Dynamic scheduling method and device for architectural decoration and finishing and medium

Abstract

The invention discloses a dynamic scheduling method and device for building decoration, and a medium, and the method comprises the steps: determining a preset neural network model group, and carrying out the deep learning training of the neural network model group according to the pre-obtained multi-dimensional construction data, so as to obtain a dynamic scheduling model group; construction tasks and worker characteristics are determined, the construction tasks and the worker characteristics are imported into the dynamic scheduling model set, so that the construction condition is determined through the prediction model, the construction condition is input into the optimization model, and a construction scheduling scheme is determined through the optimization model; construction is conducted according to the construction scheduling scheme, a construction site is monitored in real time to obtain real-time data of the construction site, and whether an emergency exists in the construction site or not is determined according to the real-time data; and if the emergency situation exists in the construction site, importing the emergency situation into the dynamic adjustment model so as to determine a scheduling adjustment scheme through the dynamic adjustment model, and re-scheduling the construction site through the scheduling adjustment scheme.

Description

Technical Field

[0001] This application relates to the technical field of building decoration, and particularly to a dynamic scheduling method, device, and medium for building decoration. Background Art

[0002] In the building decoration industry, construction scheduling is crucial for project quality, cost control, and project duration management. Currently, most decoration scheduling lacks intelligent optimization capabilities, resulting in low construction efficiency and serious resource waste. Although traditional project management software can perform task arrangements, it is usually based on static schedules and relies on the experience and intuition of project managers. When facing complex multi-task parallelism, resource conflicts, and emergencies, it often appears powerless, unable to dynamically optimize, and difficult to adapt to complex and changing construction environments. Summary of the Invention

[0003] To solve the above problems, this application proposes a dynamic scheduling method for building decoration, including: determining a pre-set neural network model group, performing deep learning training on the neural network model group according to pre-acquired multi-dimensional construction data to obtain a dynamic scheduling model group, the dynamic scheduling model group including a prediction model, an optimization model, and a dynamic adjustment model, the multi-dimensional construction data including environmental parameters, equipment operation data, personnel locations, and material inventory information; determining construction tasks and worker characteristics, importing the construction tasks and worker characteristics into the dynamic scheduling model group to determine the construction situation through the prediction model, and inputting the construction situation into the optimization model to determine a construction scheduling plan through the optimization model; performing construction according to the construction scheduling plan, and monitoring the construction site in real time to obtain real-time data of the construction site, and determining whether there are emergencies at the construction site according to the real-time data; if there are emergencies at the construction site, importing the emergencies into the dynamic adjustment model to determine a scheduling adjustment plan through the dynamic adjustment model, and re-scheduling the construction site through the scheduling adjustment plan.

[0004] In one example, performing deep learning training on the neural network model group according to pre-acquired multi-dimensional construction data specifically includes: preprocessing the multi-dimensional construction data, converting the preprocessed multi-dimensional construction data to obtain feature vectors, and inputting the feature vectors into an LSTM neural network model for training to obtain the dynamic scheduling model, the expression of the dynamic scheduling model being:

[0005] Wherein, is the activation value of the input gate, is the weight matrix, is the bias, is the value at time step t−1 in the hidden state, is the input feature vector at time step t, is the sigmoid function, is the activation value of the forget gate, is the weight matrix, is the bias, is the value of the cell state at time step t, is the old cell state, is the weight matrix, is the bias, is the activation value of the output gate, is the weight matrix, is the bias, is the value of the hidden state at time step t, is the hyperbolic tangent activation function.

[0006] In one example, deep learning training is performed on the neural network model group according to pre-acquired multi-dimensional construction data. Specifically, it further includes: preprocessing the multi-dimensional construction data, converting the preprocessed multi-dimensional construction data to obtain a feature vector, and inputting the feature vector into the MDP model for training to obtain the optimization model. The expression of the optimization model is:

[0007] where, is the core value function in the learning algorithm, representing the expected cumulative reward value of performing action a in the current state s, is the learning rate, is the reward function, is the discount factor, represents taking the maximum value of the Q values corresponding to all actions a′ in the new state

[0008] In one example, inputting the construction situation into the optimization model specifically includes: determining the corresponding chromosome encoding according to the construction situation to obtain the optimization model. The expression of the optimization model is:

[0009] where, is the fitness value, , are the weight coefficients, is the total construction period, is the minimum value of the resource utilization rate, is the maximum value of the resource utilization rate.

[0010] ​In one example, the construction tasks and worker characteristics are determined, specifically including: determining the pre-set work types and skill levels, and determining the pre-set work intensity; determining the urgency corresponding to the construction tasks, and determining the working hours corresponding to the construction tasks according to historical construction data; determining the skill specialties corresponding to the workers according to the work types and skill levels, and determining the working years, fatigue levels, and task loads of the workers.

[0011] In one example, the method further includes: preprocessing the construction tasks and the worker characteristics to obtain an input vector, and importing the input vector into a pre-set matching model; obtaining the construction situation from the prediction model through the matching model, and determining the worker allocation plan according to the construction situation; inputting the worker allocation plan into the optimization model, so that the optimization model outputs the construction scheduling plan according to the worker allocation plan; and inputting the worker allocation plan into the dynamic adjustment model, so that the dynamic adjustment model outputs the scheduling adjustment plan according to the worker allocation plan.

[0012] In one example, the method further includes: determining the input layer, hidden layer, and output layer of the matching model; obtaining the input vector through the input layer, and outputting the matching probability through the output layer to determine the worker allocation plan according to the matching probability; the expression of the matching model is:

[0013] where represents the matching probability, and are weight matrices, is the input vector, is the activation function, is the Sigmoid activation function.

[0014] In one example, the method further includes: determining the number of tasks corresponding to the construction equipment, if the number of tasks is greater than 1, then determining the construction map, dividing the construction map to determine multiple grid cells, and determining multiple weights corresponding to the multiple grid cells; marking the regions of the construction map according to the multiple weights to obtain a weight map, and obtaining the weight map through a pre-set planning model to determine the scheduling path of the construction equipment according to the weight map.

[0015] On the other hand, the present application also proposes a dynamic scheduling device for building decoration, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the dynamic scheduling device for building decoration can execute: the method described in any one of the above examples.

[0016] On the other hand, the present application also proposes a non-volatile computer storage medium storing computer-executable instructions, and the computer-executable instructions are set as: the method described in any one of the above examples.

[0017] The present application integrates three types of models: prediction, optimization, and dynamic adjustment, forming a "prediction - decision - correction" closed loop. The LSTM neural network accurately captures the time-series characteristics of the construction state. The MDP model uses reinforcement learning to achieve iterative evolution of the scheduling strategy. The genetic algorithm dynamically optimizes the response plan for emergency scenarios to ensure the full-cycle effectiveness of the scheduling plan. Four-dimensional data such as environmental parameters, equipment operation, personnel status, and material inventory are integrated, and after being standardized, a digital twin is constructed, significantly improving the perception accuracy of the construction state and providing a reliable data foundation for intelligent decision-making. A dual decision-making path of a worker-task matching probability matrix and chromosome encoding is established. The task matching degree is predicted through deep learning, and the genetic algorithm is used to globally search for the optimal solution to balance local accurate matching and overall scheduling efficiency. Real-time monitoring triggers the dynamic adjustment mechanism, quickly responds to construction changes through feature vector reconstruction, combines with the path planning algorithm to optimize equipment scheduling, and the material prediction model drives intelligent replenishment to form an adaptive construction ecosystem. The dynamic threshold combined with online learning continuously optimizes the early warning model, and the historical data drives the iteration of the knowledge base, enabling the system to have the ability of self-evolution and continuously improving the quality of scheduling decisions. The present application realizes dynamic optimization of scheduling by real-time analyzing the construction progress, resource allocation, and environmental changes through intelligent algorithms, improving the construction efficiency and resource utilization rate. Description of the Drawings

[0018] The drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings: Figure 1 It is a schematic flowchart of a dynamic scheduling method for building decoration in an embodiment of the present application; Figure 2 It is a schematic diagram of a dynamic scheduling device for building decoration in an embodiment of the present application. Detailed Embodiments

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Apparently, the described embodiments are only a part of the embodiments of this application, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts belong to the scope of protection of this application.

[0020] The following will detail the technical solutions provided by each embodiment of this application in conjunction with the drawings.

[0021] In the building decoration industry, construction scheduling has a decisive impact on project efficiency, cost control, and project duration management. Currently, the industry mainly relies on manual experience to formulate schedules, lacking intelligent optimization capabilities, resulting in low construction efficiency and serious resource waste. Although traditional project management software can achieve basic task arrangements, it cannot dynamically optimize schedules and is difficult to adapt to complex and changing construction environments. Therefore, developing an intelligent scheduling system has become an urgent need to improve construction efficiency, reduce costs, and enhance scheduling flexibility.

[0022] Traditional building decoration scheduling methods are mostly based on static schedules and highly rely on the personal experience of project managers. In the face of parallel multi-tasks, resource conflicts, and emergencies (such as material delays, improper worker scheduling, equipment failures, etc.), traditional methods often struggle to cope, easily leading to project delays and cost overruns. At the same time, traditional methods lack in-depth mining of historical data, unable to form experience precipitation, and difficult to achieve continuous optimization.

[0023] Although existing project management software has improved task arrangement efficiency, its functions are mostly limited to task assignment and progress tracking, lacking dynamic response capabilities. When delays occur in the construction process, traditional software cannot automatically adjust resource allocation or task priorities, affecting the overall progress. When dealing with complex scheduling problems of multiple projects and multiple resources, traditional software more obviously exposes the shortcomings of insufficient computing power and limited optimization effects.

[0024] With the development of artificial intelligence and machine learning technologies, intelligent scheduling systems provide solutions to the industry's pain points. By introducing AI algorithms, the system can learn scheduling rules from historical data, predict risks, and generate optimal solutions. Machine learning technology can also dynamically adjust schedules based on real-time data to ensure efficient use of resources and precise control of project duration. For example, the system can automatically optimize worker task allocation, material supply plans, and equipment scheduling according to the real-time status of the construction site, minimizing resource waste and project delays to the greatest extent.

[0025] In addition, the intelligent scheduling system is combined with Internet of Things (IoT) technology. Through sensors and intelligent devices, it monitors the construction environment, equipment status and personnel dynamics in real time, providing real-time data support for dynamic adjustment. The application of cloud computing and big data technologies enables the system to process massive amounts of data, provide real-time analysis and visualization reports, and assist managers in making scientific decisions.

[0026] In summary, the traditional scheduling methods for building decoration are no longer able to meet the high requirements of modern projects for efficiency, cost and flexibility. The dynamic optimization scheduling system based on artificial intelligence and machine learning algorithms can significantly improve construction efficiency, reduce costs and enhance project quality through intelligent and dynamic scheduling solutions, providing key technical support for the digital transformation of the industry.

[0027] As Figure 1 shown, to solve the above problems, an embodiment of the present application provides a dynamic scheduling method for building decoration, which is applied in a dynamic scheduling system for building decoration. The method includes: S101. Determine a pre-set neural network model group, and perform in-depth learning training on the neural network model group according to pre-acquired multi-dimensional construction data to obtain a dynamic scheduling model group. The dynamic scheduling model group includes a prediction model, an optimization model, and a dynamic adjustment model. The multi-dimensional construction data includes environmental parameters, equipment operation data, personnel positions, and material inventory information.

[0028] The data collection and processing module relies on Internet of Things (IoT) devices, sensors and intelligent terminals to collect multi-dimensional data on the construction site in real time, covering construction progress, personnel dynamics, equipment status, material inventory and environmental parameters (such as temperature, humidity, etc.). At the same time, the system is compatible with external data sources such as project management software, design drawings, and contract documents, and supports the import of structured and unstructured data. After collection, the system uses big data processing technology to clean, integrate and store the data, building a high-quality data foundation for subsequent intelligent analysis. In the early stage of construction, the system collects data on the construction site in all directions and multi-dimensionally through IoT devices, including environmental parameters, equipment operation status, personnel positions and activity trajectories, material inventory, etc. At the same time, structured and unstructured data are imported from project management software, design drawings and contract documents. The collected raw data needs to be cleaned, de-duplicated and standardized. For example, for environmental data collected by sensors, the mean imputation method or a time-series-based prediction model is used to fill in missing values, and the Z-score standardization or Min-Max standardization method is used to convert data with different dimensions to a unified numerical range.

[0029] The artificial intelligence and machine learning algorithm engine, as the core of the system, integrates multiple advanced algorithms to drive the dynamic optimization of construction scheduling. Through the prediction model constructed with historical data and real-time data, it accurately estimates the construction progress, resource requirements, and potential risks. The optimization model generated based on optimization techniques such as reinforcement learning and genetic algorithms continuously outputs the optimal scheduling plan with high resource efficiency and the shortest construction period. The dynamic adjustment model responds in real time to sudden changes on the construction site, such as sudden delays or resource conflicts, and automatically adjusts the plan to ensure the steady progress of the project.

[0030] S102. Determine the construction tasks and worker characteristics, and import the construction tasks and worker characteristics into the dynamic scheduling model group to determine the construction situation through the prediction model, and input the construction situation into the optimization model to determine the construction scheduling plan through the optimization model.

[0031] In one embodiment, the system uses historical data and real-time data to train a deep learning model (such as an LSTM neural network) to predict the construction progress, resource requirements, and potential risks. The mathematical expression of the prediction model is:

[0032] is the activation value of the input gate. At time step t, the input gate determines which new information will be written into the cell state This value is obtained by calculating the linear combination of the previous hidden state and the current input through the weight matrix and the bias , and then mapping it to the range of 0 to 1 through the sigmoid function σ. For example, if has an element value of 0.8, it means that there is an 80% probability that the new information at that position will be written into the cell state.

[0033] is the activation value of the forget gate. At time step t, the forget gate determines which old information in the cell state will be discarded. It also calculates the linear combination of and the bias through the weight matrix and , and then obtains a value between 0 and 1 through the sigmoid function. For example, if has an element of 0.2, it means that there is a 20% probability that the old information at that position will be retained and an 80% probability of being forgotten.

[0034] is the value of the cell state at time step t. It is a key part for storing long-term information in the LSTM neural network. The update of the cell state consists of two parts: one part is the retention degree of the old cell state determined by the forget gate ft, that is, ; the other part is the write degree of the new information determined by the input gate , that is, . Among them, the tanh function maps the new information to the range from -1 to 1, ensuring that the value of the cell state is within a reasonable interval. .

[0035] is the activation value of the output gate. At time step t, the output gate determines which information is read from the cell state as the output of the hidden state . It calculates the linear combination of and the bias through the weight matrix and , and then obtains a value between 0 and 1 through the sigmoid function. For example, if a certain element of is 0.9, it means that there is a 90% probability that the cell state information at that position is read and output.

[0036] is the value of the hidden state at time step t. It is the output of the LSTM neuron at this time step and is used to be passed to the next time step and other network layers. The hidden state is jointly determined by the output gate and the cell state . This means that only the information allowed by the output gate is read and output from the cell state, and the output information is remapped by the tanh function to ensure the stability of the output.

[0037] σ is the sigmoid activation function. Its mathematical expression is σ(x)=1+e−x1. This function maps the input value to the range from 0 to 1 and is often used in probability calculations and gating mechanisms. In the LSTM, it is used to determine the probabilities of writing, retaining, and reading information. By adjusting the weights and biases, the network can learn which information is important and which can be ignored.

[0038] , , , are weight matrices. These matrices correspond to the weight parameters of the input gate, forget gate, cell state update, and output gate respectively. They are continuously optimized and adjusted during the model training process by learning the patterns and rules in the data. For example, is a two-dimensional matrix, where the number of rows corresponds to the dimension of the input features, and the number of columns corresponds to the dimension of the hidden state. During training, the gradients of the loss function with respect to these weights are calculated using the backpropagation algorithm, and then the weight values are updated using an optimization algorithm (such as gradient descent) to minimize the prediction error.

[0039] , , , are bias vectors. These vectors correspond to the bias parameters of the input gate, forget gate, cell state update, and output gate respectively. The bias term is used to add a constant term in the neuron calculation, which helps to improve the expressive power and fitting ability of the model. For example, is a one-dimensional vector with the same dimension as the hidden state. When calculating the activation value of the input gate, the bias is added to the linear combination of the weight matrix and the input, enabling the model to better adapt to different data distributions.

[0040] is the value of the hidden state at time step t−1. As the output of the network at the previous moment, it, together with the input at the current moment, serves as the input to the LSTM neuron, reflecting the temporal sequence and dependency relationship of the time series data. The hidden state ht−1 contains key information such as the construction progress and resource usage at the previous moment, which will affect the prediction result at the current moment.

[0041] is the input feature vector at time step t. It contains relevant data for predicting construction progress, resource requirements, and potential risks at the current moment. For example, can include construction progress data such as the amount of completed tasks, the deviation between the actual construction period and the planned construction period; resource usage conditions such as the man-hours input, material consumption rate, and equipment operation duration; and environmental data such as the temperature and humidity at the construction site. These data are collected in real time through Internet of Things devices and are preprocessed to form a structured feature vector.

[0042] ⊙ represents the element-wise multiplication operation. That is, the elements at the corresponding positions of two vectors or matrices with the same dimension are multiplied to obtain a new vector or matrix. In LSTM, element-wise multiplication is used to screen and combine information in the gating mechanism and cell state update. For example, when calculating , the activation value of the forget gate is multiplied element-wise with the old cell state to determine which old information will be retained in the new cell state.

[0043] Tanh is the hyperbolic tangent activation function. Its mathematical expression is tanh(x) = (e^x - e^(-x)) / (e^x + e^(-x)). This function maps the input value to the range of -1 to 1 and is commonly used for the update of cell states and the calculation of hidden states. Compared with the sigmoid function, the tanh function can provide a richer output range, which helps to enhance the non-linear fitting ability of the neural network, enabling the model to better capture the complex mapping relationship between the input data and the output.

[0044] The core input data of the LSTM neural network model , stems from the deep integration of the on-site construction real-time monitoring system and the project management system. Among them, the real-time monitoring system continuously collects multi-dimensional data through Internet of Things devices, including sensors, cameras, etc., covering environmental parameters such as temperature and humidity; equipment operation status such as operation duration and failure frequency; and personnel dynamics such as worker positioning and working hours statistics. The project management system supplements structured data such as task lists and construction period plans; and unstructured data such as construction logs and communication documents. These data are transformed into input feature vectors adapted to the LSTM model through processes of cleaning, deduplication, and standardization. .

[0045] Through the deep calculation of the LSTM network, the finally output hidden state , not only encodes the key features of the current construction state, but also integrates the evolution laws and trends in historical data. This state vector serves as the input foundation for the subsequent prediction module, providing data support for construction progress estimation, resource demand calculation, and risk warning, and driving the intelligent and dynamic decision-making of the entire scheduling system. For example, when predicting risks such as material shortages or equipment failures, the system can based on the construction progress and resource consumption patterns contained, combined with the pre-trained prediction model, generate warning signals and corresponding suggestions to assist managers in making timely decisions and ensuring the orderly progress of the project.

[0046] In one embodiment, the optimization model is based on the reinforcement learning algorithm. The system simulates different construction scenarios to generate the optimal scheduling plan. The system regards the construction project as a Markov decision process (MDP), defining the state space S, the action space A, and the reward function R(s, a). The agent continuously tries different actions through interaction with the environment, updates the policy parameter π(a|s) according to the received reward signal, and the goal is to maximize the cumulative reward. Its update formula is:

[0047] Q(s,a) is the core value function in the Q-learning algorithm, representing the expected cumulative reward for taking action a in state s. Its value reflects the long-term return expectation of taking a specific action in a specific state. Initially, the values of Q(s,a) are usually initialized to zero or small random numbers. As the agent continuously interacts with and learns from the environment, these values are gradually updated and optimized, eventually converging to the optimal values.

[0048] α is the learning rate. This is a hyperparameter with a value range between 0 and 1. It determines the proportion of newly learned information in each update of the Q value. For example, when α = 0.1, it means the new information accounts for 10% and the old information accounts for 90%. A larger α value means the agent adapts to new information faster, but it may also lead to an unstable learning process; a smaller α value makes the learning process smoother, but the convergence speed may be slower.

[0049] R(s,a) is the reward function. It is a function of state s and action a, used to measure the immediate reward obtained by the agent after taking action a in state s. The design of the reward function is crucial for the success of reinforcement learning because it directly guides the learning direction and goal of the agent. In the building decoration scheduling system, the reward function may be defined according to different project goals. For example, R(s,a) = w1⋅duration reward + w2⋅cost reward + w3⋅quality reward Among them, w1, w2, and w3 are weight coefficients used to balance the importance of different goals. The duration reward can be defined as giving a positive reward if action a enables the project to be completed on time or ahead of schedule, otherwise giving a negative reward; the cost reward gives corresponding rewards or penalties according to the impact of action a on the project cost; the quality reward determines the reward value based on whether the construction quality meets the standards.

[0050] γ is the discount factor. It is also a hyperparameter with a value range between 0 and 1. The discount factor is used to balance the weights of immediate rewards and future rewards. For example, when γ = 0.9, it means the weight of future rewards is 90% of the immediate reward. A larger γ value means the agent pays more attention to long-term returns, while a smaller γ value makes the agent tend to pursue immediate benefits.

[0051] is to take the maximum value of the Q values corresponding to all possible actions in state . This means that in the next state , the agent selects the action that can obtain the maximum expected cumulative reward according to the currently known Q values. This value is used to calculate the update of the Q value for the current state-action pair, reflecting the optimality principle in reinforcement learning, that is, the agent always hopes to choose the action that can obtain the maximum return in the future.

[0052] s is the current state. In a building decoration project, the state s can be a multi-dimensional vector, including the set of tasks that have been completed, the remaining resources, such as the remaining available amounts of labor, materials, and equipment; the current construction progress, and environmental conditions, such as the temperature and humidity at the construction site. This information is collected in real time through Internet of Things devices and imported into the project management system, and after preprocessing, a structured state description is formed. For example, the state s can be represented as: s = [list of completed tasks, remaining resources, current construction progress, environmental conditions] Among them, the list of completed tasks can be represented by a vector, where each element corresponds to whether a task is completed, 1 indicates completion, and 0 indicates non-completion; the remaining resources are also a vector, representing the remaining amounts of different types of resources; the current construction progress can be represented by a percentage value, such as 0.6 indicating that 60% of the construction period has been completed; the environmental conditions can be represented by multiple values, such as temperature, humidity, etc.

[0053] a is the current action. The action a is an operation that the agent can perform in the state s. For a building decoration scheduling system, the action space a may include adjusting task priorities, allocating resource quantities, changing the construction sequence, etc. For example, the action a can be a vector representing the priority adjustment and resource allocation amount for each task: a = [priority adjustment for task 1, priority adjustment for task 2,..., priority adjustment for task n, resource allocation amount] Among them, the priority adjustment of a task can be represented by a numerical value, such as +1 indicating an increase in priority and -1 indicating a decrease in priority; the resource allocation amount represents the resource quantities allocated to each task, such as the amount of labor and materials allocated to task 1.

[0054] is the new state transferred to after executing the action a. The new state is updated according to the execution result of the current state s and the action a. For example, if the action a is to increase the priority of a certain task and allocate more resources, then in the new state , the completion progress of this task will be accelerated, the remaining resources will be reduced accordingly, the construction progress will be updated, and the environmental conditions may also change due to the changes in construction activities, such as the noise level and dust concentration at the construction site.

[0055] In the building decoration scheduling system, the data basis of state s originates from the integration of the on-site real-time monitoring system and the project management system. The real-time monitoring system uses Internet of Things devices such as sensors and cameras to collect multi-dimensional data in real time, including environmental parameters such as temperature and humidity, equipment operation data such as working hours and failure rates, and personnel dynamics such as worker positioning and working hour distribution. The project management system provides structured data such as task lists and project duration nodes, and unstructured data such as construction logs and communication documents. After being cleaned, de-duplicated, and standardized, these data are transformed into the state vector s that can be used by the reinforcement learning model.

[0056] The design of action a relies on project management professional logic and construction practical experience. Based on project characteristics and requirements, the system pre-defines a set of executable actions that cover typical adjustments and optimizations in construction scheduling. For example, task priority adjustment needs to follow task dependencies and urgency; resource allocation actions need to match resource availability and task requirements.

[0057] The expression of the reward function R(s, a) is customized according to the core objectives of the project. In a project oriented to project duration control, the reward function can focus on quantifying the contribution value of action a to project duration compression.

[0058]

[0059] At the same time, cost and quality factors can also be combined to construct a more comprehensive reward function, such as the multi-objective reward function described above.

[0060] By continuously applying the Q-learning update formula, the system can gradually learn the optimal action a in different states s, thereby generating an optimal construction scheduling plan. This optimal strategy will guide project managers to make scientific and reasonable decisions during the actual construction process, ensuring that the project can be completed efficiently and with high quality, while minimizing costs and risks.

[0061] S103. Perform construction according to the construction scheduling plan, and conduct real-time monitoring of the construction site to obtain real-time data of the construction site, and determine whether there are any emergencies at the construction site based on the real-time data.

[0062] S104. If there is an emergency at the construction site, import the emergency into the dynamic adjustment model to determine a scheduling adjustment plan through the dynamic adjustment model, and re-schedule the construction site through the scheduling adjustment plan.

[0063] In one embodiment, the dynamic adjustment model monitors the changes at the construction site in real time and uses the genetic algorithm to dynamically adjust the scheduling plan. When sudden delays, resource conflicts and other changes occur at the construction site, the current scheduling plan is represented in the form of chromosome coding. By defining the fitness function, the genetic algorithm performs operations such as selection, crossover and mutation to generate a new generation of scheduling plan population. The fitness function can be expressed as:

[0064] Fit is the fitness value, which is used to measure the quality of a certain scheduling plan. The higher the fitness value, the better the scheduling plan.

[0065] 、 are weight coefficients, which are used to balance the proportion of different objectives in the fitness function. For example, if the project pays more attention to the construction period, then takes a larger value; if it pays more attention to resource utilization rate, then takes a larger value. Usually, + = 1.

[0066] T is the total construction period, which refers to the total time required to complete all construction tasks.

[0067] is the minimum value of resource utilization rate, which refers to the minimum value of resource utilization rate in the scheduling plan to ensure that resources can be fully utilized in all time periods.

[0068] is the maximum value of resource utilization rate, which refers to the maximum value of resource utilization rate in the scheduling plan to avoid over-concentration or idleness of resources.

[0069] In the building decoration scheduling system, the construction of the fitness function is based on the fusion analysis of multi-source data, and the data mainly comes from the collaboration of the real-time monitoring system at the construction site and the project management system. Among them, the value of the total construction period T is dynamically calculated by combining the construction period plan provided by the project management system with the construction progress. The system forms a real-time evaluation of the total construction period of the current scheduling plan by continuously tracking the completed tasks and estimating the construction period of the remaining tasks.

[0070] The acquisition of the resource utilization rate parameters Rmin and Rmax depends on the Internet of Things device network of the real-time monitoring system, including sensors, intelligent terminals, etc. These devices continuously collect the spatio-temporal usage data of manpower, materials and equipment. For example, in a certain construction scenario, the monitoring system captures that the equipment utilization rate fluctuates in the range of 50% to 80%, then Rmin takes 0.5 and Rmax takes 0.8, which accurately reflects the resource utilization boundary.

[0071] Based on the genetic algorithm framework, the system uses a fitness function to quantitatively evaluate the scheduling plan and drives the iterative optimization of the plan through genetic operations such as selection, crossover, and mutation. The ultimate goal is to generate an optimal scheduling adjustment plan that not only meets the construction constraints but also can dynamically adapt to on-site changes, ensuring the continuity and efficiency of the construction process.

[0072] In one embodiment, the visualization interface intuitively presents construction progress, resource allocation, and risk warning information in various forms such as Gantt charts, heat maps, and dashboards, helping project managers to have an overall grasp. At the same time, the system is built with an intelligent decision-making engine that can automatically generate response plans for abnormal situations. For example, when it is detected that the task progress is lagging, the system will automatically calculate adjustment plans, such as increasing the manpower input or extending the working hours, and push the suggestions to the manager's terminal for decision-making reference.

[0073] In one embodiment, to achieve the intelligent allocation of worker tasks, feature extraction is performed on tasks and workers respectively.

[0074] For the task part, in the building decoration industry, according to the type of work and skill level standards, hierarchical management is implemented for the skill requirements of carpenters. For example, junior carpenters should be able to complete the installation of simple wood products. A typical task is to install a standard door or window within 2 hours, with the precision error controlled within ±5mm; intermediate carpenters should be able to independently complete the production and installation of complex wood products. For example, complete the entire operation of a customized wardrobe within 4 hours, with the precision reaching within ±3mm; senior carpenters need to master the production process of high-precision and artistic wood products, such as making and installing carved doors and windows within 6 hours, with the precision requirement increased to within ±1mm.

[0075] The work intensity is classified into three levels: light, medium, and heavy, based on the national labor intensity standard (GB / T 12350-2008) combined with the actual construction situation. Light-intensity work refers to the average daily energy consumption being lower than 1.5 times the basal metabolic rate (BMR), covering activities such as tool handling where the tool weight is less than 10kg; short-distance walking where the speed is less than 1km / h and other low-energy-consuming activities; medium-intensity work has an energy consumption between 1.5 and 4.0 times the BMR, including activities such as walking while carrying 10-30kg heavy objects, drilling with power tools, etc., which are medium-load operations; heavy-intensity work exceeds 4.0 times the BMR, involving high-energy-consuming scenarios such as carrying heavy objects over 30kg and long-term high-altitude operations.

[0076] In actual construction, the intensity level is evaluated comprehensively through multiple indicators. For example, when the worker's heart rate is maintained at 60-80 beats / minute, oxygen consumption is lower than 10ml / kg / min and the movements are easy, it is judged as light intensity; when the heart rate is 80-100 beats / minute, oxygen consumption is 10-20ml / kg / min and the movements are frequent and strenuous, it is medium intensity; when the heart rate exceeds 100 beats / minute, oxygen consumption is higher than 20ml / kg / min and the movements are intense and continuous, it is judged as heavy intensity work.

[0077] The urgency of tasks is divided into two levels: emergency and normal according to the project duration requirements and task dependencies. Emergency tasks must be started within 48 hours, and there is no buffer for completion time. They usually involve key path nodes or tasks that have major constraints on subsequent construction. For example, if the construction of the foundation waterproof layer is not completed within 48 hours, it will directly block the progress of ground construction. Normal tasks are allowed to flexibly arrange the start time between 48 hours and the project general construction, with a certain amount of time redundancy. For example, the painting of wall latex paint can be carried out within one week after the completion of ground construction.

[0078] Based on historical construction data and industry standard construction period quotas, combined with specific construction environments and conditions, the completion time of each task is estimated. For example, for a 20-square-meter room wall latex paint painting task, under normal construction conditions, that is, an ambient temperature of 20-25°C and a humidity of 40%-60%, the estimated working time is 2 working days, that is, 8 working hours per day; if the construction environment humidity exceeds 60%, the estimated working time increases to 3 working days.

[0079] For workers, the building decoration and renovation industry implements professional assessment and classification of workers' skill levels based on skill level standards. Taking carpentry skills as an example, the assessment system covers three core dimensions: wood product manufacturing accuracy, installation stability, and decorative aesthetics. The skill level is determined through a dual mechanism of practical assessment and project historical performance. For example, junior carpenters need to master basic installation skills, be able to complete the assembly of conventional wood products, and control the production accuracy error within ±5mm; intermediate carpenters need to have the ability to operate the entire process of complex wood products, and be able to independently complete the entire process from customized product design to installation, with an accuracy of within ±3mm; senior carpenters need to be proficient in high-precision craftsmanship and artistic processing, and be able to produce special wood products such as complex carved components, with the accuracy requirement increased to within ±1mm.

[0080] The construction decoration and renovation industry accumulates workers' working hours on an annual basis, and the accumulated working years are positively correlated with construction response capabilities. Taking carpentry as an example, a skilled worker with 5 years of work experience tends to show more sophisticated technical control and problem-solving capabilities when dealing with complex wooden structure installations compared to a worker with 2 years of work experience.

[0081] It is evaluated by the cumulative working hours and the recent task intensity. The cumulative working hours refer to the total working hours of the worker in the past 7 days, and the recent task intensity refers to the average number of tasks per day in the past 3 days. For example, when the cumulative working hours of the worker in the past 7 days exceed 50 hours and the average number of tasks per day in the past 3 days is greater than 3, the fatigue level is determined to be high; when the cumulative working hours are between 30 and 50 hours and the average number of tasks per day is between 1 and 3, the fatigue level is medium; when the cumulative working hours are less than 30 hours and the average number of tasks per day is less than 1, the fatigue level is low.

[0082] Count the number of tasks that the worker has currently been assigned but not completed, as well as the sum of the estimated working hours of these tasks. For example, if the worker currently has 2 uncompleted tasks with estimated working hours of 4 hours and 6 hours respectively, the current task load is 10 hours. According to the task load situation, the workers are divided into three states: light load, that is, the task load is less than 16 hours; medium load, that is, the task load is between 16 and 32 hours; heavy load, that is, the task load is greater than 32 hours, so that the system can reasonably allocate new tasks.

[0083] In one embodiment, the matching model is a neural network model based on deep learning. Its core function is to calculate the probability value of each worker being suitable for a certain task according to the task characteristics and worker characteristics. The model learns the patterns in the historical task allocation data to predict the optimal match for the current task allocation. The prediction model is used to predict the construction progress, resource requirements, and potential risks. The matching model can utilize these prediction results to better understand the urgency of the tasks and resource requirements, so as to allocate workers more accurately. The optimization model is used to generate the optimal construction scheduling plan. The result of the matching model, that is, the matching probability between the worker and the task, is used as one of the inputs of the optimization model to help the optimization model arrange tasks more reasonably. The dynamic adjustment model is used to adjust the scheduling plan in real time. The matching model can dynamically update the matching probability between the worker and the task, providing real-time matching suggestions for the dynamic adjustment model.

[0084] The input task characteristics are represented in vector form, including skill requirements, work intensity, urgency, and estimated working hours, etc.

[0085] For example, the task characteristics with feature vectorization are represented as: Skill requirements: One-hot encoding is adopted, and intermediate carpenter corresponds to the vector [0, 1, 0].

[0086] Work intensity: Numerically represented, 1 - light, 2 - medium, 3 - heavy.

[0087] Urgency: Binary identification, 1 - urgent, 0 - normal.

[0088] Estimated duration: Directly record the number of hours, such as 8 hours.

[0089] Worker characteristics are represented as: Skill expertise: One-hot encoding, intermediate carpentry corresponds to the vector [0, 1, 0].

[0090] Years of experience: Numerically recorded, e.g., 5 years.

[0091] Degree of fatigue: Quantified in three levels, 1 - low, 2 - medium, 3 - high.

[0092] Task load: Cumulative working hours, e.g., 10 hours.

[0093] The way of combining feature vectors is that the task vector xt = [0, 1, 0, 2, 1, 8], and the worker vector xw = [0, 1, 0, 5, 2, 10].[[]END]]

[0094] The combined input vector x = [xt, xw] = [0, 1, 0, 2, 1, 8, 0, 1, 0, 5, 2, 10].[[]END]]

[0095] The model outputs a probability value between 0 and 1, representing the probability that the worker is suitable for the task. For example, if the model outputs 0.85, it means the worker has an 85% probability of being suitable for the task.

[0096] The matching model is a simple multi-layer perceptron (MLP), whose structure includes an input layer that receives the combined vector of task features and worker features; a hidden layer that contains multiple neurons and uses the ReLU activation function; and an output layer, which is a single neuron that uses the Sigmoid activation function to output the matching probability. The mathematical expression of the matching model can be represented as:

[0097] Where, represents the matching probability, 、 are weight matrices, is the combined vector of task features and worker features, is the activation function, ReLU(x) = max(0, x), is the Sigmoid activation function, σ(x)=1+e-x1.

[0098] In one embodiment, the loss function uses the cross-entropy loss function to measure the difference between the predicted probability distribution and the true label distribution. The expression of the cross-entropy loss function is:

[0099] N is the number of samples, yi is the true label, taking values 0 or 1, indicating whether the worker is suitable for the task. Pi is the matching probability predicted by the model.

[0100] Suppose there is a task that requires intermediate carpentry skills, with a medium work intensity, an urgent urgency level, and an expected working duration of 8 hours. The task feature vector is xt = [0, 1, 0, 2, 1, 8]. At the same time, there is a worker with intermediate carpentry skills, 5 years of work experience, a medium current fatigue level, and a current task load of 10 hours. The worker feature vector is xw = [0, 1, 0, 5, 2, 10]. Combining these two vectors and inputting them into the matching model, that is, x = [xt, xw] = [0, 1, 0, 2, 1, 8, 0, 1, 0, 5, 2, 10]. After the forward propagation calculation of the model, assume that the model outputs a matching probability of 0.85, indicating that the worker has an 85% probability of being suitable for the task. In this way, the matching model can dynamically calculate the matching probability based on the characteristics of the task and the worker, providing a scientific basis for subsequent task allocation.

[0101] After obtaining the matching probability matrix of all workers and tasks, to achieve the optimal overall task allocation, the Hungarian algorithm is introduced. This algorithm finds the minimum-cost matching in the matrix to ensure that, on the premise of meeting the task skill requirements, workers are assigned to the most suitable tasks, aiming to improve the overall work efficiency and quality.

[0102] The matching probability matrix is a two-dimensional matrix. Its rows represent workers, and its columns represent tasks. Each element in the matrix represents the probability value that the corresponding worker is suitable for the corresponding task. Suppose there are m workers and n tasks in the system, and the matching probability matrix P can be expressed as:

[0103] Among them, pij represents the probability value that the i-th worker is suitable for the j-th task, and the range is between 0 and 1. For example, if p12 = 0.85, it means that the first worker has an 85% probability of being suitable for the second task.

[0104] In an embodiment, the Hungarian algorithm is a classic bipartite graph matching algorithm used to solve the assignment problem, especially to find the minimum-cost matching in the cost matrix. The matching probability matrix P is transformed into a cost matrix C, where the cost is defined as 1 - pij. In this way, the problem of maximizing the matching probability is transformed into the problem of minimizing the cost. The cost matrix C can be expressed as:

[0105] Create a marking matrix M of the same size as the cost matrix C, with an initial value of 0. The marking matrix is used to record which workers are assigned to which tasks. Create two sets X and Y, representing the set of unmatched workers and the set of unmatched tasks respectively. Initially, all workers are in set X and all tasks are in set Y. Select an unmatched worker i from set X. Use depth-first search (DFS) or breadth-first search (BFS) to find an augmenting path starting from worker i. An augmenting path is an alternating path that starts from an unmatched worker and reaches an unmatched task through alternating matched and unmatched edges. If an augmenting path is found, update the matching result by flipping the matching status on the path, that is, changing the unmatched edges on the path to matched edges and the matched edges to unmatched edges. During the process of finding the augmenting path, update the marking matrix M. If worker i is assigned to task j, set Mij to 1, otherwise set it to 0. Repeat the above steps until all workers in set X are matched or no more augmenting paths can be found. Finally, the positions with value 1 in the marking matrix M represent the optimal matching result, that is, which workers are assigned to which tasks.

[0106] For example, there are 3 workers and 3 tasks, and the matching probability matrix P is P = 0.85 0.70 0.65 0.60 0.80 0.75 0.40 0.50 0.90. Convert the matching probability matrix P into a cost matrix C, C = 0.15 0.30 0.35 0.40 0.20 0.25 0.60 0.50 0.10. Apply the Hungarian algorithm, initialize the marking matrix M as a matrix of all zeros. Start from the first worker and find an augmenting path. Suppose the path 1→1 is found, update the marking matrix M = 1 0 0 0 0 0 0 0 0. Start from the second worker and find an augmenting path. Suppose the path 2→2 is found, update the marking matrix M = 1 0 0 0 1 0 0 0 0. Start from the third worker and find an augmenting path. Suppose the path 3→3 is found, update the marking matrix M = 1 0 0 0 1 0 0 0 1. Finally, the marking matrix M represents the optimal matching result: worker 1 is assigned to task 1, worker 2 is assigned to task 2, and worker 3 is assigned to task 3. In this way, the Hungarian algorithm can effectively find the optimal task assignment scheme from the matching probability matrix, ensuring that each worker is assigned to the most suitable task, thereby improving the overall work efficiency and quality.

[0107] In one embodiment, the operating state of the device is monitored in real time, including the current usage of the device, i.e., idle, in operation, in failure; location information; remaining fuel or power, and maintenance records, etc. Construct these state information into a device state vector as the basis for subsequent scheduling decisions. Integrate the above-mentioned monitored state information into a device state vector as the basis for subsequent scheduling decisions. The device state vector can be expressed as:

[0108] Among them, U represents the current usage of the device, which is represented by an integer: 0 indicates idle, 1 indicates in operation, and 2 indicates in failure. L represents the location information of the device, which is represented by a two-dimensional coordinate: (x, y) represents the horizontal and vertical coordinate positions of the device at the construction site. F represents the remaining fuel or power of the device, which is represented by a percentage. For example, 80 means the remaining fuel or power is 80%. P represents the performance index of the device, which is represented by a numerical value. For example, 1.0 indicates that the device has good performance, 0.5 indicates that the device's performance has declined, and 0 indicates that the device has completely failed. M represents the maintenance record of the device, which is represented by a timestamp and a text description: For example, 2024-03-25: Replace the filter, and the device is in good condition.

[0109] For each construction task, analyze its specific requirements for the device, such as the device type, i.e., crane, excavator, etc.; the required quantity, usage duration, and the priority of the task. High-priority tasks will be preferentially allocated device resources.

[0110] In one embodiment, when multiple tasks have demands for the same device, the system starts a path planning algorithm to optimize the scheduling path of the device. Taking the current position of the device as the starting point and the task position as the ending point, a weight map is constructed by comprehensively considering factors such as terrain obstacles and traffic restrictions at the construction site. Through an evaluation function, that is, the sum of the actual cost function g(n) and the heuristic function h(n), g(n) represents the actual movement cost from the starting point to the current node, and h(n) estimates the optimal movement cost from the current node to the ending point. Search for the optimal path in the weight map so that the device can move from the current position to the demand point in the shortest time and with the lowest energy consumption, improving the turnover efficiency of the device and reducing the project duration delay caused by unreasonable device scheduling.

[0111] The weight map is the basis of path planning. It comprehensively considers factors such as terrain obstacles and traffic restrictions at the construction site and provides the necessary environmental information for the path planning algorithm.

[0112] The construction site is divided into multiple grid cells according to terrain obstacles, and the weight value of each cell represents the difficulty for the device to pass through the cell. Passable area: The weight value is relatively low, for example, 1, indicating that the device can pass through normally. Obstacle area: The weight value is relatively high, for example, 100, indicating that the device cannot pass through. Restricted area: The weight value is between the passable area and the obstacle area, for example, 5, indicating that the device can pass through but requires additional time or energy consumption.

[0113] Consider the traffic rules at the construction site, such as one-way traffic, speed limit areas, etc. One-way traffic area: The weight value is set to a relatively high value, for example, 10, indicating that the device can only pass through in one direction. Speed limit area: The weight value is adjusted according to the speed limit ratio. For example, the weight value of the speed limit area is twice that of the normal area.

[0114] Consider the dynamic changes at the construction site, such as temporary obstacles, the movement paths of other equipment, etc. Temporary obstacles: Dynamically adjust the weight values. For example, temporarily increase the weight value of the affected area to 50. Other equipment paths: Avoid conflicts with the paths of other equipment and increase the weight value of other equipment paths to 20.

[0115] Search for the optimal path in the weight map through the evaluation function f(n)=g(n)+h(n). Create two sets, namely the open set Open and the closed set Closed. Add the starting point S to the open set Open and set its g(S)=0 and h(S) as the heuristic estimated value from the starting point to the end point. Select the node n with the smallest f(n) from the open set Open. If n is the end point G, the search is successful, and the path is constructed by backtracking the parent nodes. Otherwise, move n from the open set Open to the closed set Closed. Perform the following operations on each neighbor node m of n: If m is impassable or already in the closed set Closed, skip this node. Calculate g(m): The actual movement cost from the starting point S to m. Calculate h(m): The heuristic estimated cost from m to the end point G. Calculate f(m)=g(m)+h(m). If m is not in the open set Open, add it to the open set and set its parent node as n. If m is already in the open set Open and the new g(m) is smaller, update the g(m) and the parent node of m. If the open set Open is empty, the search fails, indicating that no path can be found. If the end point G is found, construct the optimal path from the starting point S to the end point G by backtracking the parent nodes.

[0116] In one embodiment, based on the consumption data of various materials in historical projects, including the material usage amounts under different construction stages and different construction areas, etc., train a long short-term memory network (LSTM) model to predict the material consumption in each stage of the current project. Use factors such as construction progress, completed workload, remaining task amount, and construction technology as model inputs. After the LSTM network learns and processes the time series data, output the estimated consumption amounts of each material within a specific future time period to provide data support for the material replenishment plan.

[0117] In one embodiment, various types of data during the construction process are continuously collected, including progress data, such as the amount of completed tasks and the deviation between the actual construction period and the planned construction period; resource data, such as the utilization rate and remaining quantity of manpower, equipment, and materials; and environmental data, such as meteorological conditions at the construction site, including temperature, humidity, illumination, and noise level. The collected data is cleaned and normalized to eliminate data noise and dimensional differences, making it suitable for subsequent analysis and modeling. Dimensionality reduction techniques, such as principal component analysis (PCA), are used to extract key features related to risks from the massive data. For example, features such as task delay rate and critical path progress deviation are extracted from the progress data; features such as resource depletion rate and resource conflict frequency are extracted from the resource data; and features such as the frequency of extreme weather occurrences and the duration of environmental factor exceedance are extracted from the environmental data. Through dimensionality reduction, not only can the data dimension be reduced and the model training efficiency be improved, but also the core information related to risks can be highlighted, enabling the model to focus more on key risk factors. Classification algorithms in supervised learning, such as support vector machines and random forests, are used to build a risk identification model. Samples marked as having risks in the historical construction data, such as data records that have caused construction period delays, quality defects, safety accidents, etc., are used as positive samples, and samples without risks are used as negative samples to train the model. The model learns the complex mapping relationship between risks and features in the data. When new construction data is input, it can accurately determine whether there are risks in the current construction status and the types of risks, such as schedule risks, quality risks, safety risks, etc., providing a basis for subsequent risk early warning and response.

[0118] For various extracted risk characteristic indicators, analyze their statistical distribution characteristics under normal construction conditions, and calculate statistical parameters such as the mean and standard deviation. According to the 3σ principle, take the mean ± 3 times the standard deviation as the upper and lower thresholds for risk warning. When the risk characteristic indicator exceeds this range, it is regarded as a possible risk, and a corresponding level of warning signal is triggered. For example, for the risk characteristic of task delay rate, if its mean is μ and the standard deviation is σ, when the task delay rate > μ + 3σ, a high-level warning is triggered, indicating that there may be a serious schedule risk. As the construction project progresses and data continues to accumulate, the distribution of risk characteristics may change. Therefore, an online learning mechanism is introduced to regularly update and optimize the risk identification model and warning thresholds. The system collects new construction data in real time, adds it to the training dataset, and retrains the model so that the model can adapt to the dynamic changes during the construction process, continuously improving the accuracy and timeliness of risk warning, and ensuring that the warning signal can timely and accurately reflect the risk status at the construction site. For example, in a large commercial building decoration project, the system monitors the construction process in real time through the above risk identification algorithm. When it comes to the indoor wall paint construction stage, continuous environmental humidity data monitoring for several days shows that the humidity remains high and is close to saturation. The system extracts the key risk characteristic of environmental humidity through principal component analysis and finds that it has exceeded the warning threshold set based on statistics, that is, the mean + 2 times the standard deviation. At the same time, combined with the schedule data, it is found that due to the high humidity, the drying time of the paint has been extended, and the actual progress of the wall paint construction task has fallen behind the planned progress, and the task delay rate indicator has also triggered the warning threshold. The risk identification model comprehensively analyzes the risk characteristics in terms of both environment and schedule, and determines that the current construction faces schedule risk and quality risk, that is, paint construction in a high humidity environment may cause quality problems such as coating blistering and peeling. The system immediately issues a warning signal and automatically generates an adjustment plan according to the pre-set risk response strategy: First, adjust the construction sequence and advance the floor laying task that is less affected by humidity to reduce the project duration loss; Second, increase the investment in dehumidification equipment to speed up the drying of the wall and ensure the quality of paint construction; Third, communicate with the material supplier to prepare sufficient paint materials in advance to prevent material supply shortages due to construction delays. Through these measures, the project team can respond to risks in a timely manner, minimize potential losses, and ensure the smooth progress of the project.

[0119] In one embodiment, by integrating natural language processing (NLP) technology, construction instructions, task assignments, and progress reports are automatically parsed and generated. For example, the system can automatically generate a daily progress report based on the construction log and push it to the relevant responsible persons. The system also supports mobile access, which is convenient for on-site personnel to update the construction status and feedback problems in real time. For example, workers can report the task completion situation or feedback construction problems through the mobile phone APP, and the system will automatically update the scheduling plan and notify the relevant personnel.

[0120] In one embodiment, before being put into use, the project team needs to receive system operation training to learn how to use the front-end interface to view scheduling information, provide feedback on construction status, and process the suggestions generated by the system. The system maintenance team is responsible for regularly updating the algorithm model, fixing system vulnerabilities, and optimizing system performance to ensure the long-term stable operation of the system.

[0121] As Figure 2 shown, an embodiment of the present application further provides a dynamic scheduling device for building decoration, including: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that a dynamic scheduling device for building decoration can execute: the method described in any one of the above embodiments.

[0122] An embodiment of the present application further provides a non-volatile computer storage medium storing computer-executable instructions, and the computer-executable instructions are configured to: the method described in any one of the above embodiments.

[0123] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software improvements (improvements to method flows). However, with the development of technology, many method flow improvements today can be regarded as direct improvements to hardware circuit structures. Almost all designers obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement to a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user programming the device. Designers can program themselves to "integrate" a digital system on a single PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compilers used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a hardware description language (HDL), and there is not just one type of HDL, but many types, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones currently are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that by simply performing some logical programming on the method flow using the above-mentioned several hardware description languages and programming it into the integrated circuit, it is easy to obtain the hardware circuit that implements the logical method flow.

[0124] The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium that stores computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to logically program the method steps to enable the controller to be implemented in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, and embedded microcontrollers to achieve the same function. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as the structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or the structures within the hardware component.

[0125] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0126] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0127] Each embodiment in this application is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments.

[0128] The device, medium, and method provided by the embodiments of the present application correspond one by one. Therefore, the device and the medium also have beneficial technical effects similar to those of the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the device and the medium will not be elaborated here.

[0129] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the specified functions in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0130] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the specified functions in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the specified functions in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0132] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and a memory.

[0133] Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0134] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0135] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0136] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A dynamic scheduling method for building decoration, characterized in that, Including: Determine a pre-set neural network model group, and perform deep learning training on the neural network model group according to multi-dimensional construction data obtained in advance to obtain a dynamic scheduling model group. The dynamic scheduling model group includes a prediction model, an optimization model, and a dynamic adjustment model. The multi-dimensional construction data includes environmental parameters, equipment operation data, personnel positions, and material inventory information; Determine construction tasks and worker characteristics, and import the construction tasks and worker characteristics into the dynamic scheduling model group to determine the construction situation through the prediction model, and input the construction situation into the optimization model to determine a construction scheduling plan through the optimization model; Carry out construction according to the construction scheduling plan, and conduct real-time monitoring of the construction site to obtain real-time data of the construction site, and determine whether there are emergencies at the construction site according to the real-time data; If there are emergencies at the construction site, import the emergencies into the dynamic adjustment model to determine a scheduling adjustment plan through the dynamic adjustment model, and re-schedule the construction site through the scheduling adjustment plan.

2. The method according to claim 1, wherein Performing deep learning training on the neural network model group according to the multi-dimensional construction data obtained in advance specifically includes: Preprocess the multi-dimensional construction data, convert the preprocessed multi-dimensional construction data to obtain a feature vector, and input the feature vector into an LSTM neural network model for training to obtain the dynamic scheduling model. The expression of the dynamic scheduling model is: Among them, is the activation value of the input gate, is the weight matrix, is the bias, is the value of the hidden state at time step t - 1, is the input feature vector at time step t, is the sigmoid function, is the activation value of the forget gate, is the weight matrix, is the bias, is the value of the cell state at time step t, is the old cell state, is the weight matrix, is the bias, is the activation value of the output gate, is the weight matrix, is the bias, is the value of the hidden state at time step t, is the hyperbolic tangent activation function.

3. The method according to claim 1, wherein Performing deep learning training on the neural network model group according to the multi-dimensional construction data obtained in advance specifically further includes: Preprocess the multi-dimensional construction data, convert the preprocessed multi-dimensional construction data to obtain a feature vector, and input the feature vector into an MDP model for training to obtain the optimization model. The expression of the optimization model is: Among them, is the core value function in the learning algorithm, representing the expected cumulative reward value of executing action a in the current state s, is the learning rate, is the reward function, is the discount factor, represents taking the maximum value of the Q values corresponding to all actions a' in the new state s'.

4. The method according to claim 1, characterized in that Inputting the construction situation into the optimization model specifically includes: Determine the corresponding chromosome encoding according to the construction situation to obtain the optimization model. The expression of the optimization model is: Among them, is the fitness value, , is the weight coefficient, is the total construction period, is the minimum value of resource utilization rate, is the maximum value of resource utilization rate.

5. The method according to claim 1, wherein Determine construction tasks and worker characteristics specifically including: Determine pre-set job types and skill levels, and determine pre-set work intensities; Determine the urgency level corresponding to the construction task, and determine the working hours corresponding to the construction task according to historical construction data; Determine the skill specialties corresponding to the workers according to the job types and skill levels, and determine the working years, fatigue levels, and task loads of the workers.

6. The method according to claim 1, wherein The method further includes: Preprocess the construction tasks and worker characteristics to obtain an input vector, and import the input vector into a pre-set matching model; Obtain the construction situation from the prediction model through the matching model, and determine a worker allocation plan according to the construction situation; Input the worker allocation plan into the optimization model so that the optimization model outputs the construction scheduling plan according to the worker allocation plan; And input the worker allocation plan into the dynamic adjustment model, so that the dynamic adjustment model outputs the scheduling adjustment plan according to the worker allocation plan.

7. The method according to claim 6, wherein The method further includes: Determine the input layer, hidden layer, and output layer of the matching model; Obtain the input vector through the input layer, and output the matching probability through the output layer to determine the worker allocation plan according to the matching probability; The expression of the matching model is: Among them, represents the matching probability, , is the weight matrix, is the input vector, is the activation function, is the Sigmoid activation function.

8. The method according to claim 1, wherein The method further includes: Determine the number of tasks corresponding to the construction equipment. If the number of tasks is greater than 1, then Determine the construction map, divide the construction map to determine a plurality of grid cells, and determine a plurality of weights corresponding to the plurality of grid cells; Mark the construction map according to the plurality of weights to obtain a weight map Obtain the weight map through a pre-set planning model to determine the scheduling path of the construction equipment according to the weight map.

9. A dynamic scheduling device for building decoration, characterized in that, Includes: At least one processor; And, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor. The instructions are executed by the at least one processor so that a dynamic scheduling device for building decoration can execute: the method according to any one of claims 1-8.

10. A non-volatile computer storage medium stores computer-executable instructions, characterized in that, The computer-executable instructions are set to: the method according to any one of claims 1-8.

Images