Method and system for predicting demand of aerial material spare parts

Abstract

The invention discloses a demand prediction method and system for aerial material spare parts, and relates to the field of aerial material management. The invention aims to provide the demand prediction system and method for the aerial material spare parts specially for the scene of demand prediction of the aerial material spare parts, the method is improved on the basis of the existing statistical analysis technology, the influence of various factors on the demand is considered in the demand change, the prediction deviation caused by human intervention is reduced, and the prediction accuracy is improved. Based on a time sequence method, a reasonable and effective classification mode and a scientific and accurate demand prediction method for different demand characteristic aviation materials are researched, a more accurate maintenance aviation material spare part consumption prediction model is obtained, an efficient prediction method suitable for aviation material management is provided, the purpose of reducing aviation material waste is achieved, and finally the reduction of aviation material cost is achieved. The method is of great significance to aviation material management and maintenance plan optimization of airline companies.

Description

Technical Field

[0001] The present invention relates to the field of aviation material management, and in particular to a method and system for forecasting demand for aviation material spare parts. Background Art

[0002] In the fiercely competitive civil aviation market, domestic airlines face the pressure of high operating costs. Especially in economic downturns, civil aviation needs to have the ability to make continuous profits in order to cope with the competitive market. This requires them to improve their management level at all stages, implement "savings" to control costs while expanding "open sources", strive to achieve refined management, and cope with market competition. In general, the cost of aviation materials is the part with a relatively high proportion of funds. At present, the models of mainstream domestic feeder lines are highly patented and the market circulation is highly specialized, but the reusable nature of aviation material turnover parts means that if the plan is unreasonable, there may be inventory backlogs. If there is too much inventory, it will lead to a significant increase in inventory costs and financial pressure. If there are missing parts in the inventory and they cannot meet the use, it will directly affect the aircraft's maintenance and warranty plan, and even cause flight delays.

[0003] Therefore, reducing the cost of aviation materials has become a cost control and enhanced confrontation strategy for airlines. Lean management requires new requirements for aviation material management, systematic management and scientific forecasting to effectively reduce the cost of aviation materials. Accurate aviation material planning is based on accurate estimation of demand, so the demand forecasting method for aviation material spare parts is crucial. The demand forecasting of aviation material spare parts for civil aviation aircraft maintenance has always been the focus of attention, which not only supports maintenance planning and inventory optimization, but is also crucial to management.

[0004] At present, aviation material management mainly relies on manual planning of maintenance aviation material spare parts demand based on historical data. Maintenance engineers collect the amount of zero aviation material spare parts required to complete maintenance tasks based on experience, which has great limitations and disadvantages. With the increase in the number of maintenance tasks and the complexity of aviation material spare parts, the seasonality and trend changes in demand are not taken into account, resulting in large errors in demand forecasting. In the scenario of aviation material spare parts consumption forecasting, the failure to conduct demand forecasting will seriously affect the accuracy of aviation forecasts, and will have a serious negative impact on the accuracy of aviation material spare parts demand forecasting, thereby affecting the rationality of aviation material costs, the effectiveness of inventory management and maintenance efficiency, and may even affect the progress of aviation safety work. Therefore, it is very important to optimize the aviation material spare parts consumption forecasting method and implement quota management. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a demand forecasting system and method for aviation materials and spare parts. The technical problem to be solved is: the traditional demand forecasting method for aviation materials and spare parts is mainly based on statistical analysis of historical consumption data, and there are large errors when applied to the scenario of aviation material management, which may affect the efficiency and safety of aircraft maintenance, especially in the scenario of large fluctuations in demand and complex and changeable scenarios; the purpose of the present invention is to specifically provide a demand forecasting system and method for aviation materials and spare parts for the scenario of aviation material and spare parts demand forecasting, which improves the method based on the existing statistical analysis technology, is based on the time series forecasting method, and considers the influence of various factors on demand in the change of demand, and comprehensively considers various demand characteristics in the forecasting process, reduces the forecast deviation caused by human intervention, and provides an efficient forecasting method suitable for aviation material management.

[0006] The purpose of the present invention is mainly achieved through the following technical solutions: A demand forecasting system for aviation material and spare parts specifically comprises an aviation material and spare parts demand classification unit and a maintenance aviation material and spare parts consumption forecasting unit which are connected in sequence.

[0007] According to the aviation material data used in airlines' historical data for route work, scheduled inspection work, non-routine work and routine work, the data on the demand for each type of aviation material is counted as the collected historical aviation material data. The aviation material spare parts demand classification unit collects the collected data according to the consumption data of each type of aviation material spare parts at equal monthly intervals to obtain the original sequence of each type of aviation material spare parts consumption, and classifies each original sequence of aviation material spare parts consumption through the Syntetos aviation material spare parts demand classification judgment standard module. The Syntetos aviation material spare parts demand classification judgment standard module adopts a two-dimensional classification method to classify the original sequence of aviation material spare parts consumption into three types of aviation material spare parts: flat demand, intermittent demand and random demand according to the square of the coefficient of variation and the set threshold of the average demand interval.

[0008] The maintenance aviation material and spare parts consumption prediction unit includes three time series modules for respectively predicting the demand for aviation material and spare parts for smoothly demanded aviation material, intermittent demanded aviation material and random demanded aviation material. Among them, the consumption of aviation material and spare parts with smoothly demand is predicted by a time series module including an exponential smoothing method, the consumption of intermittent aviation material and spare parts is predicted by a time series module including a solution to a non-periodic model, and the consumption of random aviation material and spare parts is predicted by a time series module including an autoregressive moving average model.

[0009] The specific steps of the method for predicting demand for aviation material spare parts of the present invention are as follows: Step S1: Construct a data set of aviation spare parts consumption based on monthly statistics of the historical consumption of each aviation spare part within a set period. The data set of aviation spare parts consumption from the first month to the nth month is used as the original sequence of aviation spare parts consumption, recorded as , a two-dimensional classification method is used to classify the original sequence of aviation material and spare parts consumption into three types of aviation material and spare parts demand: smooth demand, intermittent demand, and random demand according to the square of the coefficient of variation and the set threshold of the average demand interval; among them, the square of the coefficient of variation is recorded as CV², which is the standard deviation of the demand for aviation material and spare parts divided by the square of the mean demand; the average demand interval is recorded as ADI, which is the average time interval between two consecutive demands for aviation material and spare parts; Step S2, using the time series forecasting method to predict the demand for each type of aviation material spare parts according to the type of aviation material spare parts; wherein, the consumption of aviation material spare parts with smooth demand is predicted by the exponential smoothing method in the time series model, the consumption of intermittent aviation material spare parts is predicted by the non-periodic solution model in the time series model, and the consumption of random aviation material spare parts is predicted by the autoregressive moving average model in the time series model.

[0010] Further optimization, the consumption of aviation materials and spare parts with smooth demand is predicted by using the exponential smoothing method in the time series model, including: According to the original sequence of aviation material and spare parts consumption, the time series of aviation material and spare parts consumption is obtained as follows: , from the first month to the nth month of the original sequence of aviation material and spare parts consumption, set the observation interval time, arbitrarily take 1 to N interval times, and the aviation material and spare parts consumption at each time is taken as the aviation material and spare parts consumption time series.

[0011] Smooth the data and calculate the exponential smoothing of the demand for aviation spare parts at time t as follows: ,in, is the smoothing coefficient, whose value is between 0 and 1. is the historical demand value of aviation spare parts at time t, is the predicted value; among them, the initial value , is the initial value.

[0012] By calculating the MAD indicator under different smoothing coefficients, it is expressed as The smoothing index is determined and the mean square error MSD is calculated, which is expressed as , using planning solution, we can get the optimal value of the planning solution target. When the mean square error MSD is the minimum, there is an independent variable smoothing coefficient The optimal solution of .

[0013] When the image of the predicted data is a straight line, the linear smoothing method is no longer applicable, so it is necessary to introduce the quadratic exponential smoothing method for prediction. The prediction model is expressed as: ,in, for The predicted value of aviation materials at the time, T is the predicted time, and is the first smoothing value of period t and t-1, and are the quadratic exponential smoothing values ​​at period t and t-1 respectively.

[0014] For further optimization, the autoregressive moving average model of the time series model is used to predict the random aviation material and spare parts consumption, wherein the autoregressive moving average model is recorded as the ARIMA (p, d, q) model, which specifically includes: obtaining the aviation material and spare parts consumption time series according to the original aviation material and spare parts consumption sequence, performing a stationarity test on the aviation material and spare parts consumption time series, using SPSS software to visualize the time series data, observing the seasonal and stationary characteristics according to the time series diagram, autocorrelation function diagram ACF and partial autocorrelation function diagram PACF of the time series, preliminarily judging the stationarity of the data, and dividing the aviation material time series into a stationary time series and a non-stationary time series.

[0015] The non-stationary time series is stabilized and the difference operation is performed on the non-stationary time series until the processed data passes the stationarity test. The number of difference operations at this time is recorded as the number of differences. In the ARIMA (p, d, q), p is the autoregressive order, d is the number of differences, and q is the sliding average order.

[0016] The ARIMA (p, d, q) model includes three parts: the autoregressive model AR (p), the single integration I (d) and the moving average model MA (q); AR is the sequence value of the time series, which is a linear combination of the observations lagged p periods; the number of single integration I (d) specifically refers to the number of difference processing required to transform a non-stationary time series into a stationary series after d differences; MA is the model value, which is expressed as: a linear function of the q-order residual term, Calculation of ARIMA(p,d,q) model: ,in, are the parameters of the autoregressive model, where The value cannot be 0. represents the observed value at time t; is a white noise sequence; is the parameter of the moving average model, where the value of q cannot be 0. is the deviation between the predicted value and the sequence value at time t-1, is the deviation between the predicted value and the sequence value at time tq, yes The weight of .

[0017] ARIMA (p, d, q) model, where the calculation of the p-order autoregressive model is: , ARIMA (p, d, q) model, where the q-order moving average model is calculated as: , Get the autoregressive moving average model ARMA(p,q): , in, represents the sequence value at any time t, represents the random error at any time t, represents the linear combination of the first p sequence values ​​at time t, represents a constant, Represents the linear combination of the cumulative sum of white noise of the first q sequences at time t.

[0018] The BIC criterion is used to determine the model order. The BIC criterion is the Bayesian Information Criterion. , where k is the number of parameters in the model, including the number of autoregressive coefficients, number of differences, and number of moving average coefficients; n is the number of samples; and L is the likelihood function.

[0019] Take an upper limit value of p and q, which is less than 10, and then traverse all p and q combinations, and calculate the BIC values ​​by constructing the BIC matrix. The p and q values ​​with the smallest BIC values ​​are the final p and q. That is, select the data with the smallest BIC information as the values ​​of p and q in the ARIMA (p, d, q) model to determine the ARIMA (p, d, q) model.

[0020] The ARIMA (p, d, q) model is tested to determine whether the residual sequence is a white noise sequence. The residual is the predicted value minus the true value, and the residual sequence is all predicted value sequences minus the true value sequence. White noise is a special weak stationary process. The residual sequence is tested for white noise. If the residual sequence is white noise, it means that the valid data has been extracted from the original sequence. The ARIMA (p, d, q) model that has passed the test is used for prediction to obtain the predicted random aviation material and spare parts consumption data.

[0021] For further optimization, the intermittent aviation material and spare parts consumption is predicted by using the non-periodic model solved by the time series model, wherein the intermittent demand forecasting Croston method or the grey forecasting method is used to predict the intermittent aviation material and spare parts consumption.

[0022] Among them, the intermittent demand forecasting Croston method is decomposed into two stages: one is the prediction interval length, and the other is the prediction of the demand in each interval, using exponential smoothing and linear regression models respectively; the grey prediction method adopts the GM (1,1) model and uses the weighted cumulative generating function to process the sequence data.

[0023] Further optimization, the intermittent demand forecasting Croston method specifically includes: According to the original sequence of aviation material and spare parts consumption, it is converted into demand interval sequence and demand quantity sequence. According to the original sequence of aviation spare parts consumption, the monthly consumption of aviation spare parts is the demand, the intervals between two adjacent non-zero demands constitute the demand interval sequence, and the non-zero demands constitute the demand sequence, and we get: Demand interval sequence ,in , Aircraft spare parts demand sequence ,in .

[0024] The demand interval series and the demand series of aviation spare parts are predicted by combining the exponential smoothing method and the linear regression model, which can be expressed as: Demand interval estimation , demand estimate ,in, is the smoothing coefficient, and its value should be between 0 and 1; The final demand forecast value for the next period is: .

[0025] Further optimization, the GM(1,1) model specifically includes: Get the original non-negative sequence based on the original sequence of aviation material and spare parts consumption ,in ; Perform accumulation operation on the original non-negative sequence , and the generated sequence is .

[0026] The generated sequence after accumulation and the original sequence satisfy the grey differential equation, that is, the whitened form equation is established: , this is a first-order differential equation model with one variable, so it is recorded as GM (1,1). in, is the development coefficient, is the ash action, .

[0027] The parameters are estimated using the least squares method, and we get: ,in, , Generate Sequence Whitening differential equations satisfying the GM(1,1) model , solve for the development coefficient Amount of ash action , and then substitute it into the above whitened differential equation to obtain the time response function of the GM(1,1) model , the prediction model is restored .

[0028] Further optimization, a two-dimensional classification method is used to divide the original sequence of aviation material spare parts consumption into three types of aviation material spare parts: smooth demand, intermittent demand and random demand according to the square of the coefficient of variation and the set threshold of the average demand interval. Specifically, the average demand interval ADI and the square of the coefficient of variation of aviation material replacement data are calculated as CV², where the square of the coefficient of variation CV² is the standard deviation of the demand divided by the square of the mean demand, expressed as: ,in, .

[0029] The average demand interval ADI is the average time interval between two consecutive demands for aviation materials and spare parts, expressed as ,in, The quantity of aviation spare parts required. is the time interval between two consecutive aviation spare parts requirements, N is the number of time intervals, is the average demand value.

[0030] Among them, the threshold CV is set 2 =0.49, ADI=1.32 is used as the dividing value between different demand types.

[0031] When ADI < 1.32, CV 2 <0.49, the demand for aviation spare parts is a flat demand; when ADI<1.32, CV 2 When ADI is ≥0.49, the demand for aviation materials and spare parts is random; when ADI is ≥1.32, the demand for aviation materials and spare parts is intermittent.

[0032] Further optimization, the system specifically includes a sequentially connected aviation material spare parts demand classification unit and a maintenance aviation material spare parts consumption prediction unit; the aviation material spare parts demand classification unit collects the collected data according to the data of each type of aviation material spare parts consumption at equal monthly intervals to obtain each type of aviation material spare parts consumption original sequence, and each type of aviation material spare parts consumption original sequence is classified into three types of aviation material spare parts, namely, flat demand, intermittent demand and random demand, according to the square of the coefficient of variation and the set threshold of the average demand interval, by a two-dimensional classification method through the Syntetos aviation material spare parts demand classification judgment standard module.

[0033] The maintenance aviation material and spare parts consumption prediction unit includes three time series modules for respectively predicting the demand for aviation material and spare parts for smoothly demanded aviation material, intermittent demanded aviation material and random demanded aviation material, wherein the consumption of smoothly demanded aviation material and spare parts is predicted by a time series module including an exponential smoothing method, wherein the consumption of intermittent aviation material and spare parts is predicted by a time series module including a solution to a non-periodic model, wherein the consumption of random aviation material and spare parts is predicted by a time series module including an autoregressive moving average model.

[0034] Further optimization is performed to solve the time series module of the non-periodic model by using the intermittent demand forecasting Croston module.

[0035] Further optimization, the time series module to solve the non-periodic model is implemented by using the grey prediction module

[0036] In summary, the present invention has the following beneficial effects compared with the prior art: The time series prediction method of the present invention is established on a rigorous mathematical theoretical basis, and has the characteristics of simple structure, fast prediction speed, and convenient operation. A two-dimensional classification method is used to further classify the demand for aviation materials and spare parts, and a more appropriate time series prediction model is adopted for different types of aviation materials and spare parts demand, so as to more accurately predict the demand for different types of aviation materials and spare parts, thereby ensuring the use demand of aviation materials and avoiding excessive procurement, issuance and storage of aviation materials, which leads to waste. Moreover, the system and method can transfer learning to other prediction scenarios and can be applied in many industrial practices. In the scenario of demand prediction, since both the target data and the source data are time series data, the same feature columns can be constructed from the time series, so as to perform sample fusion training, thereby training a model for the target data. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings: Figure 1 This is a flow chart of a demand forecasting system for aviation spare parts; Figure 2 A diagram of a demand forecasting method for aviation spare parts; Figure 3 It is a classification diagram of the demand characteristics of aviation materials and spare parts; Figure 4 This is a diagram of the consumption forecasting method for different types of aviation materials and spare parts using the time series forecasting method. DETAILED DESCRIPTION

[0038] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0039] Traditional methods for forecasting demand for aviation materials and spare parts are mainly based on statistical analysis of historical consumption data. When applied to aviation material management scenarios, there are large errors, which may affect the efficiency and safety of aircraft maintenance, especially in scenarios with large fluctuations in demand and complex and changeable situations. The purpose of the present invention is to specifically target scenarios for forecasting demand for aviation materials and spare parts. In view of this, the present invention provides the following embodiments to solve the above-mentioned technical problems.

[0040] like Figure 1 As shown, the principle of the present invention is: A demand forecasting system for aviation material and spare parts specifically comprises a sequentially connected aviation material and spare parts demand classification unit and a maintenance aviation material and spare parts consumption forecasting unit.

[0041] Based on the aviation material usage data for route work, scheduled inspection work, non-routine work and routine work in the airline's historical data, the data on the demand for each type of aviation material is statistically compiled as the collected historical aviation material data.

[0042] The aviation material and spare parts demand classification unit collects the collected data according to the consumption data of each type of aviation material and spare parts at equal intervals on a monthly basis to obtain the original sequence of each type of aviation material and spare parts consumption. Each original sequence of aviation material and spare parts consumption is classified into three types of aviation material and spare parts, namely, smooth demand, intermittent demand and random demand, according to the square of the coefficient of variation and the set threshold of the average demand interval through the Syntetos aviation material and spare parts demand classification judgment standard module using a two-dimensional classification method; the maintenance aviation material and spare parts consumption prediction unit includes three time series modules for respectively predicting the demand for aviation material and spare parts for smooth demand aviation material, intermittent demand aviation material and random demand aviation material, wherein the consumption of aviation material and spare parts for smooth demand is predicted by a time series module including an exponential smoothing method, wherein the consumption of intermittent aviation material and spare parts is predicted by a time series module including a solution to a non-periodic model, wherein the consumption of random aviation material and spare parts is predicted by a time series module including an autoregressive moving average model.

[0043] The specific operation method of the demand forecasting system for aviation material spare parts of the present invention is as follows: Figure 2 , a demand forecasting method for aviation spare parts is as follows: Step S1: Construct a data set of aviation spare parts consumption based on monthly statistics of the historical consumption of each aviation spare part within a set period. The data set of aviation spare parts consumption from the first month to the nth month is used as the original sequence of aviation spare parts consumption, recorded as , a two-dimensional classification method is used to divide the original sequence of aviation material and spare parts consumption into three types of aviation material and spare parts demand: smooth demand, intermittent demand and random demand according to the square of the coefficient of variation and the set threshold of the average demand interval; among them, the square of the coefficient of variation is denoted as CV², which is the standard deviation of the demand for aviation material and spare parts divided by the square of the mean demand; the average demand interval is denoted as ADI, which is the average time interval between two consecutive demands for aviation material and spare parts.

[0044] Step S2, using the time series forecasting method to predict the demand for each type of aviation material spare parts according to the type of aviation material spare parts; wherein, the consumption of aviation material spare parts with smooth demand is predicted by the exponential smoothing method in the time series model, the consumption of intermittent aviation material spare parts is predicted by the non-periodic solution model in the time series model, and the consumption of random aviation material spare parts is predicted by the autoregressive moving average model in the time series model.

[0045] In specific embodiment 1: Step S1: Classify the demand for aviation material spare parts based on the collected historical aviation material data.

[0046] In actual applications, the properties of aviation materials are complex, so there are many classification methods, and different methods target different properties. Traditional classification methods include ABC classification, repairability classification, importance classification, and life time limit classification. However, these traditional methods have some problems in aviation material management. First, the evaluation dimension of traditional classification is relatively single, mainly reflecting the characteristics of aviation materials at a certain technical level, and has limited guiding role in aviation material management. The demand patterns and economic characteristics of the same type of aviation materials may also vary greatly, resulting in a lack of reference value when forecasting demand. Secondly, there are no clear boundaries between the categories of some traditional classification methods, and most of them are based on qualitative analysis of specific conditions. Finally, these classification methods are independent of each other but the results overlap, which makes it difficult for aviation material departments to operate and find the focus when managing. Therefore, traditional classification methods are difficult to effectively apply in aviation material management and demand forecasting.

[0047] The research on aviation material classification is an important foundation for aviation material management, forecasting, decision-making and other aspects. According to the emphasis of different influencing factors, a variety of classification principles can be adopted, aiming to more accurately predict the demand for the use of various aviation materials in the maintenance work card. Therefore, the classification should be carried out according to the demand characteristics of maintenance aviation materials. The demand characteristics of aviation material spare parts include a wide variety of categories, large quantities, and significant differences in the demand patterns of different categories. Based on the randomness of the time and number of occurrences of aviation material replacement, the present invention adopts the two-dimensional classification method proposed by Syntetos et al., and based on the mining and research of a large amount of data, directly finds the classification boundaries from the data, and classifies the original sequence of aviation material spare parts consumption.

[0048] When conducting a quantitative analysis of the demand for aviation materials and spare parts, randomness is reflected in two aspects. First, the time of occurrence of the demand for aviation materials and spare parts is random. In a specific time period, the demand may appear or not appear at any time. Secondly, the number of times the demand occurs is also random, that is, in a certain period of time, even if there is demand, the specific number of times it occurs is uncertain, so a two-dimensional classification method is adopted. When using the two-dimensional classification method, based on the average demand interval ADI and the square of the coefficient of variation of aviation material replacement data is CV², the original sequence of aviation material and spare parts consumption can be divided into four categories according to the values ​​of these two indicators: flat demand, intermittent demand, block demand and random demand.

[0049] Calculate the average demand interval ADI and the square of the coefficient of variation of the material replacement data as CV², where the square of the coefficient of variation CV² is the standard deviation of the demand divided by the square of the mean demand, expressed as ,in, , The average demand interval ADI is the average time interval between two consecutive demands for aviation materials and spare parts, expressed as: ,in, The quantity of aviation spare parts required. is the time interval between two consecutive aviation spare parts requirements, N is the number of time intervals, is the average demand value.

[0050] Among them, the threshold values ​​CV²=0.49 and ADI=1.32 are set as the dividing values ​​between different demand types; smooth demand: ADI<1.32, CV²<0.49, the intervals between demand and the changes in demand are relatively stable; intermittent demand: ADI≥1.32, CV²<0.49, the intervals between demand are random, and there is no demand for a long time, that is, the demand is equal to 0, but the changes in demand are relatively stable; random demand: ADI<1.32, CV²≥0.49, the changes in demand are irregular, and the intervals between demand are relatively stable; block demand: ADI≥1.32, CV²≥0.49, the intervals between demand are random, and the changes in demand are irregular, and there is no demand for many time periods, that is, the demand is 0. Among them, smooth demand and intermittent demand belong to regular demand; according to the actual use and consumption data of aviation materials in the maintenance tasks of the maintenance work card, and for the convenience of subsequent research on different categories, the block demand is regarded as intermittent demand; the four types are simplified into three categories: smooth, non-smooth / random, and intermittent. This method is very intuitive and easier to apply in practice, such as Figure 3:When ADI<1.32, CV²<0.49, the demand for aviation materials and spare parts is a flat demand; when ADI<1.32, CV²≥0.49, the demand for aviation materials and spare parts is a random demand; when ADI≥1.32, the demand for aviation materials and spare parts is an intermittent demand; the demand for aviation materials with a flat demand is relatively stable, with small ADI and CV², stable data, and easy prediction; the demand for aviation materials with an intermittent demand is zero in some time periods, and the demand fluctuates greatly; the average demand interval for aviation materials with a random demand is small, the demand is relatively unstable, and the data fluctuates greatly.

[0051] Step S2: Use time series prediction method to predict the consumption of different types of aviation materials and spare parts, such as Figure 4 : Maintenance spare parts consumption forecasting is a comprehensive analysis process that takes into account the characteristics of aviation materials, the maintenance plans, maintenance capabilities and usage environment of each unit, and combines historical data to classify the demand for aviation materials and spare parts through a two-dimensional classification method. For different types of aviation materials, a variety of forecasting methods are used to infer the quantity of aviation materials required for the next mission of the maintenance base.

[0052] Time series data will show obvious trends and patterns of change, which can be used to extrapolate and predict the data in the time series. In a set of time series data, later data often have a certain correlation with previous data. The previous data can be used to infer the range of later data and thus make predictions.

[0053] Step S21, the consumption of smooth aviation materials and spare parts is predicted by exponential smoothing method. Exponential smoothing method is a commonly used demand forecasting method, a method of time series forecasting, a data set arranged in chronological order, and continuous demand is analyzed by weighted average. The exponential smoothing value is the weighted average of the actual value at this moment and the exponential smoothing value at the previous moment. It is a weighted average of the past value and the sequence value and an adjustment of the current total to offset the random influence of the statistical value, thereby obtaining a smoothed time series and obtaining the weighted demand of historical data as the forecast result. Common exponential smoothing methods include primary exponential smoothing, secondary exponential smoothing and triple exponential smoothing. There is a certain regularity in the forecast accuracy of the three exponential smoothing methods and the three trends of consumption data. The primary exponential smoothing method is suitable for predicting time series with large fluctuations, the secondary exponential smoothing method is suitable for predicting time series with an upward trend, and the triple exponential smoothing method is suitable for predicting time series with a downward trend.

[0054] For aircraft maintenance materials with flat demand, the demand usually fluctuates around a steady level, so the first-order exponential smoothing method is used to forecast demand.

[0055] The time series of spare parts consumption is obtained based on the original sequence of spare parts consumption. The time series is a sequence of observations of a phenomenon at different times arranged in chronological order. The time series of spare parts consumption is recorded as , from the first month to the nth month of the original sequence of aviation material and spare parts consumption, set the observation interval time, arbitrarily take 1 to N interval times, and the aviation material and spare parts consumption at each time is taken as the aviation material and spare parts consumption time series.

[0056] Smooth the data and calculate the exponential smoothing of the demand for aviation spare parts at time t as follows:

[0057] in, is the smoothing coefficient, whose value is between 0 and 1. is the historical demand value of aviation spare parts at time t, is the predicted value; among them, the initial value , is the initial value.

[0058] When the sample size is large, the first value can be used instead. When the sample size is small, less than 20, the average value of the previous periods can be used instead.

[0059] for The selection of should refer to the volatility of historical data. When the data fluctuates greatly, the value of the previous actual consumption value for the forecast will become larger accordingly, so a higher smoothing coefficient should be selected. On the contrary, when the historical data fluctuates less, a lower smoothing coefficient can be selected. .specific The selection of the smoothing coefficient can also be determined based on the mean absolute error MAD generated by different smoothing coefficients during the simulation process.

[0060] By calculating the MAD indicator under different smoothing coefficients, it is expressed as , determine the smoothing index and calculate the mean square error MSD, expressed as , using planning solution, we can get the optimal value of the planning solution target. When the mean square error MSD is the minimum, there is an independent variable smoothing coefficient The optimal solution of .

[0061] The deviation between the model prediction value and the measured value is often difficult to compare. In this implementation, the Excel planning solver function is used to solve the problem with the minimum mean square error (MSD) value as the objective function and the smoothing coefficient is an optimization problem with independent variables.

[0062] Using planning solution, we can get the optimal value of planning solution. When the mean square error (MSD) is the minimum, we can get the corresponding smoothing coefficient of the independent variable. is the optimal value.

[0063] When the predicted data has an obvious linear trend, the graph is a straight line, and the first-order linear smoothing method is no longer applicable. Therefore, it is necessary to introduce the second-order exponential smoothing method for prediction.

[0064] Step S22: The intermittent demand forecasting Croston method is used to forecast the consumption of intermittent aviation materials and spare parts. Usually, the time series of aviation maintenance materials has two main characteristics: one is the sporadic nature of demand, with zero demand in some time periods; the other is the variability of demand, with large fluctuations in demand. These two characteristics are consistent with intermittent demand. Since the data contains zero values ​​and the demand data is a non-equal interval sequence, the intermittent demand forecasting Croston method is used to construct non-equal intervals for forecasting.

[0065] There are a lot of zero values ​​in intermittent demand, and historical data is scattered and lacks regularity. The Croston method for intermittent demand forecasting is an extension and development of the smoothing index method. It is mainly aimed at an algorithm that overestimates demand development when the exponential smoothing method predicts periodic intermittent demand. The Croston method for intermittent demand forecasting distinguishes the demand interval from the demand amount, and then uses the average demand amount to predict future demand: if demand occurs, the demand interval and demand amount are estimated, and then the demand amount is divided by the demand interval to get the average demand, and the average demand is used for forecasting; if demand does not occur, the original forecast is maintained, and only the demand interval from the last demand to the present is updated.

[0066] The Croston method for intermittent demand forecasting performs exponential smoothing on both non-zero demand and the time interval between non-zero demand, separates the demand interval from the demand, and uses the exponential smoothing method to calculate the demand interval and demand separately. This can greatly eliminate the forecast deviation after block demand. Practice has proved that this method has good adaptability to intermittent demand and uncertain demand.

[0067] For aviation spare parts whose demand follows normal distribution and whose demand intervals are relatively stable and follow geometric distribution, the prediction accuracy is generally high. In addition, this method can not only obtain more accurate prediction values, but also has a significant advantage that when there is demand in each time period, this method is similar to the exponential smoothing method. Therefore, this method can not only be used for demand forecasting of intermittent aviation spare parts, but also has a certain prediction effect for continuous aviation spare parts.

[0068] First, according to the original sequence of aviation material and spare parts consumption, the original sequence of a certain type of intermittent demand maintenance aviation material is converted into a demand interval sequence and a demand quantity sequence according to the elements in the original sequence of aviation material and spare parts consumption. The consumption of aviation material and spare parts each month is the demand quantity, and the intervals between two adjacent non-zero demand quantities constitute the demand interval sequence, and the non-zero demand quantity constitutes the demand quantity sequence, so we get: demand interval sequence ,in , Aircraft spare parts demand sequence ,in .

[0069] The demand interval series and the demand series of aviation spare parts are predicted by combining the exponential smoothing method and the linear regression model, which can be expressed as: Demand interval estimation , Demand Estimation , in, is the smoothing coefficient, its value should be between 0 and 1.

[0070] The final demand forecast value for the next period is: .

[0071] Step S23: The random aviation material spare parts consumption is predicted using the autoregressive moving average model ARIMA (p, d, q) model.

[0072] The demand characteristics of random aviation spare parts are small average demand interval and unstable demand. For this type of aviation material, the autoregressive moving average model, denoted as ARIMA(p,d,q) model, can be used to forecast demand. In ARIMA(p,d,q), p is the autoregressive order, d is the number of differences, and q is the sliding average order.

[0073] The ARIMA (p, d, q) model is a statistical method used to analyze and predict time series data. If the analyzed time series has a trend, it is necessary to describe the random sequence processing and generate a stationary sequence. Then, according to the statistical characteristics of the sample's autocorrelation function and partial autocorrelation function, determine the model type suitable for the random sequence, determine the order of the model, and establish a model of the random process of the random sequence. Finally, use the model of the best monomer to predict future demand through the past and present time series settings.

[0074] The specific steps of ARIMA (p, d, q) model modeling and prediction are as follows: The first step is stationarity test. The aviation spare parts consumption prediction model based on the ARIMA (p, d, q) model first performs a stationarity test on the aviation spare parts consumption time series, and divides the aviation material time series into stationary time series and non-stationary time series according to the test results. The stationarity of the data is preliminarily determined based on the time series diagram, autocorrelation function diagram, and partial autocorrelation function diagram of the time series.

[0075] For the consumption of maintenance aviation spare parts with random demand, the consumption data usually shows non-stationarity. The stationarity test of time series is the first step in predicting the consumption of aviation spare parts with ARIMA (p, d, q) model. First, the SPSS software is used to visualize the time series data, and the seasonality and stationarity characteristics are observed according to the time series time series diagram, autocorrelation function diagram ACF and partial autocorrelation function diagram PACF, and the stationarity of the data is preliminarily judged.

[0076] Judging from the autocorrelation function graph ACF and the partial autocorrelation function graph PACF, an important characteristic of the autocorrelation function graph ACF of a stationary series is that the autocorrelation coefficient will gradually converge to zero as the delay order increases; since the mean, variance or other statistical characteristics of a non-stationary time series change over time, its autocorrelation coefficient may not converge to zero as the delay order increases.

[0077] The autocorrelation function and partial autocorrelation function of the maintenance spare parts consumption with random demand quickly fall into the random interval, so the time series trend is basically eliminated and has randomness. The difference sequence is basically evenly distributed on both sides of the 0 scale line, so it can be considered that the difference sequence is stable, and the autocorrelation function graph ACF and partial autocorrelation function graph PACF of the difference sequence are both tailing, so it is suitable to use the ARIMA (p, d, q) model, and the ARIMA (p, d, q) model (p, 1, q) model can be established for the original sequence.

[0078] The second step is to stabilize the non-stationary time series. Stabilize the non-stationary time series data and perform differential operations on the non-stationary time series until the processed data can pass the stationarity test. Where d≠0, d is the number of differences.

[0079] Because the difference operation is performed to reduce the influence of time on the data, the difference result can be obtained by subtracting the current observation value from the previous moment value. Then the first-order seasonal difference is performed. Judging from the autocorrelation diagram (autocorrelation function diagram ACF) and partial correlation diagram (partial autocorrelation function diagram PACF), the time series tends to be stable after the first-order seasonal difference processing, and D=1 is taken.

[0080] The third step is model identification and order determination. The ARIMA (p, d, q) model is selected as the model for data analysis. At this time, d is taken as d=0. The BIC matrix of the sequence is calculated as the basis for model order determination. The model order determination includes determining the specific values ​​of p and q of the ARIMA (p, d, q) model. Among them, the ARIMA (p, d, q) model (p, 0, q) is equivalent to ARMA (p, q). (p, d, q) is a key step to ensure the accuracy and predictive performance of the ARIMA (p, d, q) model.

[0081] The ARIMA (p, d, q) model includes three parts: autoregressive model AR(p), single integration I(d) and moving average model MA(q); AR is the sequence value of the time series, which is a linear combination of the observations lagged p periods; the number of single integration I(d) specifically refers to the number of difference processing required to transform a non-stationary time series into a stationary series after d differences; MA is the model value, which is expressed as a linear function of the q-order residual term.

[0082] Calculation of ARIMA(p,d,q) model: , in, It is the deviation between the predicted value and the sequence value at time tq, and the values ​​at each moment are independently and simultaneously distributed.

[0083] The calculation of the p-order autoregressive model in the ARIMA (p, d, q) model is: , The calculation of the q-order moving average model in the ARIMA (p, d, q) model is: , Get the autoregressive moving average model ARMA(p,q): , in, represents the sequence value at any time t, represents the random error at any time t, represents the linear combination of the first p sequence values ​​at time t, represents a constant, Represents the linear combination of the cumulative sum of white noise of the first q sequences at time t.

[0084] The BIC criterion is used to determine the model order. The BIC criterion is the Bayesian Information Criterion. , determine the optimal model.

[0085] Where k is the number of parameters in the model, including the number of autoregressive coefficients, number of differences, and number of moving average coefficients; n is the number of samples; and L is the likelihood function.

[0086] Take an upper limit value of p and q, which is generally less than 10, and then traverse all p and q combinations, and calculate the BIC values ​​by constructing the BIC matrix. The p and q values ​​with the smallest BIC values ​​are the final p and q. That is, select the data with the smallest BIC information as the values ​​of p and q in the ARIMA (p, d, q) model.

[0087] Using SPSS software, on the basis of determining d=1, P=1, D=1, Q=1, the tailing and truncation of the autocorrelation function ACF and the partial autocorrelation function PACF are used as the basic judgment, and (p, q) is taken as different value combinations to compare the test results of different values. In general, p and q are determined according to the attenuation amplitude of the autocorrelation coefficient and the partial correlation coefficient. When the attenuation amplitude is very large, it is gradually fitted from p=1, q=1. When the attenuation amplitude is small, it is fitted from larger p and q values.

[0088] Step 4: Model parameter estimation. The coordinates corresponding to the minimum value in the BIC matrix are the optimal p and q values. Once the p and q values ​​are determined, the ARIMA model can be obtained.

[0089] Step 5: Model test. Assume the model and determine whether the residual sequence is a white noise sequence. The residual is the predicted value minus the true value, and the residual sequence is the sequence of all predicted values ​​minus the true value sequence. White noise is a special weak stationary process. Perform a white noise test on the residual sequence. If the residual sequence is white noise, it means that the valid data has been extracted from the original sequence.

[0090] Step 6: Model prediction: Use the ARIMA (p, d, q) model that has been tested to predict and obtain the predicted random aviation material and spare parts consumption data.

[0091] In a specific embodiment, it also includes step S3, formulating the usage quota standard and error range limit of each aviation material according to the result of the maintenance aviation material and spare parts consumption prediction model. The expert quota unit uses the Delphi method to statistically consult the expert group, collects and summarizes the revised opinions, and continuously repeats the above steps, feedbacks and modifies until the opinions of each expert are stable and unchanged, and calculates the usage quota standard and error range of aviation materials.

[0092] Firstly, the demand for aviation spare parts is predicted by time series forecasting method. Then, considering the complexity of maintenance, the Delphi method is used to determine the demand for maintenance engineers with different experience levels and set error range limits.

[0093] The core task of aviation material and spare parts consumption quota management is to optimize the use of aviation materials and provide solutions to aviation material management and supply issues. Strengthening this management can improve the efficiency of aviation material use and achieve the best inventory plan. At present, the research on aviation material and spare parts consumption quota is still insufficient, mainly focusing on the formulation, approval, execution, evaluation and revision of quotas. Aviation material and spare parts consumption quota is the basis for maintenance engineers to apply for aviation materials, helping aviation material departments to make purchase and allocation decisions, reduce blindness, and improve support efficiency. Reasonable aviation material and spare parts consumption quotas are conducive to rapid support and reasonable allocation of management funds, so it is necessary to conduct in-depth research on the consumption rules of aviation material and spare parts.

[0094] The aviation material and spare parts consumption quota in the maintenance link refers to the aviation material and spare parts consumption standard in the maintenance work card. The maintenance work card contains maintenance procedures, tools and material information, and engineers apply for aviation materials based on this. This embodiment is based on the established prediction model and combines the Delphi method to formulate the usage quota and error range limit for maintenance engineers at all levels, thereby forming the best maintenance work card quota plan. The maintenance work card, also known as the maintenance work order, is a maintenance work document compiled, compiled, checked and reviewed and published by the aircraft operator or maintenance unit. It is the basic basis for aircraft maintenance work. Continuous airworthiness maintenance plans, various maintenance technical documents provided by aircraft and aircraft component manufacturers, airworthiness directives, service bulletins or letters, and maintenance items determined by the operator can be used as the basis for compiling work orders.

[0095] Technicians of different ages and experiences may lead to differences in the amount of aviation materials used, and it is necessary to analyze the reasons for the engineers who use less. Based on the model prediction results, the Delphi method of expert scoring method is used to formulate the usage quota standards and set the error range limits for each category of maintenance engineers, and carry out the maintenance work card quota plan.

[0096] The Delphi method is to invite m experts, scholars, engineers and managers in the aviation field as an expert group, send the problems to be solved to each expert separately for consultation, then summarize the opinions of all experts and compile a comprehensive opinion, and then feed back the comprehensive opinion and predicted problems to the experts separately, solicit opinions again, and then summarize and feed back to the experts again, repeat the above steps, and gradually obtain a more accurate decision-making method of collective judgment.

[0097] In order to quantify the evaluation, we first need to set an experience level for each maintenance engineer, and divide them according to the number of historical maintenance operations, including: engineers with less than 10 maintenance operations are rated as level one, engineers with 10 to 50 maintenance operations are rated as level two, and engineers with more than 50 maintenance operations are rated as level three.

[0098] A group of m experts is formed, and a discussion outline is drawn up to determine the issues that need to be resolved, namely, determine the aviation materials that need to be rationed. At the same time, data on the consumption of each type of aviation material spare parts, the forecast results of the demand for each type of aviation material spare parts, and the historical data of the maintenance work card of each maintenance engineer are prepared and sent to each expert for consultation. Among them, the maintenance work card includes maintenance procedures, tools and material information, and the maintenance engineer applies for aviation materials based on this. Among them, the historical data includes the basic information of the maintenance engineer and the usage of various types of aviation materials.

[0099] After analyzing the historical usage of aviation materials in each maintenance work card, each expert will make reasonable usage recommendations for maintenance engineers of different levels. The opinions of all experts will be summarized and compiled into a comprehensive opinion. The first expert opinions will be summarized and presented in the form of charts for comparison. The comprehensive opinions will then be fed back to the experts separately for another round of consultation, allowing them to review different viewpoints and adjust their judgments.

[0100] After collecting and summarizing the revised opinions, they are distributed to the experts again for a second round of revisions. The above steps are repeated until the opinions of each expert remain stable.

[0101] According to the data provided by the experts, combined with the experience level, classification and comprehensive analysis are carried out to determine the reasonable usage of each aviation material under different conditions. m experts are invited, and the i-th expert believes that the usage quota standard of aviation material A for a maintenance card is , the error range is , then the reasonable dosage range S can be calculated as: .

[0102] In a specific embodiment, before step 1, it also includes collecting and statistically analyzing the consumption data of aviation materials and spare parts.

[0103] Aviation materials are an indispensable foundation in aircraft maintenance. Due to the complexity of aircraft maintenance, the demand and consumption sources of aviation materials are diverse. According to the randomness of consumption, it can be divided into non-routine replacement and routine replacement. Among them, non-routine replacement refers to the demand for aviation material spare parts when faults are found in daily maintenance route work and regular inspection work; among them, routine replacement refers to known planned work performed in routes and regular inspections, such as modification, regular replacement of control components, and other planned tasks of executing airline maintenance plans. According to the source of work, aviation material spare parts consumption can be divided into route replacement and regular inspection replacement. Route replacement involves aviation materials used in daily route maintenance of aircraft. These aviation materials may come from routine work or parts replaced due to faults. Regular inspection replacement refers to aviation materials required in regular inspections, which may be routine or non-routine use. According to the aviation material data used in route work, regular inspection work, non-routine work and routine work in the airline's historical data, the data on the demand for each type of aviation material spare parts are counted as the collected historical aviation material data.

[0104] In Example 2, step S22 is different from the example, and the other steps are consistent with the example. Step S22 uses the grey prediction method to predict the consumption of intermittent aviation materials and spare parts. There are a lot of zero values ​​in intermittent demand, and historical data is scattered and lacks regularity. Therefore, gray prediction is considered to be an effective method for predicting intermittent demand. Gray method prediction is a random time prediction technology and belongs to the system prediction method. The typical model is the GM (1,1) model. Its basic principle is to generate a sequence with obvious trend by accumulation when there is no obvious trend in the time series, and then establish a prediction model based on the growth trend of the sequence, while considering the influence of gray factors. Finally, the inverse operation is performed by the cumulative subtraction method to restore the original time series to obtain the prediction result.

[0105] The time series of aviation maintenance materials has two main characteristics: one is the sporadic nature of demand, with zero demand in some time periods; the other is the variability of demand, with large fluctuations in demand. These two characteristics are consistent with intermittent demand. Since the data contains zero values ​​and the demand data is a non-equally spaced series, the gray prediction method is used to construct a non-equally spaced gray GM (1,1) model for prediction.

[0106] First, the non-equal interval consumption sequence of a certain type of intermittent maintenance aircraft material is converted into an equal interval sequence, and then accumulated and generated. The generated sequence is analyzed to obtain the regular prediction value, and finally restored by the cumulative reduction method to obtain the prediction value of the original sequence, and the prediction is completed when the disgust factor condition is met.

[0107] Get the original non-negative sequence based on the original sequence of aviation material and spare parts consumption ,in .

[0108] Perform accumulation operation on the original non-negative sequence , and the generated sequence is .

[0109] The generated sequence after accumulation and the original sequence satisfy the grey differential equation, that is, the whitened form equation is established: , this is a first-order differential equation model with one variable, so it is recorded as GM (1,1). in, is the development coefficient, is the ash action, .

[0110] The parameters are estimated using the least squares method, and we get: , in, , , Generate Sequence The whitened differential equation that satisfies the GM(1,1) model is: , solve for the development coefficient Amount of ash action , and then bring it into the above whitened differential equation to get the time response function of the GM(1,1) model: , Restored prediction model .

[0111] Then, the accuracy of the model is tested, the residual test is performed, and the average relative error is calculated. , when the mean relative error The model accuracy test is level 1 and is recorded as excellent prediction accuracy, when the average relative error; When the mean relative error The model accuracy test is the second level and is recorded as good prediction accuracy, when the average relative error; When the mean relative error The model accuracy test is graded as level three and the prediction accuracy is qualified, when the average relative error; When the mean relative error If the model accuracy test is level 4, it is recorded as unqualified prediction accuracy.

[0112] In actual testing, the standard deviation of the original sequence and the standard deviation of the residual sequence between the predicted fitting value and the original sequence can also be calculated. The ratio of the standard deviation of the residual sequence between the original sequence and the standard deviation of the original sequence can be used to evaluate the model accuracy C. The small error probability P can also be used to evaluate the model accuracy. The residual is then used to correct the deviation value. After the model passes the accuracy test, if the accuracy is high, the model can be used directly for prediction. If the accuracy is low, the residual model is used for correction to improve the accuracy of the model.

[0113] Finally, the model is used for prediction. The GM (1,1) model is essentially an exponential prediction model established using a linearization method. Therefore, when the coefficient changes exponentially, the prediction accuracy is higher.

[0114] Grey system models are mostly established based on equidistant sequences. Therefore, the premise of using the GM (1,1) model is that the data sequence must meet the requirement of equidistant intervals. In the field of engineering technology, there are often a large number of data fitting and prediction problems of non-equidistant sequences. If the data is incomplete, the data has strong jumps, and the data has different signs, the data is formed according to non-equidistant independent variables, which will lead to the appearance of non-equidistant sequences. Therefore, establishing the GM (1,1) model of non-equidistant sequences has broad practical significance.

[0115] Therefore, for the non-equally spaced data sequences encountered in actual situations, they must first be transformed into equally spaced sequences. After one accumulation generation, they can be approximately fitted into the first-order differential equation, namely the GM (1,1) model.

[0116] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for forecasting demand for aviation spare parts, characterized in that: include: Step S1: Construct a data set of aviation spare parts consumption based on monthly statistics of the historical consumption of each aviation spare part within a set period. The data set of aviation spare parts consumption from the first month to the nth month is used as the original sequence of aviation spare parts consumption, recorded as , a two-dimensional classification method is used to classify the original sequence of aviation material and spare parts consumption into three types of aviation material and spare parts demand: smooth demand, intermittent demand, and random demand according to the square of the coefficient of variation and the set threshold of the average demand interval; among them, the square of the coefficient of variation is recorded as CV², which is the standard deviation of the demand for aviation material and spare parts divided by the square of the mean demand; the average demand interval is recorded as ADI, which is the average time interval between two consecutive demands for aviation material and spare parts; Step S2, using the time series forecasting method to predict the demand for each type of aviation material spare parts according to the type of aviation material spare parts; wherein, the consumption of aviation material spare parts with smooth demand is predicted by the exponential smoothing method in the time series model, the consumption of intermittent aviation material spare parts is predicted by the non-periodic solution model in the time series model, and the consumption of random aviation material spare parts is predicted by the autoregressive moving average model in the time series model.

2. The method for forecasting demand for aviation materials and spare parts according to claim 1, characterized in that: The consumption of aviation materials and spare parts with flat demand is predicted by using the exponential smoothing method in the time series model, which specifically includes: According to the original sequence of aviation material and spare parts consumption, the time series of aviation material and spare parts consumption is obtained as follows: , from the first month to the nth month, the original sequence of aviation material and spare parts consumption is set, and the observation interval is arbitrarily selected from 1 to N intervals, and the aviation material and spare parts consumption at each time is taken as the aviation material and spare parts consumption time series; Smooth the data and calculate the exponential smoothing of the demand for aviation spare parts at time t as follows: ,in, is the smoothing coefficient, whose value is between 0 and 1. is the historical demand value of aviation spare parts at time t, is the predicted value; among them, the initial value , is the initial value; By calculating the MAD indicator under different smoothing coefficients, it is expressed as The smoothing index is determined and the mean square error MSD is calculated, which is expressed as , using planning solution, we can get the optimal value of the planning solution target. When the mean square error MSD is the minimum, there is an independent variable smoothing coefficient The optimal solution of ; When the image of the predicted data is a straight line, the linear smoothing method is no longer applicable, so it is necessary to introduce the quadratic exponential smoothing method for prediction. The prediction model is expressed as: ,in, for The predicted value of aviation materials at the time, T is the predicted time, and is the first smoothing value of period t and t-1, and are the quadratic exponential smoothing values ​​at period t and t-1 respectively.

3. The method for forecasting demand for aviation materials and spare parts according to claim 1, characterized in that: The random aviation material spare parts consumption is predicted by the autoregressive moving average model in the time series model, wherein the autoregressive moving average model is recorded as the ARIMA (p, d, q) model, which specifically includes: The time series of aviation material and spare parts consumption is obtained based on the original series of aviation material and spare parts consumption. The stability of the time series of aviation material and spare parts consumption is tested. The time series data is visualized using SPSS software. The seasonality and stability characteristics are observed based on the time series diagram, autocorrelation function diagram ACF and partial autocorrelation function diagram PACF of the time series. The stability of the data is preliminarily determined, and the time series of aviation materials is divided into a stable time series and a non-stationary time series. The non-stationary time series is stabilized and the difference operation is performed on the non-stationary time series until the processed data passes the stationarity test. The number of difference operations at this time is recorded as the number of differences. In the ARIMA (p, d, q), p is the autoregressive order, d is the number of differences, and q is the sliding average order; The ARIMA (p, d, q) model includes three parts: the autoregressive model AR (p), the single integration I (d) and the moving average model MA (q); AR is the sequence value of the time series, which is a linear combination of the observations lagged p periods; the number of single integration I (d) specifically refers to the number of difference processing required to transform a non-stationary time series into a stationary series after d differences; MA is the model value, which is expressed as: a linear function of the q-order residual term, Calculation of ARIMA(p,d,q) model: ,in, are the parameters of the autoregressive model, where The value cannot be 0. represents the observed value at time t; is a white noise sequence; is the parameter of the moving average model, where the value of q cannot be 0. is the deviation between the predicted value and the sequence value at time t-1, is the deviation between the predicted value and the sequence value at time tq, yes The weight of ARIMA (p, d, q) model, where the calculation of the p-order autoregressive model is: , ARIMA (p, d, q) model, where the q-order moving average model is calculated as: , Get the autoregressive moving average model ARMA(p,q): , in, represents the sequence value at any time t, represents the random error at any time t, represents the linear combination of the first p sequence values ​​at time t, represents a constant, Represents the linear combination of the cumulative sum of white noise of the first q sequences at time t; The BIC criterion is used to determine the model order. The BIC criterion is the Bayesian Information Criterion. , where k is the number of parameters in the model, including the number of autoregressive coefficients, number of differences, and number of moving average coefficients; n is the number of samples; L is the likelihood function; Take an upper limit value of p and q, which is less than 10, and then traverse all p and q combinations, and calculate the BIC values ​​by constructing the BIC matrix. The p and q values ​​with the smallest BIC values ​​are the final p and q. That is, the data with the smallest BIC information amount is selected as the p and q values ​​in the ARIMA (p, d, q) model to determine the ARIMA (p, d, q) model. The ARIMA (p, d, q) model is tested to determine whether the residual sequence is a white noise sequence. The residual is the predicted value minus the true value, and the residual sequence is all predicted value sequences minus the true value sequence. White noise is a special weak stationary process. The residual sequence is tested for white noise. If the residual sequence is white noise, it means that the valid data has been extracted from the original sequence. The ARIMA (p, d, q) model that has passed the test is used for prediction to obtain the predicted random aviation material and spare parts consumption data.

4. The method for forecasting demand for aviation materials and spare parts according to claim 1, characterized in that: The intermittent aviation material and spare parts consumption is predicted by using a non-periodic model in a time series model, wherein the non-periodic model uses an intermittent demand forecasting Croston method or a grey forecasting method to predict the intermittent aviation material and spare parts consumption; Among them, the intermittent demand forecasting Croston method is decomposed into two stages: one is the prediction interval length, and the other is the prediction of the demand in each interval, using exponential smoothing and linear regression models respectively; the grey prediction method adopts the GM (1,1) model and uses the weighted cumulative generating function to process the sequence data.

5. The method for forecasting demand for aviation materials and spare parts according to claim 4, characterized in that: The intermittent demand forecasting Croston method specifically includes: According to the original sequence of aviation material and spare parts consumption, it is converted into demand interval sequence and demand quantity sequence. According to the original sequence of aviation spare parts consumption, the monthly consumption of aviation spare parts is the demand, the intervals between two adjacent non-zero demands constitute the demand interval sequence, and the non-zero demands constitute the demand sequence, and we get: Demand interval sequence ,in , Aircraft spare parts demand sequence ,in ; The demand interval series and the demand series of aviation spare parts are predicted by combining the exponential smoothing method and the linear regression model, which can be expressed as: Demand interval estimation , demand estimate ,in, is the smoothing coefficient, and its value should be between 0 and 1; The final demand forecast value for the next period is: .

6. The method for forecasting demand for aviation materials and spare parts according to claim 4, characterized in that: The GM (1,1) model specifically includes: Get the original non-negative sequence based on the original sequence of aviation material and spare parts consumption ,in ; Perform accumulation operation on the original non-negative sequence , and the generated sequence is ; The generated sequence after accumulation and the original sequence satisfy the grey differential equation, that is, the whitened form equation is established: , this is a first-order differential equation model with one variable, so it is recorded as GM (1,1). in, is the development coefficient, is the ash action, ; The parameters are estimated using the least squares method, and we get: ,in, , Generate Sequence Whitening differential equations satisfying the GM(1,1) model , solve for the development coefficient Amount of ash action , and then substitute it into the above whitened differential equation to obtain the time response function of the GM(1,1) model , the prediction model is restored .

7. The method for forecasting demand for aviation materials and spare parts according to claim 1, characterized in that: The two-dimensional classification method is used to classify the original sequence of aviation material and spare parts consumption into three types of aviation material and spare parts, namely, smooth demand, intermittent demand and random demand, according to the square of the coefficient of variation and the set threshold of the average demand interval. Specifically, the following are included: Calculate the average demand interval ADI and the square of the coefficient of variation of the aircraft material replacement data as CV². The square of the coefficient of variation, CV², is the standard deviation of the demand divided by the square of the mean demand and is expressed as: ,in, , The average demand interval ADI is the average time interval between two consecutive demands for aviation materials and spare parts, expressed as ,in, The quantity of aviation spare parts required. is the time interval between two consecutive aviation spare parts requirements, N is the number of time intervals, is the average demand value; Among them, the threshold CV is set 2 =0.49, ADI=1.32 as the dividing value between different demand types, When ADI < 1.32, CV 2 <0.49, the demand for aviation materials and spare parts is a flat demand; When ADI < 1.32, CV 2 When ≥0.49, the demand for aviation spare parts is random; When ADI≥1.32, the demand for aviation materials and spare parts is intermittent.

8. A demand forecasting system for aviation spare parts, characterized in that: The system applies a method for forecasting demand for aviation materials and spare parts as claimed in any one of claims 1 to 7, and the system specifically comprises a sequentially connected aviation material and spare parts demand classification unit and a maintenance aviation material and spare parts consumption prediction unit; the aviation material and spare parts demand classification unit collects the collected data according to the consumption data of each type of aviation material and spare parts at equal monthly intervals to obtain the original consumption sequence of each type of aviation material and spare parts, and each original consumption sequence of aviation material and spare parts is classified into three types of aviation material and spare parts, namely, flat demand, intermittent demand and random demand, by using a two-dimensional classification method through the Syntetos aviation material and spare parts demand classification judgment standard module according to the square of the coefficient of variation and the set threshold of the average demand interval; The maintenance aviation material and spare parts consumption prediction unit includes three time series modules for respectively predicting the demand for aviation material and spare parts for smoothly demanded aviation material, intermittent demanded aviation material and random demanded aviation material, wherein the consumption of smoothly demanded aviation material and spare parts is predicted by a time series module including an exponential smoothing method, wherein the consumption of intermittent aviation material and spare parts is predicted by a time series module including a solution to a non-periodic model, wherein the consumption of random aviation material and spare parts is predicted by a time series module including an autoregressive moving average model.

9. The demand forecasting system for aviation material spare parts according to claim 8, characterized in that: The time series module for solving the non-periodic model adopts the intermittent demand forecasting Croston module.

10. The demand forecasting system for aviation material and spare parts according to claim 8, characterized in that: The time series module for solving the non-periodic model is implemented by using a grey prediction module.

Images