A method and system for predicting net procurement volume based on wind turbine component data

Abstract

The application discloses a kind of based on wind turbine spare part data's net purchase quantity prediction method and system, time series prediction model is established by historical spare part consumption monthly or quarterly consumption data, can quickly and accurately predict the future demand data of different spare parts, calculate safety stock, combine the quantity of material in stock and in-transit quantity, output the prediction result of net purchase quantity, can quickly realize the recommendation of spare parts and spare parts net purchase quantity, and then provide reference guidance for condition maintenance decision;The data required by the application is easy to obtain, and the model training and prediction operation is simple, under the premise of ensuring that the spare parts meet the demand, the quarterly demand of spare parts and spare parts can be accurately predicted, and the compatibility is high, which can be used for quarterly demand prediction and net purchase quantity calculation of different spare parts.

Description

Technical Field

[0001] This invention relates to the technical field of wind turbine spare parts procurement quantity forecasting, and in particular to a method and system for forecasting net procurement quantity based on wind turbine spare parts data. Background Technology

[0002] Forecasting net spare parts procurement plays a crucial role in inventory management, impacting not only the maintenance efficiency of wind turbines but also the operating costs of wind farms. Forecasting methods for spare parts can be categorized into qualitative and quantitative forecasting. Quantitative forecasting methods include causal analysis forecasting, time series analysis-based forecasting, and single-model or combined model forecasting based on machine learning. While qualitative forecasting provides some reference, it is heavily influenced by human subjectivity, often leading to significant biases in the recommended results. Causal analysis forecasting uses linear regression, and the choice of regression variables directly affects forecast accuracy. Furthermore, this method is unsuitable for handling nonlinear data structures. Machine learning includes grey prediction methods, neural network prediction methods, and support vector machine methods, but achieving high prediction accuracy requires large datasets. Since spare parts demand typically has a small sample size, the prediction results are prone to bias.

[0003] In order to objectively and reasonably recommend the inventory structure of spare parts, reduce the high inventory costs or material stagnation costs caused by high purchase volume, and reduce the power generation loss costs caused by unit operation with faults or long-term unit shutdown due to low purchase volume (i.e., low availability), it is urgent to accurately predict the future demand for spare parts, ensure a high availability rate, minimize inventory, reduce material stagnation, and achieve intelligent management of spare parts. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a net purchase quantity prediction method and system based on wind turbine spare parts data. By establishing a time series prediction model through monthly or quarterly consumption data of historical spare parts, it can quickly and accurately predict the future demand data of different spare parts, calculate safety stock, and output the net purchase quantity prediction by combining the quantity of materials in stock and in transit. It can quickly realize the recommendation of net purchase quantity of spare parts, and thus provide reference guidance for condition-based maintenance decisions.

[0005] The objective of this invention is achieved through the following technical solution: a method for predicting net purchase volume based on wind turbine spare parts data, comprising the following steps:

[0006] S1. Establish a spare parts database and read historical consumption data of spare parts;

[0007] S2. Clean and remove outliers from the historical consumption data of spare parts read from the spare parts database.

[0008] S3. Perform feature analysis on the pre-processed historical consumption data of spare parts, and select a time series prediction model based on the features;

[0009] S4. Input the preprocessed historical consumption data of spare parts into the time series prediction model selected in step S3 for training, and obtain the trained spare parts demand prediction model.

[0010] S5. Use the trained spare parts demand prediction model to predict the spare parts demand data at future times, and quantify the accuracy of the spare parts demand prediction model by calculating error statistics. Select the spare parts demand prediction model to be applied based on the error statistics.

[0011] S6. Calculate safety stock data, and based on the safety stock data, the predicted spare parts demand data, and the spare parts database data, calculate the net spare parts purchase quantity data.

[0012] Furthermore, step S1 includes:

[0013] Establish a spare parts database, collect historical spare parts consumption data, and perform quarterly statistics on historical spare parts consumption data based on the historical spare parts consumption data; wherein, the spare parts database also includes spare parts in stock data and spare parts in transit data.

[0014] Furthermore, step S3 includes:

[0015] Seasonal, trend and autocorrelation analyses were performed on the pre-processed historical consumption data of spare parts, and the appropriate time series forecasting model was selected based on the data characteristics obtained from the analysis.

[0016] Among them, the time series graph is used to determine whether the historical consumption data of spare parts is seasonal. If the time series graph shows periodic changes, that is, the data value changes greatly within a preset period and changes little within other periods, then the current historical consumption data of spare parts is considered to be seasonal; otherwise, it is not seasonal.

[0017] By drawing a time series graph, it can be determined whether the historical consumption data of spare parts has a trend. If the time series graph shows a trend, that is, the data value continues to increase or decrease within a preset period, then the historical consumption data of spare parts is considered to have a trend; otherwise, it does not have a trend.

[0018] The DW test is used to evaluate the autocorrelation of historical spare parts consumption data, with the following criteria:

[0019] a) When the test value DW∈(1.5,2.5), the data residuals are independent of the independent variable;

[0020] b) When the test value DW < 1, the residuals of adjacent data points are positively correlated;

[0021] c) When the test value DW > 3, the residuals of adjacent data points are negatively correlated;

[0022] The time series forecasting models include simple moving average model, weighted moving average model, first exponential smoothing model, second exponential smoothing model, Holt-Winter model, and autoregressive difference moving average model.

[0023] Furthermore, step S3 includes:

[0024] If the historical consumption data of spare parts has the following characteristics: the data volume is less than the preset data volume or the volatility of the historical consumption data of spare parts is less than the preset volatility, then a simple moving average model is adopted, as shown in the following formula (1):

[0025]

[0026] Among them, S t +S t-1 +…+S t-n-1 It represents the observations from the past n time points, where n is the window size of the moving average, i.e., the number of past data points;

[0027] If the historical consumption data of spare parts has multiple variable factors and no trend or seasonality, then the weighted moving average model is adopted, as shown in the following formula (2):

[0028] S t+1 =ω0S t +ω1S t-1 +…+ω n-1 S t-n-1 (2);

[0029] Where, ω i For the observed value S t-i The corresponding weights are i, which is a constant, and n, which is the window size of the moving average.

[0030] Furthermore, step S3 includes:

[0031] If the historical consumption data of spare parts does not exhibit trends or seasonality, then a single exponential smoothing model is adopted, as shown in the following formula (3):

[0032] S t+1 =αX t+1 +(1-α)S t (3);

[0033] Among them, S t+1 Let X represent the predicted point at time t+1. t+1S represents the actual observed value at time point t+1. t This represents the predicted value at time point t, where α is the smoothing coefficient.

[0034] If the historical consumption data of spare parts has a trend but no seasonality, then a quadratic exponential smoothing model is adopted, as shown in the following formula (4):

[0035]

[0036] Among them, S t+1 Let X represent the predicted point at time t+1. t+1 S represents the actual observed value at time point t+1. t Let b represent the predicted value at time point t. t+1 S represents the predicted value at time point t+1. t+1 The predicted value S at time point t t The difference is where α represents the smoothing coefficient and β represents the trend smoothing coefficient.

[0037] Furthermore, step S3 includes:

[0038] If the historical consumption data of spare parts exhibits trends and seasonality, then the Holt-Winter model is adopted, as shown in the following formula (5):

[0039]

[0040] Among them, S t+1 Let X represent the predicted point at time t+1. t+1 S represents the actual observed value at time point t+1. t Let L represent the predicted value at time t, where L is the season length, S is the stable component, b is the linear component, I is the seasonal component, α is the smoothing parameter representing the level, β is the smoothing parameter representing the trend, and γ is the smoothing parameter representing the seasonality.

[0041] Furthermore, step S3 includes:

[0042] If the historical consumption data of spare parts exhibits autocorrelation and seasonality, then the autoregressive moving average model is adopted, as shown in the following formula (6):

[0043]

[0044] Among them, S t+1 S represents the predicted point at time t+1. t Let p represent the predicted value at time point t, where p is the autoregressive order and q is the moving average order. Both β and ε are non-zero undetermined coefficients, ε is an independent error term, and i and j are constants.

[0045] Furthermore, in step S5, the error statistics include:

[0046] The statistical indicators mean absolute percentage error (MAPE), mean absolute value error (MAE), and symmetric mean absolute percentage error (SMPAE) are shown in the following formula (7):

[0047]

[0048] Among them, S i The actual value at time i. Let be the predicted value at time i.

[0049] Furthermore, step S6 includes:

[0050] Safety stock is calculated based on the formula of the periodic ordering model, assuming periodic ordering and replenishment. The safety stock of each spare part is calculated according to the procurement lead time, quarterly average consumption, quarterly consumption standard deviation and service level. The safety stock is shown in the following formula (8):

[0051] SS=Zσ T +L (8);

[0052] Where z is the standard deviation multiple for a specific service probability, σ T+L It is the standard deviation of the inventory count cycle and the lead time demand, where T represents the number of days between two consecutive inventory counts and L represents the lead time in days;

[0053] The purchase quantity is calculated by subtracting the sum of the safety stock and the predicted quarterly demand from the spare parts inventory and in transit quantities read from the spare parts database, ultimately yielding the recommended net purchase quantity of spare parts.

[0054] A net purchase quantity forecasting system based on wind turbine spare parts data, used to implement the aforementioned net purchase quantity forecasting method based on wind turbine spare parts data, includes:

[0055] The spare parts database module is used to store historical consumption data, in-transit data, and in-stock data of spare parts.

[0056] The preprocessing module is used to read historical consumption data from the spare parts database module and perform cleaning and outlier removal operations.

[0057] The feature analysis and model selection module is used to perform seasonality, trend and autocorrelation feature analysis on the preprocessed historical consumption data, and finally select the appropriate time series prediction model based on the features.

[0058] The model building module is used to input historical consumption data into the selected time series prediction model for training.

[0059] The model validation module is used to calculate error statistics to quantify the accuracy of the trained model.

[0060] Safety stock calculation module, used to calculate spare parts safety stock;

[0061] The forecast procurement quantity calculation module calculates the net procurement quantity of spare parts based on safety stock data, forecasted spare parts demand data, and spare parts database data.

[0062] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0063] 1. This invention can accurately predict the quarterly demand for spare parts while ensuring that the demand for spare parts is met.

[0064] 2. This invention has high compatibility and can be used for quarterly demand forecasting and net purchase calculation of different spare parts;

[0065] 3. The data required by this invention is easy to obtain, and the model training and prediction operations are simple. It can quickly recommend the net purchase quantity of spare parts, thereby providing reference guidance for condition-based maintenance decisions. Attached Figure Description

[0066] Figure 1 This is a flowchart of the present invention.

[0067] Figure 2 This is a seasonal and trend breakdown chart of historical spare parts consumption data over a quarterly period.

[0068] Figure 3 This is a comparison chart of the predicted values ​​and actual values ​​using the Holt-Went model.

[0069] Figure 4 This is a comparison chart of the predicted values ​​and actual values ​​using the autoregressive moving average model. Detailed Implementation

[0070] The present invention will be further described below with reference to specific embodiments.

[0071] Example 1

[0072] See Figure 1 As shown in the figure, the net purchase quantity forecasting method based on wind turbine spare parts data provided in this embodiment forecasts the demand for high-strength stud spare parts for wind turbines in the first quarter of 2024 and calculates the safety stock. Combining the inventory and in-transit quantities of spare parts, it outputs the recommended net purchase quantity, including the following steps:

[0073] S1. Establish a spare parts database and read historical consumption data for high-strength studs; among them, the historical consumption data of the read spare parts should be statistically analyzed on a quarterly basis.

[0074] S2. Clean and remove outliers from the historical consumption data of spare parts read from the spare parts database; then, statistically analyze the data by quarter to obtain the historical dataset, as shown in Table 1 below:

[0075] time quantity 2020-03-31 9 2020-06-30 14 2020-09-30 9 2020-12-31 9 2021-03-31 4 2021-06-30 9 2021-09-30 6 2021-12-31 17 2022-03-31 5 2022-06-30 9 2022-09-30 3 2022-12-31 10 2023-03-31 4 2023-06-30 8 2023-09-30 3

[0076] Table 1. Dataset of historical consumption of high-strength studs after preprocessing.

[0077] S3. Perform feature analysis on the preprocessed historical consumption data of spare parts, and select a time series prediction model based on the features. The time series prediction models include simple moving average model, weighted moving average model, first exponential smoothing model, second exponential smoothing model, Holt-Winter model, and autoregressive difference moving average model. The applicability of each model is shown in Table 2 below:

[0078] Model Applicable conditions Simple moving average model Limited historical data or low volatility Weighted moving average model Historical data is subject to various variables and lacks trends and seasonality. First exponential smoothing model Historical data lacks trends and seasonality. Quadratic exponential smoothing model Historical data shows trends but not seasonality. Holtwint model Historical data exhibits trends and seasonality. Autoregressive moving average difference model Historical data exhibits autocorrelation and seasonality.

[0079] Table 2 Applicability of Each Model

[0080] See Figure 2 As shown, seasonality and trend decomposition are performed on the preprocessed historical dataset to observe the seasonal and trend characteristics of the data. Figure 2 It can be seen that the quarterly data on the historical consumption of this spare part has seasonality and trends.

[0081] An autocorrelation test was performed on the quarterly data, and the calculated DW test value was 0.5426, indicating that the series has a positive correlation.

[0082] Therefore, based on the characteristic analysis of historical consumption data, the Holt-Winter model and the autoregressive moving average model can be selected.

[0083] S4. Input the preprocessed historical consumption data of spare parts into the Holt-Went model and the autoregressive difference moving average model selected in step S3 for training, and obtain the trained spare parts demand prediction model.

[0084] The Holt-Went model is adopted, as shown in the following formula (5):

[0085]

[0086] Among them, S t+1 Let X represent the predicted point at time t+1. t+1S represents the actual observed value at time point t+1. t Let L represent the predicted value at time t, where L is the season length, S is the stable component, b is the linear component, I is the seasonal component, α is the smoothing parameter representing the level, β is the smoothing parameter representing the trend, and γ is the smoothing parameter representing the seasonality.

[0087] The autoregressive moving average model is adopted, as shown in the following formula (6):

[0088]

[0089] Among them, S t+1 S represents the predicted point at time t+1. t Let p represent the predicted value at time point t, where p is the autoregressive order and q is the moving average order. Both β and ε are non-zero undetermined coefficients, ε is an independent error term, and i and j are constants.

[0090] S5. Use the trained spare parts demand prediction model to predict the spare parts demand data at future moments, and quantify the accuracy of the spare parts demand prediction model by calculating error statistics.

[0091] See Figures 3 to 4 As shown, the Holt-Winter model and the autoregressive moving average model were used to predict the spare parts consumption in the fourth quarter of 2023, respectively. The chart compares the predicted and actual values ​​of historical data.

[0092] The comparison between the Holt-Went model and the autoregressive moving average model for the predicted and actual values ​​of the fourth quarter of 2023 is shown in Table 3 below:

[0093] time actual value Holt-Winter Autoregressive moving average difference method 2023Q4 8 9 9.6505

[0094] Table 3. Comparison of Actual and Forecast Values ​​of the Holt-Winter Model and the Autoregressive Moving Average Model in the Fourth Quarter of 2023

[0095] The mean absolute error (MAE), mean absolute percentage error (MAPE), and symmetric mean absolute percentage error (SMAPE) between the predicted and actual values ​​for the fourth quarter of 2023 are calculated as shown in the following formula (7):

[0096]

[0097] Among them, S i The actual value at time i. Let be the predicted value at time i.

[0098] The calculation results are shown in Table 4 below:

[0099] method MAE MAPE SMAPE Holt-Winter 1 0.125 0.08 Autoregressive moving average difference method 1.6505 0.2063 0.1287

[0100] Table 4. Error statistics of the Holt-Winter model and the autoregressive moving average model in the fourth quarter of 2023.

[0101] Therefore, based on the comparison of the error statistics of the two methods between the actual and predicted values ​​in the fourth quarter of 2023, the Holtwent model was selected as the model for predicting spare parts consumption in the first quarter of 2024.

[0102] The results of the Holt-Went model's predictions for the first quarter of 2024 are shown in Table 5 below:

[0103] Quarterly data Holt-Winter Q1 of 2024 1.5626

[0104] Table 5 shows the results of the first quarter of 2024 predicted using the Holt-Winter model.

[0105] S6. Calculate safety stock data, and based on the safety stock data, the predicted spare parts demand data, and the spare parts database data, calculate the net spare parts purchase quantity data.

[0106] Based on the periodic ordering model formula and assumptions, according to the lead time of different materials, the average quarterly demand, the quarterly standard deviation and the service level, the safety stock of spare parts is calculated according to the safety stock formula. The safety stock is shown in the following formula (8):

[0107] SS=Zσ T+L (8);

[0108] Where z is the standard deviation multiple for a specific service probability, σ T+L It is the standard deviation of the inventory count cycle and the lead time demand, where T represents the number of days between two consecutive inventory counts and L represents the lead time in days;

[0109] The results are shown in Table 6 below:

[0110]

[0111] Table 6 Safety Stock Results

[0112] Based on the read inventory and in-transit quantities of spare parts, combined with the safety stock and the projected quarterly demand, the purchase quantity of spare parts is calculated, and the results are shown in Table 7 below:

[0113] Safety Stock Number of items in transit Number of items in stock Forecast demand Net purchases 3.4 1 42 1.5626 0

[0114] Table 7 Net Purchases Results for the First Quarter of 2024

[0115] Example 2

[0116] This embodiment discloses a net purchase volume prediction system based on wind turbine spare parts data, used to implement the net purchase volume prediction method based on wind turbine spare parts data described in Embodiment 1, including:

[0117] The spare parts database module is used to store historical consumption data, in-transit data, and in-stock data of spare parts.

[0118] The preprocessing module is used to read historical consumption data from the spare parts database module and perform cleaning and outlier removal operations.

[0119] The feature analysis and model selection module is used to perform seasonal and trend feature analysis on the preprocessed historical consumption data, and finally select the appropriate time series prediction model based on the features.

[0120] The model building module is used to input historical consumption data into the selected time series prediction model for training.

[0121] The model validation module is used to calculate error statistics to quantify the accuracy of the trained model.

[0122] Safety stock calculation module, used to calculate spare parts safety stock;

[0123] The forecast procurement quantity calculation module calculates the net procurement quantity of spare parts based on safety stock data, forecasted spare parts demand data, and spare parts database data.

[0124] Example 3

[0125] This embodiment discloses a non-transitory computer-readable medium storing instructions that, when executed by a processor, perform the steps of the net purchase quantity prediction method based on wind turbine spare parts data as described in Embodiment 1.

[0126] In this embodiment, the non-transitory computer-readable medium can be a disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), USB flash drive, portable hard drive, etc.

[0127] Example 4

[0128] This embodiment discloses a computing device, including a processor and a memory for storing processor-executable programs. When the processor executes the program stored in the memory, it implements the net purchase quantity prediction method based on wind turbine spare parts data described in Embodiment 1.

[0129] The computing device described in this embodiment may be a desktop computer, laptop computer, smartphone, PDA handheld terminal, tablet computer, programmable logic controller (PLC), or other terminal device with processor function.

[0130] The above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Therefore, any changes made in accordance with the shape and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting net purchase volume based on wind turbine spare parts data, characterized in that, Includes the following steps: S1. Establish a spare parts database and read historical consumption data of spare parts; S2. Clean and remove outliers from the historical consumption data of spare parts read from the spare parts database. S3. Perform feature analysis on the pre-processed historical consumption data of spare parts, based on... Feature-selective time series prediction models include: Seasonal, trend and autocorrelation analyses were performed on the pre-processed historical consumption data of spare parts, and the appropriate time series forecasting model was selected based on the data characteristics obtained from the analysis. Among them, the time series graph is used to determine whether the historical consumption data of spare parts is seasonal. If the time series graph shows periodic changes, that is, the data value changes greatly within a preset period and changes little within other periods, then the current historical consumption data of spare parts is considered to be seasonal; otherwise, it is not seasonal. By drawing a time series graph, it can be determined whether the historical consumption data of spare parts has a trend. If the time series graph shows a trend, that is, the data value continues to increase or decrease within a preset period, then the historical consumption data of spare parts is considered to have a trend; otherwise, it does not have a trend. The DW test is used to evaluate the autocorrelation of historical spare parts consumption data, with the following criteria: a) When the test value DW∈(1.5,2.5), the data residuals are independent of the independent variable; b) When the test value DW < 1, the residuals of adjacent data points are positively correlated; c) When the test value DW>3, the residuals of adjacent data points are negatively correlated; The time series prediction models include simple moving average model, weighted moving average model, first exponential smoothing model, second exponential smoothing model, Holt-Winter model, and autoregressive difference moving average model. If the historical spare parts consumption data has the following characteristics: the data volume is less than the preset data volume or the volatility of the historical spare parts consumption data is less than the preset volatility, then a simple moving average model is used; if the historical spare parts consumption data has the following characteristics: there are multiple changing factors, but no trend or seasonality, then a weighted moving average model is used; if the historical spare parts consumption data has the following characteristics: there is no trend or seasonality, then a single exponential smoothing model is used; if the historical spare parts consumption data has the following characteristics: there is a trend but no seasonality, then a double exponential smoothing model is used; if the historical spare parts consumption data has the following characteristics: there is a trend and seasonality, then a Holt-Winter model is used; if the historical spare parts consumption data has the following characteristics: there is autocorrelation and seasonality, then an autoregressive difference moving average model is used. S4. Input the preprocessed historical consumption data of spare parts into the time series prediction model selected in step S3 for training, and obtain the trained spare parts demand prediction model. S5. Use the trained spare parts demand prediction model to predict the spare parts demand data at future times, and quantify the accuracy of the spare parts demand prediction model by calculating error statistics. Select the spare parts demand prediction model to be applied based on the error statistics. S6. Calculate safety stock data, and based on the safety stock data, predicted spare parts demand data, and spare parts database data, calculate the net spare parts procurement data, including: Safety stock is calculated based on the formula of the periodic ordering model, assuming periodic ordering and replenishment. The safety stock of each spare part is calculated according to the procurement lead time, quarterly average consumption, quarterly consumption standard deviation and service level. The safety stock is shown in the following formula (8): SS=Zσ T +L (8); Where Z is the standard deviation multiple for a specific service probability, and σ T+L It is the standard deviation of the inventory count cycle and the lead time demand, where T represents the number of days between two consecutive inventory counts and L represents the lead time in days; The purchase quantity is calculated by subtracting the sum of the safety stock and the predicted quarterly demand from the spare parts inventory and in transit quantities read from the spare parts database, ultimately yielding the recommended net purchase quantity of spare parts.

2. The method for predicting net purchase volume based on wind turbine spare parts data according to claim 1, characterized in that, Step S1 includes: Establish a spare parts database, collect historical spare parts consumption data, and perform quarterly statistics on historical spare parts consumption data based on the historical spare parts consumption data; wherein, the spare parts database also includes spare parts in stock data and spare parts in transit data.

3. The method for predicting net purchase volume based on wind turbine spare parts data according to claim 1, characterized in that, Step S3 includes: If the historical consumption data of spare parts has the following characteristics: the data volume is less than the preset data volume or the volatility of the historical consumption data of spare parts is less than the preset volatility, then a simple moving average model is adopted, as shown in the following formula (1): Among them, S t +S t-1 +…+S t-n-1 It represents the observations from the past n time points, where n is the window size of the moving average, i.e., the number of past data points; If the historical consumption data of spare parts has multiple variable factors and no trend or seasonality, then the weighted moving average model is adopted, as shown in the following formula (2): S t+1 =ω0S t +ω1S t-1 +…+oh n-1 S t-n-1 (2); Where, ω i For the observed value S t-i The corresponding weights are i, which is a constant, and n, which is the window size of the moving average.

4. The method for predicting net purchase volume based on wind turbine spare parts data according to claim 1, characterized in that, Step S3 includes: If the historical consumption data of spare parts does not exhibit trends or seasonality, then a single exponential smoothing model is adopted, as shown in the following formula (3): S t+1 =αX t+1 +(1-α)S t (3); Among them, S t+1 Let X represent the predicted point at time t+1. t+1 S represents the actual observed value at time point t+1. t This represents the predicted value at time point t, where α is the smoothing coefficient. If the historical consumption data of spare parts has a trend but no seasonality, then a quadratic exponential smoothing model is adopted, as shown in the following formula (4): Among them, S t+1 Let X represent the predicted point at time t+1. t+1 S represents the actual observed value at time point t+1. t Let b represent the predicted value at time point t. t+1 S represents the predicted value at time point t+1. t+1 The predicted value S at time point t t The difference is where α represents the smoothing coefficient and β represents the trend smoothing coefficient.

5. The method for predicting net purchase volume based on wind turbine spare parts data according to claim 1, characterized in that, Step S3 includes: If the historical consumption data of spare parts exhibits trends and seasonality, then the Holt-Winter model is adopted, as shown in the following formula (5): Among them, S t+1 Let X represent the predicted point at time t+1. t+1 S represents the actual observed value at time point t+1. t Let L represent the predicted value at time t, where L is the season length, S is the stable component, b is the linear component, I is the seasonal component, α is the smoothing parameter representing the level, β is the smoothing parameter representing the trend, and γ is the smoothing parameter representing the seasonality.

6. The method for predicting net purchase volume based on wind turbine spare parts data according to claim 1, characterized in that, Step S3 includes: If the historical consumption data of spare parts exhibits autocorrelation and seasonality, then the autoregressive moving average model is adopted, as shown in the following formula (6): Among them, S t+1 S represents the predicted point at time t+1. t Let p represent the predicted value at time point t, where p is the autoregressive order and q is the moving average order. Both β and ε are non-zero undetermined coefficients, ε is an independent error term, and i and j are constants.

7. The method for predicting net purchase volume based on wind turbine spare parts data according to claim 1, characterized in that, In step S5, the error statistics include: The statistical indicators mean absolute percentage error (MAPE), mean absolute value error (MAE), and symmetric mean absolute percentage error (SMPAE) are shown in the following formula (7): Among them, S i The actual value at time i. Let be the predicted value at time i.

8. A net purchase quantity prediction system based on wind turbine spare parts data, characterized in that, The method for predicting net purchase volume based on wind turbine spare parts data as described in any one of claims 1-7 includes: The spare parts database module is used to store historical consumption data, in-transit data, and in-stock data of spare parts. The preprocessing module is used to read historical consumption data from the spare parts database module and perform cleaning and outlier removal operations. The feature analysis and model selection module is used to perform seasonality, trend and autocorrelation feature analysis on the preprocessed historical consumption data, and finally select the appropriate time series prediction model based on the features. The model building module is used to input historical consumption data into the selected time series prediction model for training. The model validation module is used to calculate error statistics to quantify the accuracy of the trained model. Safety stock calculation module, used to calculate spare parts safety stock; The forecast procurement quantity calculation module calculates the net procurement quantity of spare parts based on safety stock data, forecasted spare parts demand data, and spare parts database data.

Images