Power grid construction subpackage management and control method based on big data

Abstract

The invention discloses a power grid construction subpackage management and control method based on big data, and the method comprises the following steps: collecting the data of a power grid construction site in real time, carrying out the preprocessing, and carrying out the storage through a distributed storage system; generating a comprehensive evaluation report by using the residual neural network and the graph convolutional network; adopting a particle swarm optimization algorithm to carry out subpackage provider selection and task allocation; predicting the construction progress through a long short-term memory network; performing image analysis on the video stream of the construction site by using a residual neural network and a single-time multi-frame detection algorithm; a Bayesian network and a multi-layer perceptron are used to predict construction risks, an isolated forest algorithm is used to carry out abnormal value detection on data, and a risk early warning report is generated and notified to related personnel. According to the invention, big data, deep learning and reinforcement learning technologies are fully combined, the power grid construction subpackage management and control method based on big data is described in detail, and the method has the advantages of high construction efficiency, accurate risk prediction and resource optimization.

Description

Technical Field

[0001] The present invention relates to the field of power grid construction, and in particular to a power grid construction subcontracting management and control method based on big data. Background Art

[0002] With the rapid development of the power industry, the scale of power grid construction continues to expand, and the complexity and management difficulty of construction tasks are also increasing. Power grid construction usually involves the collaborative work of multiple subcontractors, and each subcontractor is responsible for different types of construction tasks with large differences in scale. In traditional power grid construction, the selection of subcontractors, task allocation, progress control, quality monitoring and other links are mostly based on manual judgment and experience, and lack systematic and intelligent management methods. This method is not only inefficient, but also easily interfered by subjective factors, resulting in project progress delays, frequent quality problems, and even increased construction safety hazards. Therefore, how to improve subcontractor management, task scheduling and construction quality monitoring during power grid construction has become an urgent problem to be solved in the industry.

[0003] At present, the traditional construction management methods of power grid construction mainly rely on manual management and traditional information technology. Most project management still uses offline coordination methods, and the tracking and handling of construction progress and quality issues rely on the judgment of on-site staff. The selection of subcontractors and the allocation of tasks usually rely on the experience and historical data of project managers. Although some power companies have also tried to use information systems for project management, these systems often fail to achieve comprehensive data integration and intelligent analysis, resulting in the inability to fully utilize data, making it difficult for project management to adjust in real time, and difficult to cope with complex and changing construction site conditions. For example, most existing systems lack monitoring and analysis of real-time data on the construction site, and only rely on regular manual inspections, making it difficult to promptly discover and solve problems that arise during the construction process. For the selection of subcontractors, existing technologies mainly rely on previous cooperation records and completion of construction periods, but these data are often one-sided and static, unable to provide a comprehensive and real-time evaluation basis, and difficult to adapt to the rapid response needs of changes in project progress and tasks.

[0004] In addition, traditional construction subcontractor evaluation methods mostly rely on manual experience to judge quality and progress, lack of scientific and effective evaluation models, and difficult to accurately optimize the scheduling of construction tasks and subcontractors. Especially in complex construction projects involving multiple subcontractors, the coordination of construction progress, resource consumption and task allocation is more prominent. How to achieve the optimal allocation of construction tasks while ensuring quality and maximize project benefits has become one of the shortcomings of traditional methods. In the process of subcontractor evaluation, traditional methods are highly subjective and cannot fully consider the potential laws and relationships in historical construction data.

[0005] Most of the existing construction risk prediction and management methods rely on empirical data and static models, lacking real-time and dynamic features. During the construction process, risk factors are complex and changeable, including safety risks, quality risks, schedule risks, etc., and the occurrence of risks is often highly accidental. Traditional methods cannot quickly respond to risk events that occur on site. Traditional risk prediction models are generally based on static data for calculations, and do not effectively utilize real-time monitoring data and dynamic feedback, resulting in low prediction accuracy. In the actual construction process, especially in complex projects or large-scale power grid construction, there are often risk problems such as project progress lag, unqualified quality, and construction delays, and these problems are often not promptly warned and handled in traditional systems.

[0006] In terms of task scheduling, traditional power grid construction project management relies on simple timetables and empirical rules, which makes project scheduling lack flexibility and adaptability when facing complex construction environments and changes. For example, in terms of construction resource allocation, construction schedule, and progress scheduling, existing technologies generally fail to make dynamic adjustments based on real-time data and are ineffective in responding to changes in the construction environment. Therefore, how to combine emerging technologies such as big data, artificial intelligence, and deep learning to optimize the scheduling of construction tasks in real time through in-depth analysis and intelligent processing of construction data has become the key to improving the efficiency and quality of power grid construction.

[0007] Traditional construction quality monitoring methods mainly rely on manual inspections and regular quality assessments, which results in the inability to timely discover and deal with quality problems at the construction site. Especially in some large-scale construction projects, the workload of quality inspections at the construction site is huge and complex. The traditional manual inspection mode cannot achieve real-time monitoring and comprehensive coverage, and is prone to missed inspections and wrong inspections, which in turn affects the construction quality. In addition, there are a large number of safety hazards in construction, such as workers' illegal operations, equipment failures, and missing materials. Traditional quality monitoring systems can usually only conduct quality assessments after the fact, lacking real-time dynamic monitoring and alarm mechanisms for the construction site, making it difficult to take effective measures at the early stage of the problem.

[0008] Based on the defects of the above-mentioned existing technologies, how to use big data technology, artificial intelligence algorithms and deep learning models to comprehensively improve the management capabilities of various links such as subcontractor control, task scheduling, progress prediction, quality monitoring and risk warning in power grid construction has become an important topic in the field of power grid construction today. The shortcomings of the existing methods have prompted the demand for innovative solutions based on advanced technologies such as big data, deep learning, and reinforcement learning. In particular, in terms of dynamic adjustment of construction tasks, real-time prediction of construction progress, automatic identification of quality problems, and early warning of construction risks, the existing technology cannot meet the needs of rapidly changing construction environments and task scheduling. Therefore, developing an intelligent construction management method based on big data analysis and artificial intelligence technology can not only solve the bottleneck problems in the existing technology, but also greatly improve the overall efficiency and quality of power grid construction. Summary of the invention

[0009] One purpose of the present invention is to propose a method for controlling subcontracts in power grid construction based on big data. The present invention improves the intelligence level of subcontract control in power grid construction by combining technologies such as big data, artificial intelligence and deep learning, and realizes accurate evaluation of subcontractors, dynamic optimization of construction tasks, accurate prediction of construction progress, automatic quality monitoring and real-time risk warning, which significantly improves construction management efficiency, construction quality and safety, effectively reduces costs and construction periods, and ensures that projects are completed on time and with high quality.

[0010] A method for controlling power grid construction subcontracting based on big data according to an embodiment of the present invention includes the following steps:

[0011] S1. Real-time collection of multi-dimensional data from the power grid construction site; pre-processing, and use a distributed storage system to store the processed data;

[0012] S2. Use residual neural network to evaluate the multi-dimensional data of subcontractors, and further process the relationship between subcontractors through graph convolution network to generate a comprehensive evaluation report;

[0013] S3. Based on the comprehensive evaluation report and project requirements, the particle swarm optimization algorithm is used to select subcontractors and allocate tasks, and the optimal subcontractor combination and corresponding task allocation plan are obtained;

[0014] S4. Apply the deep Q network model to dynamically adjust and optimize the construction tasks and generate an optimized task scheduling plan;

[0015] S5. Use multi-dimensional data to predict the construction progress through a long short-term memory network to obtain real-time prediction results of the construction progress;

[0016] S6. Use residual neural network and single multi-frame detection algorithm to perform real-time image analysis on the video stream of the construction site to automatically identify construction quality problems;

[0017] S7. Based on the real-time prediction results of construction progress and construction quality issues, Bayesian networks and multi-layer perceptrons are used to predict construction risks, and the isolation forest algorithm is used to detect outliers in real-time data, generate risk warning reports and notify relevant personnel.

[0018] Optionally, the multi-dimensional data includes construction site environment data, worker attendance, equipment status, material consumption, and construction quality, and the preprocessing includes removing noise, standardizing data, and filling missing values;

[0019] Optionally, the S2 specifically includes:

[0020] S21. Training the multi-dimensional data of the subcontractor through a residual neural network model, and using the trained residual neural network model to evaluate the multi-dimensional data of each subcontractor to obtain an evaluation score for each subcontractor, where the evaluation score is used to measure the comprehensive construction capability of the subcontractor;

[0021] S22. Construct a relationship graph of subcontractors, where the nodes of the relationship graph are the subcontractors, and the edges represent the relationships between the subcontractors. The weights of the edges are weighted according to the historical cooperation data and the project success rate index. The subcontractor relationship graph is convolved through a graph convolution network, and the node information is updated through the information of the adjacent nodes:

[0022]

[0023] in, is the representation of the i-th node at the l+1th layer, N(i) is the set of neighbor nodes of the i-th node, c ij is the normalization coefficient, W (l) is the weight matrix of the lth layer, b (l) is the bias term, σ is the activation function;

[0024] S23. The subcontractor relationship information processed by the graph convolutional network is integrated with the evaluation score of the residual neural network model to generate a comprehensive evaluation report, which includes the comprehensive capability score of each subcontractor and the overall performance under the network effect.

[0025] Optionally, the S3 specifically includes:

[0026] S31. Determine the scope of subcontractor selection based on the subcontractor evaluation scores provided in the comprehensive evaluation report;

[0027] S32. Setting the objective function of the project, wherein the objective function is to comprehensively evaluate the selection of subcontractors and the task allocation plan to achieve the optimal construction goal:

[0028] f(x)=w1C(x)+w2T(x)+w3Q(x);

[0029] Among them, C(x) is the total cost under the selected plan, T(x) is the construction period, Q(x) is the quality, w1, w2, w3 are the weights of each goal in the objective function, x is the task allocation plan, and f(x) is the objective function;

[0030] S33, defining the particle position in the particle swarm optimization algorithm, wherein the particle position represents the decision variable for subcontractor selection and task allocation, and the particle position vector can be expressed as:

[0031] x i =(x i1 ,x i2 ,…,x in );

[0032] Among them, x i is the current position of the ith particle, x ij is the allocation of the i-th particle to the j-th task. The velocity vector of each particle represents the rate of change of task allocation:

[0033] v i =(v i1 ,v i2 ,…,v in );

[0034] Among them, v i is the current velocity of the ith particle, v ij is the allocation of the i-th particle to the j-th task;

[0035] S34, initialize the particle swarm, randomly generate the initial position and velocity for each particle according to the constraints given in the comprehensive evaluation report, and ensure that the decision variables of each particle meet the resource constraints of the project;

[0036] S35. Update the speed and position of each particle through information exchange and local search between particles. The update formula of each particle is:

[0037]

[0038]

[0039] in, is the velocity of the ith particle in the dth dimension, is the current position of the ith particle in the dth dimension, is the historical optimal position of the ith particle in the dth dimension, is the global optimal position, r1 and r2 are random numbers, c1 and c2 are acceleration constants, and w is the inertia weight;

[0040] S36. Generate the optimal subcontractor selection scheme and task allocation scheme based on the results of particle swarm optimization. Each subcontractor is assigned a corresponding task based on the comprehensive evaluation score. The task allocation scheme needs to meet all constraints of the project and optimize the overall objective function.

[0041] S37. Generate a corresponding task allocation report based on the optimal subcontractor selection and task allocation plan, wherein the task allocation report includes the tasks, resource requirements, construction schedule, and related costs of each subcontractor.

[0042] Optionally, the S4 specifically includes:

[0043] S41. According to the current state of the construction task, define the state space in the deep Q network model. The state space is all possible states of the construction task, including the work progress, task completion status, resource consumption, and construction environment changes of each subcontractor. The state space is:

[0044] S={s1,s2,…,s n};

[0045] Where S is the state space, s i is the state of the construction task at the i-th moment;

[0046] S42, define the action space in the deep Q network model, the action space is all possible adjustment operations of the construction task, including assigning new tasks, adjusting the construction sequence, and reallocating resources. The action space is:

[0047] A={a1,a2,…,a m};

[0048] Among them, A is the action space, a j is the adjustment operation of the construction task at the jth moment;

[0049] S43. According to the specific requirements of the construction task and the project goals, a reward function of the deep Q network model is constructed. The reward function is the benefit after each action is executed, including a comprehensive score of construction progress, quality, and resource utilization. The reward function is:

[0050] R(s t ,a t )=αP t +βQ t -γC t ;

[0051] Among them, R(s t ,a t ) is the reward function, P t is the current task progress, Q t Score the construction quality, C t is the cost of task execution, α, β, γ are weighting coefficients;

[0052] S44, initialize the policy function in the deep Q network model, the policy function determines which action to select in the current state, and the policy function is:

[0053]

[0054] Among them, π(s t ) is in state s t The optimal action to be performed under t ,a) indicates that in state s t The expected reward after executing action a;

[0055] S45. Based on the deep Q network model, through multiple rounds of experiments and feedback, the state-action value function is updated. This function represents the long-term benefit of performing a certain action in a given state. The update formula is:

[0056]

[0057] Among them, α is the learning rate, which determines the influence of new information on known information, γ is the discount factor, which indicates the influence of future rewards on current decisions, and a ′ For possible actions at the next moment;

[0058] S46, by iteratively updating the strategy function and the state-action value function, the scheduling scheme of the construction tasks is gradually optimized. After each update, the system re-evaluates the effect of the current task scheduling scheme, optimizes task allocation, resource scheduling and construction period arrangement, and maximizes the overall benefit of the construction project;

[0059] S47. Based on the final updated strategy function, an optimized construction task scheduling plan is generated, wherein the scheduling plan includes a specific task execution sequence, resource allocation arrangement, and construction schedule, and the plan is applied to the actual construction process for execution.

[0060] Optionally, the S5 specifically includes:

[0061] S51, generating a multidimensional data set and performing preprocessing, wherein the preprocessing includes normalizing the data, removing outliers, and filling in missing data;

[0062] S52. Design and train a long short-term memory network model. The long short-term memory network model updates the network state through a recursive calculation formula:

[0063] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0064] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0065] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0066] C t =f t *C t-1 +i t *tanh(W C ·[h t-1 ,x t ]+b C );

[0067] h t =o t *tanh(C t );

[0068] Among them, f t For the forget gate, i t is the input gate, o t is the output gate, C t is the unit state, h t is the output state, x t is the input data, W f , W i , W o , W C is the weight matrix, W f , W i , W o , W C is the weight matrix, b f , b i , b o , b C is the bias term, σ is the activation function;

[0069] S54. During the training of the long short-term memory network model, the gradient descent method is used to update the parameters, and the optimization goal is to minimize the error between the predicted value and the actual value:

[0070]

[0071] in, is the predicted construction progress, y i is the actual construction progress, L(θ) is the loss function;

[0072] S55. Use the trained long short-term memory network model to predict the construction progress;

[0073] S56. Generate a construction progress forecast report, wherein the construction progress forecast report includes the expected construction progress, task completion status, and resource requirements in the future.

[0074] Optionally, the S6 specifically includes:

[0075] S61, collecting a video stream of the construction site, and preprocessing each frame of the collected video stream, wherein the video stream is captured in real time by a plurality of high-definition cameras installed at the construction site, wherein the cameras cover the main area of ​​the construction site and provide high-quality video data;

[0076] S62, extracting features from each frame of the processed video image using a residual neural network;

[0077] S63, inputting the features extracted by the residual neural network into a fully connected neural network to identify and classify construction quality problems, where the construction quality problems include worker violations, material defects, and equipment failures;

[0078] S64, using a single multi-frame detection algorithm to locate the location of a specific problem in the image, and generating multiple candidate frames, each candidate frame corresponding to a possible construction quality problem;

[0079] S65. Further confirm the construction quality problem based on the candidate boxes generated by the multi-box detection algorithm and the corresponding categories of the construction quality problem.

[0080] Optionally, the S7 specifically includes:

[0081] S71. Based on the construction progress data and the risk identification data, the risk identification data includes construction quality problems, equipment failures, and worker violations, a construction risk prediction model is constructed to normalize the input data;

[0082] S72. Use Bayesian network to predict construction risks and calculate the probability of each risk event by analyzing the conditional dependency between construction progress and risk identification data:

[0083]

[0084] Among them, P(ri |d) is the risk event r under the condition of given construction data d i The probability of occurrence, P(d|r i ) is the risk event r i The likelihood of observing data d under i ) is the prior probability of the risk event, P(d) is the marginal probability of data d;

[0085] S73. Use multi-layer perceptron to classify and evaluate construction risks;

[0086] S74. Use the isolation forest algorithm to detect outliers in real-time data from construction sites:

[0087]

[0088] Where S(x) is the anomaly score of data point x, E(h(x)) is the average path length of the data point in the constructed decision tree, and c(n) is a constant for the number of samples n;

[0089] S75. Combining the results of the Bayesian network, the multi-layer perceptron and the isolation forest algorithm, a comprehensive assessment report of the construction risk is generated, wherein the comprehensive assessment report lists in detail the predicted probability, risk level, possible impact and expected solution measures of each risk event, and the comprehensive assessment report also includes the priority order of different risk categories;

[0090] S76. Based on the risk levels and priorities in the comprehensive assessment report, the system will automatically trigger the early warning mechanism. When certain risk levels exceed the predetermined threshold, a risk early warning report will be generated and the report will be notified to relevant personnel via SMS, email or other means;

[0091] S77. The system updates the risk information to the construction management platform, and adjusts the construction progress and resource scheduling in real time, providing decision support so that construction management personnel can take effective measures to deal with potential construction risks.

[0092] The beneficial effects of the present invention are:

[0093] First, the present invention builds a comprehensive data collection system by collecting multi-dimensional data of the power grid construction site in real time. The real-time acquisition of these data provides sufficient basis for subsequent in-depth analysis, avoiding the lag and limitations of traditional manual monitoring. Especially when the construction site environment is complex and the construction tasks are heavy, comprehensive data collection ensures that project managers can grasp the progress of the project in real time, providing reliable data support for decision-making.

[0094] Secondly, in terms of data preprocessing, the present invention adopts multiple processing methods such as denoising, standardization, and filling missing values, and combines distributed storage technology to ensure efficient data storage and scalability. The optimization of this process not only improves data quality, but also provides a more accurate and comprehensive data foundation for subsequent intelligent analysis and modeling, thereby providing more accurate information support for various decisions in construction management.

[0095] Through the combination of residual neural network and graph convolutional network, the present invention realizes the evaluation of historical construction data of subcontractors and in-depth mining of the relationship between subcontractors. This evaluation method is more objective and scientific than the traditional manual experience evaluation. It can comprehensively examine the comprehensive capabilities of subcontractors and make the best subcontractor selection and task allocation plan based on the historical construction data. This method effectively avoids the errors caused by subjective judgment in traditional evaluation methods, ensures the accuracy of subcontractor selection and optimization of task allocation, and improves the construction efficiency of the entire power grid construction project.

[0096] In terms of dynamic adjustment and optimization of construction tasks, the present invention applies a deep Q network model to dynamically schedule the progress, resources and tasks of construction tasks. This enables construction management to be flexibly adjusted according to the real-time changing construction environment and task requirements, and optimizes the construction progress and resource allocation. Compared with the traditional static task scheduling method, the present invention can respond to various changes in the construction process in a timely manner, avoid problems such as progress delays and resource waste, and ensure that the project is completed efficiently on schedule.

[0097] In addition, the present invention also uses a long short-term memory network to predict the construction progress, ensuring the real-time and accuracy of the construction progress prediction. Compared with the traditional experience-based progress estimation method, this deep learning-based progress prediction method has a stronger real-time adjustment capability, which can help project managers to timely discover potential progress problems during the construction process and take measures to intervene in advance, thereby effectively avoiding the negative impact of construction delays.

[0098] In terms of construction quality monitoring, the present invention uses residual neural networks and multi-frame detection algorithms to perform real-time image analysis on the video stream of the construction site to automatically identify construction quality problems. This automated quality monitoring method can monitor the construction site around the clock and continuously, detect quality problems in a timely manner, and generate relevant warnings to ensure that construction quality problems can be dealt with in the shortest possible time. Compared with traditional manual inspection methods, the present invention not only improves the efficiency of quality inspection, but also reduces human omissions and improves the accuracy and reliability of quality management.

[0099] In terms of construction risk management, the present invention predicts construction risks through Bayesian networks and multi-layer perceptrons, and uses the isolation forest algorithm to detect outliers in real-time data. This method can monitor various risk factors in real time during the construction process, predict potential risks in advance and generate early warning reports, and promptly notify relevant personnel to handle emergencies. This innovative risk early warning mechanism not only improves construction safety, but also reduces project delays and cost increases caused by risk factors.

[0100] In general, the present invention proposes a new method for controlling and managing subcontracts of power grid construction by combining cutting-edge technologies such as big data, artificial intelligence, and deep learning, which significantly improves the intelligent level of construction management. Whether in terms of subcontractor selection and task allocation, construction progress prediction, quality monitoring, or prediction and early warning of construction risks, the present invention has achieved remarkable results. Through the application of these technologies, power grid construction projects can be managed more efficiently and accurately, the construction quality of the project is effectively guaranteed, and the construction risks are effectively predicted and controlled, thereby achieving a reduction in construction costs, a shortening of construction period, and an improvement in quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0102] Figure 1 This is a flow chart of a method for controlling power grid construction subcontracting based on big data proposed by the present invention;

[0103] Figure 2 The present invention provides a flowchart of subcontractor task allocation for a power grid construction subcontract management and control method based on big data. DETAILED DESCRIPTION

[0104] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0105] refer to Figure 1 , a power grid construction subcontracting management and control method based on big data, comprising the following steps:

[0106] S1. Real-time collection of multi-dimensional data from the power grid construction site; pre-processing, and use a distributed storage system to store the processed data;

[0107] S2. Use residual neural network to evaluate the multi-dimensional data of subcontractors, and further process the relationship between subcontractors through graph convolution network to generate a comprehensive evaluation report;

[0108] S3. Based on the comprehensive evaluation report and project requirements, the particle swarm optimization algorithm is used to select subcontractors and allocate tasks, and the optimal subcontractor combination and corresponding task allocation plan are obtained;

[0109] S4. Apply the deep Q network model to dynamically adjust and optimize the construction tasks and generate an optimized task scheduling plan;

[0110] S5. Use multi-dimensional data to predict the construction progress through a long short-term memory network to obtain real-time prediction results of the construction progress;

[0111] S6. Use residual neural network and single multi-frame detection algorithm to perform real-time image analysis on the video stream of the construction site to automatically identify construction quality problems;

[0112] S7. Based on the real-time prediction results of construction progress and construction quality issues, Bayesian networks and multi-layer perceptrons are used to predict construction risks, and the isolation forest algorithm is used to detect outliers in real-time data, generate risk warning reports and notify relevant personnel.

[0113] In this embodiment, the multi-dimensional data includes construction site environment data, worker attendance, equipment status, material consumption, and construction quality, and the preprocessing includes removing noise, standardizing data, and filling missing values;

[0114] In this implementation, S2 specifically includes:

[0115] S21. Training the multi-dimensional data of the subcontractor through a residual neural network model, and using the trained residual neural network model to evaluate the multi-dimensional data of each subcontractor to obtain an evaluation score for each subcontractor, where the evaluation score is used to measure the comprehensive construction capability of the subcontractor;

[0116] S22. Construct a relationship graph of subcontractors, where the nodes of the relationship graph are the subcontractors, and the edges represent the relationships between the subcontractors. The weights of the edges are weighted according to the historical cooperation data and the project success rate index. The subcontractor relationship graph is convolved through a graph convolution network, and the node information is updated through the information of the adjacent nodes:

[0117]

[0118] in, is the representation of the i-th node at the l+1th layer, N(i) is the set of neighbor nodes of the i-th node, c ij is the normalization coefficient, W (l) is the weight matrix of the lth layer, b (l) is the bias term, σ is the activation function;

[0119] S23. The subcontractor relationship information processed by the graph convolutional network is integrated with the evaluation score of the residual neural network model to generate a comprehensive evaluation report, which includes the comprehensive capability score of each subcontractor and the overall performance under the network effect.

[0120] In this implementation, S3 specifically includes:

[0121] S31. Determine the scope of subcontractor selection based on the subcontractor evaluation scores provided in the comprehensive evaluation report;

[0122] S32. Setting the objective function of the project, wherein the objective function is to comprehensively evaluate the selection of subcontractors and the task allocation plan to achieve the optimal construction goal:

[0123] f(x)=w1C(x)+w2T(x)+w3Q(x);

[0124] Among them, C(x) is the total cost under the selected plan, T(x) is the construction period, Q(x) is the quality, w1, w2, w3 are the weights of each goal in the objective function, x is the task allocation plan, and f(x) is the objective function;

[0125] S33, defining the particle position in the particle swarm optimization algorithm, wherein the particle position represents the decision variable for subcontractor selection and task allocation, and the particle position vector can be expressed as:

[0126] x i =(x i1 ,x i2 ,…,x in );

[0127] Among them, x i is the current position of the ith particle, x ij is the allocation of the i-th particle to the j-th task. The velocity vector of each particle represents the rate of change of task allocation:

[0128] v i =(v i1 ,v i2 ,…,v in );

[0129] Among them, v i is the current velocity of the ith particle, v ij is the allocation of the i-th particle to the j-th task;

[0130] S34, initialize the particle swarm, randomly generate the initial position and velocity for each particle according to the constraints given in the comprehensive evaluation report, and ensure that the decision variables of each particle meet the resource constraints of the project;

[0131] S35. Update the speed and position of each particle through information exchange and local search between particles. The update formula of each particle is:

[0132]

[0133] in, is the velocity of the ith particle in the dth dimension, is the current position of the ith particle in the dth dimension, is the historical optimal position of the ith particle in the dth dimension, is the global optimal position, r1 and r2 are random numbers, c1 and c2 are acceleration constants, and w is the inertia weight;

[0134] S36. Generate the optimal subcontractor selection scheme and task allocation scheme based on the results of particle swarm optimization. Each subcontractor is assigned a corresponding task based on the comprehensive evaluation score. The task allocation scheme needs to meet all constraints of the project and optimize the overall objective function.

[0135] S37. Generate a corresponding task allocation report based on the optimal subcontractor selection and task allocation plan, wherein the task allocation report includes the tasks, resource requirements, construction schedule, and related costs of each subcontractor.

[0136] In this implementation, S4 specifically includes:

[0137] S41. According to the current state of the construction task, define the state space in the deep Q network model. The state space is all possible states of the construction task, including the work progress, task completion status, resource consumption, and construction environment changes of each subcontractor. The state space is:

[0138] S={s1,s2,…,s n};

[0139] Where S is the state space, s i is the state of the construction task at the i-th moment;

[0140] S42, define the action space in the deep Q network model, the action space is all possible adjustment operations of the construction task, including assigning new tasks, adjusting the construction sequence, and reallocating resources. The action space is:

[0141] A={a1,a2,…,a m};

[0142] Among them, A is the action space, a j is the adjustment operation of the construction task at the jth moment;

[0143] S43. According to the specific requirements of the construction task and the project goals, a reward function of the deep Q network model is constructed. The reward function is the benefit after each action is executed, including a comprehensive score of construction progress, quality, and resource utilization. The reward function is:

[0144] R(s t ,a t )=αP t +βQ t -γC t ;

[0145] Among them, R(s t ,a t ) is the reward function, P t is the current task progress, Q t Score the construction quality, C t is the cost of task execution, α, β, γ are weighting coefficients;

[0146] S44, initialize the policy function in the deep Q network model, the policy function determines which action to select in the current state, and the policy function is:

[0147]

[0148] Among them, π(s t ) is in state s t The optimal action to be performed under t ,a) indicates that in state s t The expected reward after executing action a;

[0149] S45. Based on the deep Q network model, through multiple rounds of experiments and feedback, the state-action value function is updated. This function represents the long-term benefit of performing a certain action in a given state. The update formula is:

[0150]

[0151] Among them, α is the learning rate, which determines the influence of new information on known information, γ is the discount factor, which indicates the influence of future rewards on current decisions, and a ′ For possible actions at the next moment;

[0152] S46, by iteratively updating the strategy function and the state-action value function, the scheduling scheme of the construction tasks is gradually optimized. After each update, the system re-evaluates the effect of the current task scheduling scheme, optimizes task allocation, resource scheduling and construction period arrangement, and maximizes the overall benefit of the construction project;

[0153] S47. Based on the final updated strategy function, an optimized construction task scheduling plan is generated, wherein the scheduling plan includes a specific task execution sequence, resource allocation arrangement, and construction schedule, and the plan is applied to the actual construction process for execution.

[0154] In this implementation manner, S5 specifically includes:

[0155] S51, generating a multidimensional data set and performing preprocessing, wherein the preprocessing includes normalizing the data, removing outliers, and filling in missing data;

[0156] S52. Design and train a long short-term memory network model. The long short-term memory network model updates the network state through a recursive calculation formula:

[0157] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0158] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0159] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0160] C t =f t *C t-1 +i t *tanh(W C ·[h t-1 ,x t ]+b C );

[0161] h t =o t *tanh(C t );

[0162] Among them, f t For the forget gate, i t is the input gate, o t is the output gate, C t is the unit state, h t is the output state, x t is the input data, W f , W i , Wo , W C is the weight matrix, W f , W i , W o , W C is the weight matrix, b f , b i , b o , b C is the bias term, σ is the activation function;

[0163] S54. During the training of the long short-term memory network model, the gradient descent method is used to update the parameters, and the optimization goal is to minimize the error between the predicted value and the actual value:

[0164]

[0165] in, is the predicted construction progress, y i is the actual construction progress, L(θ) is the loss function;

[0166] S55. Use the trained long short-term memory network model to predict the construction progress;

[0167] S56. Generate a construction progress forecast report, wherein the construction progress forecast report includes the expected construction progress, task completion status, and resource requirements in the future.

[0168] Optionally, the S6 specifically includes:

[0169] S61, collecting a video stream of the construction site, and preprocessing each frame of the collected video stream, wherein the video stream is captured in real time by a plurality of high-definition cameras installed at the construction site, wherein the cameras cover the main area of ​​the construction site and provide high-quality video data;

[0170] S62, extracting features from each frame of the processed video image using a residual neural network;

[0171] S63, inputting the features extracted by the residual neural network into a fully connected neural network to identify and classify construction quality problems, where the construction quality problems include worker violations, material defects, and equipment failures;

[0172] S64, using a single multi-frame detection algorithm to locate the location of a specific problem in the image, and generating multiple candidate frames, each candidate frame corresponding to a possible construction quality problem;

[0173] S65. Further confirm the construction quality problem based on the candidate boxes generated by the multi-box detection algorithm and the corresponding categories of the construction quality problem.

[0174] In this implementation manner, the S7 specifically includes:

[0175] S71. Based on the construction progress data and the risk identification data, the risk identification data includes construction quality problems, equipment failures, and worker violations, a construction risk prediction model is constructed to normalize the input data;

[0176] S72. Use Bayesian network to predict construction risks and calculate the probability of each risk event by analyzing the conditional dependency between construction progress and risk identification data:

[0177]

[0178] Among them, P(r i |d) is the risk event r under the condition of given construction data d i The probability of occurrence, P(d|r i ) is the risk event r i The likelihood of observing data d under i ) is the prior probability of the risk event, P(d) is the marginal probability of data d;

[0179] S73. Use multi-layer perceptron to classify and evaluate construction risks;

[0180] S74. Use the isolation forest algorithm to detect outliers in real-time data from construction sites:

[0181]

[0182] Where S(x) is the anomaly score of data point x, E(h(x)) is the average path length of the data point in the constructed decision tree, and c(n) is a constant for the number of samples n;

[0183] S75. Combining the results of the Bayesian network, the multi-layer perceptron and the isolation forest algorithm, a comprehensive assessment report of the construction risk is generated, wherein the comprehensive assessment report lists in detail the predicted probability, risk level, possible impact and expected solution measures of each risk event, and the comprehensive assessment report also includes the priority order of different risk categories;

[0184] S76. Based on the risk levels and priorities in the comprehensive assessment report, the system will automatically trigger the early warning mechanism. When certain risk levels exceed the predetermined threshold, a risk early warning report will be generated and the report will be notified to relevant personnel via SMS, email or other means;

[0185] S77. The system updates the risk information to the construction management platform, and adjusts the construction progress and resource scheduling in real time, providing decision support so that construction management personnel can take effective measures to deal with potential construction risks.

[0186] Embodiment 1:

[0187] This embodiment is applied to power grid construction projects, aiming to optimize construction subcontracting management and control, improve construction efficiency, ensure project quality, and realize real-time risk warning through data-driven. The project is located in a large power grid construction area in a certain province. The project background is due to the complex power grid construction tasks in the area, the large number of construction subcontractors, the huge construction tasks and engineering volume involved, and the project has a total of 35 sub-projects, including 10 major construction tasks. Different subcontractors undertake different construction tasks in each sub-project. Managers need to coordinate the work progress, quality and resource consumption of each subcontractor, and evaluate and adjust each construction stage at the same time. Therefore, the management of construction quality, progress control and construction risks faces huge challenges.

[0188] In order to achieve efficient management of the project, the project party decided to adopt a construction subcontracting control method based on big data, through real-time monitoring of multi-dimensional data of the construction site, using deep learning, particle swarm optimization algorithm and deep Q network model to improve the decision support, risk warning and task optimization capabilities during the construction process. During the implementation of the project, by using the method of the present invention, the construction efficiency of the project has been significantly improved, and the construction risks have been effectively controlled.

[0189] Table 1 Comparison of changes in key indicators before and after implementation

[0190]

[0191]

[0192] First, from the perspective of the total construction progress, the total progress of the project before implementation was 315 days, and after implementation, by optimizing subcontractor selection, task scheduling and construction progress forecasting, it was advanced by 25 days to 290 days, an advance rate of about 8%. This means that the present invention effectively reduces the construction time in task allocation and progress control, and improves the execution efficiency of the construction project.

[0193] In terms of construction quality qualification rate, the quality qualification rate was 80% before implementation, and increased to 93% after implementation, an increase of 13%. This shows that real-time monitoring and quality control of the construction site through deep neural networks, image analysis and other technologies have effectively improved the construction quality, reduced the incidence of quality problems, and ensured higher project acceptance standards.

[0194] The number of construction risk events has also shown a significant reduction. There were 6 risk events before implementation, but it was reduced to 3 after implementation, a reduction of 50%. This is due to the accurate prediction of construction risks by the present invention through Bayesian networks and multi-layer perceptrons, which can timely discover potential problems and issue early warnings, greatly reducing the risk rate during the construction process and ensuring the smooth progress of the project.

[0195] In terms of subcontractor task scheduling optimization efficiency, before implementation, due to traditional manual scheduling and lack of effective dynamic adjustment mechanism, the task scheduling efficiency was 0%. After the introduction of particle swarm optimization algorithm and deep Q network model to dynamically optimize construction tasks, the task scheduling efficiency was increased to 15%. This improvement effectively ensures the reasonable allocation and scheduling of construction resources, avoiding resource waste and time delays.

[0196] The reduction in downtime due to accidents also reflects the significant advantages of the technical solution of the present invention in terms of construction safety. Before implementation, the downtime due to accidents was 10 days, and after implementation it was reduced to 4 days, a reduction of 60%. This shows that through accurate risk prediction and timely early warning systems, safety accidents in construction have been effectively prevented, significantly reducing downtime caused by accidents and ensuring the smooth completion of the construction period.

[0197] From the perspective of the construction delay ratio, the construction delay ratio of the project was 16.5% before implementation, and it dropped to 8% after implementation, a reduction of 8.5%. This further verifies the advantages of the present invention in dynamic scheduling and real-time early warning, which can effectively control the construction progress, reduce construction delays caused by various factors, and ensure the project is delivered on time.

[0198] In summary, through the implementation of the present invention, construction projects have achieved significant improvements in progress, quality, safety, cost and risk management. In particular, in terms of construction risk warning and construction progress optimization, the present invention has demonstrated its powerful advantages and effectively promoted the intelligent, refined and efficient project management.

[0199] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for controlling power grid construction subcontracting based on big data, characterized in that: The steps include: S1. Real-time collection of multi-dimensional data from the power grid construction site; pre-processing, and use a distributed storage system to store the processed data; S2. Use residual neural network to evaluate the multi-dimensional data of subcontractors, and further process the relationship between subcontractors through graph convolution network to generate a comprehensive evaluation report; S3. Based on the comprehensive evaluation report and project requirements, the particle swarm optimization algorithm is used to select subcontractors and allocate tasks, and the optimal subcontractor combination and corresponding task allocation plan are obtained; S4. Apply the deep Q network model to dynamically adjust and optimize the construction tasks and generate an optimized task scheduling plan; S5. Use multi-dimensional data to predict the construction progress through a long short-term memory network to obtain real-time prediction results of the construction progress; S6. Use residual neural network and single multi-frame detection algorithm to perform real-time image analysis on the video stream of the construction site to automatically identify construction quality problems; S7. Based on the real-time prediction results of construction progress and construction quality issues, Bayesian networks and multi-layer perceptrons are used to predict construction risks, and the isolation forest algorithm is used to detect outliers in real-time data, generate risk warning reports and notify relevant personnel.

2. According to a method for controlling power grid construction subcontracting based on big data according to claim 1, it is characterized in that: The multi-dimensional data includes construction site environment data, worker attendance, equipment status, material consumption, and construction quality. The preprocessing includes removing noise, standardizing data, and filling missing values.

3. According to a method for controlling power grid construction subcontracting based on big data according to claim 1, it is characterized in that: The S2 specifically includes: S21. Training the multi-dimensional data of the subcontractor through a residual neural network model, and using the trained residual neural network model to evaluate the multi-dimensional data of each subcontractor to obtain an evaluation score for each subcontractor, where the evaluation score is used to measure the comprehensive construction capability of the subcontractor; S22. Construct a relationship graph of subcontractors, where the nodes of the relationship graph are the subcontractors, and the edges represent the relationships between the subcontractors. The weights of the edges are weighted according to the historical cooperation data and the project success rate index. The subcontractor relationship graph is convolved through a graph convolution network, and the node information is updated through the information of the adjacent nodes: in, is the representation of the i-th node at the l+1th layer, N(i) is the set of neighbor nodes of the i-th node, c ij is the normalization coefficient, W (l) is the weight matrix of the lth layer, b (l) is the bias term, σ is the activation function; S23. The subcontractor relationship information processed by the graph convolutional network is integrated with the evaluation score of the residual neural network model to generate a comprehensive evaluation report, which includes the comprehensive capability score of each subcontractor and the overall performance under the network effect.

4. According to a method for controlling power grid construction subcontracting based on big data according to claim 1, it is characterized in that: The S3 specifically includes: S31. Determine the scope of subcontractor selection based on the subcontractor evaluation scores provided in the comprehensive evaluation report; S32. Setting the objective function of the project, wherein the objective function is to comprehensively evaluate the selection of subcontractors and the task allocation plan to achieve the optimal construction goal: f(x)=w1C(x)+w2T(x)+w3Q(x); Among them, C(x) is the total cost under the selected plan, T(x) is the construction period, Q(x) is the quality, w1, w2, w3 are the weights of each goal in the objective function, x is the task allocation plan, and f(x) is the objective function; S33, defining the particle position in the particle swarm optimization algorithm, wherein the particle position represents the decision variable for subcontractor selection and task allocation, and the particle position vector can be expressed as: x i =(x i1 ,x i2 ,…,x in ); Among them, x i is the current position of the ith particle, x ij is the allocation of the i-th particle to the j-th task. The velocity vector of each particle represents the rate of change of task allocation: v i =(v i1 ,v i2 ,…,v in ); Among them, v i is the current velocity of the ith particle, v ij is the allocation of the i-th particle to the j-th task; S34, initialize the particle swarm, randomly generate the initial position and velocity for each particle according to the constraints given in the comprehensive evaluation report, and ensure that the decision variables of each particle meet the resource constraints of the project; S35. Update the speed and position of each particle through information exchange and local search between particles. The update formula of each particle is: in, is the velocity of the ith particle in the dth dimension, is the current position of the ith particle in the dth dimension, is the historical optimal position of the ith particle in the dth dimension, is the global optimal position, r1 and r2 are random numbers, c1 and c2 are acceleration constants, and w is the inertia weight; S36. Generate the optimal subcontractor selection scheme and task allocation scheme based on the results of particle swarm optimization. Each subcontractor is assigned a corresponding task based on the comprehensive evaluation score. The task allocation scheme needs to meet all constraints of the project and optimize the overall objective function. S37. Generate a corresponding task allocation report based on the optimal subcontractor selection and task allocation plan, wherein the task allocation report includes the tasks, resource requirements, construction schedule, and related costs of each subcontractor.

5. A method for controlling power grid construction subcontracting based on big data as described in claim 1, characterized in that: The S4 specifically includes: S41. According to the current state of the construction task, define the state space in the deep Q network model. The state space is all possible states of the construction task, including the work progress, task completion status, resource consumption, and construction environment changes of each subcontractor. The state space is: S={s1,s2,…,s n }; Where S is the state space, s i is the state of the construction task at the i-th moment; S42, define the action space in the deep Q network model, the action space is all possible adjustment operations of the construction task, including assigning new tasks, adjusting the construction sequence, and reallocating resources. The action space is: <h2 style=";text-align:left;direction:ltr">A = {a1,a2,…,a<h2 style=";text-align:left;direction:ltr"> m <h2 style=";text-align:left;direction:ltr">}; Among them, A is the action space, a j is the adjustment operation of the construction task at the jth moment; S43. According to the specific requirements of the construction task and the project goals, a reward function of the deep Q network model is constructed. The reward function is the benefit after each action is executed, including a comprehensive score of construction progress, quality, and resource utilization. The reward function is: R(s t ,a t )=αP t +βQ t -γC t ; Among them, R(s t ,a t ) is the reward function, P t is the current task progress, Q t Score the construction quality, C t is the cost of task execution, α, β, γ are weighting coefficients; S44, initialize the policy function in the deep Q network model, the policy function determines which action to select in the current state, and the policy function is: Among them, π(s t ) is in state s t The optimal action to be performed under t ,a) indicates that in state s t The expected reward after executing action a; S45. Based on the deep Q network model, through multiple rounds of experiments and feedback, the state-action value function is updated. This function represents the long-term benefit of performing a certain action in a given state. The update formula is: Among them, α is the learning rate, which determines the influence of new information on known information, γ is the discount factor, which indicates the influence of future rewards on current decisions, and a ′ For possible actions at the next moment; S46, by iteratively updating the strategy function and the state-action value function, the scheduling scheme of the construction tasks is gradually optimized. After each update, the system re-evaluates the effect of the current task scheduling scheme, optimizes task allocation, resource scheduling and construction period arrangement, and maximizes the overall benefit of the construction project; S47. Based on the final updated strategy function, an optimized construction task scheduling plan is generated, wherein the scheduling plan includes a specific task execution sequence, resource allocation arrangement, and construction schedule, and the plan is applied to the actual construction process for execution.

6. A method for controlling power grid construction subcontracting based on big data according to claim 1, characterized in that: The S5 specifically includes: S51, generating a multidimensional data set and performing preprocessing, wherein the preprocessing includes normalizing the data, removing outliers, and filling in missing data; S52. Design and train a long short-term memory network model. The long short-term memory network model updates the network state through a recursive calculation formula: f t =σ(W f ·[h t-1 ,x t ]+b f ); i t =σ(W i ·[h t-1 ,x t ]+b i ); the t =σ(W o ·[h t-1 ,x t ]+b o ); C t =f t *C t-1 +i t *tanh(W C ·[h t-1 ,x t ]+b C ); h t =o t *tanh(C t ); Among them, f t For the forget gate, i t is the input gate, o t is the output gate, C t is the unit state, h t is the output state, x t is the input data, W f , W i , W o , W C is the weight matrix, W f , W i , W o , W C is the weight matrix, b f , b i , b o , b C is the bias term, σ is the activation function; S54. During the training of the long short-term memory network model, the gradient descent method is used to update the parameters, and the optimization goal is to minimize the error between the predicted value and the actual value: in, is the predicted construction progress, y i is the actual construction progress, L(θ) is the loss function; S55. Use the trained long short-term memory network model to predict the construction progress; S56. Generate a construction progress forecast report, wherein the construction progress forecast report includes the expected construction progress, task completion status, and resource requirements in the future.

7. A method for controlling power grid construction subcontracting based on big data according to claim 1, characterized in that: The S6 specifically includes: S61, collecting a video stream of the construction site, and preprocessing each frame of the collected video stream, wherein the video stream is captured in real time by a plurality of high-definition cameras installed at the construction site, wherein the cameras cover the main area of ​​the construction site and provide high-quality video data; S62, extracting features from each frame of the processed video image using a residual neural network; S63, inputting the features extracted by the residual neural network into a fully connected neural network to identify and classify construction quality problems, where the construction quality problems include worker violations, material defects, and equipment failures; S64, using a single multi-frame detection algorithm to locate the location of a specific problem in the image, and generating multiple candidate frames, each candidate frame corresponding to a possible construction quality problem; S65. Further confirm the construction quality problem based on the candidate boxes generated by the multi-box detection algorithm and the corresponding categories of the construction quality problem.

8. A method for controlling power grid construction subcontracting based on big data according to claim 1, characterized in that: The S7 specifically includes: S71. Based on the construction progress data and the risk identification data, the risk identification data includes construction quality problems, equipment failures, and worker violations, a construction risk prediction model is constructed to normalize the input data; S72. Use Bayesian network to predict construction risks and calculate the probability of each risk event by analyzing the conditional dependency between construction progress and risk identification data: Among them, P(r i |d) is the risk event r under the condition of given construction data d i The probability of occurrence, P(d|r i ) is the risk event r i The likelihood of observing data d under i ) is the prior probability of the risk event, P(d) is the marginal probability of data d; S73. Use multi-layer perceptron to classify and evaluate construction risks; S74. Use the isolation forest algorithm to detect outliers in real-time data from construction sites: Where S(x) is the anomaly score of data point x, E(h(x)) is the average path length of the data point in the constructed decision tree, and c(n) is a constant for the number of samples n; S75. Combining the results of the Bayesian network, the multi-layer perceptron and the isolation forest algorithm, a comprehensive assessment report of the construction risk is generated, wherein the comprehensive assessment report lists in detail the predicted probability, risk level, possible impact and expected solution measures of each risk event, and the comprehensive assessment report also includes the priority order of different risk categories; S76. Based on the risk levels and priorities in the comprehensive assessment report, the system will automatically trigger the early warning mechanism. When certain risk levels exceed the predetermined threshold, a risk early warning report will be generated and the report will be notified to relevant personnel via SMS, email or other means; S77. The system updates the risk information to the construction management platform, and adjusts the construction progress and resource scheduling in real time, providing decision support so that construction management personnel can take effective measures to deal with potential construction risks.

Images