the interference of loading and unloading vibration; second, a Butterworth low-pass filter (cut-off frequency 50Hz) is used to filter the high-frequency vibration of the engine and tires, focusing on the 0-50Hz low-frequency band. The root mean square (RMS) of vibration acceleration is used as the core index, supplemented by peak factor and kurtosis parameters, to construct a multicharacteristic rating system and improve the accuracy of roughness evaluation. 3.6 Weather Factor Interface Weather indirectly changes the fuel consumption by affecting road conditions and engine environment, so a weather factor collection module is added. By building a weather station in the mining area or establishing a data interface with weather service, the weather data such as rain/snow or fog, ambient temperature, humidity and wind speed are collected. The weather is classified into different impact levels to provide a basis for fair energy efficiency under different weather conditions. 4. CONSTRUCTION OF ENERGY EFFICIENCY MODEL BASED ON MULTI-SOURCE DATA 4.1 Selection Idea of Energy Efficiency Model During the mine transportation operation, fuel consumption is affected by multiple factors such as mining trucks operation status, load change, road conditions, environmental factors and driving behavior. There is a significant non-linear coupling relationship between the influencing factors, and it also has obvious time-series characteristics. Traditional linear regression models or static machine learning models are often difficult to fully describe the dynamic law of fuel consumption with the evolution of working conditions in such complex scenarios. Therefore, on the basis of comprehensively comparing common modeling methods such as multiple linear regression and support vector regression, this paper selects the deep learning model as the core modeling tool for energy efficiency, and simplifies the model selection process into a background explanation of method comparison. Considering the continuity and dependence of fuel consumption data in the time dimension, this paper finally chooses the Long Short-Term Memory (LSTM) network to construct the energy efficiency model. By introducing a gating mechanism, LSTM effectively alleviates the gradient disappearance problem of traditional recurrent neural networks in long-sequence modeling, and can capture both short-term fluctuation characteristics and long-term change trends, which is suitable for the joint modeling needs of multi-source time-series data in the mine transportation process. While ensuring strong non-linear fitting ability, the model also has good stability and scalability, providing a reliable model basis for subsequent energy efficiency analysis. 4.2 Model Input and Sample Organization In terms of model structure design, this study employs the LSTM network as the main structure of the energy efficiency model. The model input is a time-series feature vector after multisource fusion, covering multi-dimensional information such as mining trucks operation status, load and road condition information, and environment and operation conditions, which are uniformly used as the time-series input of the model. The model input is a time-series feature vector after multi-source fusion, which includes 26 core dimensions in total:
RkJQdWJsaXNoZXIy MTM0Mzk2