Research On Sales Forecast Of Commodity Based On WaveNet-LSTM Network

The accuracy of commodity sales forecast is related to the profit of all stakeholders.But the prevalence of shortages and overstocking presents retailers with a dilemma: they need to balance the loss of stock and the cost of keeping safe stocks.In this case,it is very necessary to carry out an in-depth study of commodity sales forecast.By using Data Mining technology,researchers can develop models to predict the sales of commodity,help retailers to dig out the potential business value from a large number of goods sales data.Among them,different from the traditional supervised learning method,the neural network can find the hidden internal rules from the data without a large number of artificial features,so as to carry out effective feature learning,which has a natural advantage in solving the problem of short-term commodity sales forecast.If the input of LSTM neural network model in each time step can contain richer historical information,the prediction performance of the model will also be improved,so it is particularly important to add a feature extractor before the LSTM module.Taking two-layer LSTM model as an example,the first LSTM module is equivalent to a feature extractor,but the instability of cells in the hidden layer will affect the quality of output features to some extent.Based on this,this paper proposes a commodity sales forecasting model based on WaveNet-LSTM network.The model uses the WaveNet network to extract the features on the time series,and then inputs them into the LSTM module for forecasting output.Specifically,features on time series are extracted by stacking multiple layers of dilated causal convolution,due to the translation invariance of convolution,the extracted features of the network are more robust;the combination of local and global features is realized by overlaying the intermediate results of each layer by Skip-connections.Taking these features as the input of LSTM module,so that the input of each time step in the LSTM module can contain abundant local and global information,which is conducive to the effective learning of the model.For the problem of commodity sales forecast,this paper applied the model of commodity sales forecast based on WaveNet-LSTM network to the real data set as anexample,and compared with LSTM and other forecasting models.In this process,we made a detailed visual analysis of the obtained data set,and described the whole process from data processing to experimental verification analysis in detail.The experimental results show that WaveNet network can effectively extract the characteristics in time series,and the prediction accuracy of the model can be improved by combining it with LSTM network.