An integrated methodology for assessing short-term and long-term driving risks of heavy-duty trucks considering vehicle load

Despite increasing use of radar-based high-dimensional data for heavy-duty trucks (HDTs) risk assessment, such data often lack vehicle load. This study addresses the critical need for real-time and trip-level driving risk assessments in HDTs, focusing on the undercharacterized relationships between short-term driving risk (SDR), which reflects real-time fluctuations, long-term driving risk (LDR), which captures trip-level patterns, and load statuses. Using 23.6 million HDT operational records, the study employed a rolling time window approach to extract SDR features from speed, and to derive LDR features via discrete wavelet transform (DWT) combined with statistical metrics from accelerate and jerk. Findings classify SDR into four and LDR into three categories, with risk proportions varying (up to 57.6% for SDR, 69.4% for LDR) across load statuses. The load–risk relationship is non-monotonic: Lowest SDR and LDR occur at 75% of rated load capacity. SDR peaks at 50% of rated load capacity, while LDR increases at 25%, 100%, and over 100% load. Load primarily influences SDR through speed volatility, and LDR through high-frequency wavelet components and wavelet energy entropy. Feature importance analysis confirms the efficacy of extracted features, with static vehicle attributes and temporal factors also contributing to risk. This methodology quantifies HDT driving risks, which can be applied to inform, e.g., risk-based expressway reservations.