Power grid infrastructure engineering material collaborative management method

Abstract

The invention provides a power grid infrastructure project material collaborative management method, which comprises the following steps: analyzing historical material use, power grid construction progress and external environment data through a reinforcement learning algorithm, dynamically adjusting model parameters, and predicting material demands; a low-power-consumption wide area network and an edge computing technology are used for monitoring and analyzing material states in a warehouse and a transportation process in real time, and a result is transmitted to the central management system; based on the prediction data, an optimized material purchasing plan is automatically generated, and the purchasing strategy is dynamically adjusted by considering the arrival time, the inventory state and the supplier capability; using an improved ant colony algorithm to optimize the transportation route and time of the materials from the warehouse to the construction site; using an intelligent sensor to record the material use condition, and carrying out preliminary data processing and analysis through edge calculation; and an information management platform is established to realize information sharing and cooperative work. According to the method, the intelligence and efficiency of material management are improved by integrating multiple technologies.

Description

Technical Field

[0001] The present invention relates to the field of power grid infrastructure projects, and in particular to a collaborative management method for materials in power grid infrastructure projects. Background Art

[0002] In power grid infrastructure projects, material management has always been one of the key links for the smooth implementation of the project. Traditional material management methods often rely on manual experience and simple information systems. This approach has many problems, such as inaccurate demand forecasts, untimely material status monitoring, unreasonable procurement plans, insufficient optimization of transportation routes, and poor information sharing. These problems not only affect the progress of the project, but may also lead to material waste and increased costs.

[0003] In the existing technology, material demand forecasting is usually based on simple linear regression or time series analysis of historical data, which fails to fully consider the complex external environment and changes in project progress, resulting in inaccurate forecast results. Material status monitoring mainly relies on manual inspection and regular inventory, which has poor real-time and accuracy and cannot reflect the actual use and inventory status of materials in a timely manner. The formulation of material procurement plans is often based on experience and fixed rules, lacking dynamic adjustment and optimization mechanisms, resulting in a mismatch between procurement volume and actual demand, and the risk of inventory backlog or shortage.

[0004] In the process of material distribution, traditional transportation route optimization methods mostly use static algorithms, such as the shortest path algorithm or heuristic algorithm, which fails to fully consider the dynamic changes of real-time traffic conditions and material demand, resulting in low transportation efficiency. The recording and management of material usage mainly rely on manual reporting, and the data is delayed and prone to errors, making it difficult to achieve refined management. In terms of information sharing, the existing material management systems are mostly independent local area networks or stand-alone systems. The information sharing efficiency between departments and links is low, and data inconsistency and information island problems are serious, affecting the overall management efficiency and the scientific nature of decision-making.

[0005] Therefore, how to provide a method for coordinated management of materials for power grid infrastructure projects is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0006] One purpose of the present invention is to propose a collaborative management method for materials in power grid infrastructure projects. The present invention makes full use of reinforcement learning algorithms, low-power wide area networks, edge computing technology, improved ant colony algorithms, and smart sensors, and describes in detail an intelligent management method for material demand forecasting, status monitoring, procurement plan optimization, transportation route optimization, usage recording, and information sharing, which has the advantages of high prediction accuracy, strong real-time monitoring, reasonable procurement optimization, high transportation efficiency, accurate data processing, and secure information sharing.

[0007] A method for collaborative management of materials for power grid infrastructure projects according to an embodiment of the present invention includes the following steps:

[0008] S1. Adopt a material demand forecasting algorithm based on reinforcement learning, analyze historical material usage data, power grid construction progress data and external environment data, dynamically adjust the forecasting model parameters, and predict the demand for various materials in power grid infrastructure projects;

[0009] S2. Use low-power wide area networks and edge computing technology to monitor the status of materials in warehouses and during transportation in real time, process and analyze monitoring data in real time, and transmit the processing results to the central management system;

[0010] S3. Automatically generate optimized material procurement plans based on demand data output by the reinforcement learning algorithm. The procurement plan takes into account the arrival time, inventory status and supplier capabilities of the materials, and optimizes the procurement strategy through dynamic adjustment.

[0011] S4. In the process of material distribution, the improved ant colony algorithm is used to optimize the transportation route and time of materials from the warehouse to the construction site;

[0012] S5. During the use of materials, RFID technology and smart sensors are used to record the use of materials in real time, and preliminary data processing and analysis are performed through edge computing technology;

[0013] S6. Establish an information management platform to achieve information sharing and collaborative work among various departments and links.

[0014] Optionally, the step S1 specifically includes:

[0015] S11. Collect historical material usage data h , Power grid construction progress data D p and external environment data D e ;

[0016] S12. Preprocess the collected data, including data cleaning to remove outliers and missing values, fill in missing data, normalize the data, generate more training data through random sampling, smoothing and aggregation; and use principal component analysis and time series decomposition methods to extract key features from the original data, decompose the data into trend, seasonality and residual components, and form a feature data set F pre ;

[0017] S13. Construct a material demand prediction model based on reinforcement learning, wherein the material demand prediction model includes a state space S, an action space A, a reward function R and a strategy π, wherein the reward function R is expressed as:

[0018]

[0019] Where M is the number of multi-objectives in the reward function, λ k is the weight of each target, Q k (s t ,a t ) is the evaluation function of the kth objective;

[0020] S14, the preprocessed feature data set F pre Input into the material demand forecasting model and initialize the model parameters θ;

[0021] S15, through the reinforcement learning algorithm, at each time step t, according to the current material demand state s t Select action a t , and according to the reward function R(s t ,a t ) and the improved gradient ascent algorithm to update the model parameters θ;

[0022] S16. Iteratively update the model parameters θ until the model converges to obtain the final demand forecasting model:

[0023]

[0024] Among them, η is the learning rate, ∈ is the noise coefficient, N(0,σ 2 ) is the normal distribution noise;

[0025] S17. Use the final demand forecasting model to forecast the demand for various materials in power grid infrastructure projects and output the forecast results

[0026] Optionally, the step S15 specifically includes:

[0027] S151. In each iteration, calculate the current material demand forecast model parameter θ t The reward function gradient under

[0028]

[0029] Among them, λ k is the weight of the kth target, is about the current parameter θ t The evaluation function gradient of

[0030] S152, according to the current state s t and action a t The feature map φ(s t ,a t ) performs local quadratic approximation to obtain an improved gradient ascent update formula;

[0031]

[0032] in, is the Hessian matrix of the reward function, Δθ is the gradient update step size;

[0033] S153. When updating model parameters, add a normally distributed noise term to enhance exploration capability;

[0034] θ t+1 =θ t+1 +ε·N(0,σ 2 );

[0035] Where ε is the noise coefficient, N(0,σ 2 ) is the normal distribution noise;

[0036] S154. After each iteration, evaluate the model performance. If the performance does not improve, reduce the learning rate η and add a momentum term to accelerate convergence.

[0037]

[0038] Among them, α is the momentum coefficient, Δθ t-1 is the parameter change of the previous iteration;

[0039] S155. Repeat steps S151-S154 until the demand forecasting model converges.

[0040] Optionally, the step S2 specifically includes:

[0041] S21. Material status monitoring in warehouses and during transportation, using sensors and RFID technology to obtain material quantity, location and environmental condition data to form a material status data set D s ;

[0042] S22. Perform data cleaning in the central management system to remove outliers and missing values ​​and fill in missing data;

[0043]

[0044] Among them, μ is the mean, σ is the standard deviation, and D si is the i-th data point in the material status data, T(D s ) is the original material status data set D s The transformed or processed data were used to remove outliers with more than 3 standard deviations.

[0045] S23, normalizing the data;

[0046]

[0047] Among them, Dnorm is the normalized data set, D si is a single data point after cleaning;

[0048] S24, extract features from the normalized data, including extracting key features from the original data using principal component analysis and time series decomposition methods, decomposing the data into trend, seasonality and residual components, and generating a feature data set F extract , combined with data enhancement method F augmented , generate more training data:

[0049]

[0050] Among them, ε is the data enhancement coefficient, N(0,σ 2 ) is the normal distribution noise, f ij Represents the enhanced feature dataset F augmented The j-th eigenvalue of the i-th sample in ;

[0051] S25, combined with the enhanced feature dataset F augmented and material status dataset D s , forming a comprehensive data set D for material management final , and the comprehensive dataset D final Transfer to demand forecasting model.

[0052] Optionally, the step S3 specifically includes:

[0053] S31. Demand data output based on reinforcement learning algorithm Constructing a material procurement planning model, wherein the model includes procurement time T, inventory status I and supplier capacity C;

[0054] S32, according to the demand data and the current inventory status I, determine the procurement time T of the materials, where the procurement time is calculated by the optimization formula:

[0055]

[0056] Among them, λ1, λ2, λ3 are weight coefficients, T lead is the lead time, C inv (t) is the inventory cost, C stockout (t) is the stock-out cost;

[0057] S33. Generate a material procurement plan P based on the optimized procurement time T, combined with the supplier capacity C and inventory status I. The procurement plan takes into account the supplier's delivery time and supply capacity. When generating the material procurement plan, the supplier's historical delivery performance and current production capacity are taken into account, and the supplier scoring model is used for evaluation;

[0058] S34. Dynamically adjust the material procurement strategy by real-time monitoring of inventory status and demand changes, and update the procurement plan using the feedback control mechanism:

[0059]

[0060] Among them, α and β are adjustment coefficients, For new demand data, I new For the new inventory status;

[0061] S35. By iterating the above steps in a loop, in each iteration, the adjusted procurement plan and actual execution results are recorded and analyzed to improve and optimize the reinforcement learning algorithm.

[0062] Optionally, the step S4 specifically includes:

[0063] S41, initializing the parameters of the ant colony algorithm, including the pheromone concentration matrix τ, the number of ants m, the number of iterations N, the pheromone importance coefficient α, the heuristic function importance coefficient β and the pheromone volatility coefficient ρ;

[0064] S42, setting a heuristic function η, which is defined as the reciprocal of the distance between the warehouse and the construction site;

[0065] S43. For each ant, choose the next path according to probability, and choose the probability p ij Defined as:

[0066]

[0067] Among them, allowed is the set of paths that ants can choose, τ ij is the pheromone concentration of path i to j;

[0068] S44. Calculate the length L of each ant's transport route and record the best route L. best :

[0069]

[0070] Among them, d i,i+1 represents the distance from the i-th node to the i+1-th node, d n,1 Indicates the distance from the last node back to the starting point, forming a closed loop path;

[0071] S45. Adaptive pheromone update mechanism:

[0072] τ ij (t+1)=(1-ρ)·τ ij (t)+Δτ ij (t);

[0073] Among them, the pheromone increment Δτ ij (t) is calculated as:

[0074]

[0075] Among them, Q is a constant, L k (t) is the path length of the kth ant, λ is the multi-objective optimization weight, D k (t) is the multi-objective function value of the kth ant, and the pheromone increment is defined as a function of the path length and multi-objective optimization values, so that the shorter paths and the paths with excellent comprehensive performance can obtain more information pheromones, guide the ants to choose high-quality paths, and adopt The contribution of all ants is accumulated in the form of , and the pheromone volatility coefficient (1-ρ) avoids unlimited accumulation of pheromones, so that the algorithm can be dynamically adjusted to adapt to environmental changes;

[0076] S46. After each round of iteration, a local search strategy is introduced to optimize the current solution, and a 2-opt or 3-opt algorithm is used to locally adjust the current optimal path;

[0077] S47, iterating the process of S41-S46 until the termination condition is met, the maximum number of iterations is reached or the optimal route length does not change within a certain number of times;

[0078] S48, output the best transportation route L best and its transportation time, so that materials can be delivered to the construction site in the best time;

[0079] S49. Input the optimized transportation route and time parameters into the material management system to realize dynamic monitoring and management of the material transportation process.

[0080] Optionally, the step S5 specifically includes:

[0081] S51. During the use of materials, RFID technology and intelligent sensors are used to record the use of materials in real time to form a material use data set U;

[0082] S52. Perform preliminary data processing on the material usage data through edge computing devices, including data cleaning, data normalization and feature extraction;

[0083] S53, using feature extraction technology, extract key features from normalized data to form a feature data set F U :

[0084]

[0085] Among them, w i is the feature extraction weight, φ(ui ) is the characteristic mapping function of material usage data;

[0086] S54. Use edge computing technology to conduct preliminary analysis of feature data, calculate statistical information on material usage and predictive model input:

[0087] Statistics calculation:

[0088]

[0089] Trend and seasonal components:

[0090]

[0091]

[0092] Where n is the number of data points in the feature dataset, t i is the timestamp of the i-th data point;

[0093] S55, transmit the analysis results and original data to the central management system, and combine with the data of other modules for comprehensive analysis and decision support;

[0094] S56. Use edge computing devices to monitor material usage in real time, update material status and send warning information.

[0095] The beneficial effects of the present invention are:

[0096] The present invention realizes the intelligent, real-time and collaborative management of materials for power grid infrastructure projects by comprehensively applying reinforcement learning algorithm, low-power wide area network, edge computing technology, improved ant colony algorithm and intelligent sensor. The reinforcement learning algorithm is used to improve the accuracy of material demand forecasting and overcome the problem of inaccurate forecasting in traditional methods. The combination of low-power wide area network and edge computing technology realizes real-time monitoring and processing of material status in warehouses and transportation processes, solving the problem of insufficient real-time and accuracy of manual monitoring. The optimized material procurement plan generated based on the reinforcement learning algorithm takes into account the arrival time, inventory status and supplier capabilities, and avoids the risk of inventory backlog or shortage by dynamically adjusting the procurement strategy, thereby improving the rationality and responsiveness of the procurement plan. The improved ant colony algorithm is used to optimize the transportation route and time of materials from the warehouse to the construction site, dynamically adjusts the transportation route according to real-time traffic conditions and material demand, improves transportation efficiency and punctuality, and reduces transportation costs. The application of intelligent sensors and edge computing technology in the use of materials improves the real-time and accuracy of data collection, reduces the errors of manual reporting, and provides reliable data support for subsequent material management and decision-making. The establishment of an information management platform has realized information sharing and collaborative work among various departments and links, ensured the transparency and immutability of data, solved the problems of information islands and data inconsistency, and improved the overall management efficiency and scientific nature of decision-making. The invention significantly improves the efficiency and accuracy of material management and ensures the smooth progress and efficient implementation of power grid infrastructure projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] 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:

[0098] Figure 1 A flow chart of a method for collaborative management of materials for power grid infrastructure projects proposed by the present invention; DETAILED DESCRIPTION

[0099] 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.

[0100] refer to Figure 1 , a method for collaborative management of materials for power grid infrastructure projects, comprising the following steps:

[0101] S1. Adopt a material demand forecasting algorithm based on reinforcement learning, analyze historical material usage data, power grid construction progress data and external environment data, dynamically adjust the forecasting model parameters, and predict the demand for various materials in power grid infrastructure projects;

[0102] In this implementation, step S1 specifically includes:

[0103] S11. Collect historical material usage data h , Power grid construction progress data D p and external environment data D e ;

[0104] S12. Preprocess the collected data, including data cleaning to remove outliers and missing values, fill in missing data, normalize the data, generate more training data through random sampling, smoothing and aggregation; and use principal component analysis and time series decomposition methods to extract key features from the original data, decompose the data into trend, seasonality and residual components, and form a feature data set F pre ;

[0105] S13. Construct a material demand prediction model based on reinforcement learning, wherein the material demand prediction model includes a state space S, an action space A, a reward function R and a strategy π, wherein the reward function R is expressed as:

[0106]

[0107] Where M is the number of multi-objectives in the reward function, λ k is the weight of each target, Q k (s t ,a t ) is the evaluation function of the kth objective;

[0108] S14, the preprocessed feature data set F pre Input into the material demand forecasting model and initialize the model parameters θ;

[0109] S15, through the reinforcement learning algorithm, at each time step t, according to the current material demand state s t Select action a t , and according to the reward function R(s t ,a t ) and the improved gradient ascent algorithm to update the model parameters θ;

[0110] S16. Iteratively update the model parameters θ until the model converges to obtain the final demand forecasting model:

[0111]

[0112] Among them, η is the learning rate, ∈ is the noise coefficient, N(0,σ 2 ) is the normal distribution noise;

[0113] S17. Use the final demand forecasting model to forecast the demand for various materials in power grid infrastructure projects and output the forecast results

[0114] S15, through the reinforcement learning algorithm, at each time step t, according to the current material demand state s t Select action a t , and according to the reward function R(s t ,a t ) and the improved gradient ascent algorithm to update the model parameters θ;

[0115] In this implementation, step S15 specifically includes:

[0116] S151. In each iteration, calculate the current material demand forecast model parameter θ t The reward function gradient under

[0117]

[0118] Among them, λ k is the weight of the kth target, is about the current parameter θ t The evaluation function gradient of

[0119] S152, according to the current state s t and action a t The feature map φ(s t ,a t ) performs local quadratic approximation to obtain an improved gradient ascent update formula;

[0120]

[0121] in, is the Hessian matrix of the reward function, Δθ is the gradient update step size;

[0122] S153. When updating model parameters, add a normally distributed noise term to enhance exploration capability;

[0123] θ t+1 =θ t+1 +ε·N(0,σ 2 );

[0124] Among them, ∈ is the noise coefficient, N(0,σ 2 ) is the normal distribution noise;

[0125] S154. After each iteration, evaluate the model performance. If the performance does not improve, reduce the learning rate η and add a momentum term to accelerate convergence.

[0126]

[0127] Among them, α is the momentum coefficient, Δθ t-1 is the parameter change of the previous iteration;

[0128] S155. Repeat steps S151-S154 until the demand forecasting model converges.

[0129] S2. Use low-power wide area networks and edge computing technology to monitor the status of materials in warehouses and during transportation in real time, process and analyze monitoring data in real time, and transmit the processing results to the central management system;

[0130] In this implementation, step S2 specifically includes:

[0131] S21. Material status monitoring in warehouses and during transportation, using sensors and RFID technology to obtain material quantity, location and environmental condition data to form a material status data set D s ;

[0132] S22. Perform data cleaning in the central management system to remove outliers and missing values ​​and fill in missing data;

[0133]

[0134] Among them, μ is the mean, σ is the standard deviation, and D si is the i-th data point in the material status data, T(D s ) is the original material status data set D s The transformed or processed data were used to remove outliers with more than 3 standard deviations.

[0135] S23, normalizing the data;

[0136]

[0137] Among them, D norm is the normalized data set, D si is a single data point after cleaning;

[0138] S24, extract features from the normalized data, including extracting key features from the original data using principal component analysis and time series decomposition methods, decomposing the data into trend, seasonality and residual components, and generating a feature data set F extract , combined with data enhancement method F augmented , generate more training data:

[0139]

[0140] Among them, ∈ is the data enhancement coefficient, N(0,σ2 ) is the normal distribution noise, f ij Represents the enhanced feature dataset F augmented The j-th eigenvalue of the i-th sample in ;

[0141] S25, combined with the enhanced feature dataset F augmented and material status dataset D s , forming a comprehensive data set D for material management final , and the comprehensive dataset D final Transfer to demand forecasting model.

[0142] S3. Automatically generate optimized material procurement plans based on demand data output by the reinforcement learning algorithm. The procurement plan takes into account the arrival time, inventory status and supplier capabilities of the materials, and optimizes the procurement strategy through dynamic adjustment.

[0143] In this implementation, step S3 specifically includes:

[0144] S31. Demand data output based on reinforcement learning algorithm Constructing a material procurement planning model, wherein the model includes procurement time T, inventory status I and supplier capacity C;

[0145] S32, according to the demand data and the current inventory status I, determine the procurement time T of the materials, where the procurement time is calculated by the optimization formula:

[0146]

[0147] Among them, λ1, λ2, λ3 are weight coefficients, T lead is the lead time, C inv (t) is the inventory cost, C stockout (t) is the stock-out cost;

[0148] S33. Generate a material procurement plan P based on the optimized procurement time T, combined with the supplier capacity C and inventory status I. The procurement plan takes into account the supplier's delivery time and supply capacity. When generating the material procurement plan, the supplier's historical delivery performance and current production capacity are taken into account, and the supplier scoring model is used for evaluation;

[0149] S34. Dynamically adjust the material procurement strategy by real-time monitoring of inventory status and demand changes, and update the procurement plan using the feedback control mechanism:

[0150]

[0151] Among them, α and β are adjustment coefficients, For new demand data, I newFor the new inventory status;

[0152] S35. By iterating the above steps in a loop, in each iteration, the adjusted procurement plan and actual execution results are recorded and analyzed to improve and optimize the reinforcement learning algorithm.

[0153] S4. In the process of material distribution, the improved ant colony algorithm is used to optimize the transportation route and time of materials from the warehouse to the construction site;

[0154] In this implementation, step S4 specifically includes:

[0155] S41, initializing the parameters of the ant colony algorithm, including the pheromone concentration matrix τ, the number of ants m, the number of iterations N, the pheromone importance coefficient α, the heuristic function importance coefficient β and the pheromone volatility coefficient ρ;

[0156] S42, setting a heuristic function η, which is defined as the reciprocal of the distance between the warehouse and the construction site;

[0157] S43. For each ant, choose the next path according to probability, and choose the probability p ij Defined as:

[0158]

[0159] Among them, allowed is the set of paths that ants can choose, τ ij is the pheromone concentration of path i to j;

[0160] S44. Calculate the length L of each ant's transport route and record the best route L. best :

[0161]

[0162] Among them, d i,i+1 represents the distance from the i-th node to the i+1-th node, d n,1 Indicates the distance from the last node back to the starting point, forming a closed loop path;

[0163] S45. Adaptive pheromone update mechanism:

[0164] τ ij (t+1)=(1-ρ)·τ ij (t)+Δτ ij (t);

[0165] Among them, the pheromone increment Δτ ij (t) is calculated as:

[0166]

[0167] Among them, Q is a constant, L k (t) is the path length of the kth ant, λ is the multi-objective optimization weight, D k (t) is the multi-objective function value of the kth ant, and the pheromone increment is defined as a function of the path length and multi-objective optimization values, so that the shorter paths and the paths with excellent comprehensive performance can obtain more information pheromones, guide the ants to choose high-quality paths, and adopt The contribution of all ants is accumulated in the form of , and the pheromone volatility coefficient (1-ρ) avoids unlimited accumulation of pheromones, so that the algorithm can be dynamically adjusted to adapt to environmental changes;

[0168] S46. After each round of iteration, a local search strategy is introduced to optimize the current solution, and a 2-opt or 3-opt algorithm is used to locally adjust the current optimal path;

[0169] S47, iterating the process of S41-S46 until the termination condition is met, the maximum number of iterations is reached or the optimal route length does not change within a certain number of times;

[0170] S48, output the best transportation route L best and its transportation time, so that materials can be delivered to the construction site in the best time;

[0171] S49. Input the optimized transportation route and time parameters into the material management system to realize dynamic monitoring and management of the material transportation process.

[0172] S5. During the use of materials, RFID technology and smart sensors are used to record the use of materials in real time, and preliminary data processing and analysis are performed through edge computing technology;

[0173] In this implementation, step S5 specifically includes:

[0174] S51. During the use of materials, RFID technology and intelligent sensors are used to record the use of materials in real time to form a material use data set U;

[0175] S52. Perform preliminary data processing on the material usage data through edge computing devices, including data cleaning, data normalization and feature extraction;

[0176] S53, using feature extraction technology, extract key features from normalized data to form a feature data set F U :

[0177]

[0178] Among them, w i is the feature extraction weight, φ(u i ) is the characteristic mapping function of material usage data;

[0179] S54. Use edge computing technology to conduct preliminary analysis of feature data, calculate statistical information on material usage and predictive model input:

[0180] Statistics calculation:

[0181]

[0182] Trend and seasonal components:

[0183]

[0184]

[0185] Where n is the number of data points in the feature dataset, t i is the timestamp of the i-th data point;

[0186] S55, transmit the analysis results and original data to the central management system, and combine with the data of other modules for comprehensive analysis and decision support;

[0187] S56. Use edge computing devices to monitor material usage in real time, update material status and send warning information.

[0188] Embodiment 1:

[0189] In power grid infrastructure projects, material management is a complex and critical link. Traditional material management methods have problems such as inaccurate predictions, untimely status monitoring, unreasonable procurement plans, insufficient optimization of transportation routes, and poor information sharing. In order to verify the effectiveness of the method of the present invention in practical applications, we selected a provincial power grid construction project as a test scenario.

[0190] In this project, the demand for materials is huge and varied, including cables, transformers, switch cabinets and other key equipment and materials. In order to ensure the timeliness and accuracy of material supply, we fully applied the method of the present invention in this project. First, by collecting the historical material usage data of the project in the past three years, the current grid construction progress data and the local weather and traffic and other external environmental data, a material demand forecasting model based on the reinforcement learning algorithm was established. The model can dynamically adjust parameters and update the forecast results in real time. In terms of material status monitoring, we used low-power wide area network and edge computing technology to install smart sensors and RFID tags on warehouses and transport vehicles. These devices can monitor the quantity, location and environmental conditions of materials in real time, and perform preliminary processing of data through edge computing devices. Based on the demand forecast data, we automatically generated an optimized material procurement plan. The plan not only takes into account the arrival time, inventory status and supplier capabilities of materials, but also optimizes the procurement strategy through dynamic adjustment. For example, in a transformer procurement, based on the comprehensive consideration of forecast demand and supplier delivery period, we changed the original monthly batch procurement to a one-time procurement, saving about 5% of procurement costs and reducing inventory backlogs.

[0191] In the process of material distribution, we use the improved ant colony algorithm to optimize the transportation route of materials from the warehouse to the construction site. The algorithm can dynamically adjust the transportation route according to the real-time traffic conditions and material demand. For example, during a material transportation process, the system monitored in real time that a traffic accident on the main road caused congestion, and automatically replanned the transportation route. In the end, the materials were delivered to the construction site 30 minutes earlier than originally planned, effectively avoiding construction stoppages caused by transportation delays.

[0192] During the use of materials, we use smart sensors and edge computing technology to record the use of materials in real time. For example, at a construction site, the system automatically records the length and location of each cable roll, uploads the data to the central management system, and generates a detailed use report. This real-time recording and analysis method not only improves the accuracy of the data, but also greatly reduces the workload and error rate of manual recording.

[0193] In order to achieve information sharing among departments and links, we have established an information management platform. The platform ensures the transparency and immutability of data. For example, each department can query the inventory status and usage of materials in real time through the platform, avoiding the problems of information asymmetry and data inconsistency. In an emergency dispatch, the project manager found out through the platform that there were still remaining cables in a warehouse, and quickly deployed them to the emergency construction site to ensure the progress of the project.

[0194] Table 1 Demand forecasting, monitoring and procurement optimization effects

[0195]

[0196] Table 2 Transportation optimization effect and data recording accuracy

[0197]

[0198]

[0199] Table 3 Information sharing effect

[0200]

[0201] As shown in Table 1 above, the present invention significantly improves the accuracy of prediction through material demand prediction based on reinforcement learning algorithm. The prediction error of key materials such as cables, transformers and switchgear has been reduced by 20%. For example, the traditional method predicts the value of cable demand is 140 kilometers, while the method of the present invention predicts the value of 150 kilometers, which is consistent with the actual demand. Through low-power wide area network and edge computing technology, real-time monitoring of material status is achieved, effectively avoiding material losses caused by adverse environmental conditions, and avoiding losses of about 200,000 yuan. In terms of procurement optimization, the method of the present invention takes into account the arrival time, inventory status and supplier capabilities of materials, dynamically adjusts the procurement strategy, and saves about 5% of procurement costs. For example, the traditional procurement cost of the transformer is 1.05 million yuan, and the cost is 1 million yuan after optimization by the method of the present invention, saving 50,000 yuan.

[0202] As shown in Table 2 above, during the transportation of materials, the improved ant colony algorithm is applied to optimize the transportation route and time, which significantly improves the transportation efficiency. For example, the traditional transportation time at a construction site is 2.5 hours, but after optimization by the method of the present invention, it only takes 2.0 hours, which is 0.5 hours earlier. Overall, the transportation time is 0.5 hours earlier than the traditional method. In addition, during the use of materials, real-time recording and analysis are performed through smart sensors and edge computing technology to improve the accuracy of data recording. The traditional method has a high recording error rate. For example, the error rate of cable usage records at a construction site is 5%, while the method of the present invention reduces the error rate to 1%, reducing the error rate by 4%.

[0203] As shown in Table 3 above, the present invention has established an information management platform to achieve information sharing and collaborative work among various departments and links. The platform can query the inventory status and usage of materials in real time, which significantly improves the response efficiency. For example, in emergency dispatch, the response time of the traditional method is 60 minutes, while the response time of the method of the present invention is 15 minutes, which improves the efficiency by 75%. Similarly, the response time of inventory query and demand adjustment is also greatly shortened, which improves the efficiency and accuracy of information sharing.

[0204] It can be seen from the above data that the method of the present invention significantly improves the performance of various key indicators in the material management of power grid infrastructure projects. It shows superiority in demand forecasting, real-time monitoring, procurement optimization, transportation route optimization, data recording accuracy and information sharing efficiency. Through these technical means, the present invention effectively solves many problems in traditional methods, improves the intelligence and efficiency of material management, and ensures the smooth progress and efficient implementation of power grid infrastructure projects.

[0205] In summary, the method of the present invention significantly improves the efficiency and accuracy of material management of power grid infrastructure projects through intelligent material demand forecasting, real-time material status monitoring, optimized procurement planning, dynamic transportation route optimization, accurate usage records and efficient information sharing, solves many problems in traditional methods, and has important practical application value.

[0206] 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 collaborative management of materials for power grid infrastructure projects, characterized in that: The following steps are involved: S1. Adopt a material demand forecasting algorithm based on reinforcement learning, analyze historical material usage data, power grid construction progress data and external environment data, dynamically adjust the forecasting model parameters, and predict the demand for various materials in power grid infrastructure projects; S2. Use low-power wide area networks and edge computing technology to monitor the status of materials in warehouses and during transportation in real time, process and analyze monitoring data in real time, and transmit the processing results to the central management system; S3. Automatically generate optimized material procurement plans based on demand data output by the reinforcement learning algorithm. The procurement plan takes into account the arrival time, inventory status and supplier capabilities of the materials, and optimizes the procurement strategy through dynamic adjustment. S4. In the process of material distribution, the improved ant colony algorithm is used to optimize the transportation route and time of materials from the warehouse to the construction site; S5. During the use of materials, RFID technology and smart sensors are used to record the use of materials in real time, and preliminary data processing and analysis are performed through edge computing technology; S6. Establish an information management platform to achieve information sharing and collaborative work among various departments and links.

2. A method for collaborative management of materials for power grid infrastructure projects according to claim 1, characterized in that: The step S1 specifically includes: S11. Collect historical material usage data h , Power grid construction progress data D p and external environment data D e ; S12. Preprocess the collected data, including data cleaning to remove outliers and missing values, fill in missing data, normalize the data, generate more training data through random sampling, smoothing and aggregation; and use principal component analysis and time series decomposition methods to extract key features from the original data, decompose the data into trend, seasonality and residual components, and form a feature data set F pre ; S13. Construct a material demand prediction model based on reinforcement learning, wherein the material demand prediction model includes a state space S, an action space A, a reward function R and a strategy π, wherein the reward function R is expressed as: Where M is the number of multi-objectives in the reward function, λ k is the weight of each target, Q k (s t ,a t ) is the evaluation function of the kth objective; S14, the preprocessed feature data set F pre Input into the material demand forecasting model and initialize the model parameters θ; S15, through the reinforcement learning algorithm, at each time step t, according to the current material demand state s t Select action a t , and according to the reward function R(s t ,a t ) and the improved gradient ascent algorithm to update the model parameters θ; S16. Iteratively update the model parameters θ until the model converges to obtain the final demand forecasting model: Among them, η is the learning rate, ∈ is the noise coefficient, N(0,σ 2 ) is the normal distribution noise; S17. Use the final demand forecasting model to forecast the demand for various materials in power grid infrastructure projects and output the forecast results 3. A method for collaborative management of materials for power grid infrastructure projects according to claim 2, characterized in that: The step S15 specifically includes: S151. In each iteration, calculate the current material demand forecast model parameter θ t The reward function gradient under Among them, λ k is the weight of the kth target, is about the current parameter θ t The evaluation function gradient of S152, according to the current state s t and action a t The feature map φ(s t ,a t ) performs local quadratic approximation to obtain an improved gradient ascent update formula; in, is the Hessian matrix of the reward function, Δθ is the gradient update step size; S153. When updating model parameters, add a normally distributed noise term to enhance exploration capability; i t+1 =θ t+1 +ε·N(0,σ 2 ); Among them, ∈ is the noise coefficient, N(0,σ 2 ) is the normal distribution noise; S154. After each iteration, evaluate the model performance. If the performance does not improve, reduce the learning rate η and add a momentum term to accelerate convergence. Among them, α is the momentum coefficient, Δθ t-1 is the parameter change of the previous iteration; S155. Repeat steps S151-S154 until the demand forecasting model converges.

4. A method for collaborative management of materials for power grid infrastructure projects according to claim 1, characterized in that: The step S2 specifically includes: S21. Material status monitoring in warehouses and during transportation, using sensors and RFID technology to obtain material quantity, location and environmental condition data to form a material status data set D s ; S22. Perform data cleaning in the central management system to remove outliers and missing values ​​and fill in missing data; Among them, μ is the mean, σ is the standard deviation, and D si is the i-th data point in the material status data, T(D s ) is the original material status dataset D s The transformed or processed data were used to remove outliers with more than 3 standard deviations. S23, normalizing the data; Among them, D norm is the normalized data set, D si is a single data point after cleaning; S24, extract features from the normalized data, including extracting key features from the original data using principal component analysis and time series decomposition methods, decomposing the data into trend, seasonality and residual components, and generating a feature data set F extract , combined with data enhancement method F augmented , generate more training data: Among them, ε is the data enhancement coefficient, N(0,σ 2 ) is the normal distribution noise, f ij Represents the enhanced feature dataset F augmented The j-th eigenvalue of the i-th sample in ; S25, combined with the enhanced feature dataset F augmented and material status dataset D s , forming a comprehensive data set D for material management final , and the comprehensive dataset D final Transfer to demand forecasting model.

5. A method for collaborative management of materials for power grid infrastructure projects according to claim 1, characterized in that: The step S3 specifically includes: S31. Demand data output based on reinforcement learning algorithm Constructing a material procurement planning model, wherein the model includes procurement time T, inventory status I and supplier capacity C; S32, according to the demand data and the current inventory status I, determine the procurement time T of the materials, where the procurement time is calculated by the optimization formula: Among them, λ1, λ2, λ3 are weight coefficients, T lead is the lead time, C inv (t) is the inventory cost, C stockout (t) is the stock-out cost; S33. Generate a material procurement plan P based on the optimized procurement time T, combined with the supplier capacity C and inventory status I. The procurement plan takes into account the supplier's delivery time and supply capacity. When generating the material procurement plan, the supplier's historical delivery performance and current production capacity are taken into account, and the supplier scoring model is used for evaluation; S34. Dynamically adjust the material procurement strategy by real-time monitoring of inventory status and demand changes, and update the procurement plan using the feedback control mechanism: Among them, α and β are adjustment coefficients, For new demand data, I new For the new inventory status; S35. By iterating the above steps in a loop, in each iteration, the adjusted procurement plan and actual execution results are recorded and analyzed to improve and optimize the reinforcement learning algorithm.

6. A method for collaborative management of materials for power grid infrastructure projects according to claim 1, characterized in that: The step S4 specifically includes: S41, initializing the parameters of the ant colony algorithm, including the pheromone concentration matrix τ, the number of ants m, the number of iterations N, the pheromone importance coefficient α, the heuristic function importance coefficient β and the pheromone volatility coefficient ρ; S42, setting a heuristic function η, which is defined as the reciprocal of the distance between the warehouse and the construction site; S43. For each ant, choose the next path according to probability, and choose the probability p ij Defined as: Among them, allowed is the set of paths that ants can choose, τ ij is the pheromone concentration of path i to j; S44. Calculate the length L of each ant's transport route and record the best route L. best : Among them, d i,i+1 represents the distance from the i-th node to the i+1-th node, d n,1 Indicates the distance from the last node back to the starting point, forming a closed loop path; S45. Adaptive pheromone updating mechanism: t ij (t+1)=(1-ρ)·τ ij (t)+Δτ ij (t); Among them, the pheromone increment Δτ ij (t) is calculated as: Among them, Q is a constant, L k (t) is the path length of the kth ant, λ is the multi-objective optimization weight, D k (t) is the multi-objective function value of the kth ant, and the pheromone increment is defined as a function of the path length and multi-objective optimization values, so that the shorter paths and the paths with excellent comprehensive performance can obtain more information pheromones, guide the ants to choose high-quality paths, and adopt The contribution of all ants is accumulated in the form of , and the pheromone volatility coefficient (1-ρ) avoids unlimited accumulation of pheromones, so that the algorithm can be dynamically adjusted to adapt to environmental changes; S46. After each round of iteration, a local search strategy is introduced to optimize the current solution, and a 2-opt or 3-opt algorithm is used to locally adjust the current optimal path; S47, iterating the process of S41-S46 until the termination condition is met, the maximum number of iterations is reached or the optimal route length does not change within a certain number of times; S48, output the best transportation route L best and its transportation time, so that materials can be delivered to the construction site in the best time; S49. Input the optimized transportation route and time parameters into the material management system to realize dynamic monitoring and management of the material transportation process.

7. A method for collaborative management of materials for power grid infrastructure projects according to claim 1, characterized in that: The step S5 specifically includes: S51. During the use of materials, RFID technology and intelligent sensors are used to record the use of materials in real time to form a material use data set U; S52. Perform preliminary data processing on the material usage data through edge computing devices, including data cleaning, data normalization and feature extraction; S53, using feature extraction technology, extract key features from normalized data to form a feature data set F U : Among them, w i is the feature extraction weight, φ(u i ) is the characteristic mapping function of material usage data; S54. Use edge computing technology to conduct preliminary analysis of feature data, calculate statistical information on material usage and predictive model input: Statistics calculation: Trend and seasonal components: Where n is the number of data points in the feature dataset, t i is the timestamp of the i-th data point; S55, transmit the analysis results and original data to the central management system, and combine with the data of other modules for comprehensive analysis and decision support; S56. Use edge computing devices to monitor material usage in real time, update material status and send warning information.

Images