Optimization method for predicting safety fire-fighting product inventory based on time sequence

Abstract

The invention discloses a time sequence prediction-based safety fire-fighting product inventory optimization method, and relates to an enterprise inventory management method. According to the method, a Hybrid Forecast hybrid prediction model is established. Through comparison with actual data, the mean absolute error of the Hybrid Forecast hybrid prediction model is 7.61%, and the accuracy is high. And finally, based on the accurate prediction of the sales volume, proposing an inventory production plan optimization strategy combining push-and-pull of a semi-finished product Push and push-and-pull of a customer demand finished product Pull. Sales are predicted by means of mathematical tools, a safety stock strategy most suitable for the company is summarized in combination with industry characteristics, market demand habits and market internal and external factors, a better stock level is achieved, a safety stock plan of the enterprise is perfected, production is fixed by sales, production is arranged, products are promoted to complete scientific value conversion, and the economic benefits of the enterprise are improved. Customer satisfaction is improved, and scale economy is realized.

Description

Technical Field

[0001] The invention relates to an enterprise inventory optimization method, in particular to a safety fire protection product inventory optimization method based on time series prediction. Background Art

[0002] Fire, as one of the most common disasters in real life, has seriously affected my country's social and economic development and natural environment.

[0003] As an important tool for monitoring fires, safety fire alarms have an increasing market share due to their ability to effectively prevent fires.

[0004] At present, in the highly competitive market environment, it is necessary to understand the changes in market demand through effective sales forecasting, improve the contradiction between production and sales, achieve the goal of optimizing enterprise inventory, and thus improve the economic benefits of the enterprise. Summary of the invention

[0005] The purpose of the present invention is to provide a method for optimizing the inventory of safety and fire protection products based on time series prediction. The present invention verifies the accuracy of the model by comparing the predicted value with the actual data error, so that the enterprise can achieve more scientific and accurate sales forecasts, reduce the inventory level, and reduce inventory and procurement costs.

[0006] The objective of the present invention is achieved through the following technical solutions:

[0007] A method for optimizing the inventory of safety and fire protection products based on time series prediction, the method comprising the following steps:

[0008] The first step is to use EViews software to determine the stationarity of the data based on the time series diagram and the unit root test (ADF), then preliminarily establish an AR model based on the autocorrelation coefficient (ACF) and the partial autocorrelation coefficient (PACF), and use the white noise method to perform regression analysis on the model to obtain the p value of the regression equation;

[0009] In the second step, according to the seasonal fluctuation of the sequence, the Holt-Winters multiplication model is used to eliminate the seasonal trend, and the data is regressed to obtain the intercept and slope; then the trend and seasonal correction exponential smoothing method is used to find the seasonal factor, and the mean square error is calculated. The nonlinear GRG method is selected by the planning solution method to find the α, β, and γ values ​​that minimize the mean square error as the optimal solution;

[0010] The third step is to use the basic conversion regression as the basis, set the variables of the conversion probability of each district system as a Markov type, generate different models by formulating the district system (a1, a2), and then use the one-step state transition probability to determine the district system in the current period; finally, use the rolling forecast method, and update the latest forecast value into the initial data every month, and delete the previous data, so that the initial data step size is the same each time, and establish 12 models;

[0011] The fourth step is to extract the data of the period months, draw a time series graph, and use the multiplication model;

[0012] Separate the seasonal component, establish a linear equation through regression analysis, multiply it by the corresponding seasonal index, and calculate the predicted value of each month; the prediction model is as follows: t =443358.8+21187.65*t

[0013] Finally, the monthly forecast values ​​of the moving average trend elimination model are substituted into the Markov regime switching model, and then a rolling forecast is performed to the HybridForecast model to obtain the final forecast results.

[0014] The method for optimizing the inventory of safety and fire protection products based on time series prediction, the fourth step adopts a multiplication model to separate the seasonal component, establishes a linear equation through regression analysis, multiplies it by the corresponding seasonal index, and calculates the predicted value for each month.

[0015] The above-mentioned method for optimizing the inventory of safety and fire protection products based on time series prediction, wherein the predicted value of each month is calculated, and the prediction model is: t =443358.8+21187.65*t.

[0016] The method for optimizing the inventory of safety and fire protection products based on time series prediction, wherein the monthly prediction value is for the period from January to April.

[0017] The advantages and effects of the present invention are:

[0018] 1. The present invention uses mathematical tools to make sales forecasts, combines industry characteristics, market demand habits, and internal and external market factors to summarize a safety stock strategy that best suits the company, achieves a better inventory level, improves the company's safety stock plan, determines production based on sales, arranges production, and promotes scientific value conversion of products, improves customer satisfaction, and achieves economies of scale.

[0019] 2. The present invention firstly establishes the ARIMA (11, 1, 0) model and the Holt-Winters model through data processing and analysis, and constructs the HybridForecast model by combining the Markov region switching model and the moving average trend elimination model, and compares and selects the optimal forecast model based on the accuracy of the model. Secondly, based on the Push / Pull production method of the safety inventory strategy, according to the prediction results of the model, a production plan for assembling finished products using a push strategy in the process of raw material procurement and parts production is formulated, and a pull strategy is adopted according to customer demand after sales are generated. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of the present invention;

[0021] Figure 2 It is the ARIMA forecasting model flow chart of the present invention;

[0022] Figure 3 It is the autocorrelation and partial autocorrelation order diagram of the ARIMA forecasting model of the present invention;

[0023] Figure 4 It is the ARIMA forecasting model parameter regression result diagram of the present invention;

[0024] Figure 5 is the residual graph of the ARIMA forecasting model of the present invention;

[0025] Figure 6 is the seasonal factor graph of the Holt-winters prediction model of the present invention;

[0026] Figure 7 It is an equation setting diagram of the Markov zone conversion prediction model of the present invention;

[0027] Figure 8 It is the regression model result of the Markov region switching prediction model of the present invention. DETAILED DESCRIPTION

[0028] In order to better understand the above technical solution, it will be described in more detail below.

[0029] The first step is to use EViews software to determine the stationarity of the data based on the time series diagram and the unit root test (ADF). If the data is not stationary, it is necessary to use methods such as difference and logarithm to make the data stable. Table 1 shows the ADF test of the ARIMA forecasting model.

[0030] Table 1 ADF test of ARIMA forecast model

[0031]

[0032] Then, the AR model was initially established based on the autocorrelation coefficient (ACF) and the partial autocorrelation coefficient (PACF), and the model was regressed using the white noise method to obtain the p value of the regression equation. On the basis that the data is stable non-white noise, an ARMA (p, q) model of appropriate order was selected to fit it and verified, and the optimal ARIMA (11, 1, 0) model was obtained to complete the analysis. The model prediction equation is as follows:

[0033] x t =0.112439x (t-1) -0.040012x (t-2) -0.028064x (t-3) +0.08617x (t-4) -0.071124x (t-5) +0.04453x (t-6) -0.008898x (t-7) +0.018561x (t-8) +0.032309x (t-9) -0.032854x (t-10) +0.131956x (t-11) +0.754987x (t-12) +7103.708

[0034] In the second step, according to the seasonal fluctuation of the series, the Holt-Winters multiplication model is used to eliminate the seasonal variation trend, and the data is regressed to obtain the intercept and slope. Table 2 shows the regression results of the Holt-Winters prediction model.

[0035] Table 2 Holt-winters prediction model regression results

[0036]

[0037] Then, the trend and seasonality modified exponential smoothing method is used to find the seasonal factor. By calculating the mean square error, the planning solution method is used to select the nonlinear GRG method to find the α, β, and γ values ​​that minimize the mean square error as the optimal solution. The final parameter values ​​are shown in Table 3.

[0038] Table 3 Holt-winters prediction model parameter values

[0039]

[0040] Finally, substitute the equation to get the prediction result.

[0041] Smoothing equation:

[0042]

[0043] T t = 0*(L t -L t-1 ) + (1 - 0)*T t-1

[0044]

[0045] Prediction equation:

[0046] X t+n = (L t *nT t )S t+n-12

[0047] In the third step, based on the basic transformation regression, the variables of the regime conversion probability in each region are set to be of the first-order Markov type. By formulating the regimes (a1, a2), different models are generated, and then the regime in the current period is determined using the one-step state transition probability. Since different ways of generating random numbers will lead to different results in repeated operations, the models established each time are not the same. At this time, it can be followed that the P-values of the parameters of each model are significant at the 1% statistical level, that is, the P-value needs to be less than 0.1 to test the significance of the model, and the appropriate model is selected. Then, according to the AIC criterion, the model with the smaller AIC value is selected as the better model.

[0048] Then, the one-step state transition probability is calculated, which is the product of the initial state and the state transition probability. Let the two state regimes be a1 and a2. When the state is 1, a1 < a2; when the state is 2, a1 > a2. The initial state is the state in the current period. The initial state in the first month can be observed through software, and the subsequent initial states are the one-step state transition probabilities of the previous period. The state transition probability matrix is the probability that the sales volume transfers from one state to another. Taking January 2015 as an example. Table 4 shows the state transition probability matrix in January 2015.

[0049] Table 4 State Transition Probability Matrix of the Markov Prediction Model in January 2015

[0050]

[0051] Finally, the rolling prediction method is used. Every time a one-month prediction is made, the latest predicted value is updated into the initial data, and the previous data is deleted to make the step length of the initial data the same each time, and 12 models are established. Table 5 shows the Markov prediction model formulas for January - December 2023.

[0052] Table 5 Markov Prediction Model Formulas

[0053]

[0054]

[0055]

[0056] The fourth step is to extract the data from January to April of 2015-2022 and draw a time series graph. Since sales volume is seasonal and has an upward trend, a multiplication model can be used. Then calculate the moving average and find the seasonal ratio.

[0057] Taking the data as an example, Table 6 shows the seasonal ratio after moving average. Table 6 Seasonal ratio of moving average trend elimination prediction model

[0058]

[0059] Finally, the seasonal component is separated, and a linear equation is established through regression analysis. The corresponding seasonal index is multiplied to calculate the forecast value of each month. The forecast model is as follows: t =443358.8+21187.65*t

[0060] Finally, since the Markov regime switching model is slightly less effective in predicting peak time periods with large fluctuations, in order to further improve the accuracy, the forecast values ​​from February to April of the moving average trend removal model are substituted into the Markov regime switching model, and a rolling forecast is performed again. The final forecast results are obtained by using the HybridForecast model as shown in Table 7 below.

[0061] Table 7 HybridForecast model prediction results

[0062]

[0063] Although mathematical models can provide a decision support system for product demand forecasting, when making the final decision, enterprises need to conduct certain human interventions based on the actual situation of the current enterprise to improve the inventory management plan. At this time, a production strategy based on safety stock Push / Pull can be adopted to better manage inventory.

[0064] Semi-finished product Push production method

[0065] Since the volume, specifications and capital occupancy ratio of raw materials and semi-finished products are small and have strong versatility, push production is adopted in the upstream of the supply chain. By analyzing the customer's order quantity in a certain period of time and establishing a model to predict the sales volume of products in the future months, the company can use this quantity as the basis for production planning and issue a production plan. When the finished product inventory quantity is lower than 25% of the predicted sales volume, the demand for semi-finished products and parts is proposed to the previous process, and production is carried out on demand. Then it is traced back to the demand for raw materials, and the purchasing department can also carry out a new procurement plan.

[0066] Customer demand pull production method

[0067] When the intermediate product is produced, its value-added activities are suspended and it is stored in the warehouse until a customer order is received. When the finished product inventory quantity is higher than 25% of the predicted sales volume, pull production is adopted, customer demand is the guide, and semi-finished products are assembled into finished products. The product delivery time is greatly shortened and the differentiated needs of customers can also be met.

[0068] In this way, the later process can accurately receive the goods from the previous process when needed, and the previous process can accurately provide the goods to the later process when needed, which emphasizes the coordination between the various production processes. Each process can produce the necessary quantity in the necessary time, avoiding inventory backlog caused by blind production. In addition, there are some factors that affect demand forecasting, such as the marketing strategy planned by the enterprise, economic changes in the market, and the actions of competitors, which require enterprises to understand in a timely manner in order to adapt to the market.

Claims

1. A method for optimizing the inventory of safety and fire protection products based on time series prediction, characterized in that: The method comprises the following steps: The first step is to use EViews software to determine the stationarity of the data based on the time series diagram and the unit root test (ADF), then preliminarily establish an AR model based on the autocorrelation coefficient (ACF) and the partial autocorrelation coefficient (PACF), and use the white noise method to perform regression analysis on the model to obtain the p value of the regression equation; In the second step, according to the seasonal fluctuation of the sequence, the Holt-Winters multiplication model is used to eliminate the seasonal trend, and the data is regressed to obtain the intercept and slope; then the trend and seasonal correction exponential smoothing method is used to find the seasonal factor, and the mean square error is calculated. The nonlinear GRG method is selected by the planning solution method to find the α, β, and γ values ​​that minimize the mean square error as the optimal solution; The third step is to use the basic conversion regression as the basis, set the variables of the conversion probability of each district system as a Markov type, generate different models by formulating the district system (a1, a2), and then use the one-step state transition probability to determine the district system in the current period; finally, use the rolling forecast method, and update the latest forecast value into the initial data every month, and delete the previous data, so that the initial data step size is the same each time, and establish 12 models; The fourth step is to extract the data of the period months, draw a time series graph, and use the multiplication model; Separate the seasonal component, establish a linear equation through regression analysis, multiply it by the corresponding seasonal index, and calculate the predicted value of each month; the prediction model is as follows: t =443358.8+21187.65*t Finally, the monthly forecast values ​​of the moving average trend elimination model are substituted into the Markov regime switching model, and a rolling forecast is performed again to the HybridForecast model to obtain the final forecast results.

2. According to claim 1, a method for optimizing the inventory of safety and fire protection products based on time series prediction is characterized in that: The fourth step adopts a multiplication model to separate the seasonal component, establishes a linear equation through regression analysis, multiplies the corresponding seasonal index, and calculates the predicted value of each month.

3. The method for optimizing the inventory of safety and fire protection products based on time series prediction according to claim 1 is characterized in that: The predicted value of each month is calculated, and its prediction model is: t =443358.8+21187.65*t.

4. The method for optimizing the inventory of safety and fire protection products based on time series prediction according to claim 1 is characterized in that: The monthly forecast value for the period mentioned is for the period from January to April.

Images