Construction period prediction network training method, application method and visual platform

Abstract

The invention provides a construction period prediction network training method, an application method and a visual platform, and belongs to the field of engineering prediction management. The training method comprises the following steps: dividing acquired historical construction period data into a plurality of staged construction period data according to project stages, and constructing a construction period prediction sub-data set of each project stage according to each staged construction period data; and sequentially inputting the construction period prediction sub-data sets into the initial construction period prediction network, carrying out adaptive amplitude parameter fine tuning on the initial construction period prediction network, and iterating the parameter fine tuning process until the model performance reaches a preset threshold value, thereby obtaining a completely trained construction period prediction network. According to the method, the historical construction period data is divided into a plurality of stage data, so that a more efficient data set can be established; through self-adaptive amplitude parameter fine tuning, training weights of data with different qualities are reasonably adjusted, and the accuracy of project period prediction is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering forecasting management, and in particular to a construction period forecasting network training method, an application method and a visualization platform. Background Art

[0002] In the construction of large-scale industrial facilities (such as cement plant construction projects), the projects involve multiple complex links and a large amount of resource coordination. The demand for accurate prediction and efficient management of construction periods is increasing. Traditional construction period prediction methods mostly rely on the experience of project managers and historical project data. However, in the face of dynamic construction environments and uncontrollable factors such as weather changes and material supply issues, real-time adjustments are often impossible, resulting in large deviations in construction periods and affecting progress and cost control.

[0003] Large language models, with their ability to process and analyze massive amounts of data, can be fine-tuned to specific domain parameters, enabling them to mine historical project data and analyze multivariate relationships, providing a new approach for construction duration prediction. However, construction project duration prediction often involves varying factors affecting different phases, and data quality can also vary. For example, during the project preparation phase, factors such as the material procurement cycle and the frequency of design changes significantly impact project duration, but these factors exhibit high uncertainty and poor data quality. During the construction phase, however, factors such as personnel skill levels and the availability of construction equipment have a significant impact, but these factors exhibit low uncertainty and, therefore, relatively high data quality. However, existing methods for fine-tuning large language models struggle to effectively account for the varying factors and data quality differences across different phases during the model adjustment process, hindering accurate project duration prediction.

[0004] Therefore, it is urgent to provide a construction duration prediction network training method, application method and visualization platform to improve the accuracy of project duration prediction. Summary of the Invention

[0005] In view of this, it is necessary to provide a construction duration prediction network training method, application method and visualization platform to solve the technical problems existing in the existing technology that are unable to effectively pay attention to the changes in influencing factors of data at different stages and the differences in data quality, resulting in the inability to accurately predict the project duration.

[0006] In a first aspect, the present invention provides a construction period prediction network training method, comprising: The historical construction duration data obtained is divided into several phased construction duration data according to the project phases, and the construction duration prediction sub-datasets of each project phase are constructed based on the phased construction duration data; Input the construction period prediction sub-dataset into the initial construction period prediction network in sequence, perform adaptive amplitude parameter fine-tuning on the initial construction period prediction network, and iterate the parameter fine-tuning process until the model performance reaches the preset threshold to obtain a well-trained construction period prediction network; Among them, the adaptive amplitude parameter fine-tuning includes: performing parameter fine-tuning with different training gradients on the initial construction period prediction network at different project stages.

[0007] In some possible implementation manners, the project stages include the preliminary preparation stage, the mid-term construction stage, and the late acceptance stage.

[0008] In some possible implementation manners, constructing the construction period prediction sub-dataset for each project stage according to the construction period data of each sub-stage includes: Performing data screening on the construction period data of each sub-stage according to the preset key influencing factors corresponding to each project stage to obtain screened data; Performing data formatting, word segmentation, data cleaning, and word segmentation encoding on the screened data in sequence to obtain the construction period prediction sub-dataset for each project stage.

[0009] In some possible implementation manners, the initial construction period prediction network is built based on a pre-trained large language model, and the historical construction period data includes project construction data in several different modalities.

[0010] In some possible implementation manners, performing adaptive amplitude parameter fine-tuning on the initial construction period prediction network includes: Determining the fine-tuning amplitude according to the project stage corresponding to the construction period prediction sub-dataset; Adjusting the parameter adjustment gradient of the low-rank adapter according to the fine-tuning amplitude, and performing low-rank adaptation parameter fine-tuning on the initial construction period prediction network according to the low-rank adapter.

[0011] In some possible implementation manners, determining the fine-tuning amplitude according to the project stage corresponding to the construction period prediction sub-dataset includes: Determining the data quality weight according to the project stage corresponding to the construction period prediction sub-dataset, and determining the data volume weight according to the data volume of the construction period prediction sub-dataset; Determining the fine-tuning amplitude according to the data quality weight and the data volume weight.

[0012] In some possible implementation manners, the preliminary preparation stage corresponds to a first fine-tuning amplitude, the mid-term construction stage corresponds to a second fine-tuning amplitude, the late acceptance stage corresponds to a third fine-tuning amplitude, and the second fine-tuning amplitude is greater than the first fine-tuning amplitude and the third fine-tuning amplitude.

[0013] On the second aspect, the present invention also provides a method for applying a construction period prediction network, including: Obtaining the multiple influencing factors of the project to be predicted; Input the multiple influencing factors into the trained construction duration prediction network, and output the construction duration prediction result; Among them, the trained construction duration prediction network is determined according to the training of the construction duration prediction network in any one of the above.

[0014] In a third aspect, the present invention also provides a visualization platform, including a processor, a memory, and a computer program stored on the memory and executable on the processor, characterized in that when the processor executes the program, it realizes the construction duration prediction network training method according to any one of the above and / or the construction duration prediction network application method according to the above.

[0015] In a fourth aspect, the present invention also provides a computer-readable storage medium, on which a computer program is stored, characterized in that when the computer program is executed by a processor, it realizes the construction duration prediction network training method according to any one of the above and / or the construction duration prediction network application method according to the above.

[0016] The beneficial effects of adopting the above embodiments are as follows: In the construction duration prediction network training method provided by the present invention, by dividing the historical construction duration data into multiple stage data, the influencing factors of each stage can be combined during the construction of the data set to establish a more efficient data set; through the adaptive amplitude parameter fine-tuning of each construction duration prediction sub-data set, the parameter fine-tuning amplitude of different quality data can be dynamically adjusted during the parameter fine-tuning of the large language model, and the training weights of different quality data can be reasonably adjusted, effectively improving the accuracy of the project construction duration prediction. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0018] Figure 1 It is a flowchart of an embodiment of the construction duration prediction network training method proposed by the present invention; Figure 2 It is a flowchart of constructing a construction duration prediction sub-data set in an embodiment of the present invention; Figure 3 It is a flowchart of the adaptive amplitude parameter fine-tuning in an embodiment of the present invention; Figure 4 It is a flowchart of determining the fine-tuning amplitude in an embodiment of the present invention; Figure 5 It is a flowchart of an embodiment of the construction duration prediction network application method provided by the present invention; Figure 6 It is a schematic structural diagram of an embodiment of the visualization platform provided by the present invention. Detailed implementation manners

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present invention.

[0020] It should be understood that the schematic drawings are not drawn to actual scale. The flowcharts used in the present invention illustrate the operations implemented according to some embodiments of the present invention. It should be understood that the operations in the flowcharts may not be implemented in sequence, and the steps without logical context relationships may be reversed or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart or remove one or more operations from the flowchart under the guidance of the content of the present invention. Some of the block diagrams shown in the drawings are functional entities, which do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor systems and / or microcontroller systems.

[0021] The descriptions such as "first" and "second" involved in the embodiments of the present invention are only for descriptive purposes, and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Therefore, the technical features defined with "first" and "second" may explicitly or implicitly include at least one such feature.

[0022] Referring to "embodiments" herein means that the specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of the present invention. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0023] The present invention provides a construction period prediction network training method, an application method, and a visualization platform, which will be described separately below.

[0024] It should be noted that parameter fine-tuning refers to the process of optimizing the model performance by adjusting the model parameters, which can effectively reduce the computational amount in the process of adjusting the parameters of a large model.

[0025] Figure 1A flowchart of an embodiment of the construction period prediction network training method proposed by the present invention is shown as follows Figure 1 The construction period prediction network training method includes: S101. Divide the obtained historical construction period data into several sub-period construction period data according to the project stage, and construct a construction period prediction sub-dataset for each project stage based on the sub-period construction period data; Among them, the historical construction period data includes several influencing factor data of each project in the historical project and the corresponding construction period data. In the embodiment, the historical construction period data is divided according to the project stage, so that the data corresponding to the relatively important influencing factors in different project stages can be adaptively retained during the construction of each sub-dataset, and the useless data can be deleted, thereby establishing a more efficient dataset.

[0026] S102. Input the construction period prediction sub-dataset into the initial construction period prediction network in sequence, perform adaptive amplitude parameter fine-tuning on the initial construction period prediction network, and iterate the parameter fine-tuning process until the model performance reaches a preset threshold to obtain a trained and complete construction period prediction network; Among them, the adaptive amplitude parameter fine-tuning includes: performing parameter fine-tuning with different training gradients on the initial construction period prediction network in different project stages.

[0027] Among them, in the embodiment, by using each construction period prediction sub-dataset to perform adaptive amplitude parameter fine-tuning on the initial construction period prediction network, the data quality differences of different stages of data can be fully considered during the parameter fine-tuning process. That is, for data with higher data quality, the parameter adjustment amplitude can be increased, and for data with lower data quality, the parameter adjustment amplitude can be appropriately reduced. Through adaptive amplitude parameter fine-tuning, in different project stages, by changing the parameters in the training process and adjusting the training gradient in the parameter fine-tuning process, the accuracy of the model in the construction period prediction task is effectively improved.

[0028] Compared with the prior art, in the construction period prediction network training method provided by the present invention, by dividing the historical construction period data into multiple stage data, the influencing factors of each stage can be combined during the construction of the dataset to establish a more efficient dataset; through adaptive amplitude parameter fine-tuning for each construction period prediction sub-dataset, the parameter fine-tuning amplitude of different quality data can be dynamically adjusted during the parameter fine-tuning of the large language model, and the training weights of different quality data can be reasonably adjusted, effectively improving the accuracy of the project construction period prediction.

[0029] In some embodiments of the present invention, the project stage includes a preliminary preparation stage, a mid-term construction stage, and a late acceptance stage.

[0030] Among them, to construct a higher-quality segmented sub-dataset, according to the characteristics of construction projects, the project phases can be divided into the preliminary preparation phase, the mid-term construction phase, and the late acceptance phase, and the key factors affecting the construction period in different phases are analyzed and determined. Among them, in the preliminary preparation phase, key attention is paid to resource allocation, building material procurement cycle, design change frequency, etc.; in the mid-term construction phase, attention is paid to personnel skill levels, construction equipment availability, environmental conditions, etc.; in the late acceptance phase, attention is paid to acceptance criteria, rectification efficiency, delivery process, etc. Through phased analysis, more targeted decision-making basis can be provided for the construction of the dataset.

[0031] In some embodiments of the present invention, Figure 2 is a schematic flowchart of the process for constructing the construction period prediction sub-dataset of the embodiment of the present invention, as Figure 2 shown, constructing the construction period prediction sub-dataset for each project phase according to the construction period data of each sub-phase, including: S201. Perform data screening on the construction period data of each sub-phase according to the preset key influencing factors corresponding to each project phase to obtain screened data; Among them, to ensure the effectiveness of the data in the construction period prediction sub-dataset for each phase, before constructing the construction period prediction sub-dataset, the embodiment performs data screening through the preset key influencing factors analyzed in advance, so as to select the important influencing factors that affect the construction period of the construction project in each phase. For example, in the mid-term construction phase, the influencing factors include personnel skill levels, the number of personnel, construction equipment availability, environmental conditions, the scale of the construction area, and the number of construction floors, etc.

[0032] S202. Perform data formatting, word segmentation processing, data cleaning, and word segmentation encoding on the screened data in sequence to obtain the construction period prediction sub-dataset for each project phase.

[0033] Among them, to ensure that the data format meets the model input requirements, the dataset construction process includes multiple steps such as data formatting, word segmentation processing, data cleaning, and word segmentation encoding.

[0034] Among them, data formatting converts the historical data file into a standardized JSON format to ensure the readability and structuring of the data; word segmentation processing uses an automatic word segmenter to perform word segmentation on the text data and defines system instructions and user questions to ensure the context coherence of the input; data cleaning removes the irrelevant parts in the input data; word segmentation encoding is for text data, which divides it into meaningful units, such as words or sub-word units, and then converts these divided units into digital representations, which may involve using a vocabulary to map each word to a unique ID.

[0035] In addition, for the convenience of model processing, the embodiments also add special markers (such as start and end) at the beginning and end of sentences; ensure that all input sequences have the same length by padding shorter sequences or truncating longer sequences; and provide corresponding labels for each input sequence, that is: if the outputs at certain positions should not be considered in the loss calculation (for example, because they correspond to padded parts), then a special label, such as -100, can be assigned to these positions to indicate that they should be ignored.

[0036] In some embodiments of the present invention, the initial construction period prediction network is built based on a pre-trained large language model, and the historical construction period data includes project construction data in several different modalities.

[0037] Specifically, in the embodiments, the initial construction period prediction network is obtained by importing the pre-trained Qianwen large language model. Among them, in the embodiments, the pre-trained large language model is further pre-trained using building materials industry corpus on the basis of a general pre-trained model to improve the model's understanding ability of industry-specific knowledge. In the model import, an adaptive tokenizer based on building materials industry-specific terms and contexts is also set to ensure the accurate parsing of professional terms.

[0038] In addition, in the data collection stage, to obtain richer historical data, the embodiments collect and organize the construction period data of completed projects and ongoing projects, and generate a comprehensive data set containing construction period influencing factors. In particular, data sources unique to the building materials industry (such as delivery records of building materials suppliers, building materials quality inspection reports, on-site construction environment monitoring data, etc.) are introduced, and a systematic construction period prediction database is established. The data sparsity problem in the building materials industry is solved through data augmentation techniques (such as synthetic data generation or data interpolation) to provide a stable and diverse source for the input data of the model. At the same time, on the basis of traditional text data, multi-modal data (such as pictures, videos, sensor data, etc.) is introduced to capture the real-time state of the construction site and the actual usage of building materials. Through multi-modal feature extraction techniques (such as image recognition, time series analysis, etc.), and convert unstructured data into structured features, and fuse them with text data to form a more comprehensive input feature set.

[0039] In some embodiments of the present invention, Figure 3 is a schematic flow diagram of the adaptive amplitude parameter fine-tuning for the embodiments of the present invention. As Figure 3 shown, performing adaptive amplitude parameter fine-tuning on the initial construction period prediction network includes: S301. Determine the fine-tuning amplitude according to the project phase corresponding to the construction period prediction sub-data set; S302. Adjust the parameter adjustment gradient of the low-rank adapter according to the fine-tuning amplitude, and perform low-rank adaptation parameter fine-tuning on the initial construction period prediction network according to the low-rank adapter.

[0040] Among them, in the process of adaptive amplitude parameter fine-tuning, the embodiment adopts the LoRA (Low-Rank Adaptation) fine-tuning method, which adjusts the behavior of the pre-trained model by introducing a small, task-specific network (i.e., the adapter), rather than directly modifying or retraining all the parameters of the entire pre-trained model. This way can significantly reduce the demand for computing resources and help maintain the knowledge already learned in the pre-trained model. The low-rank adapter formula is expressed as:

[0041] Among them, is the original weight matrix, and are low-rank matrices of rank . In the embodiment, by selecting the attention layer and the feed-forward layer in the construction period prediction, low-rank decomposition is applied to them.

[0042] In the low-rank adaptation parameter fine-tuning, it is necessary to adjust the adapter parameters, including: setting the neuron inactivation ratio of LoRA to prevent overfitting, using the LoRA adapter to integrate the low-rank matrices and into the model, and using backpropagation for gradient update. The update formula can be expressed as , where represents the learning rate, is the loss function. The embodiment changes the training gradient in the parameter fine-tuning process by adaptively adjusting the adapter parameters, so as to set different fine-tuning amplitudes for different project phases.

[0043] By adopting the above method, for the dataset with relatively low data quality, a relatively low fine-tuning amplitude can be set, which can reduce the impact of low-quality data on the model accuracy. For the data with relatively high data quality, a larger fine-tuning amplitude is set, so as to make full use of high-quality data in the model training process.

[0044] In some embodiments of the present invention, Figure 4 is a schematic flow chart for determining the fine-tuning amplitude in the embodiment of the present invention. As shown in Figure 4 , determining the fine-tuning amplitude according to the project phase corresponding to the construction period prediction sub-dataset includes: S401. Determine the data quality weight according to the project phase corresponding to the construction period prediction sub-dataset, and determine the data volume weight according to the data volume of the construction period prediction sub-dataset; S402. Determine the fine-tuning amplitude according to the data quality weight and the data volume weight.

[0045] In some embodiments of the present invention, the preliminary preparation stage corresponds to a first fine-tuning amplitude, the mid-term construction stage corresponds to a second fine-tuning amplitude, and the final acceptance stage corresponds to a third fine-tuning amplitude. The second fine-tuning amplitude is greater than the first fine-tuning amplitude and the third fine-tuning amplitude.

[0046] Among them, in order to determine more reasonable fine-tuning amplitudes for each stage, for each project stage, the embodiment considers two aspects: data quality and data volume, respectively sets data quality weights and data volume weights, and comprehensively considers the data quality weights and data volume weights to adjust the fine-tuning amplitude.

[0047] In the embodiment, for the preliminary preparation stage of the project, since the uncertainty of the preliminary data is relatively high and the data volume is also relatively small, the fine-tuning amplitude is set to be small. At this time, it focuses on the adaptability of the model to macro factor influencing factors such as resource allocation and design changes. For the mid-term construction stage of the project, considering that the mid-term data is relatively rich and stable, compared with other stages, a larger fine-tuning amplitude is set in the mid-term stage. At this time, it focuses on the accurate prediction ability of the model for the project duration using real-time factor influencing factors such as construction progress, personnel efficiency, and equipment availability. For the final acceptance stage of the project, at this time, the late-stage data tends to be stable, but the data volume is not large, so a small fine-tuning amplitude is also set, focusing on the optimization of the model for detail factors such as acceptance criteria and rectification efficiency. Among the fine-tuning amplitudes of the three stages, the fine-tuning amplitude set in the mid-term stage is significantly higher than that in the preliminary stage and the late stage. In addition, although both the preliminary stage and the late stage set small fine-tuning amplitudes, if the data volumes are the same, considering the data quality, the fine-tuning amplitude in the late stage is higher than that in the preliminary stage.

[0048] Adopting the above-mentioned phased dynamic fine-tuning method can effectively ensure that the model can provide high-precision prediction results in different stages.

[0049] In summary, in the construction duration prediction network training method provided by the present invention, by dividing the historical construction duration data into multiple stage data, the influencing factors of each stage can be combined in the process of constructing the data set to establish a more efficient data set; through the adaptive amplitude parameter fine-tuning of each construction duration prediction sub-data set, the parameter fine-tuning amplitude of different quality data can be dynamically adjusted in the process of fine-tuning the large language model parameters, and the training weights of different quality data can be reasonably adjusted, effectively improving the accuracy of the project duration prediction.

[0050] In addition, the present invention also provides a construction duration prediction network application method, combined with Figure 5 Looking at Figure 5 is a schematic flowchart of an embodiment of the construction duration prediction network application method provided by the present invention. As Figure 5 shown, the construction duration prediction network application method includes: S501. Obtain the multiple influencing factors of the project to be predicted; S502. Input the multiple influencing factors into the well-trained construction duration prediction network, and output the duration prediction result; Among them, the well-trained construction duration prediction network is determined according to the above-mentioned construction duration prediction network training.

[0051] In an embodiment of the present invention, first, effectively obtain the multiple influencing factors of the project to be predicted, and then use the above-mentioned well-trained construction duration prediction network to effectively predict the duration, and the duration prediction result can be output.

[0052] As Figure 6 shown, the present invention also correspondingly provides a visualization platform 600, which includes a processor 601, a memory 602 and a display 603. Figure 6 Only some components of the visualization platform 600 are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0053] The memory 602 can be an internal storage unit of the visualization platform 600 in some embodiments, such as the hard disk or memory of the visualization platform 600. The memory 602 can also be an external storage device of the visualization platform 600 in other embodiments, such as a plug-in hard disk equipped on the visualization platform 600, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc.

[0054] The processor 601 can be a Central Processing Unit (CPU), a microprocessor or other data processing chips in some embodiments, and is used to run the program code stored in the memory 602 or process data, such as the construction duration prediction network training method and / or the construction duration prediction network application method in the present invention.

[0055] The display 603 can be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. in some embodiments. The display 603 is used to display the information of the visualization platform 600 and to display the visual user interface. The components 601-603 of the visualization platform 600 communicate with each other through a system bus.

[0056] In an embodiment, when the processor 601 executes the construction duration prediction program in the memory 602, the following steps can be implemented: The obtained historical construction duration data is divided into several sub-stage construction duration data according to the project stages, and the duration prediction sub-datasets for each project stage are constructed based on the sub-stage construction duration data. The duration prediction sub-datasets are sequentially input into the initial construction duration prediction network, and the adaptive amplitude parameter of the initial construction duration prediction network is finely tuned. The parameter fine-tuning process is iterated until the model performance reaches a preset threshold, and a well-trained construction duration prediction network is obtained.

[0057] and / or implement the following steps: Obtain the multiple influencing factors of the project to be predicted. Input the multiple influencing factors into the well-trained construction duration prediction network, and the duration prediction result is output.

[0058] It should be understood that when the processor 601 executes the construction duration prediction program in the memory 602, in addition to the above functions, other functions can also be implemented. For details, refer to the description of the corresponding method embodiments above.

[0059] Correspondingly, an embodiment of the present application further provides a computer-readable storage medium, which is used to store computer-readable programs or instructions. When the programs or instructions are executed by a processor, the steps or functions in the construction duration prediction network training method and / or the construction duration prediction network application method provided by the above method embodiments can be implemented.

[0060] Those skilled in the art can understand that all or part of the processes for implementing the methods in the above embodiments can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program. The computer program can be stored in a computer-readable storage medium. The computer-readable storage medium is a magnetic disk, an optical disk, a read-only memory, or a random access memory, etc.

[0061] The above has introduced in detail the construction duration prediction network training method, application method and visualization platform provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for training a construction period prediction network, characterized in that Including: Dividing the obtained historical construction duration data into several sub - stage construction duration data according to the project stages, and constructing a duration prediction sub - dataset for each project stage based on each of the sub - stage construction duration data; Sequentially inputting the duration prediction sub - dataset into the initial construction duration prediction network, performing adaptive amplitude parameter fine - tuning on the initial construction duration prediction network, and iterating the parameter fine - tuning process until the model performance reaches a preset threshold to obtain a trained - complete construction duration prediction network; Among them, the adaptive amplitude parameter fine - tuning includes: performing parameter fine - tuning with different training gradients on the initial construction duration prediction network in different project stages.

2. The construction period prediction network training method according to claim 1, wherein, The project stages include the preliminary preparation stage, the mid - term construction stage, and the late acceptance stage.

3. The construction period prediction network training method according to claim 1, characterized in that The constructing a duration prediction sub - dataset for each project stage based on each of the sub - stage construction duration data includes: Performing data screening on each of the sub - stage construction duration data according to the preset key influencing factors corresponding to each project stage to obtain screened data; Sequentially performing data formatting, word segmentation, data cleaning, and word segmentation encoding on the screened data to obtain a duration prediction sub - dataset for each project stage.

4. The construction period prediction network training method according to claim 1, wherein, The initial construction duration prediction network is built based on a pre - trained large - language model, and the historical construction duration data includes project construction data in several different modalities.

5. The construction period prediction network training method according to claim 1, wherein, The performing adaptive amplitude parameter fine - tuning on the initial construction duration prediction network includes: Determining the fine - tuning amplitude according to the project stage corresponding to the duration prediction sub - dataset; Adjusting the parameter adjustment gradient of the low - rank adapter according to the fine - tuning amplitude, and performing low - rank adaptation parameter fine - tuning on the initial construction duration prediction network according to the low - rank adapter.

6. The construction period prediction network training method according to claim 5, wherein The determining the fine - tuning amplitude according to the project stage corresponding to the duration prediction sub - dataset includes: Determining the data quality weight according to the project stage corresponding to the duration prediction sub - dataset, and determining the data volume weight according to the data volume of the duration prediction sub - dataset; Determining the fine - tuning amplitude according to the data quality weight and the data volume weight.

7. The construction period prediction network training method according to claim 2, wherein The preliminary preparation stage corresponds to a first fine - tuning amplitude, the mid - term construction stage corresponds to a second fine - tuning amplitude, the late acceptance stage corresponds to a third fine - tuning amplitude, and the second fine - tuning amplitude is greater than the first fine - tuning amplitude and the third fine - tuning amplitude.

8. A method for applying a construction period prediction network, characterized in that, Including: Obtaining the multiple influencing factors of the project to be predicted; Inputting the multiple influencing factors into the trained - complete construction duration prediction network and outputting a duration prediction result; Among them, the trained - complete construction duration prediction network is determined by training according to the construction duration prediction network described in any one of claims 1 to 7.

9. A visualization platform, comprising a processor, a memory, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the construction duration prediction network training method described in any one of claims 1 to 7 and / or the construction duration prediction network application method described in claim 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the construction duration prediction network training method described in any one of claims 1 to 7 and / or the construction duration prediction network application method described in claim 8.

Images