Closed-loop dual-network fusion prediction method for household appliance sales recovery amount

Abstract

The invention relates to a household appliance sales recovery closed-loop dual-network fusion prediction method, and the method comprises the following steps: S1, building a closed-loop sales recovery fusion prediction network based on the metering economic relation between the sales volume and the recovery volume of a household appliance product; s2, on the basis of the historical social inventory of the household electrical appliance products, predicting the future social inventory of the household electrical appliance products by adopting a deep learning model; s3, on the basis of the historical sales volume and the historical recovery volume of the household electrical appliance product, predicting the future recovery volume of the household electrical appliance product by adopting a Steiner estimation model; and S4, on the basis of the historical social inventory, the predicted social inventory and the predicted recovery amount of the household electrical appliances, predicting the future sales volume of the household electrical appliances by adopting a Karinner-Milon model. The method provided by the invention can improve the prediction precision of the sales recovery amount of the household electrical appliance products.

Description

Technical Field

[0001] The present invention belongs to the fields of deep learning, econometrics, and sales recovery prediction of household appliances, and relates to a closed-loop dual-network fusion prediction method for sales recovery of household appliances based on a deep learning model and a Stanford-Carnegie Mellon model. Background Art

[0002] The sales and service network of household appliances mainly includes manufacturers, wholesalers, retailers and after-sales service providers. These participants jointly build a complete supply chain, from production to final consumer purchase and use, and then to subsequent maintenance and warranty services; the recycling and reuse network involves the collection, transportation, treatment and reuse of discarded household appliances. This network is usually composed of recycling stations, processing centers and remanufacturing companies. Accurately predicting the sales volume of household appliances can help manufacturers, wholesalers and retailers better formulate production plans, inventory management and marketing strategies, thereby improving resource utilization efficiency and corporate competitiveness; accurately predicting the recycling volume of household appliances can help the government and related companies plan the construction and operation of recycling facilities in advance, and improve the efficiency of waste household appliances treatment and resource reuse.

[0003] Traditional home appliance sales forecasts mostly use time series analysis, which can effectively capture the seasonality and trend in sales data. However, it is highly dependent on data. When the data fluctuates greatly or there are emergencies, the prediction accuracy will decrease. Traditional home appliance recycling forecasts mostly use life cycle models, which can better reflect the entire process from sales to scrapping of products. However, it is difficult to obtain home appliance life data, and the assumption of life distribution may be inaccurate. In addition, it is difficult to share data between the sales service network and the recycling network of home appliance products. Traditional methods often predict sales or recycling volume separately, only considering the trend of historical sales or historical recycling volume, ignoring the factors that affect each other between sales and recycling volume, resulting in reduced sales and recycling forecast accuracy of home appliance products. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a closed-loop dual-network fusion prediction method for home appliance sales and recovery volume, that is, to predict the recovery volume through sales volume, and the recovery volume through closed-loop dual-network fusion prediction of sales volume, which is more in line with the changing environment of the home appliance market.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A closed-loop dual-network fusion prediction method for household appliance sales recovery volume includes the following steps:

[0007] S1: Based on the econometric relationship between the sales volume and recycling volume of household appliances, a closed-loop sales and recycling fusion prediction network is constructed;

[0008] S2: Based on the historical social ownership of home appliances, a deep learning model is used to predict the future social ownership of home appliances;

[0009] S3: Based on the historical sales and recycling volumes of home appliances, the Stanford estimation model is used to predict the future recycling volumes of home appliances.

[0010] S4: Based on the historical social ownership, predicted social ownership and predicted recycling volume of home appliances, the Carnegie Mellon model is used to predict the future sales of home appliances.

[0011] Further, the step S1 specifically includes the following contents:

[0012] S11: Predicting recycling volume through sales volume

[0013] The econometric relationship between sales volume and recycling volume follows the Stanford estimation model, that is, the total recycling volume of the appliance in year i, Q i Equal to the previous annual sales volume of household appliances s t Multiply by the scrap rate f of this batch of household appliances in year i t (i), and sum:

[0014]

[0015] S12: Predicting sales volume by recycling volume

[0016] The econometric relationship between recycling volume and sales volume follows the Carnegie Mellon model, that is, the sales volume in year t+1 s t+1 Subtract the recycling amount Q in year t+1 t+1 Equal to the social holdings x in year t+1 t+1 Subtract the social holdings x in year t t :

[0017] s t+1 -Q t+1 =x t+1 -x t

[0018] S13: Predict social ownership based on deep learning models, and combine the deep learning model with the above-mentioned Stanford-Carnegie Mellon model to form a closed-loop fusion prediction network in which sales volume predicts recycling volume and recycling volume predicts sales volume.

[0019] Further, the step S2 specifically includes the following contents:

[0020] S21: Collect the social ownership of a certain household appliance product in a certain area in the previous T years X T , x t It represents the social ownership of this household appliance product in year t.

[0021] S22: Use the long short-term memory network (LSTM) in the deep learning model to predict the future social ownership of the household appliance. Assume that the output of the LSTM network is y t , the input is X T =[x1,x2,...,x t ], then the expression for LSTM network update is:

[0022]

[0023] in, z and are the input values ​​of the forget gate, input gate, input gate and output gate respectively, y t-1 is the last output value, σ(·) is the sigmoid function, tanh(·) is the hyperbolic tangent function, W f , W i , W and W o are the weights of the forget gate, input gate, input gate and output gate respectively, C t For memory cells, It is Hadamard.

[0024] Further, the step S3 specifically includes the following contents:

[0025] S31: Collect the sales volume of a household appliance product in a certain region in the previous T years, that is, S T =[s1,s2,...,s t ],s t Indicates the sales volume of the home appliance product in year t. Collect the annual recycling volume of the first batch of home appliances s1 sold in T years, that is, R T =[r1,r2,...,r t ].

[0026] S32: Calculate the annual recycling rate u of the household appliance t ,Right now:

[0027]

[0028] Assuming that the proportion of scrapped household appliances entering the formal recycling system is g, the annual scrapping ratio (i.e. life ratio) of the household appliances is l t for:

[0029]

[0030] S33: The distribution f of the appliance life ratio is fitted using Weibull distribution, which can be expressed as:

[0031]

[0032] Among them, β is the shape parameter, η is the scale parameter, and γ is the location parameter. The shape parameter β and scale parameter η of the Weibull distribution can be calculated through the cumulative distribution function and median equation of the Weibull distribution. The location parameter γ is generally set to 0. Then the cumulative distribution function F(t) and median equation m of the Weibull distribution can be expressed as:

[0033]

[0034] Among them, F(t) and m can be obtained from the life ratio l t It can be concluded that F(t) is the sum of all life expectancy ratios in the first t years, and m is the maximum value of the life expectancy ratio in the first t years. By combining the above two equations, we can obtain the parameters β and η of the Weibull distribution.

[0035] S34: Use the Stanford Estimation Model (SEM) to calculate the total recycling volume Q of the household appliance in year i i The Stanford estimation model is the total recycling volume Q of the household appliance in the i-th year i Equal to the previous annual sales volume of household appliances s t Multiply by the scrap rate f of this batch of household appliances in year i t (i), and sum:

[0036]

[0037] Further, the step S4 specifically includes the following contents:

[0038] Based on the social ownership of the household appliance in year t x t 、The social ownership of the household appliance in year t+1 x t+1 and the amount of social holdings withdrawn in year t+1, assuming that the amount of social holdings withdrawn is equal to the amount of recycling, i.e., Q t+1 , using the Carnegie Mellon model to predict the sales volume of the appliance in year t+1 t+1 , which can be expressed as:

[0039] s t+1 -Q t+1 =x t+1 -x t

[0040] The Carnegie Mellon model is actually a stock-flow model based on the principle of mass conservation. The stock change is the difference between the inflow and outflow of household appliances. t+1 Subtract the recycling amount Q in year t+1 t+1is the amount of new products added to the market in year t+1, that is, the social ownership in year t+1 minus the social ownership in year t. Therefore, the sales volume of this home appliance in year t+1 is s t+1 It can be expressed as:

[0041] s t+1 =Q t+1 +x t+1 -x t

[0042] Beneficial effects of the present invention:

[0043] The closed-loop dual-network fusion forecasting method, which predicts recycling volume by sales volume and sales volume by recycling volume, is more in line with the changing environment of the home appliance market. Compared with a single deep learning model, it can accurately capture the trend of the inflow and outflow of home appliance products; compared with traditional econometric models such as time series analysis and life cycle models, it can obtain higher prediction accuracy and can adapt to the rapidly changing home appliance sales and recycling market.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] 1. High prediction accuracy: The present invention adopts a method that combines machine learning and econometric models. Based on the historical sales volume and historical recycling volume of home appliances, the Stanford model is used to predict the future recycling volume of home appliances; based on the historical social holdings of home appliances, the LSTM model is used to predict the future social holdings of home appliances; based on the historical social holdings of home appliances, the predicted social holdings and the predicted recycling volume, the Carnegie Mellon model is used to predict the sales volume of home appliances, thereby improving the prediction accuracy of the sales volume and recycling volume of home appliances.

[0046] 2. Strong adaptability: The present invention adopts a method that combines machine learning and econometric models. The machine learning model can quickly respond to market changes and external shocks. By combining the econometric model, it can make stable predictions of market fluctuations and long-term trends, thereby providing flexibility and adaptability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A step diagram of the sales recovery prediction method of the present invention;

[0048] Figure 2 This is a performance improvement diagram of the sales recovery prediction method according to an embodiment of the present invention compared with a control method. DETAILED DESCRIPTION

[0049] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] like Figure 1As shown, the present invention relates to a closed-loop dual-network fusion prediction method for household appliance sales recovery volume, comprising the following steps:

[0051] Step S1: Construct a closed-loop sales and recovery fusion prediction network

[0052] S11: Predicting recycling volume through sales volume

[0053] The econometric relationship between sales volume and recycling volume follows the Stanford estimation model, that is, the total recycling volume of the appliance in year i, Q i Equal to the previous annual sales volume of household appliances s t Multiply by the scrap rate f of this batch of household appliances in year i t (i), and sum:

[0054]

[0055] S12: Predicting sales volume by recycling volume

[0056] The econometric relationship between recycling and sales follows the Carnegie Mellon model, that is, sales in year t+1 minus recycling in year t+1 equals social holdings in year t+1 minus social holdings in year t:

[0057] s t+1 -Q t+1 =x t+1 -x t

[0058] S13: Predict social ownership based on deep learning models. Through deep learning models and the Stanford-Carnegie Mellon model, a closed-loop fusion prediction network is formed in which sales volume predicts recycling volume and recycling volume predicts sales volume.

[0059] Step S2: Predict the social ownership of home appliances

[0060] S21: Collect the social ownership of a certain household appliance product in a certain area in the previous T years X T , x t It represents the social ownership of this household appliance product in year t.

[0061] S22: Use LSTM to predict the social ownership of the appliance y t Assume that the output of the LSTM network is y t , the input is X T =[x1,x2,...,x t ], then the expression for LSTM network update is:

[0062]

[0063] in, z and are the input values ​​of the forget gate, input gate, input gate and output gate respectively, y t-1 is the last output value, σ(·) is the sigmoid function, tanh(·) is the hyperbolic tangent function, W f , W i , W and W o are the weights of the forget gate, input gate, input gate and output gate respectively, C t For memory cells, It is Hadamard.

[0064] Step S3: Predict the recycling volume of home appliances

[0065] S31: Collect the sales volume of a household appliance product in a certain region in the previous T years, that is, S T =[s1,s2,...,s t ],s t Indicates the sales volume of the home appliance product in year t. Collect the annual recycling volume of the first batch of home appliances s1 sold in T years, that is, R T =[r1,r2,...,r t ].

[0066] S32: Calculate the annual recycling rate u of the household appliance t ,Right now:

[0067]

[0068] Assuming that the proportion of scrapped household appliances entering the formal recycling system is g (for simplicity, g is set to 1 in the embodiment, that is, all scrapped household appliances enter the formal recycling system), the annual scrapping ratio (i.e., life ratio) of the household appliances is l t for:

[0069]

[0070] S33: The distribution f of the appliance life ratio is fitted using Weibull distribution, which can be expressed as:

[0071]

[0072] Among them, β is the shape parameter, η is the scale parameter, and γ is the location parameter. The shape parameter β and scale parameter η of the Weibull distribution can be calculated through the cumulative distribution function and median equation of the Weibull distribution. The location parameter γ is generally set to 0. Then the cumulative distribution function F(t) and median equation m of the Weibull distribution can be expressed as:

[0073]

[0074] Among them, F(t) and m can be obtained from the life ratio lt It can be concluded that F(t) is the sum of all life expectancy ratios in the first t years, and m is the maximum value of the life expectancy ratio in the first t years. By combining the above two equations, we can obtain the parameters β and η of the Weibull distribution.

[0075] S34: Use the Stanford Estimation Model (SEM) to calculate the total recycling volume Q of the household appliance in year i i , that is, the total recycling volume of the household appliance in year i, Q i Equal to the previous annual sales volume of household appliances s t Multiply by the scrap rate f of this batch of household appliances in year i t (i)

[0076]

[0077] Step S4: Predict the sales volume of home appliances

[0078] Based on the social ownership of the household appliance in year t x t 、The social ownership of the household appliance in year t+1 x t+1 and the amount of stocks withdrawn from society in year t+1 (assuming that the amount of stocks withdrawn from society is equal to the amount of stocks recycled, i.e., Q t+1 ), and use the Carnegie Mellon model to predict the sales volume s of the appliance in year t+1 t+1 , which can be expressed as:

[0079] s t+1 =x t+1 +Q t+1 -x t

[0080] Algorithm performance comparison: The prediction accuracy of the present invention's home appliance sales and recycling closed-loop dual-network fusion prediction model based on deep learning and Stanford-Carnegie Mellon (LSTM-SEM-CMM) is compared with that of the LSTM network, GRU network, TCN network, and ARIMA network, as follows:

[0081] 1) Split the collected historical social ownership data, historical sales data and historical recycling data, of which 70% is used for the training set and 30% is used for the test set.

[0082] 2) The training set is imported into the home appliance sales and recycling prediction model based on deep learning and Stanford-Carnegie Mellon of the present invention to predict the sales volume and recycling volume data, and then the sales volume and recycling volume of the test set are compared to calculate the prediction accuracy.

[0083] 3) Import the training set into the LSTM network, GRU (Gate Recurrent Unit) network, TCN (Temporal Convolutional Network) network, and ARIMA (Autoregressive Integrated Moving Average) network to predict the sales volume and recovery volume data respectively, and then compare them with the sales volume and recovery volume of the test set to calculate the prediction accuracy.

[0084] 4) Comparison results of prediction accuracy are as follows Figure 2 It can be seen that the prediction accuracy of the sales volume and recycling volume of household appliances by the method of the present invention is significantly higher than that of the comparison method.

[0085] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed by the present invention, and these modifications or substitutions should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A closed-loop dual-network fusion prediction method for home appliance sales recovery volume, characterized in that: The following steps are involved: S1: Based on the econometric relationship between the sales volume and recycling volume of household appliances, a closed-loop sales and recycling fusion prediction network is constructed; S2: Based on the historical social ownership of home appliances, a deep learning model is used to predict the future social ownership of home appliances; S3: Based on the historical sales and recycling volumes of home appliances, the Stanford estimation model is used to predict the future recycling volumes of home appliances. S4: Based on the historical social ownership, predicted social ownership and predicted recycling volume of home appliances, the Carnegie Mellon model is used to predict the future sales of home appliances.

2. A closed-loop dual-network fusion prediction method for home appliance sales recovery volume according to claim 1, characterized in that: The step S1 specifically includes the following steps: S11: Predicting recycling volume through sales volume The econometric relationship between sales volume and recycling volume follows the Stanford estimation model, that is, the total recycling volume of the appliance in year i, Q i Equal to the previous annual sales volume of household appliances s t Multiply by the scrap rate f of this batch of household appliances in year i t (i), and sum: S12: Predicting sales volume by recycling volume The econometric relationship between recycling volume and sales volume follows the Carnegie Mellon model, that is, the sales volume in year t+1 s t+1 Subtract the recycling amount Q in year t+1 t+1 Equal to the social holdings x in year t+1 t+1 Subtract the social holdings x in year t t : s t+1 -Q t+1 =x t+1 -x t S13: Predict social ownership based on deep learning models, and combine the deep learning model with the above-mentioned Stanford-Carnegie Mellon model to form a closed-loop fusion prediction network in which sales volume predicts recycling volume and recycling volume predicts sales volume.

3. A closed-loop dual-network fusion prediction method for home appliance sales recovery volume according to claim 1, characterized in that: The step S2 specifically includes the following steps: S21: Collect the social ownership of a certain household appliance product in a certain area in the previous T years X T , x t It represents the social ownership of the household appliance product in year t. S22: Use the long short-term memory network (LSTM) in the deep learning model to predict the social ownership of the appliance y t ; Assume that the output of the LSTM network is y t , the input is X T =[x1,x2,...,x t ], then the expression for LSTM network update is: z=tanh(W·[y t-1 ,x t ]), in, z and are the input values ​​of the forget gate, input gate, input gate and output gate respectively, y t-1 is the last output value, σ(·) is the sigmoid function, tanh(·) is the hyperbolic tangent function, W f , W i , W and W o are the weights of the forget gate, input gate, input gate and output gate respectively, C t For memory cells, It is Hadamard.

4. A closed-loop dual-network fusion prediction method for home appliance sales recovery volume according to claim 1, characterized in that: The step S3 specifically includes the following steps: S31: Collect the sales volume of a household appliance product in a certain region in the previous T years, that is, S T =[s1,s2,...,s t ],s t represents the sales volume of the household appliance product in the tth year; collects the annual recycling volume of the first batch of household appliances s1 sold in T years, that is, R T =[r1,r2,...,r t ]; S32: Calculate the annual recycling rate u of the household appliance t ,Right now: Assuming that the ratio of scrapped home appliances to those entering the formal recycling system is g, the annual life ratio of the home appliance is l t for: S33: Use Weibull distribution to fit the distribution f of the appliance life ratio, expressed as: Among them, β is the shape parameter, η is the scale parameter, and γ is the location parameter. The shape parameter β and scale parameter η of the Weibull distribution are calculated through the cumulative distribution function and median equation of the Weibull distribution, and the location parameter γ is set to 0; then the cumulative distribution function F(t) and median equation m of the Weibull distribution are expressed as: Among them, F(t) and m can be obtained from the life ratio l t It can be concluded that F(t) is the sum of all life expectancy ratios in the first t years, and m is the maximum value of the life expectancy ratio in the first t years; the above two equations can be combined to obtain the parameters β and η of the Weibull distribution; S34: Use the Stanford estimation model to calculate the total recycling volume Q of the household appliance in year i i , that is, the total recycling volume of the household appliance in year i, Q i Equal to the previous annual sales volume of household appliances s t Multiply by the scrap rate f of this batch of household appliances in year i t (i), and sum:

5. A closed-loop dual-network fusion prediction method for home appliance sales recovery volume according to claim 1, characterized in that: The step S4 specifically includes the following steps: Based on the social ownership of the household appliance in year t x t 、The social ownership of the household appliance in year t+1 x t+1 and the amount of social holdings withdrawn in year t+1, assuming that the amount of social holdings withdrawn is equal to the amount of recycling, i.e., Q t+1 , using the Carnegie Mellon model to predict the sales volume of the appliance in year t+1 t+1 ,as follows: s t+1 =x t+1 +Q t+1 -x t 。

Images