIM engineering cost dynamic data carrying template annotation data, and the template annotation data of each second template BIM engineering cost dynamic data includes the actual cost description label of each actual cost component element of the second template BIM engineering cost dynamic data.

[0121] Taking multiple completed commercial construction projects as an example, the template annotation data of each second template BIM engineering cost dynamic data includes the actual cost description label of each actual cost component element of the project. For example, in a commercial office building project, for the beam component element, its actual cost description label is "rectangular cross-section reinforced concrete beam, C30 concrete, beam height 1 meter, width 0.5 meters, cost per meter is XXX yuan", and there is also a corresponding accurate actual cost description label for the column component element. These data from different commercial construction projects together constitute the second sample learning data sequence, providing a data basis for the training of the annotation network.

[0122] Step S202: extracting cost component elements from each second template BIM engineering cost dynamic data through a annotation network with initialized parameters according to each second template BIM engineering cost dynamic data, and generating a second estimated cost component element for each second template BIM engineering cost dynamic data.

[0123] For example, when processing data of a commercial construction project, the annotation network may have deviations in identifying cost component elements due to initialization parameters. For example, in the structural part of the building, some beams with special shapes may be mistakenly identified as ordinary beams, resulting in inaccurate second estimated cost component elements generated for each second template BIM engineering cost dynamic data.

[0124] Step S203: Based on each second template BIM engineering cost dynamic data and the second estimated cost component element, the second estimated cost component element is labeled through a labeling network with initialized parameters to generate a second estimated cost description label for the second estimated cost component element; based on the second estimated cost component element and the second estimated cost description label for each second template BIM engineering cost dynamic data, a labeling result for each second template BIM engineering cost dynamic data is generated.

[0125] For example, for a special-shaped beam that was previously mistakenly identified as an ordinary beam, an inaccurate second estimated cost description label may be generated based on the characteristics of the ordinary beam during annotation, such as "ordinary beam (actually a special-shaped beam), C30 concrete, cost per meter is YYY yuan (wrong value)". Then, based on the second estimated cost component elements and second estimated cost description labels of each second template BIM engineering cost dynamic data, an annotation result of each second template BIM engineering cost dynamic data is generated, which contains the error information in the extraction and annotation process.

[0126] Step S204, determining a second network learning error based on a characteristic distance between a second estimated cost description label and an actual cost description label of each second template BIM engineering cost dynamic data, and training a labeling network of initialization parameters based on the second network learning error.

[0127] For example, for beam component elements, the beam type, cost value, etc. in the inaccurate second estimated cost description label are compared with the correct information in the actual cost description label, and the characteristic distances in terms of description accuracy and cost value accuracy are calculated. These characteristic distances are combined to determine the second network learning error. Based on this second network learning error, the annotation network with initialized parameters is trained. During the training process, the annotation network will adjust its internal parameters, such as adjusting the algorithm parameters related to the annotation of beam component elements, to improve the accuracy of the annotation of cost component elements. With the learning of more dynamic data of the second template BIM engineering cost, the annotation network gradually improves its annotation capabilities.

[0128] The training steps of the standardized deep learning network include:

[0129] Step S301, obtain a third sample learning data sequence, wherein the third sample learning data sequence includes multiple third template BIM engineering cost dynamic data carrying template annotation data, and the template annotation data of each third template BIM engineering cost dynamic data includes the actual cost component elements of the third template BIM engineering cost dynamic data and the actual cost description labels of each actual cost component element.

[0130] Taking various types of completed construction projects (including commercial buildings, residential buildings, etc.) as examples, the template annotation data of each third template BIM engineering cost dynamic data includes the actual cost component elements of the project and the actual cost description labels of each actual cost component element. For example, in a large shopping mall renovation project, the actual cost component elements include facade decoration components, interior decoration components, etc., and their actual cost description labels record in detail the type, specifications, materials, and corresponding cost values ​​of the components. The data of these different projects constitute the third sample learning data sequence.

[0131] Step S302: Based on each third template BIM engineering cost dynamic data, cost component elements are extracted from each third template BIM engineering cost dynamic data through a standardized deep learning network with initialized parameters to generate a third estimated cost component element of each third template BIM engineering cost dynamic data.

[0132] Step S303: Based on each third template BIM engineering cost dynamic data and the third estimated cost component element, the third estimated cost component element is labeled through a standardized deep learning network with initialized parameters to generate a third estimated cost description label of the third estimated cost component element; based on the third estimated cost component element and the third estimated cost description label of each third template BIM engineering cost dynamic data, a labeling result of each third template BIM engineering cost dynamic data is generated.

[0133] For example, when processing the data of a shopping mall renovation project, the standardized deep learning network may miss some small decorative components or misestimate the size of some components when extracting the facade decorative components due to the inaccuracy of the initialization parameters, thereby obtaining the third estimated cost component elements with some errors. Then, based on each third template BIM engineering cost dynamic data and the third estimated cost component element, the third estimated cost component element is annotated through the standardized deep learning network with initialization parameters to generate a third estimated cost description label. For example, for the facade decorative components with incorrectly estimated sizes, an inaccurate third estimated cost description label will be generated according to the incorrect size when annotating. Then, based on the third estimated cost component element and the third estimated cost description label of each third template BIM engineering cost dynamic data, the annotation result of each third template BIM engineering cost dynamic data is generated, and this result contains the inaccurate information in the extraction and annotation process.

[0134] Step S304, determining the third network learning error based on the third annotation results of each third template BIM engineering cost dynamic data and the characteristic distance between the template annotation data, and training the standardized deep learning network of the initialization parameters based on the third network learning error.

[0135] For example, for the facade decoration components, the wrong component elements and inaccurate cost description labels in the third annotation results are compared with the correct actual cost component elements and actual cost description labels in the template annotation data, and the feature distances are calculated from aspects such as component integrity and cost numerical accuracy. These feature distances are combined to determine the third network learning error. According to this third network learning error, the standardized deep learning network with initialization parameters is trained. During the training process, the standardized deep learning network will adjust various parameters within it, such as neuron connection weights, coefficients in the algorithm, etc., to improve the accuracy of cost component element extraction and annotation. With the learning of more third template BIM engineering cost dynamic data, the performance of the standardized deep learning network continues to improve, and it can more accurately handle the cost component element extraction and annotation tasks in the target BIM engineering cost dynamic data.

[0136] Figure 2 The hardware structure of the BIM-based engineering cost dynamic data annotation system 100 for implementing the BIM-based engineering cost dynamic data annotation method provided in an embodiment of the present invention is shown as follows: Figure 2 As shown, the engineering cost dynamic data annotation system 100 based on BIM technology may include a processor 110 , a machine-readable storage medium 120 , a bus 130 and a communication unit 140 .

[0137] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the BIM-based engineering cost dynamic data annotation system 100 uses to execute or use to complete the exemplary method described in the present invention.

[0138] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the engineering cost dynamic data labeling method based on BIM technology in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0139] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned engineering cost dynamic data annotation system 100 based on BIM technology. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0140] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the above-mentioned method for dynamic data labeling of engineering cost based on BIM technology is implemented.

[0141] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, various features are sometimes combined into one embodiment, drawing or description thereof.

Claims

1. A method for dynamic data annotation of engineering cost based on BIM technology, characterized in that: The method comprises: Acquire target BIM engineering cost dynamic data to be annotated, extract cost component elements from the target BIM engineering cost dynamic data through a standardized deep learning network, generate a first confidence sequence associated with a first preset cost mapping knowledge set, extract cost component elements from the target BIM engineering cost dynamic data through a customized deep learning network, generate a second confidence sequence associated with the first preset cost mapping knowledge set, wherein the first preset cost mapping knowledge set includes a plurality of reference cost component elements; According to the first confidence sequence and the second confidence sequence, a target cost component element in the target BIM engineering cost dynamic data is obtained; According to the target BIM engineering cost dynamic data and the target cost component element, the target cost component element is annotated by the standardized deep learning network to generate a third confidence sequence associated with a second preset cost mapping knowledge set, and the target cost component element is annotated by the annotation network to generate a fourth confidence sequence associated with the second preset cost mapping knowledge set, wherein the second preset cost mapping knowledge set includes a plurality of reference cost description tags; Obtaining a target cost description label of the target cost component element according to the third confidence sequence and the fourth confidence sequence; Based on the target cost component element and the target cost description tag of the target cost component element, the annotation result of the target BIM engineering cost dynamic data is generated.

2. The method for dynamic data annotation of construction cost based on BIM technology according to claim 1 is characterized in that: The customized deep learning network is generated by learning network parameters based on the characteristic distance between the first estimated cost component element and the actual cost component element of each first template BIM engineering cost dynamic data; the annotation network is generated by learning network parameters based on the characteristic distance between the second estimated cost description label and the actual cost description label of each second template BIM engineering cost dynamic data; The standardized deep learning network is generated by learning network parameters based on the characteristic distance between the third annotation results of each third template BIM engineering cost dynamic data and the template annotation data, the third annotation results include the third estimated cost component elements and the third cost description labels, and the template annotation data include the actual cost component elements and the actual cost description labels of the third template BIM engineering cost dynamic data.

3. The method for dynamic data annotation of construction cost based on BIM technology according to claim 2 is characterized in that: The method further comprises: According to the target BIM engineering cost dynamic data, cost component elements are extracted from the target BIM engineering cost dynamic data through a labeling network to generate a fifth confidence sequence associated with the first preset cost mapping knowledge set; According to the target BIM engineering cost dynamic data and the target cost component element, the target cost component element is labeled by the customized deep learning network to generate a sixth confidence sequence associated with the second preset cost mapping knowledge set; Wherein, obtaining the target cost component element in the target BIM engineering cost dynamic data according to the first confidence sequence and the second confidence sequence includes: Determining a first extraction loss of the second confidence sequence according to a feature distance between the first confidence sequence and the second confidence sequence, and determining a second extraction loss of the fifth confidence sequence according to a feature distance between the first confidence sequence and the fifth confidence sequence; Determining a first revised confidence for the first confidence sequence according to a deviation value between the second confidence sequence and the first extraction loss, and revising the first confidence sequence according to the first revised confidence; Determine a second target confidence sequence according to the deviation value between the modified first confidence sequence and the second extraction loss, and obtain the target cost component element according to the second target confidence sequence; Wherein, obtaining the target cost description label of the target cost component element according to the third confidence sequence and the fourth confidence sequence includes: The target cost description label corresponding to the target cost component element is obtained according to the third confidence sequence, the fourth confidence sequence and the sixth confidence sequence.

4. The method for dynamic data annotation of construction cost based on BIM technology according to claim 1 is characterized in that: The method further comprises: Obtain auxiliary guidance data for dynamic data annotation requests for engineering cost; The auxiliary guidance data includes component extraction guidance data and component annotation guidance data, wherein the component extraction guidance data is used to guide the extraction of cost component elements of the target cost knowledge domain in the target BIM engineering cost dynamic data, and the component annotation guidance data is used to guide the annotation of the extracted cost component elements; The first confidence sequence is obtained by extracting cost component elements from the target BIM engineering cost dynamic data and component extraction guidance data through a standardized deep learning network; the second confidence sequence is obtained by extracting cost component elements from the target BIM engineering cost dynamic data and component extraction guidance data through a customized deep learning network; The third confidence sequence is obtained by annotating the target cost component elements through a standardized deep learning network based on the target BIM project cost dynamic data, target cost component elements and component annotation guidance data; the fourth confidence sequence is obtained by annotating the target cost component elements through a annotation network based on the target BIM project cost dynamic data, target cost component elements and component annotation guidance data.

5. The method for dynamic data annotation of construction cost based on BIM technology according to claim 4 is characterized in that: The step of obtaining auxiliary guidance data for the engineering cost dynamic data annotation request includes: Acquire target data from the source data node corresponding to the engineering cost dynamic data annotation request, wherein the target data includes the type of engineering project involved, the usage scenario of the cost data, and the final application purpose of the annotation data; Determining a target data source according to the target data, wherein the target data source includes an existing engineering cost database, a project document library, an industry standard specification library, and a related expert experience knowledge library; Establishing data screening rules according to the marking target of the engineering cost dynamic data marking request, and extracting an initial engineering cost related data set from the target data source based on the data screening rules; Performing feature analysis on the initial engineering cost related data set to generate data feature analysis results, the data feature analysis results including structural features, semantic features, time series features and numerical features of the initial engineering cost related data set, as well as representation methods and association relationships of cost component elements in different data sources, semantic changes and consistency requirements of cost description labels in different contexts, and differences in description labels of the same cost component in different engineering projects. The time series features include dynamic change features of cost with project progress, and the impact of the dynamic change features on the extraction and annotation of cost component elements; According to the data feature analysis result, determining the type of auxiliary guidance data of the engineering cost dynamic data annotation request, the type including: rule-based guidance data, example-based guidance data, and association-based guidance data; According to the type of the auxiliary guidance data, a structural framework of the guidance data is constructed, wherein the rule-based guidance data is classified and organized according to the cost knowledge domain, the example-based guidance data is hierarchically stored according to the component type and the project type, and the association-based guidance data is constructed in the form of a relationship map; Determining the hierarchical structure of the auxiliary boot data to establish dependency relationships and calling sequences between data at different levels; According to the determined guide data type and the hierarchical structure of the auxiliary guide data, the corresponding target data content is extracted from the initial engineering cost related data set, and the extracted target data content of different types is integrated according to the constructed structural framework to construct the auxiliary guide data.

6. The method for dynamic data annotation of construction cost based on BIM technology according to claim 1 is characterized in that: The step of obtaining the target cost component element in the target BIM engineering cost dynamic data according to the first confidence sequence and the second confidence sequence includes: Determining a first extraction loss of the second confidence sequence according to a feature distance between the first confidence sequence and the second confidence sequence; Determining a first modified confidence for the first confidence sequence according to a deviation value between the second confidence sequence and the first extraction loss; Correcting the first confidence sequence according to the first corrected confidence to generate a first target confidence sequence; The target cost component element is obtained according to the first target confidence sequence.

7. The method for dynamic data annotation of construction cost based on BIM technology according to any one of claims 1 to 6, characterized in that: The step of determining the first confidence sequence and the second confidence sequence comprises: According to the target BIM engineering cost dynamic data, cost component elements are extracted from the target BIM engineering cost dynamic data through a standardized deep learning network to generate a first basic confidence sequence associated with a first preset cost mapping knowledge set, and cost component elements are extracted from the target BIM engineering cost dynamic data through a customized deep learning network to generate a second basic confidence sequence associated with the first preset cost mapping knowledge set; Wherein, the first basic confidence sequence includes the first basic confidence of each reference cost component element, and the second basic confidence includes the second basic confidence of each reference cost component element; According to the first basic confidence sequence and the second basic confidence sequence, extracting target reference component elements whose first basic confidence and second basic confidence are both greater than or equal to a set confidence from the reference cost component elements; The confidences corresponding to the non-target reference component elements in the first basic confidence sequence and the second basic confidence sequence are respectively eliminated to generate the first confidence sequence and the second confidence sequence.

8. The method for dynamic data annotation of construction cost based on BIM technology according to claim 3 is characterized in that: The target BIM engineering cost dynamic data includes a plurality of target cost component elements, and the annotation result includes a plurality of target cost component elements and target cost description tags of the target cost component elements; The annotation result is generated by enabling x rounds of deep learning process, each round of deep learning process is a round of deep learning component extraction process or a round of deep learning component annotation process, and the deep learning result of each round of deep learning process is a target cost component element or a target cost description label; Among them, the specific steps of the a-th round of deep learning process include: According to the deep learning result of the a-1th round of deep learning process, determine the output category of the result to be output, the output category is the component extraction category or the component annotation category, and the output category corresponding to the first round of deep learning process is the component extraction category; According to the target BIM engineering cost dynamic data and the deep learning results of the previous a-1 round of deep learning process, a standardized confidence sequence is obtained through a standardized deep learning network, a customized confidence sequence is obtained through a customized deep learning network, and a labeled confidence sequence is obtained through a labeled network; Wherein, the standardized confidence sequence is the first confidence sequence or the third confidence sequence, the customized confidence sequence is the second confidence sequence or the sixth confidence sequence, and the annotated confidence sequence is the fourth confidence sequence or the fifth confidence sequence; Generate a deep learning result of the a-th round of deep learning process based on the standardized confidence sequence, the customized confidence sequence and the labeled confidence sequence.

9. The method for dynamic data annotation of construction cost based on BIM technology according to any one of claims 3 to 6, characterized in that: The training steps of the customized deep learning network include: Acquire a first sample learning data sequence, wherein the first sample learning data sequence includes a plurality of first template BIM engineering cost dynamic data carrying template annotation data, wherein the template annotation data of each first template BIM engineering cost dynamic data includes an actual cost component element of the first template BIM engineering cost dynamic data; According to each first template BIM engineering cost dynamic data, extract the cost component element of each first template BIM engineering cost dynamic data through a customized deep learning network with initialized parameters to generate a first estimated cost component element of each first template BIM engineering cost dynamic data; According to each of the first template BIM engineering cost dynamic data and the first estimated cost component element, the first estimated cost component element is labeled through a customized deep learning network with initialization parameters to generate a first estimated cost description label of the first estimated cost component element; according to the first estimated cost component element and the first estimated cost description label of each of the first template BIM engineering cost dynamic data, a labeling result of each of the first template BIM engineering cost dynamic data is generated; Determine a first network learning error based on a characteristic distance between a first estimated cost component element and an actual cost component element of each first template BIM engineering cost dynamic data, and train a customized deep learning network of initialization parameters based on the first network learning error; The training steps of the labeling network include: Acquire a second sample learning data sequence, wherein the second sample learning data sequence includes a plurality of second template BIM engineering cost dynamic data carrying template annotation data, wherein the template annotation data of each second template BIM engineering cost dynamic data includes an actual cost description label of each actual cost component element of the second template BIM engineering cost dynamic data; According to each second template BIM engineering cost dynamic data, extract the cost component element of each second template BIM engineering cost dynamic data through the annotation network of the initialization parameter, and generate the second estimated cost component element of each second template BIM engineering cost dynamic data; According to each second template BIM engineering cost dynamic data and the second estimated cost component element, the second estimated cost component element is labeled through a labeling network with initialized parameters to generate a second estimated cost description label of the second estimated cost component element, and according to the second estimated cost component element and the second estimated cost description label of each second template BIM engineering cost dynamic data, a labeling result of each second template BIM engineering cost dynamic data is generated; Determine a second network learning error according to a characteristic distance between a second estimated cost description label and an actual cost description label of each second template BIM engineering cost dynamic data, and train a labeling network of initialization parameters according to the second network learning error; The training steps of the standardized deep learning network include: Acquire a third sample learning data sequence, wherein the third sample learning data sequence includes a plurality of third template BIM engineering cost dynamic data carrying template annotation data, wherein the template annotation data of each third template BIM engineering cost dynamic data includes an actual cost component element of the third template BIM engineering cost dynamic data and an actual cost description label of each actual cost component element; According to each third template BIM engineering cost dynamic data, extract the cost component element of each third template BIM engineering cost dynamic data through a standardized deep learning network with initialized parameters to generate a third estimated cost component element of each third template BIM engineering cost dynamic data; According to each third template BIM engineering cost dynamic data and the third estimated cost component element, the third estimated cost component element is labeled by a standardized deep learning network with initialized parameters to generate a third estimated cost description label of the third estimated cost component element, and according to the third estimated cost component element and the third estimated cost description label of each third template BIM engineering cost dynamic data, a labeling result of each third template BIM engineering cost dynamic data is generated; The third network learning error is determined based on the third annotation results of each third template BIM engineering cost dynamic data and the characteristic distance between the template annotation data, and the standardized deep learning network of the initialization parameters is trained based on the third network learning error.

10. A dynamic data annotation system for engineering cost based on BIM technology, characterized in that: The engineering cost dynamic data labeling system based on BIM technology includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the engineering cost dynamic data labeling method based on BIM technology as described in any one of claims 1 to 9.

Images