A supply and demand management method and system for electric power materials warehousing

Abstract

The application relates to the technical field of power material storage management, and discloses a supply-demand management method and system for power material storage, which comprises the following steps: obtaining material consumption time sequence data and real-time operation data of various power equipment; obtaining a material demand benchmark prediction value according to the material consumption time sequence data and a demand benchmark prediction model, and correcting the material demand benchmark prediction value according to the real-time operation data to obtain a material demand prediction value; calculating a stockout risk index according to the current inventory and the material demand prediction value, and calculating a material priority according to the equipment types, fault frequencies and supply cycles of the various power equipment; and formulating and executing a material supply scheduling strategy according to the stockout risk index and the material priority. Through the whole-process management of data-driven prediction, dynamic resource regulation and closed-loop feedback optimization, the application realizes fine management of material supply and demand, improves the operation and maintenance efficiency of a power system, and improves the safety and stability of power system operation.

Description

Technical Field

[0001] This invention relates to the field of power material warehousing management technology, and in particular to a supply and demand control method and system for power material warehousing. Background Technology

[0002] In power equipment warehousing management, accurate forecasting and dynamic adjustment of supply plans are crucial. This not only affects the continuity of power grid operation but also the efficient utilization of warehousing resources. Especially in environments with frequent power grid load fluctuations and variable equipment maintenance needs, equipment management must be closely integrated with equipment operating status, maintenance plans, and load changes to ensure the availability and efficiency of materials at critical moments.

[0003] The operating status of power grid equipment is highly uncertain. For example, equipment failure may cause maintenance windows to be brought forward or delayed, which makes it difficult to accurately quantify the time-series constraints of material demand. Most existing methods of power material storage management rely on single time series models based on historical consumption data, resulting in insufficient prediction accuracy. This can easily lead to inventory backlog or sudden shortages. In addition, the inventory monitoring indicators of existing methods only focus on static inventory levels and cannot respond in a timely manner to the material demand caused by the dynamic changes in the operating status of power grid equipment.

[0004] Therefore, there is an urgent need for a management method for power material storage that can improve the responsiveness and risk resistance of the power grid material supply system. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a supply and demand management method and system for power material warehousing, which can solve the problems of supply and demand mismatch, low warehousing efficiency and delayed response in existing methods, thereby achieving the technical effect of improving the refined management of power material warehousing supply and demand.

[0006] In a first aspect, the present invention provides a supply and demand management method for power material storage, the method comprising: Acquire time-series data on material consumption and real-time operation data of various power equipment in the corresponding power material storage area; Based on the material consumption time series data and the preset demand benchmark prediction model, the material demand benchmark prediction value is obtained, and the material demand benchmark prediction value is corrected based on the real-time operation data to obtain the material demand prediction value. The demand benchmark prediction model is constructed based on the time series prediction model. Based on the current inventory of power materials and the forecast of demand for the materials, a shortage risk index is calculated, and the material priority is calculated based on the equipment type, failure frequency and supply cycle of various power equipment. Based on the shortage risk index and the material priority, a material supply scheduling strategy is formulated and implemented. After the material supply scheduling strategy is implemented, material supply feedback data is obtained. Based on the material supply feedback data, the inventory structure of power material storage is optimized.

[0007] Furthermore, the step of correcting the baseline forecast value of material demand based on the real-time operational data to obtain the forecast value of material demand includes: Calculate the data deviation rate based on the real-time and historical operating data; Based on the data deviation rate, the material types of various power equipment are obtained, including emergency materials and regular materials; Based on the data deviation rate, the baseline forecast value of the emergency supplies demand is corrected to obtain the forecast value of the emergency supplies demand. The baseline forecast value of the material demand for the conventional materials is used as the forecast value of the material demand for the conventional materials.

[0008] Furthermore, the step of correcting the baseline forecast value of material demand based on the real-time operational data to obtain the forecast value of material demand includes: Acquire equipment status data, fault repair data, and external impact data of various power equipment, and extract features from the real-time operation data, equipment status data, fault repair data, and external impact data to obtain real-time status features and dynamic features; The real-time state features and the dynamic features are input into a preset demand fluctuation prediction model to obtain the predicted value of material demand fluctuation. The demand fluctuation prediction model is constructed based on a long short-term memory neural network model with an attention mechanism. Feature extraction is performed on the equipment status data and the external influence data to obtain multi-factor cross features. The baseline forecast value of material demand, the forecast value of material demand fluctuation, and the multi-factor cross features are then concatenated to obtain comprehensive features. The comprehensive features are input into a preset demand correction prediction model to obtain the predicted value of material demand. The demand correction prediction model is constructed based on a support vector regression model.

[0009] Further, the step of extracting features from the real-time operating data, the equipment status data, the fault repair data, and the external impact data to obtain real-time status features and dynamic features includes: Based on the real-time operating data and the device status data, the real-time status characteristics are obtained; Based on the external impact data, external weather characteristics are obtained; Based on the fault repair data and the external impact data, specific event days are identified, including maintenance days, peak load days, and fault days. If the specific event day exists, then based on the specific event day, an event time sequence feature is generated, and the external weather feature and the event time sequence feature are used as dynamic features; If the specific event day does not exist, the external weather characteristics will be treated as dynamic characteristics.

[0010] Furthermore, the step of generating event time sequence features based on the specific event date includes: Within the time window from the maintenance date to the current period, the number of maintenance days, maintenance type, and historical maintenance consumption rate for the same period are encoded to generate maintenance event time sequence characteristics. Within the time window of the load peak day, the peak day marker, peak intensity normalized value, and peak duration are encoded to generate peak event time sequence characteristics; Within the time window from the fault date to the current period, the number of fault days, fault type, and historical fault consumption rate for the same period are encoded to generate fault event time sequence characteristics.

[0011] Furthermore, the attention mechanism of the demand fluctuation prediction model calculates the attention weights using the following steps: Based on the dynamic characteristics, determine whether a specific scenario exists; If it does not exist, the first attention weight of the hidden state at each time step is calculated based on the real-time state features and the dynamic features using a preset ordinary scene weight calculation formula. If it exists, then according to the event sequence characteristics and the preset specific scene weight calculation formula, the second attention weight of the hidden state corresponding to the time step of the specific scene is calculated, and according to the real-time state characteristics and the external weather characteristics, the third attention weight of the hidden state corresponding to the time step of the non-specific scene is calculated using the ordinary scene weight calculation formula.

[0012] Furthermore, the step of calculating the shortage risk index based on the current inventory of power material storage and the predicted demand for the material includes: Based on the current inventory of power materials and the forecast of demand for the materials, calculate the demand fluctuation range and the peak demand time. Calculate the time urgency based on the peak demand time and the preset supply cycle; Calculate the stockout risk index based on the magnitude of demand fluctuations and the time urgency.

[0013] Furthermore, the step of calculating material priority based on the equipment type, failure frequency, and supply cycle of various types of power equipment includes: Based on the equipment type of various power equipment, determine the equipment importance score for each power equipment; The fault factor is obtained based on the fault frequency and the preset frequency threshold. The supply factor is obtained based on the supply cycle and the preset cycle threshold; The importance score of the equipment is corrected based on the failure factor and the supply factor to obtain the material priority of each power equipment.

[0014] Furthermore, the step of obtaining material supply feedback data and optimizing the inventory structure of power material warehousing based on the material supply feedback data includes: Obtain material supply feedback data, and based on the material supply feedback data, obtain management efficiency indicators, including inventory accuracy rate and delivery timeliness rate; The energy efficiency management indicators and indicator thresholds are compared, and based on the comparison results, it is determined whether the inventory structure of power material storage needs to be optimized. If so, the mean time between failures (MTBF) is obtained based on the fault repair data of the power equipment, and the operating stability coefficient is obtained based on the MTBF and total operating time. Based on the aforementioned operational stability coefficient, the safety stock of power equipment in the power material storage is adjusted.

[0015] Secondly, the present invention provides a supply and demand management system for power material storage, the system comprising: The data acquisition module is used to acquire time-series data on the consumption of various types of power equipment and real-time operation data in the corresponding area of ​​the power material storage area; The demand forecasting module is used to obtain the material demand baseline forecast value based on the material consumption time series data and the preset demand baseline forecasting model, and to correct the material demand baseline forecast value based on the real-time operation data to obtain the material demand forecast value. The demand baseline forecasting model is constructed based on the time series forecasting model. The stockout analysis module is used to calculate the stockout risk index based on the current inventory of power materials and the predicted demand for the materials, and to calculate the material priority based on the equipment type, failure frequency and supply cycle of various power equipment. The material supply module is used to formulate and execute material supply scheduling strategies based on the shortage risk index and the material priority. After the material supply scheduling strategy is executed, it obtains material supply feedback data and optimizes the inventory structure of power material storage based on the material supply feedback data.

[0016] This invention provides a supply and demand management method and system for power material warehousing. By employing a demand forecasting method combining medium- and long-term linear benchmark forecasting with short-term nonlinear correction, the invention improves the accuracy of material demand forecasting. Through comparing inventory levels with predicted material demand and considering equipment supply cycles, it enables accurate quantitative assessment of stockout risks. Furthermore, by using a priority calculation method based on equipment importance scores, it enhances the accuracy of prioritization. In addition, the invention continuously optimizes the warehousing inventory structure based on material supply feedback, improving the rationality of the power material warehousing inventory structure design. Through data-driven forecasting, dynamic resource regulation, and closed-loop feedback optimization, this invention achieves refined management of material supply and demand, improves the operation and maintenance efficiency of the power system, and further enhances the safety and stability of power system operation. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the supply and demand management method for power material warehousing in an embodiment of the present invention; Figure 2 This is a schematic diagram of the supply and demand management system for power material storage in an embodiment of the present invention; Figure label: 10. Data Acquisition Module; 20. Demand Forecasting Module; 30. Stockout Analysis Module; 40. Material Supply Module. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 The first embodiment of the present invention proposes a supply and demand management method for power material storage, comprising steps S10 to S40: Step S10: Obtain the material consumption time-series data and real-time operation data of various power equipment in the corresponding area of ​​the power material storage; Step S20: Based on the material consumption time series data and the preset demand benchmark prediction model, obtain the material demand benchmark prediction value, and correct the material demand benchmark prediction value based on the real-time operation data to obtain the material demand prediction value. The demand benchmark prediction model is constructed based on the time series prediction model. Step S30: Calculate the shortage risk index based on the current inventory of power materials and the predicted demand for the materials, and calculate the material priority based on the equipment type, failure frequency and supply cycle of various power equipment. Step S40: Based on the shortage risk index and the material priority, formulate and execute a material supply scheduling strategy. After the material supply scheduling strategy is executed, obtain material supply feedback data and optimize the inventory structure of power material storage based on the material supply feedback data.

[0020] This invention provides a supply and demand management method for power equipment warehousing. The power equipment warehousing stores power equipment for a specific area to meet the equipment supply needs for maintenance and repair in that area. First, it acquires time-series data on material consumption and real-time operational data for various types of power equipment in the current area. The time-series data includes historical consumption records, equipment lifespan, maintenance cycles, failure cycles, and seasonal factors. Seasonal factors refer to the varying demands on power equipment in different seasons; for example, the frequency of lightning protection equipment maintenance is higher in spring than in winter, while the material consumption of transformer cooling systems increases significantly in summer. Therefore, the seasonal factor for a particular type of power equipment can be represented by a seasonal coefficient. Equipment lifespan refers to the ratio of the equipment's operational time to its designed service life. Real-time operational data includes operational data for various types of power equipment (such as transformers, circuit breakers, and disconnectors), such as transformer voltage and load values. Then, based on the acquired data, a baseline forecast of power equipment demand is performed.

[0021] In a preferred embodiment, the present invention employs a demand benchmark forecasting model based on a time series forecasting model for forecasting material demand. Preferably, the time series forecasting model can be an Autoregressive Integrated Moving Average (ARIMA) model or a Prophet model to forecast demand based on the input material consumption time series data. Taking the ARIMA model as an example, using material consumption time series data as input data and inventory turnover rate as a constraint, the ARIMA model forecasts the material demand value for a future period, such as the next month. The inventory turnover rate constraint can limit the fluctuation range of the forecast value. Furthermore, during the adjustment of forecast weights, if the maintenance cycle overlaps with a seasonal peak, the forecast weight for that period is increased by a preset proportion to ensure sufficient material supply during critical periods. The model parameters are updated monthly through a rolling forecasting method to ensure the accuracy of the forecast results. It should be noted that the demand benchmark forecasting model in this embodiment can be constructed using either an ARIMA model or a Prophet model. The input data can be flexibly adjusted according to actual conditions. The specific forecasting steps can refer to the conventional forecasting models and are not specifically limited here.

[0022] As can be seen from the above embodiments, the core logic of the demand baseline forecasting model is to use the differential autoregressive moving average method to fit the trend term (such as increased demand due to equipment lifespan growth), seasonal term (such as a surge in demand for lightning protection equipment in spring), and cyclical term (such as equipment failure cycle and maintenance cycle) of historical demand, thereby predicting the material demand in future periods. The predicted baseline forecast values ​​of material demand for various types of power equipment focus on medium- and long-term trends. However, in reality, power grid material demand is affected by both long-term trend factors (such as seasons and equipment aging) and short-term random factors (such as sudden failures and real-time load fluctuations). The ARIMA model can only solve the uncertainty of long-term trends and cannot solve the uncertainty of short-term conditions. Therefore, this embodiment, based on the material demand baseline forecast values ​​obtained through the demand baseline forecasting model, uses real-time operating data of power equipment to perform equipment status analysis, thereby correcting the material demand baseline forecast values. The specific steps include: Calculate the data deviation rate based on the real-time and historical operating data; Based on the data deviation rate, the material types of various power equipment are obtained, including emergency materials and regular materials; Based on the data deviation rate, the baseline forecast value of the emergency supplies demand is corrected to obtain the forecast value of the emergency supplies demand. The baseline forecast value of the material demand for the conventional materials is used as the forecast value of the material demand for the conventional materials.

[0023] In this embodiment, abnormal states of equipment are identified through real-time operational data, thereby classifying the material types of power equipment. Specifically, the data deviation rate is calculated using real-time and historical operational data. Since the number of various types of power equipment in the area is not unique, the real-time operational data in this embodiment uses parameters with strong correlation. Taking transformers as an example, when calculating the data deviation rate, the real-time operational data uses the total real-time load value of all transformers in the area (summed), and the average value of the real-time operational data of the most recent week is used as the real-time operational data for calculation to avoid the impact of a single data anomaly on the accuracy of the calculation results. The data deviation rate of the transformer is obtained by calculating the difference between the real-time total load value and the average historical total load value of the area, dividing the difference by the historical average value, and then multiplying by 100%. Based on the comparison between the data deviation rate and the deviation rate threshold, the material type of the power equipment is determined. For example, if the deviation rate threshold is set to 15%, when the data deviation rate is greater than 15%, the material type of the transformer is set as emergency material; otherwise, it is set as regular material. For power equipment without strong correlation parameters, the data deviation rate is calculated based on the average real-time operation value and the average historical operation value of the power equipment in the area, thereby determining the material type.

[0024] This embodiment compares real-time and historical operating data to reflect the overall current operating status of various power equipment. Based on the current operating status, the short-term status of the equipment is determined, and the predicted value of material demand is corrected accordingly. Specifically, a large data deviation rate indicates a high risk of abnormal operating status for the equipment, potentially leading to sudden material demand. Since the demand baseline forecasting model predicts medium- to long-term trends, the predicted value of material demand can meet medium- to long-term trend material demand, but it cannot guarantee that it can meet short-term demand fluctuations. Therefore, this embodiment corrects the predicted value of material demand for emergency materials based on the data deviation rate. Preferably, a correction coefficient is set based on the difference between the data deviation rate and the deviation rate threshold. The predicted value of material demand is then adjusted based on this correction coefficient. For example, if the data deviation rate of a power equipment is 30%, the correction coefficient is 1.5. Multiplying 1.5 by the predicted value of material demand yields the corrected predicted value. For conventional materials, i.e., equipment with stable operating status, no correction is required; the predicted value of material demand for conventional materials can be directly used as the predicted value of material demand.

[0025] Furthermore, to improve the accuracy of material demand forecasts and avoid excessive stockpiling, when correcting the baseline material demand forecast based on the correction coefficient, only the time-series forecast data for a preset period in the baseline material demand forecast can be corrected. For example, the baseline data for the previous week or the first half of the month can be corrected. It should be noted that although the demand baseline forecast model outputs a monthly baseline material demand forecast, the actual output is time-series forecast data on a daily basis. By statistically analyzing the time-series forecast data, the total baseline material demand forecast can be obtained. Therefore, the baseline material demand forecast for the previous week can also be statistically analyzed, corrected, and then added to the uncorrected data for the following period to obtain the corrected material demand forecast.

[0026] This embodiment uses a time series forecasting model to predict the baseline forecast value of material demand, and corrects the baseline forecast value using static rules based on data deviation rate, which improves forecasting efficiency while ensuring the accuracy of forecasting results.

[0027] In another preferred embodiment, the present invention also provides a method for forecasting material demand using a hybrid model based on time series forecasting, in order to improve the accuracy of demand forecasting. The specific forecasting steps include: Acquire equipment status data, fault repair data, and external impact data of various power equipment, and extract features from the real-time operation data, equipment status data, fault repair data, and external impact data to obtain real-time status features and dynamic features; The real-time state features and the dynamic features are input into a preset demand fluctuation prediction model to obtain the predicted value of material demand fluctuation. The demand fluctuation prediction model is constructed based on a long short-term memory neural network model with an attention mechanism. Feature extraction is performed on the equipment status data and the external influence data to obtain multi-factor cross features. The baseline forecast value of material demand, the forecast value of material demand fluctuation, and the multi-factor cross features are then concatenated to obtain comprehensive features. The comprehensive features are input into a preset demand correction prediction model to obtain the predicted value of material demand. The demand correction prediction model is constructed based on a support vector regression model.

[0028] This embodiment designs a multi-layered cascaded hybrid model based on the time series forecasting model to correct the baseline forecast of material demand. This hybrid model includes a baseline demand forecasting model, a demand fluctuation forecasting model, and a revised demand forecasting model. The baseline and demand fluctuation forecasting models are connected in parallel. The outputs of the two models are concatenated and then input into the revised demand forecasting model to obtain the predicted material demand. Specifically, the baseline demand forecasting model is built based on the time series forecasting model, the demand fluctuation forecasting model is built based on a long short-term memory neural network model with an attention mechanism, and the revised demand forecasting model is built based on a support vector regression model. The data processing steps of the model are described in detail below, taking into account its architecture.

[0029] In this embodiment, the demand baseline forecasting model still uses the ARIMA model as an example. The material consumption time-series data of the previous period is used as input, and the ARIMA model is used to predict the baseline forecast value of material demand for future periods. The ARIMA model is a linear forecasting model based on time series analysis. It predicts by identifying trends and periodic patterns through historical data. However, the material demand of the power system is not completely linear. The ARIMA model can only predict linear trends and cannot accurately predict nonlinear fluctuations caused by the state of power equipment and the external environment. Therefore, this embodiment designs a bidirectional long short-term memory (BiLSTM) neural network model based on an attention mechanism to construct a demand fluctuation forecasting model. This model captures nonlinear relationships such as equipment failures and sudden load increases.

[0030] This embodiment extracts features from the real-time operating data, equipment status data, fault repair data, and external impact data of power equipment over the past week to obtain real-time status features and dynamic features. These features are then used as input data for a demand fluctuation prediction model to obtain predicted material demand fluctuations. Specifically, taking transformer equipment as an example, real-time operating data includes voltage and load values, while equipment status data is represented by a health score. The daily health score can be determined based on the number of historical faults and abnormal operating data in the previous month. For example, points can be deducted based on the number of historical faults and abnormal operating data to obtain the health score. The real-time operating data and health score are then used as real-time status features. It should be noted that in this embodiment, the data is preprocessed during feature extraction, including time alignment, data filtering, missing value imputation, and standardization. Specific preprocessing steps can be referenced from conventional preprocessing methods and will not be elaborated further here.

[0031] Dynamic features are extracted from fault maintenance data and external impact data. The fault maintenance data includes the previous week's fault maintenance records, including whether a fault or maintenance occurred. The external impact data includes power grid load data and external weather data. The power grid load data of the previous week and the external weather data (weather forecast) of the following week are extracted from the external impact data. The extracted power grid load data and external weather data are used as dynamic features. The real-time status features and dynamic features are encoded and concatenated and then input into the BiLSTM model to predict the material demand fluctuation forecast for the following week.

[0032] After capturing long-term linear trends using the ARIMA model and short-term nonlinear fluctuations using the BiLSTM model, this embodiment encodes and concatenates the outputs of the two models, then inputs them into the demand correction prediction model for iterative optimization and error correction to obtain the final predicted value. Preferably, the demand correction prediction model is constructed using a Support Vector Regression (SVR) model.

[0033] Furthermore, to improve the accuracy of the prediction results, in addition to encoding and concatenating the outputs of the two models, this embodiment also extracts cross-features from equipment status data and external influence data to obtain multi-factor cross-features, including cross-features such as grid load × external weather, health score × grid load, and health score × external weather. Then, the baseline forecast of material demand output by the ARIMA model is concatenated with the material demand fluctuation forecast output by the BiLSTM model, along with the multi-factor cross-features, to obtain a comprehensive feature. This comprehensive feature is then input into the SVR model. The SVR model considers the effects of linear trends, nonlinear fluctuations, and multi-dimensional interactions, and iteratively optimizes the model using a radial basis function kernel to obtain the final material demand forecast. In essence, the SVR model corrects the baseline forecast of material demand output by the ARIMA model based on short-term fluctuations, but the correction only applies to the time period corresponding to the short-term fluctuations. Through periodic iterative optimization, the baseline forecast is continuously updated and corrected, ensuring the timeliness and accuracy of the prediction results.

[0034] This embodiment designs a cascaded architecture hybrid model, which uses an ARIMA model to capture linear trends, a BiLSTM model to handle nonlinear fluctuations, and an SVR model to fuse multi-source features and iteratively correct errors. This solves the problem that a single model cannot take into account both trends and fluctuations, and improves the accuracy of material demand forecasting.

[0035] In a preferred embodiment, to improve the accuracy of the model in predicting short-term nonlinear fluctuations, this invention provides a dynamic adjustment mechanism for attention weights based on different scenarios. When the BiLSTM model using this mechanism predicts demand fluctuations, data processing is divided into two stages: the first stage is the feature extraction stage before inputting into the model, and the second stage is the weight calculation stage within the model. The specific steps of the first stage include: Based on the real-time operating data and the device status data, the real-time status characteristics are obtained; Based on the external impact data, external weather characteristics are obtained; Based on the fault repair data and the external impact data, specific event days are identified, including maintenance days, peak load days, and fault days. If the specific event day exists, then based on the specific event day, an event time sequence feature is generated, and the external weather feature and the event time sequence feature are used as dynamic features; If the specific event day does not exist, the external weather characteristics will be treated as dynamic characteristics.

[0036] In this embodiment, the extraction steps for real-time status features are the same as those in the previous embodiment, and will not be repeated here. For dynamic features, they are also extracted from fault maintenance data and external impact data. Specifically, the predicted weather data for the following week (i.e., the week to be predicted) is extracted from the external impact data as the external weather feature. In addition, specific event days are identified from the fault maintenance data and external impact data. These specific event days include maintenance days, peak load days, and fault days. The reason for identifying these specific events is that they may affect subsequent material demand. If these specific events are identified within the most recent week, event time-series features are generated based on the existing specific events, and these features, together with the external weather features, serve as the dynamic features.

[0037] In this embodiment, event sequence features are used as scene recognition features in the BiLSTM model. The BiLSTM model's attention mechanism determines whether a specific scene exists based on the input features and dynamically adjusts the attention weights adaptively based on the determination result. Specifically, the attention weight expression used by the BiLSTM model's attention mechanism is as follows: In the formula, Score(t) represents the original importance score at time step t, exp(*) represents the exponential function, n represents the total number of time steps, and Attention(t) represents the normalized attention weight at time step t. This formula normalizes the Score by taking the exponent of the Score for all time steps, transforming the Score into a probability distribution between 0 and 1, ensuring that the sum of the weights for each time step is 1. The time step can be an hour or a day.

[0038] For the score function at each time step, the commonly used dot product scoring method can be used for calculation. This involves taking the dot product of the hidden state at each time step of the input sequence with the learnable weight matrix to obtain the score for each time step. In this embodiment, when there is no specific event day, the weights for the real-time state features and dynamic features of the input model can be directly calculated using dot product attention. When there is a specific event day, the weights are adjusted according to the temporal features of the input event.

[0039] To improve the computational efficiency of the model, this embodiment uses binary event time sequence features. Specifically, for a specific event day that is identified, the specific event day and the two days immediately following it are encoded as 1, while other dates are encoded as 0, thereby generating binary event time sequence features. In order to represent the event type of the event time sequence feature, an event identifier code is also added to the event time sequence feature to indicate whether the event is a maintenance day, a peak load day, or a fault day.

[0040] Event sequence features, as explicit identifiers of specific events, directly impact the weighting of the hidden states in BiLSTM. When a value is 1, the model dynamically increases the weight of the hidden state in the corresponding time window. Preferably, the score of the hidden state at a time step marked as 1 is increased by 0.1, thereby increasing the attention weight of the hidden state at that time step. For example, when a device experiences a sudden malfunction (binary marker = 1), the hidden state will be given a higher weight in that time window, forcing the model to prioritize learning the mapping relationship between the malfunction and increased demand. In practical applications, there is generally only one specific event day. When multiple specific event days exist in a given period, the score of the hidden state at a time step marked as 1 will be superimposed based on the existing specific event days, but the upper limit of the superposition is 0.2. That is, when there are three specific event days at a certain time step, its score will increase by a maximum of 0.2 to avoid the excessive weight of a certain time step affecting the accuracy of the prediction results.

[0041] This embodiment calculates attention weights representing conventional importance using real-time state features and external weather features, and activates abnormal time windows through event time sequence features, forming a weight superposition with the attention weights of conventional importance, thereby enhancing the model's attention and realizing dynamic adjustment of attention weights based on different scenarios, thus improving the accuracy of the model's prediction results.

[0042] In a preferred embodiment, the present invention also provides another method for extracting event temporal features, the specific steps of which include: Within the time window from the maintenance date to the current period, the number of maintenance days, maintenance type, and historical maintenance consumption rate for the same period are encoded to generate maintenance event time sequence characteristics. Within the time window of the load peak day, the peak day marker, peak intensity normalized value, and peak duration are encoded to generate peak event time sequence characteristics; Within the time window from the fault date to the current period, the number of fault days, fault type, and historical fault consumption rate for the same period are encoded to generate fault event time sequence characteristics.

[0043] In this embodiment, critical time windows are defined based on these specific event days. For maintenance days, the critical time window is from the maintenance day to the current period. Since the input data is from the previous week, the critical time window is at most 7 days and at least 1 day. For peak load days, the critical time window is the peak load day itself. For fault days, the critical time window is from the fault day to the current period. Of course, this time window setting is only a preferred option. Depending on the actual situation, only the specific event day or the specific event day and the following day or two can be selected as the time window. The time window can be understood as a time step.

[0044] Within the time window from the maintenance date to the current period, the number of maintenance days, maintenance type identifier, and historical consumption rate are encoded to generate the time series characteristics of the maintenance event. The number of maintenance days reflects the urgency of the time; the closer to the maintenance date, the more urgent the demand. The maintenance type identifier includes planned maintenance and temporary maintenance to distinguish between rigid demand. The historical consumption rate refers to the average material consumption of the equipment during the past three maintenance periods, serving as a reference benchmark, thus obtaining the time series of the maintenance event. For peak load days, the time window of that day is encoded, including a peak day marker, a peak intensity normalized value, and a peak duration. The peak day marker directly identifies key time points, the peak intensity normalized value quantifies the degree of load anomaly, and the peak duration reflects the cumulative effect of high load on the equipment. The time step of the peak day carries these characteristics, enabling the model to capture the correlation between high load, high loss, and high demand. Within the time window from the fault date to the current period, similar to maintenance days, the number of fault days, fault type, and historical fault consumption rate for the same period are encoded. The number of fault days reflects the urgency of the situation; the closer to the fault date, the more urgent the time requirement. The fault type reflects the severity of the fault. The historical fault consumption rate refers to the average material consumption of the equipment during the past three similar faults, thus obtaining the fault event time series. It should be noted that in this embodiment, the data is standardized during feature extraction to overcome the problem of inconsistent units; this will not be repeated here or in subsequent processes.

[0045] After obtaining the event time series features, they are input together with real-time state features and external weather features into the demand fluctuation prediction model. In this embodiment, the specific steps of the attention mechanism in the demand fluctuation prediction model to calculate the attention weight according to the dynamic adjustment mechanism for different scenarios include: Based on the dynamic characteristics, determine whether a specific scenario exists; If it does not exist, the first attention weight of the hidden state at each time step is calculated based on the real-time state features and the dynamic features using a preset ordinary scene weight calculation formula. If it exists, then according to the event sequence characteristics and the preset specific scene weight calculation formula, the second attention weight of the hidden state corresponding to the time step of the specific scene is calculated, and according to the real-time state characteristics and the external weather characteristics, the third attention weight of the hidden state corresponding to the time step of the non-specific scene is calculated using the ordinary scene weight calculation formula.

[0046] In this embodiment, the existence of a specific scenario is determined based on the input dynamic features. If the dynamic features only include external weather features, it indicates that no specific scenario exists. In this case, the weight calculation can be performed by calculating the attention weights for the real-time state features and external weather features using the dot product attention calculation method in the previous embodiment. If the input dynamic features also include event sequence features, it indicates that a specific scenario exists. Then, according to the preset weight calculation formulas for different scenarios, the score of the hidden state at the time step corresponding to the specific scenario is calculated. For example, dot product attention, scaled dot product attention, or additive attention can be used to calculate the score for the event sequence features. Preferably, the weight formula for the specific scenario can be the additive attention calculation formula. For time steps without event sequence features, i.e., time steps corresponding to ordinary scenarios, the score is calculated based on the real-time state features and external weather features, thereby realizing the dynamic adjustment of the weights of the hidden state of the time step based on different scenarios. When there are multiple event sequence features at a certain time step, the average score of each specific event is taken as the final score, and finally, the attention weight is obtained through normalization.

[0047] In a preferred embodiment, another attention weight calculation method can also be used. In this embodiment, for a time step with a specific scenario, the score of the hidden state is calculated based on the event sequence characteristics, and another score is calculated based on the real-time state characteristics and external weather characteristics of the time step. Finally, the two scores are weighted and summed to obtain the final score. It can be understood that this embodiment corrects the conventional attention score by using the event sequence characteristics of a specific scenario, thereby realizing the dynamic adjustment of the attention weight of the time step of the specific scenario, and thus improving the model's ability to capture nonlinear fluctuations.

[0048] In this embodiment, after obtaining the predicted value of material demand fluctuation through the demand fluctuation prediction model, the processing steps of the aforementioned embodiment can be referred to to perform cross-feature extraction on the equipment status data and external influence data, and then concatenate them with the material demand baseline prediction value and the material demand fluctuation prediction value. After iterative optimization through the SVR model, the final material demand prediction value is obtained.

[0049] This embodiment achieves dynamic adjustment of attention weights based on different scenarios by fusing multi-source heterogeneous data and constructing dynamic features, thereby improving the model's ability to capture nonlinear fluctuations. Through a cascaded architecture of medium- and long-term linear benchmark prediction, short-term nonlinear correction, and multi-feature fusion iterative optimization, it not only ensures the independence of feature capture but also achieves synergistic optimization of results, effectively improving the accuracy of material demand forecasting and thus providing accurate data support for subsequent warehouse supply and demand management.

[0050] After obtaining accurate material demand forecasts, and combining them with the current inventory of power material warehouses, it can be determined whether there is a risk of stockouts. If the current inventory is less than the material demand forecast and cannot be replenished within a preset timeframe, such as 2 days, then there is a risk of stockouts.

[0051] To accurately quantify stockout risk, in a preferred embodiment, the present invention provides a method for calculating a stockout risk index, the specific steps of which include: Based on the current inventory of power materials and the forecast of demand for the materials, calculate the demand fluctuation range and the peak demand time. Calculate the time urgency based on the peak demand time and the preset supply cycle; Calculate the stockout risk index based on the magnitude of demand fluctuations and the time urgency.

[0052] In this embodiment, the demand fluctuation range and peak demand time are first calculated based on the current inventory level and the forecasted demand. Specifically, the difference between the forecasted demand and the current inventory level is compared with the current inventory level to obtain the demand fluctuation range. If the demand fluctuation range is less than or equal to zero, it means that the current inventory level can meet the forecasted demand, and there is no risk of stockout. Warehouse management can be carried out according to the conventional material supply scheduling strategy. If the demand fluctuation range is greater than zero, it means that the current inventory level cannot meet the forecasted demand. In this case, the peak demand time is calculated based on the current inventory level and the forecasted demand. The peak demand time refers to the time when an inventory gap occurs.

[0053] In the above embodiments, although the predicted material demand values ​​obtained by the model are based on the total monthly data, the model actually outputs daily time-series forecast data. Monthly material demand forecast values ​​can be statistically obtained from this time-series forecast data. Therefore, when calculating the demand fluctuation range, the total material demand forecast value can be used, while when calculating the peak demand time point, daily time-series forecast data is used. During the calculation, the daily material demand difference is obtained by subtracting the forecast data for each day from the current inventory level. When the material demand difference for a certain day is negative, it indicates that there is an inventory gap, and that day is the peak demand time point.

[0054] After obtaining the peak demand time, the time urgency is calculated based on the preset supply cycle. The supply cycle refers to the time from equipment procurement to warehousing. This cycle can be set based on the shortest delivery time or the average delivery time of each supply channel. When calculating time urgency, the peak demand time is first subtracted from the supply cycle. If the difference is greater than or equal to zero, it means the supply cycle can fill the inventory gap, and therefore the time urgency is zero. If the difference is less than zero, it means the supply cycle cannot fill the inventory gap. In this case, the absolute value of the difference is divided by the supply cycle to obtain the time urgency. For example, if the peak demand time is day 5 and the supply cycle is 7 days, then the time urgency is 2 / 7 ≈ 0.29. The higher the time urgency, the more urgent the situation. Finally, the demand fluctuation range is multiplied by the time urgency to obtain the stockout risk index. This embodiment, by comparing inventory levels with predicted values ​​and combining the equipment's supply cycle, can accurately quantify stockout risk, thus providing accurate data support for subsequent warehouse management strategies.

[0055] In this embodiment, the forecast of material demand and the quantification of stockout risk are based on a specific type of power equipment. In reality, there are many types of power equipment stored in warehouses. Therefore, for the supply scheduling of equipment with stockout risk, in addition to considering the stockout risk, it is also necessary to consider the importance of different power equipment in actual applications, so as to determine the material priority of the equipment. Among them, the material priority of main equipment is the highest, and the material priority of auxiliary equipment is lower. For example, the material priority of transformers is set to 1, and the material priority of measuring instruments is set to 0.6, etc. The specific priority settings can be flexibly set according to the actual situation, and no further restrictions are imposed here.

[0056] In a preferred embodiment, to improve the accuracy of setting material priorities for various power devices, the present invention also provides a method for calculating material priorities based on the importance of the equipment, combined with the equipment's supply cycle and operating status. The specific steps include: Based on the equipment type of various power equipment, determine the equipment importance score for each power equipment; The fault factor is obtained based on the fault frequency and the preset frequency threshold. The supply factor is obtained based on the supply cycle and the preset cycle threshold; The importance score of the equipment is corrected based on the failure factor and the supply factor to obtain the material priority of each power equipment.

[0057] In this embodiment, firstly, the equipment importance score is determined according to the equipment type. For example, the equipment importance score of the main equipment is set to 1, and the equipment importance score of the auxiliary equipment is set to 0.6. Then, the ratio of the equipment failure frequency to a preset frequency threshold is used as the failure factor. The failure frequency can be the monthly average of the actual failure count in the previous month or the historical actual failure count in the previous year. The failure factor characterizes the operating status of the equipment. The supply cycle is compared with a preset safe supply cycle, and the supply factor is set according to the comparison result. The supply factor is used to characterize the risk characteristics of the equipment supply chain. The supply cycle can be determined with reference to the supply cycle in the previous embodiment. In a preferred embodiment, multiple cycle thresholds are preset, and then the supply cycle is compared with each cycle threshold in turn. The corresponding supply factor is determined according to the cycle threshold to which the supply cycle belongs. For example, the preset cycle thresholds include less than 10 days, 11 to 20 days, and more than 20 days. Each cycle threshold corresponds to a different supply factor, such as 1, 3, and 5. The supply factor of the equipment can be determined according to the cycle threshold to which the supply cycle belongs.

[0058] In another preferred embodiment, to enhance sensitivity to long supply cycles, the present invention also provides a method for calculating the supply factor, the specific steps of which include: The supply cycle is compared with a preset cycle threshold. If the supply cycle is less than the cycle threshold, the ratio of the supply cycle to the cycle threshold is used as the supply factor. Otherwise, the difference between the supply cycle and the cycle threshold is calculated and the difference is exponentially enhanced to obtain the supply factor.

[0059] In this embodiment, the relationship between the supply cycle and a preset cycle threshold is first determined. A safe cycle threshold, such as 10 days, is then set. When the supply cycle is less than the cycle threshold, it indicates that the supply chain has a relatively small impact on the priority of material demand. In this case, the ratio of the supply cycle to the cycle threshold is used as the supply factor. When the supply cycle is greater than or equal to the cycle threshold, it indicates that there is risk in the supply chain. To enhance the sensitivity of long cycles, this embodiment uses an exponential function for data augmentation. Specifically, the difference between the supply cycle and the cycle threshold is calculated, with the natural number e as the base and this difference as the exponent, thus obtaining the supply factor. Furthermore, to control the exponential growth rate, a risk amplification coefficient can be set. Assuming α represents the risk amplification coefficient, the exponent is α multiplied by the difference. Preferably, α is 0.1. The longer the supply cycle exceeds the cycle threshold, the more significantly the supply factor will increase, thereby achieving non-linear amplification of supply chain risk.

[0060] After calculating the fault factor and supply factor, these are used as correction factors and multiplied by the equipment importance score to obtain the material priority of the power equipment. It should be noted that when a certain equipment has no faults in a historical period, the fault factor may be zero. In order to avoid the situation where the fault factor is zero and the material priority is zero, the fault factor is added to 1 and then multiplied by the equipment importance score during the correction, so as to ensure the accuracy of the material priority calculation.

[0061] This material priority calculation method in this embodiment ensures that the risk characteristics of equipment operation and supply chain are accurately reflected in the priority, thereby improving the efficiency and effectiveness of power grid material storage resource scheduling.

[0062] After accurately quantifying the shortage risk index and material priority of various power equipment, a material supply scheduling strategy can be formulated based on the shortage risk index and material priority. Preferably, the equipment is first sorted according to its material priority, that is, high-priority equipment is prioritized. For equipment with the same priority, it is sorted according to its shortage risk index. The material supply scheduling strategy is formulated according to the sorting. The principle of the supply scheduling strategy is based on the supply scheduling capacity, prioritizing the scheduling needs of high-priority and high-shortage-risk equipment. Preferably, emergency procurement is carried out for high-priority and high-shortage-risk equipment, such as activating the emergency response mechanism for expedited procurement. For high-priority and medium-shortage-risk equipment, regular procurement or cross-regional scheduling can be carried out. For low-priority equipment, the alternative material mechanism can be activated to determine whether there are alternative materials, or to negotiate delays with the demanding departments. Of course, if the supply chain can meet the scheduling needs of all equipment, the material supply scheduling can be carried out according to the normal scheduling process. The specific scheduling strategy can be set according to the actual supply scheduling capacity of power material warehousing, and is not specifically limited here.

[0063] After the material supply scheduling strategy is completed, feedback data on the strategy execution can be obtained, and the inventory structure of power material storage can be optimized and adjusted based on the feedback data. For example, based on the accuracy of inventory and the timeliness of delivery, the equipment inventory and equipment storage layout in the next stage can be optimized. If the inventory accuracy is low, the inventory plan for the next stage can be adjusted. If the delivery delay is mainly concentrated on high-frequency equipment, while adjusting the inventory, it is also necessary to check whether the storage area layout is reasonable and optimize the area layout.

[0064] In a preferred embodiment, the present invention also provides a method for optimizing inventory structure based on material supply feedback data, the specific steps of which include: Obtain material supply feedback data, and based on the material supply feedback data, obtain management efficiency indicators, including inventory accuracy rate and delivery timeliness rate; The energy efficiency management indicators and indicator thresholds are compared, and based on the comparison results, it is determined whether the inventory structure of power material storage needs to be optimized. If so, the average fault-free operating time and fault probability are obtained based on the fault inspection data of the power equipment, and the operating stability coefficient is obtained based on the average fault-free operating time and the fault probability. Based on the aforementioned operational stability coefficient, the safety stock of power equipment in the power material storage is adjusted.

[0065] In this embodiment, the material supply feedback data includes inventory accuracy and delivery timeliness. Inventory accuracy indicates whether the inventory level can meet the demand. It is the ratio of the actual inventory level in the previous period, such as the previous month, to the actual demand. If the inventory level is greater than the actual demand, the inventory accuracy is 1. Delivery timeliness indicates whether the delivery time of the equipment meets the demand. It is the ratio of the average delivery time required for the equipment to the actual average delivery time. If the average delivery time required is greater than the actual average delivery time, the delivery timeliness is 1. That is, both inventory accuracy and delivery timeliness are values ​​between 0 and 1. Then, the inventory accuracy and delivery timeliness are weighted and summed to obtain the management efficiency index of the previous stage. Preferably, the weight of inventory accuracy is 0.4 and the weight of delivery timeliness is 0.6.

[0066] Based on the comparison between the management energy efficiency index and the preset index threshold, it is determined whether the inventory structure of the power material storage needs to be optimized. Preferably, the index threshold is set to 0.85. If the management energy efficiency index is less than this threshold, it indicates that there is a problem with the inventory structure of the storage and optimization is needed. In order to accurately and efficiently calculate the inventory adjustment amount, this embodiment adjusts the equipment inventory by analyzing the operating status of the equipment, that is, adjusting the safety stock of the equipment in the plan. Specifically, based on recent timeframes, such as fault maintenance data from the previous year or six months, the single fault-free operating time and the number of faults for the equipment are determined. The average fault-free operating time is obtained by summing the individual fault-free operating times and dividing by the number of faults. The total annual operating time is calculated in days or hours. For example, in hours, the total annual operating time is 8760 hours. The total operating time includes the fault-free operating time and the fault period. The average fault-free operating time is obtained by summing the normal operating time before each fault, dividing by the total number of faults, and obtaining the normal operating time of the equipment, which is the average fault-free operating time. This embodiment objectively reflects the reliability level of the equipment by calculating the average value and can avoid the interference of extreme values ​​caused by accidental factors in the single fault-free time, thus eliminating the randomness of single data and reflecting long-term stability.

[0067] Then, the ratio of mean time between failures (MTBF) to total uptime is used as the operating stability coefficient, which is the numerator of the operating stability coefficient. The longer the MTBF, the stronger the equipment's continuous and stable operation capability. In other words, the higher the operating stability coefficient, the lower the probability of failure of the equipment, the longer the normal operation period, and the more stable the equipment operation.

[0068] Finally, based on the operating stability coefficient of the equipment, the adjustment coefficient of the inventory is determined. Preferably, the operating stability coefficient is compared with the coefficient threshold. If it is higher than the coefficient threshold, it means that the operating stability of the equipment is low. In this case, the adjustment coefficient is set higher, such as 1.5. Conversely, if it is lower than the threshold, it means that the equipment has a certain degree of operating stability. In this case, the adjustment coefficient can be appropriately reduced, such as 1.2. Then, the adjustment coefficient is multiplied by the planned inventory to obtain the adjusted safety stock. The next stage of inventory structure adjustment is then carried out based on the adjusted safety stock.

[0069] This embodiment verifies the dynamic adaptability of inventory from the perspective of operational efficiency by managing energy efficiency indicators, avoiding resource waste, and corrects the safety baseline of inventory from the perspective of equipment reliability by using the operating stability coefficient, ensuring that the stored materials can cope with the risk of failure. This embodiment achieves a three-dimensional balance of equipment stability, inventory safety and supply efficiency, enabling the management and control of power material storage to meet the needs of emergency response and have efficient resource turnover capabilities, thereby improving the rationality of the inventory structure.

[0070] This embodiment provides a supply and demand management method for power material warehousing. This method employs a medium-to-long-term linear benchmark forecasting and a short-term nonlinear correction method for material demand forecasting. This ensures the independence of feature capture while achieving synergistic optimization of results, effectively improving the accuracy of material demand forecasting. By comparing inventory levels with predicted material demand and combining this with equipment supply cycles, accurate quantitative assessment of stockout risks is achieved. Furthermore, a priority calculation method based on equipment importance scores ensures that the risk characteristics of equipment operation and the supply chain are accurately reflected in the priority calculation, providing accurate data support for subsequent resource scheduling. In addition, this embodiment continuously optimizes the warehousing inventory structure based on material supply feedback, further improving the overall responsiveness of power material supply.

[0071] Please see Figure 2 Based on the same inventive concept, the second embodiment of this invention proposes a supply and demand management system for power material storage, comprising: Data acquisition module 10 is used to acquire time-series data on material consumption and real-time operation data of various power equipment in the corresponding area of ​​power material storage; The demand forecasting module 20 is used to obtain a baseline forecast value for material demand based on the material consumption time series data and a preset demand baseline forecasting model, and to correct the baseline forecast value for material demand based on the real-time operating data to obtain a forecast value for material demand. The baseline forecasting model for material demand is constructed based on a time series forecasting model. The stockout analysis module 30 is used to calculate the stockout risk index based on the current inventory of power materials and the predicted demand for the materials, and to calculate the material priority based on the equipment type, failure frequency and supply cycle of various power equipment. The material supply module 40 is used to formulate and execute a material supply scheduling strategy based on the shortage risk index and the material priority. After the material supply scheduling strategy is executed, it obtains material supply feedback data and optimizes the inventory structure of power material storage based on the material supply feedback data.

[0072] The technical features and effects of the supply and demand management system for power material warehousing proposed in this invention are the same as those of the method proposed in this invention, and will not be repeated here. Each module in the above-mentioned supply and demand management system for power material warehousing can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0073] In summary, the present invention proposes a supply and demand management method and system for power material storage. The method acquires time-series data and real-time operational data of various power equipment in the corresponding area of ​​the power material storage; obtains a baseline demand forecast value based on the time-series data and a preset demand baseline prediction model; and corrects the baseline demand forecast value based on the real-time operational data to obtain a final demand forecast value. The demand baseline prediction model is constructed based on a time-series prediction model. A shortage risk index is calculated based on the current inventory level of the power material storage and the predicted demand value. Material priorities are calculated based on the equipment type, failure frequency, and supply cycle of various power equipment. A material supply scheduling strategy is formulated and executed based on the shortage risk index and the material priorities. After the material supply scheduling strategy is executed, material supply feedback data is acquired, and the inventory structure of the power material storage is optimized based on the material supply feedback data. This invention achieves refined management of material supply and demand through data-driven prediction, dynamic resource regulation, and closed-loop feedback optimization throughout the entire process, improving the operation and maintenance efficiency of the power system and further enhancing the safety and stability of power system operation.

[0074] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0075] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A supply and demand management method for power material warehousing, characterized in that, include: Acquire time-series data on material consumption and real-time operation data of various power equipment in the corresponding power material storage area; Based on the material consumption time series data and the preset demand benchmark prediction model, the material demand benchmark prediction value is obtained. The demand benchmark prediction model is constructed based on the time series prediction model. Acquire equipment status data, fault repair data, and external impact data of various power equipment; extract real-time status features from the real-time operation data and the equipment status data; and extract dynamic features from the fault repair data and the external impact data. The real-time state features and the dynamic features are input into a preset demand fluctuation prediction model to obtain the predicted value of material demand fluctuation. The demand fluctuation prediction model is constructed based on a long short-term memory neural network model with an attention mechanism. Feature extraction is performed on the equipment status data and the external influence data to obtain multi-factor cross features. The baseline forecast value of material demand, the forecast value of material demand fluctuation, and the multi-factor cross features are then concatenated to obtain comprehensive features. The comprehensive features are input into a preset demand correction prediction model to obtain the predicted value of material demand. The demand correction prediction model is constructed based on a support vector regression model. Based on the current inventory of power materials and the forecast of demand for the materials, a shortage risk index is calculated, and the material priority is calculated based on the equipment type, failure frequency and supply cycle of various power equipment. Based on the shortage risk index and the material priority, a material supply scheduling strategy is formulated and implemented. After the material supply scheduling strategy is implemented, material supply feedback data is obtained. Based on the material supply feedback data, the inventory structure of power material storage is optimized.

2. The supply and demand management method for power material warehousing according to claim 1, characterized in that, The steps of extracting real-time status features from the real-time operation data and the equipment status data, and extracting dynamic features from the fault repair data and the external impact data, include: Based on the real-time operating data and the device status data, the real-time status characteristics are obtained; Based on the external impact data, external weather characteristics are obtained; Based on the fault repair data and the external impact data, specific event days are identified, including maintenance days, peak load days, and fault days. If the specific event day exists, then based on the specific event day, an event time sequence feature is generated, and the external weather feature and the event time sequence feature are used as dynamic features; If the specific event day does not exist, the external weather characteristics will be treated as dynamic characteristics.

3. The supply and demand management method for power material warehousing according to claim 2, characterized in that, The step of generating event time sequence features based on the specific event date includes: Within the time window from the maintenance date to the current period, the number of maintenance days, maintenance type, and historical maintenance consumption rate for the same period are encoded to generate maintenance event time sequence characteristics. Within the time window of the load peak day, the peak day marker, peak intensity normalized value, and peak duration are encoded to generate peak event time sequence characteristics; Within the time window from the fault date to the current period, the number of fault days, fault type, and historical fault consumption rate for the same period are encoded to generate fault event time sequence characteristics.

4. The supply and demand management method for power material warehousing according to claim 2, characterized in that, The attention mechanism of the demand fluctuation prediction model calculates the attention weights using the following steps: Based on the dynamic characteristics, determine whether a specific scenario exists; If it does not exist, the first attention weight of the hidden state at each time step is calculated based on the real-time state features and the dynamic features using a preset ordinary scene weight calculation formula. If it exists, then according to the event sequence characteristics and the preset specific scene weight calculation formula, the second attention weight of the hidden state corresponding to the time step of the specific scene is calculated, and according to the real-time state characteristics and the external weather characteristics, the third attention weight of the hidden state corresponding to the time step of the non-specific scene is calculated using the ordinary scene weight calculation formula.

5. The supply and demand management method for power material warehousing according to claim 1, characterized in that, The step of calculating the shortage risk index based on the current inventory of power materials and the predicted demand for the materials includes: Based on the current inventory of power materials and the forecast of demand for the materials, calculate the demand fluctuation range and the peak demand time. Calculate the time urgency based on the peak demand time and the preset supply cycle; Calculate the stockout risk index based on the magnitude of demand fluctuations and the time urgency.

6. The supply and demand management method for power material warehousing according to claim 1, characterized in that, The steps for calculating material priority based on the equipment type, failure frequency, and supply cycle of various power equipment include: Based on the equipment type of various power equipment, determine the equipment importance score for each power equipment; The fault factor is obtained based on the fault frequency and the preset frequency threshold. The supply factor is obtained based on the supply cycle and the preset cycle threshold; The importance score of the equipment is corrected based on the failure factor and the supply factor to obtain the material priority of each power equipment.

7. The supply and demand management method for power material warehousing according to claim 1, characterized in that, The steps of obtaining material supply feedback data and optimizing the inventory structure of power material warehousing based on the material supply feedback data include: Obtain material supply feedback data, and based on the material supply feedback data, obtain management efficiency indicators, including inventory accuracy rate and delivery timeliness rate; The energy efficiency management indicators and indicator thresholds are compared, and based on the comparison results, it is determined whether the inventory structure of power material storage needs to be optimized. If so, the mean time between failures (MTBF) is obtained based on the fault repair data of the power equipment, and the operating stability coefficient is obtained based on the MTBF and total operating time. Based on the aforementioned operational stability coefficient, the safety stock of power equipment in the power material storage is adjusted.

8. A supply and demand management system for power material warehousing, characterized in that, The system is applied to the method as described in any one of claims 1 to 7, comprising: The data acquisition module is used to acquire time-series data on the consumption of various types of power equipment and real-time operation data in the corresponding area of ​​the power material storage area; The demand forecasting module is used to obtain the material demand baseline forecast value based on the material consumption time series data and the preset demand baseline forecasting model, and to correct the material demand baseline forecast value based on the real-time operation data to obtain the material demand forecast value. The demand baseline forecasting model is constructed based on the time series forecasting model. The stockout analysis module is used to calculate the stockout risk index based on the current inventory of power materials and the predicted demand for the materials, and to calculate the material priority based on the equipment type, failure frequency and supply cycle of various power equipment. The material supply module is used to formulate and execute material supply scheduling strategies based on the shortage risk index and the material priority. After the material supply scheduling strategy is executed, it obtains material supply feedback data and optimizes the inventory structure of power material storage based on the material supply feedback data.

Images