Building construction progress prediction method based on machine learning

Abstract

The invention provides a building construction progress prediction method based on machine learning, and relates to the technical field of machine learning. The invention provides a Build Proform prediction model, which comprises a data preprocessing and multi-dimensional feature modeling module, a state flow modeling module and a propulsion mapping and progress prediction module, a quantile converter method is introduced to improve the stability of data preprocessing, and in the multi-dimensional feature modeling stage, the stability of the data preprocessing is improved. Embedding features are constructed by adopting measurement space mapping, an algebraic topological structure, measurement transformation and a quantum state resource representation mode; in the state flow modeling module, a construction task is abstracted into a multi-dimensional state vector, a nonlinear progress evolution function and a resource consumption function are designed, and the natural propulsion process of the task along with time is simulated; and in the propulsion mapping and prediction module, a task propulsion potential energy function is constructed and aggregated into system potential energy, and all the modules operate cooperatively, so that accurate prediction and intelligent optimization of the building construction progress are realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and particularly relates to a method for predicting construction progress based on machine learning. Background Art

[0002] With the continuous expansion of the scale of construction projects and the increasing tightness of construction periods, the accurate prediction of construction progress is of great significance for project managers in project planning, resource allocation, and cost control. However, most of the existing construction progress prediction methods rely on manual experience and rule-based algorithms, lacking full consideration of the dynamic changes of various complex factors during the construction process, and it is difficult to provide accurate progress predictions, resulting in an increasingly serious problem of construction period delays and cost overruns. Therefore, there is an urgent need for a new type of construction progress prediction method to improve the controllability during the construction process and the accuracy of decision-making.

[0003] In recent years, with the rapid development of information technology and artificial intelligence technology, machine learning has been widely applied in the construction field, especially showing great potential in construction progress prediction. Machine learning algorithms can automatically learn through a large amount of historical data to identify complex factors and potential laws affecting construction progress. Compared with traditional methods, the construction progress prediction method based on machine learning can better adapt to the dynamic changes on the construction site and provide high-precision progress predictions.

[0004] However, the current construction progress prediction based on machine learning still faces many challenges. Construction progress is affected by various factors, and these factors have a high degree of uncertainty and complexity. How to select appropriate features and effectively integrate different types of data is the key to achieving high-precision prediction. At the same time, there are certain difficulties in data collection and processing during the construction process, especially in large-scale projects, where the real-time and accuracy of data are often difficult to guarantee. Therefore, the construction progress prediction method based on machine learning is still in the stage of continuous exploration and optimization, and there is an urgent need to propose an innovative technical solution to better meet the needs of construction project management. Summary of the Invention

[0005] The present invention provides a method for predicting construction progress based on machine learning. Aiming at complex and multi-variable construction data, the BuildProformer model is proposed, which consists of a data preprocessing and feature extraction module, a time series modeling module, and a prediction and optimization module.

[0006] The technical solutions adopted by the present invention to achieve the above object specifically include the following steps: S1. Collect construction-related data, that is, multi-source data, which is generated during the actual construction process; S2. Preprocess the construction-related data using the quantile transformer method and divide the dataset; S3. Propose the global task status feature, global time period feature, and resource dynamic feature, and construct a multi-dimensional feature modeling and embedding module. The specific steps are as follows: S31. Input the current status of the construction task. Propose to define the construction task status data in the measure space (X, μ), use algebraic topology theory to construct the topological mapping of the task status, and finally obtain the global task status feature H task ; S32. Propose to model the time period using harmonic analysis and measure transformation, define the Fourier measure transformation of the time period, calculate the time state using the Radon-Nikodym derivative, and finally obtain the global time period feature H period ; S33. Propose to map the dynamic changes of resources during the construction process to a low-dimensional embedding space, introduce the quantum state resource mapping method, and use the quantum weight adjustment factor to control the contribution of each resource, and finally obtain the resource dynamic feature H resource ; S4. Construct a state flow modeling module. The specific steps are as follows: S41. Abstract the construction task into a multi-dimensional state vector, design a non-linear progress evolution function, combine the resource intensity, perturbation factor, and construction depth to construct the task state flow trajectory, and simulate the natural evolution process of the construction task over time; S42. Define the resource consumption update strategy, and form a state trajectory sequence by the evolution of the task state on the time axis; S5. Construct a propulsion mapping and progress prediction module, design a task propulsion potential energy function and a progress prediction function, and perform dynamic prediction of the overall construction progress driven by the system potential energy, and finally obtain the construction progress prediction result

[0007] Preferably, in S1, the construction-related data includes historical construction progress data, construction plans, resource allocation situations, and external factor data. The historical construction progress data includes the start time, end time, duration, and completion percentage of each construction task; the construction plan data includes the sequence of tasks, milestone dates, and resource allocation plans; the resource data includes the usage and availability of manpower, equipment, and materials; the external factor data includes weather conditions and holiday information to comprehensively understand various data information in the construction process.

[0008] Preferably, in S2, use the quantile transformer method to preprocess the construction-related data. First, sort the input data to obtain X sorted，t, calculate the quantile rank of each data point, and the specific formula is: rank(x i , t) = arh(x i , t in X sorted，t ); In the formula, arg is the position of the data point in the sorted array, and then the quantile rank is mapped to the uniform distribution U(0, 1) to obtain y i，t , and the specific mathematical model is: In the formula, l is the number of data points in each time step. Finally, the data set is divided into a training set and a test set according to a certain ratio to ensure the effectiveness of model training and evaluation.

[0009] Preferably, the quantile converter method can effectively preprocess the data related to building construction, convert the data into a uniform distribution. This method not only improves the robustness of the model to outliers but also improves the performance of time series modeling.

[0010] Preferably, in S3 and S31, input the building construction data, extract the key state features of each construction task, including construction progress, task type, and task priority. Regard the construction task state as a distribution in the measure space, and its change can be modeled through measure transformation. First, define the construction task state data in the measure space (X, μ), where X is the feature space of the construction task, and μ is the measure of the task state, which is used to represent the probability density distribution of the task state. Then, propose the Cauchy measure transformation to obtain the global distribution of the task state. The specific mathematical model is: In the formula, μ T (x) is the measure value of the task state, representing the state distribution at the position, x is the input feature point of the task state, x0 is the central state of the task, and γ is the parameter controlling the width of the measure distribution. To consider the dependency relationship of the construction tasks, use algebraic topology theory to construct the topological mapping of the task state. The specific mathematical model is: In the formula, H k (X, R) is the homology group in algebraic topology, representing the topological structure of the task state. k is the dimension of the homology group, representing the mutual relationship between task states, is the boundary operator in algebraic topology, representing the connection method between task states, is the kernel of the boundary operator, representing the topological connection characteristics of the task state, is the image of the boundary operator, representing the topological relationship of the task state. Finally, obtain the global characteristics of the task state. The specific mathematical model is: Htask = ∑ i ω i H k (X, R); Where, ω i is the contribution weight of each homology group in the task state, and H task is the global task state feature.

[0011] Preferably, the construction task state data is defined on a measure space, and a topological mapping of the task state is constructed, which can characterize the complex dependence relationship between construction tasks from a global perspective. At the same time, by introducing measure transformation and topological structure extraction means, double modeling of the probability distribution and spatial structure of the construction task state is realized. The introduction of the topological structure makes the model more robust to local perturbations of the task state, and improves the modeling adaptability to complex and non-linear construction processes. Especially in a construction environment with multi-task parallelism and frequent dynamic changes, this method can effectively improve the understanding ability of the subsequent prediction module for the coordination and progress evolution between tasks, providing a solid structured feature basis for building construction progress prediction.

[0012] Preferably, in the S32 part of S3, the time information in construction usually appears in the form of date, hour, and minute, and contains periodic characteristics. First, these time information need to be converted into time feature vectors. Given that t is a time point in construction, the information of year, month, day, hour, and minute can be extracted. By using harmonic analysis and measure transformation to model the time period, the construction time data is transformed onto the measure space (T, v), where T is the time axis and v is the measure of the time distribution. Then, the measure transformation of the time period is defined, and the specific mathematical model is: v(ξ) = ∫ T e -2πieξ dv(x); Where, v(ξ) is the transformation result of the time measure, capturing the time change pattern in the construction progress, e is the time feature point, representing specific time data, ξ is the time frequency, representing the frequency of periodic changes, and dv is the probability distribution of the time feature. Then, in order to more accurately represent the construction time period and calculate the time state, the specific mathematical model is: Where, is the harmonic derivative of the time state, representing the rate of change of the time state relative to the standard Lebesgue measure, and λ is the standard Lebesgue measure, describing the change of the construction time period relative to the standard time axis. Finally, the global state of the construction time is obtained, and the specific mathematical model is: Where, H period is the global time period feature, through Fourier transform and weight ωi Calculated

[0013] Preferably, it shows significant advantages in the modeling of construction time cycle characteristics, mainly reflected in the high-precision capture ability of the periodicity and dynamic changes of time series. By introducing harmonic analysis and measure transformation, this method can not only reveal the periodic fluctuations in the construction progress from the frequency domain perspective, but also use measure transformation to extract the implicit laws in the time change pattern, and calculate the time state at the same time, enabling the model to accurately depict the change rate of the time distribution relative to the standard time axis, strengthening the dynamic understanding ability of time characteristics. In addition, the introduction of global time cycle characteristics enables the model to maintain a stable modeling effect when facing long-term construction plans or cross-stage construction arrangements, which has an important improvement on the accuracy and robustness of actual construction progress prediction.

[0014] Preferably, in part S33 of S3, the input construction resource information is combined into a preliminary resource feature vector R t =[L t ,E t ,M t , where R t is a feature vector containing human resources, equipment, and materials, describing the resource state at time point t. L t is the human resource required at the current moment, E t is the equipment resource required at the current moment, M t is the material resource required at the current moment. Then R t is mapped to a low-dimensional embedding space through an embedding layer. The specific mathematical model is: R′ t =Embedding(R t ); In the formula, R′ t is the representation of the resource feature vector in the low-dimensional embedding space. During the construction process, the dynamic changes of resources will change over time. Therefore, all the resource feature vectors at all time points are aggregated to obtain a complete resource dynamic feature matrix. The specific mathematical model is: In the formula, E′ res is a t×d matrix, where d is the dimension of each embedding vector. By introducing the quantum state resource mapping method, the dynamic characteristics of resources are further enhanced. The quantum state of each resource is represented as a quantum vector. The specific mathematical model is: |R i (t)>=α i (t)|L t >+β i (t)|E t >+γi (t)|M t >; Wherein, |R i (t)> is the quantum state representation of the i-th type of resource, α i , β i , γ i are the quantum amplitudes of the resource state, reflecting the weights of different types of resources at a given time. Then, a quantum weight adjustment factor is used to control the contribution of each resource to the total resources. The specific mathematical model is: Wherein, is the quantum weight factor, and the updated mathematical model is: Wherein, C i is the usage cost of the resource, is used to control the sensitivity of resource optimization to the costs of different resources. Finally, the dynamic characteristics of the resources are obtained. The specific mathematical model is: H resource =Σ i ω i E res ; Wherein, H resource is the dynamic characteristic of the resource after being processed by the fully connected layer.

[0015] Preferably, by mapping the multi-dimensional resource feature vector to a low-dimensional embedding space, this method effectively reduces the redundancy and noise brought by the feature dimension. At the same time, a quantum state resource mapping method is introduced, and the state weights of different resources at different time points are expressed by quantum amplitudes, enabling the resource features to have non-linear expression ability and dynamic adjustment characteristics. Further, the sensitivity and controllability of the model to the changes in resource allocation are strengthened through the quantum weight adjustment factor, thereby enhancing the support of the resource features for the prediction of the overall construction progress. Especially in the face of complex situations such as unbalanced resource supply and demand and frequent fluctuations in the construction rhythm, it can still maintain the stability of feature expression and high responsiveness to changes in key resources, providing more accurate resource dynamic information support for subsequent progress modeling.

[0016] Preferably, in the S41 part of the S4, the construction task data processed in the S3 step is input, and a construction task state vector is defined. Each construction task corresponds to a five-dimensional state vector at time t. The specific mathematical model is: Wherein, θ i (t) is the construction progress value, ρ i (t) is the resource intensity index, δ i is the logical depth of the construction task, is the sensitivity to external disturbances, Φi For the process complexity score, then define the construction state evolution function, take the task state as the input, and construct the evolution rule of the construction progress. The specific mathematical model is as follows: θ i (t + 1)=θ i (t)+Δθ i (t); In the formula, Δθ i (t) is the progress increment, and the mathematical model is: In the formula, α is the progress evolution rate adjustment coefficient.

[0017] Preferably, by modeling the construction task as a multi-dimensional state vector, this method not only comprehensively describes key factors such as progress, resources, disturbances, and complexity, but also breaks the limitations of traditional single-variable modeling, enabling the model to have stronger information-carrying capacity and task comparability, providing a structured mathematical basis for subsequent task evolution.

[0018] Preferably, in the part S42 of S4, resources will gradually decrease during the progress of the construction task. Define the resource consumption update strategy, and the specific mathematical model of its change process is: ρ i (t + 1)=ρ i (t)-μ·θ i (t)·log(1 + φ i ); In the formula, μ is the resource consumption rate adjustment coefficient. Finally, construct the state trajectory, and form a trajectory sequence according to the state flow. The specific mathematical model is: Γ i ={S i (0), S i (1),..., S i (T)}; In the formula, Γ i is the state vector of each task. Finally, the trajectories of all tasks form a state set. The specific mathematical model is: A={Γ1, Γ2,..., Γ N}; In the formula, is the trajectory sequence of each task, and N is the total number of tasks.

[0019] Preferably, this method uses a custom resource consumption update function to integrate key factors such as resource efficiency, external disturbance suppression, construction depth, and complexity attenuation, which can more realistically reflect the progress behavior of construction tasks under actual conditions, has high controllability and physical interpretability, and helps to dynamically simulate the natural evolution trajectory of the construction process.

[0020] Preferably, in S5, first construct a potential energy function, and the specific mathematical model is: In the formula, K1, K2, and K3 are weight coefficients, and then aggregate the global construction potential energy. The specific mathematical model is: In the formula, ε i is the priority weight factor of the task. Finally, drive the overall progress prediction of the system with the propulsion potential energy. The specific mathematical model is: In the formula, is the prediction of the overall construction completion rate, τ is the step size adjustment factor, is the total logical depth of all tasks.

[0021] Preferably, this method constructs a construction propulsion potential energy function, introduces the tension relationship between the remaining progress of the task and the resource availability, and at the same time considers the influence of perturbations and complexity to form the self-driving force at the task level; further constructs a system potential energy aggregation index to predict the overall progress evolution trend, avoiding relying on traditional time series regression models, and significantly improving the model transparency and engineering decision-making value.

[0022] In summary, due to adopting this technical solution, the present invention proposes a composite modeling method for building construction progress prediction, which generally includes a data preprocessing and multi-dimensional feature modeling module, a state flow modeling module, and a propulsion mapping and progress prediction module. By introducing the quantile converter method, the stability of data preprocessing is improved. In the multi-dimensional feature modeling stage, starting from the task state, time period, and resource dynamics respectively, embedding features are constructed by using measure space mapping, algebraic topology structure, measure transformation, and quantum state resource representation methods to achieve the global modeling of key elements; in the state flow modeling module, the construction tasks are abstracted into multi-dimensional state vectors, and a non-linear progress evolution function and a resource consumption function are designed to simulate the natural progress process of tasks over time; in the propulsion mapping and prediction module, a task propulsion potential energy function is constructed and aggregated into a system potential energy, and combined with the logical depth adjustment factor, the dynamic prediction of the overall construction progress is realized. Compared with traditional prediction methods based on time series or black box models, the present invention has the advantages of strong interpretability, clear modeling logic, high prediction robustness, etc., and can effectively improve the intelligent and refined level of building construction progress management. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flowchart of the steps of a building construction progress prediction method.

[0024] Figure 2 It is a structural diagram of the BuildProformer prediction model.

[0025] Figure 3 It is a data preprocessing, multi-dimensional feature modeling and embedding module.

[0026] Figure 4 It is the structure diagram of the state flow modeling module.

[0027] Figure 5 It is the BuildProformer prediction model to realize the fitting effect diagram of the construction progress. Specific implementation manner

[0028] 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 of ordinary skill in the art without making creative efforts belong to the scope of protection of the present invention.

[0029] Please refer to Figures 1 - 5 , the present invention provides a technical solution: a construction progress prediction method based on machine learning, including a data preprocessing and multi-dimensional feature modeling module, a state flow modeling module, and a propulsion mapping and progress prediction module. By introducing the quantile transformer method, the stability of data preprocessing is improved. In the multi-dimensional feature modeling stage, starting from the task state, time period, and resource dynamics respectively, embedding features are constructed by using measure space mapping, algebraic topological structure, measure transformation, and quantum state resource representation methods to achieve global modeling of key elements; in the state flow modeling module, the construction tasks are abstracted into multi-dimensional state vectors, and non-linear progress evolution functions and resource consumption functions are designed to simulate the natural advancement process of tasks over time; in the propulsion mapping and prediction module, a task propulsion potential energy function is constructed and aggregated into a system potential energy, and combined with the logical depth adjustment factor, the dynamic prediction of the overall construction progress is realized. The specific steps are as Figure 1 shown.

[0030] Construct the BuildProformer prediction model, and its structure is as Figure 2 shown, and the specific steps are as follows:

[0031] S1. Collect relevant data on the construction progress, including historical construction progress data, construction plans, resource allocation situations, and external factor data. The dataset of the present invention contains 8,000 construction progress data.

[0032] Furthermore, the historical construction progress data includes the start time, end time, duration, and completion percentage of each construction task; the construction plan data includes the sequence of tasks, milestone dates, and resource allocation plans; the resource data includes the usage and availability of labor, equipment, and materials; the external factor data includes weather conditions and holiday information.

[0033] S2. Preprocess the data related to building construction using the quantile transformer method, perform outlier handling and standardization, and finally divide the dataset into a training set and a test set according to a ratio of 7:3.

[0034] Furthermore, use the quantile transformer method to preprocess the data related to building construction. First, sort the input data to obtain X sorted，t , calculate the quantile rank of each data point. The specific mathematical model is: rank(x i , t) = arg(x i , t in X sorted，t ); In the formula, arg is the position of the data point in the sorted array. Then map the quantile rank to the uniform distribution U(0, 1) to obtain y i，t , and the specific mathematical model is: In the formula, l is the number of data points at each time step. The time step is initially defined as one day, and l takes the value of 24. Finally, divide the dataset into a training set and a test set according to a ratio of 7:3 to ensure the effectiveness of model training and evaluation.

[0035] S3. Construct a multi-dimensional feature modeling and embedding module, the structure of which is as Figure 3 shown. Input the building construction data, extract the key state features of the construction tasks, including construction progress, task type, and task priority. Consider the construction task state as a distribution in the measure space, and its change can be modeled through measure transformation. First, define the construction task state data in the measure space (X, μ), where X is the feature space of the construction task, and μ is the measure of the task state, which is used to represent the probability density distribution of the task state. Then propose the Cauchy measure transformation to obtain the global distribution of the task state. The specific mathematical model is: In the formula, μ T (x) is the measure value of the task state, representing the state distribution at the position, x is the input feature point of the task state, x0 is the central state of the task, γ is the parameter controlling the width of the measure distribution, and the initial value is set to 0.2, and the value range is [0, 1]. To consider the dependency relationship of the construction tasks, use algebraic topology theory to construct the topological mapping of the task state. The specific mathematical model is: In the formula, H k (X, R) is the homology group in algebraic topology, representing the topological structure of the task state, and k is the dimension of the homology group, representing the mutual relationship between task states. is the boundary operator in algebraic topology. The initial value is taken as 1, representing the connection method between task states. is the kernel of the boundary operator, representing the topological connection characteristics of task states. is the image of the boundary operator, representing the topological relationship of task states, and finally obtaining the global characteristics of task states. The specific mathematical model is: H task = ∑ i ω i H k (X, R); In the formula, ω i is the contribution weight of each homology group in the task state. The initial value is 0.3, and the value range is (0, 1]. H task is the global task state characteristic.

[0036] Furthermore, the input t is a time point during construction. Harmonic analysis and measure transformation are used to model the time period, and the construction time data is transformed onto the measure space (T, v), where T is the time axis and v is the measure of time distribution. Then, the Fourier measure transformation of the time period is defined. The specific mathematical model is: v(ξ) = ∫ T e -2πieξ dv(x); In the formula, v(ξ) is the Fourier transform result of the time measure, capturing the time change pattern in the construction progress. e is the time characteristic point, representing the specific time data. ξ is the time frequency, representing the frequency of periodic changes. The initial value is set to 5, indicating 5 periodic changes per hour. dv is the probability distribution of time characteristics. Then, in order to more accurately represent the construction time period, the Radon - Nikodym derivative is used to calculate the time state. The specific mathematical model is: In the formula, is the harmonic derivative of the time state, representing the rate of change of the time state with respect to the standard Lebesgue measure. λ is the standard Lebesgue measure, describing the change of the construction time period with respect to the standard time axis. Finally, the global state of the construction time is obtained. The specific mathematical model is: In the formula, H period is the global time period characteristic, calculated through Fourier transform and weight ω i The initial value of ω i is 0.3, and the value range is (0, 2].

[0037] Furthermore, the usage of various resources during construction is dynamically changing and will fluctuate with the change of construction progress. The input building construction resource information is combined into a preliminary resource feature vector Rt = [L t , E t , M t , R t is a feature vector containing labor, equipment, and materials, describing the resource state at time point t. L t is the human resources required at the current moment, E t is the equipment resources required at the current moment, M t is the material resources required at the current moment. Then R t is mapped to a low-dimensional embedding space through the embedding layer. The specific mathematical model is: R' t = Embedding(R t ); In the formula, R' t is the representation of the resource feature vector in the low-dimensional embedding space. During the construction process, the dynamic changes of resources will change over time. Therefore, all the resource feature vectors at all time points are aggregated to obtain a complete resource dynamic feature matrix. The mathematical model is: In the formula, E' res is a t×d matrix, a 30X5 matrix, and d is the dimension of each embedding vector. By introducing the quantum state resource mapping method, the dynamic features of resources are further enhanced. The quantum state of each resource is represented as a quantum vector. The specific mathematical model is: |R i (t)> = α i (t)|L t > + β i (t)|E t > + γ i (t)|M t >; In the formula, |R i (t)> is the quantum state representation of the i-th resource, α i , β i , γ i are the quantum amplitudes of the resource state, reflecting the weights of different types of resources at a given time. Then, the quantum weight adjustment factor is used to control the contribution of each resource to the total resources. The specific mathematical model is: In the formula, is the quantum weight factor, with an initial value of 0.25 and a value range of (0, 0.5]. The updated mathematical model is: In the formula, C i is the usage cost of the resource, Used to control the sensitivity of resource optimization to different resource costs, with an initial value set to 0.1 and a value range of (0, 0.4], and finally obtain the resource dynamic characteristics. The specific mathematical model is as follows: H resource = Σ i ω i E res ; In the formula, H resource is the resource dynamic characteristic after being processed by the fully connected layer. The key code of the multi-dimensional feature modeling and embedding module is as follows: # Input: Construction data (X_task, X_time, X_resource) # Output: Global task status feature H_task, global time period feature H_period, resource dynamic feature H_resource def FeatureEmbeddingModule(X_task, X_time, X_resource): # S31 Task status feature extraction μ = get_task_center(X_task) σ = 0.2 μ_X = compute_measure(X_task) C_μ = cauchy_transform(μ_X, μ, σ) T_X = build_topological_mapping(μ_X) H_task = weighted_sum(T_X, weight = 0.3) # S32 Time period feature modeling λ_t = compute_time_measure(X_time) _λ = fourier_transform(λ_t, ω = 5) D_t = radon_nikodym_derivative(_λ, lebesgue) H_period = weighted_sum(D_t, weight = 0.3) # S33 Resource dynamic feature modeling R_t = embed_resource_vector(X_resource) Q_resource = [] for r in R_t: Ψ_r = compute_quantum_state(r, amplitude=True) Q_resource.append(Ψ_r) W_q = adjust_quantum_weights(Q_resource, γ = 0.25, η = 0.1) H_resource = fully_connected(W_q) return H_task, H_period, H_resource

[0038] S41. Construct a state flow modeling module, whose structure is as Figure 4 shown, and use the historical construction progress data to model the dynamic change process of construction tasks.

[0039] Input the construction task data processed in step S3, define the construction task state vector, and each construction task corresponds to a five-dimensional state vector at time t. The specific mathematical model is as follows: In the formula, θ i (t) is the construction progress value, and its value range is [0, 1]. 0 indicates that the task has not started, and 1 indicates that the task is completed. ρ i (t) is the resource intensity index, which represents the current available resource capacity, and the initial value is set to 0.9. δ i is the logical depth of the construction task, which represents the level of the task in the construction topology structure. is the sensitivity to external disturbances, and its value range is [0, 1]. The initial value is set to 0.4. Φ i is the process complexity score, which represents the difficulty level of the construction process. Then define the construction state evolution function, take the task state as the input, and construct the evolution rule of the construction progress. The specific mathematical model is as follows: θ i (t + 1) = θ i (t) + Δθ i (t); In the formula, Δθ i (t) is the progress increment, and the mathematical model is as follows: In the formula, α is the progress evolution rate adjustment coefficient, and the initial value is set to 0.62. is the square root of the resource intensity, indicating that the more resources there are, the faster the task progresses. exp(-Φ i 2 ) indicates that the higher the complexity, the slower the progress.

[0040] Furthermore, resources will gradually decrease during the progress of construction tasks. Define a resource consumption update strategy, and the specific mathematical model of its change process is as follows: ρ i (t + 1) = ρ i (t) - μ·θ i (t)·log(1 + Φ i ); In the formula, μ is the resource consumption rate adjustment coefficient, and its value range is (0, 1]. The initial value is set to 0.5. θ i (t) indicates that the more progress has been completed, the greater the resource consumption. log(1 + Φ i ) indicates that the higher the complexity, the more significant the resource consumption. Finally, construct the state trajectory, and form a trajectory sequence according to the state flow. The specific mathematical model is as follows: Γ i = {S i (0), S i (l),..., S i (T)}; In the formula, Γ i is the state vector of each task. Finally, the trajectories of all tasks form a state set. The specific mathematical model is as follows: A = {Γ1, Γ2,..., Γ N}; In the formula, is the trajectory sequence of each task, N is the total number of tasks, and its value is 5600. The key code is as follows:

[0041] S5. Construct the advancement mapping and progress prediction module. First, construct a potential energy function. The specific mathematical model is as follows: In the formula, K1, K2, and K3 are weight coefficients, which are used to adjust the importance of different factors. The initial values are set to 0.2, 0.3, and 0.4 respectively. Then, aggregate the global construction potential energy. The specific mathematical model is as follows: In the formula, ε i is the priority weight factor of the task, and its value range is (0, 1]. It is determined according to the task importance, engineering quantity, and priority. The most important, the largest engineering quantity, and the highest priority take 1, and in other cases, it decreases by 0.2 step by step. Finally, drive the overall progress prediction of the system with the advancement potential energy. The specific mathematical model is as follows: In the formula, For the prediction of the overall construction completion rate, τ is the step size adjustment factor, with a value range of (0, 1], and the initial value is set to 0.44. It is the sum of the logical depths of all tasks and serves as an inhibitory factor for progress growth. The key code is as follows:

[0042] Furthermore, the BuildProformer prediction model is written in Python and runs in a Linux operating system environment. It is constructed with the help of the PyTorch deep learning framework. Model training is carried out on an NVIDIA A100 40GB GPU, using a preprocessed building construction dataset that covers multi-source data of multiple projects at different construction stages. The training batch size is set to 64, and the Adam optimizer is used with a learning rate set to 5e-5.

[0043] Furthermore, the fitting effect diagram of the construction progress prediction achieved by the BuildProformer prediction model is as Figure 5 shown. The horizontal axis in the figure is the date, and the vertical axis is the progress prediction value. The black solid line and dots represent the actual prediction values, and the gray dashed line and crosses represent the predicted predicted values. It can be seen from the figure that the change trends of the actual values and the predicted values are roughly similar. Especially at the peaks and valleys, the change directions of the two are the same. The experimental results show that the BuildProformer model can effectively capture the change trend of the construction progress prediction and can better predict the construction progress especially in the time periods with large fluctuations.

Claims

1. A method for predicting the construction progress of a building based on machine learning, characterized in that, It includes the following steps: S1. Collect relevant multi-source data of building construction; S2. Use the quantile converter method to preprocess the building construction-related data and divide the data set; S3. Propose global task status features, global time period features, and resource dynamic features, and construct a multi-dimensional feature modeling and embedding module. The specific steps are as follows: S31. Construct a topological map of the task status, and finally obtain the global task status feature H task ; S32. Propose to model the time period by using harmonic analysis and measure transformation, define the measure transformation and calculate the time state, and finally obtain the global time period feature H period ; S33. Propose to map the resource feature vector to a low-dimensional embedding space, introduce a quantum state resource mapping method, and use a quantum weight adjustment factor to control the contribution of each resource, and finally obtain the resource dynamic feature H resource ; S4. Construct a state flow modeling module. The specific steps are as follows: S41. Abstract the construction task into a multi-dimensional state vector, design a non-linear progress evolution function, and combine the resource intensity, disturbance factor, and construction depth to construct a task state flow trajectory to simulate the natural evolution process of the construction task over time; S42. Define a resource consumption update strategy, and form a state trajectory sequence by the evolution of the task state on the time axis; S5. Build a promotion mapping and progress prediction module, design a task promotion potential energy function and a progress prediction function, dynamically predict the overall construction progress driven by the system potential energy, and finally obtain the construction progress prediction result 2. The method for predicting the construction progress based on machine learning according to claim 1, characterized in that In S31, input the building construction data, extract the key state features of the construction task, and define the construction task state data in the measure space (X, μ), where X is the feature space of the construction task, and μ is the measure of the task state. Then, propose a Cauchy measure transformation to obtain the global distribution of the task state. The specific mathematical model is: where, μ T (x) is the measure value of the task state, representing the state distribution at the position, x is the input feature point of the task state, x0 is the central state of the task, γ is the parameter controlling the width of the measure distribution, and then a topological mapping of the task state is constructed. The specific mathematical model is as follows: where H k (X, R) is the homology group in algebraic topology, representing the topological structure of the task state, and k is the dimension of the homology group, representing the mutual relationship between task states. is the boundary operator in algebraic topology, representing the connection mode between task states. is the kernel of the boundary operator, representing the topological connection characteristics of the task state. is the image of the boundary operator, representing the topological relationship of the task state, and finally the global characteristics of the task state are obtained. The specific mathematical model is as follows: H task = ∑ i ω i H k (X, R); where ω i is the contribution weight of each homology group in the task state, and H task is the global task state feature.

3. The method for predicting the construction progress based on machine learning according to claim 2, characterized in that, In S32, input t as a time point during construction. First, use harmonic analysis and measure transformation to model the time period, and convert the construction time data to the measure space (T, v), where T is the time axis and v is the measure of the time distribution. Then, define the measure transformation of the time period. The specific mathematical model is: v(ξ) = ∫ T e -2πieξ dv(x); In the formula, v(ξ) is the transformation result of the time measure, capturing the time change pattern in the construction progress, e is the time feature point, representing the specific time data, ξ is the time frequency, representing the frequency of periodic changes, dv is the probability distribution of the time feature. Then, calculate the time state. The specific mathematical model is: In the formula, is the harmonic derivative of the time state, representing the rate of change of the time state with respect to the standard Lebesgue measure. λ is the standard Lebesgue measure, which describes the change of the construction time period with respect to the standard time axis, and finally the global state of the construction time is obtained. The specific mathematical model is as follows: H period = ∑ i ω i v(ξ); where H period is the global time period feature, calculated through Fourier transform and weight ω i .

4. A method for predicting the construction progress based on machine learning according to claim 3, characterized in that In S33, the input construction resource information is combined into a preliminary resource feature vector R t =[[L t ,[[E t ,[[M t ,[[R t is a feature vector containing human resources, equipment, and materials, describing the resource state at time point t. L t is the human resources required at the current moment. E t is the equipment resources required at the current moment. M t is the material resources required at the current moment. Then R t is mapped to a low-dimensional embedding space through an embedding layer. The specific mathematical model is as follows: R′ t = Embedding(R t ); where R' t is the representation of the resource feature vector in the low-dimensional embedding space. During the construction process, the dynamic changes of resources will vary over time. Therefore, all resource feature vectors at all time points are aggregated to obtain a complete resource dynamic feature matrix. The specific mathematical model is as follows: where, E′ res is a t×d matrix, d is the dimension of each embedding vector. By introducing the quantum state resource mapping method, the dynamic characteristics of resources are further enhanced. The quantum state of each resource is represented as a quantum vector, and the specific mathematical model is: |Ri(t)> = α i (t)|L t > + β i (t)|E t > + γ i (t)|M t >; where, |R i (t)> is the quantum state representation of the i-th type of resource, α i , β i , γ i are the quantum amplitudes of the resource state, reflecting the weights of different types of resources at a given time. Then, a quantum weight adjustment factor is used to control the contribution of each resource to the total resources. The specific mathematical model is as follows: In the formula, is the quantum weight factor, and the updated mathematical model is: Where, C i is the usage cost of resources, is used to control the sensitivity of resource optimization to different resource costs, and finally obtain the dynamic characteristics of resources. The specific mathematical model is as follows: H resource = Σ i ω i E res ; Where, H resource is the dynamic feature of the resource after being processed by the fully connected layer.

5. A method for predicting the construction progress based on machine learning according to claim 4, characterized in that, In step S41, input the construction task data processed in step S3, and define the construction task state vector. Each construction task corresponds to a five-dimensional state vector at time t. The specific mathematical model is: where θ i (t) is the construction progress value, ρ i (t) is the resource intensity index, δ i is the logical depth of the construction task, is the sensitivity to external disturbances, Φ i is the process complexity score. Then, define the construction state evolution function, use the task state as the input, and construct the evolution rule of the construction progress. The specific mathematical model is as follows: θ i (t + 1) = θ i (t) + Δθ i (t); where Δθ i (t) is the progress increment, and the mathematical model is: In the formula, α is the progress evolution rate adjustment coefficient.

6. The method for predicting the construction progress based on machine learning according to claim 5, wherein In step S42, the resources will gradually decrease during the progress of the construction task. Define a resource consumption update strategy. The specific mathematical model of its change process is: ρ i (t + 1) = ρ i (t) - μ·θ i (t)·log(1 + Φ i ); In the formula, μ is the resource consumption rate adjustment coefficient. Finally, construct a state trajectory, and form a trajectory sequence according to the state flow. The specific mathematical model is: Γ i = {S i (0), S i (1),..., S i (T)}; where Γ i is the state vector of each task, and the trajectories of all tasks ultimately form a state set. The specific mathematical model is as follows: A = {Γ1, Γ2,..., Γ N}; In the formula, is the trajectory sequence of each task, and N is the total number of tasks.

7. A method for predicting the construction progress based on machine learning according to claim 6, characterized in that In step S5, first construct a potential energy function. The specific mathematical model is: wherein, K1, K2 and K3 are weight coefficients, and then the global construction potential energy is aggregated. The specific mathematical model is as follows: where ε i is the priority weight factor of the task, and finally drives the overall progress prediction of the system with the propulsion potential energy. The specific mathematical model is as follows: In the formula, is the prediction of the overall construction completion rate, τ is the step adjustment factor, is the total logical depth of all tasks.

Images