A Real-time, Robust Control Strategy for Plug-in Hybrid Electric Vehicles

Plug-in hybrid electric vehicles (PHEVs) are hybrid electric vehicles (HEVs) that can be recharged by connecting to the electric grid. PHEVs can operate in all-electric mode or in a blended mode where the battery and the internal combustion engine work simultaneously to propel the vehicle. The state-of-the-art PHEV control uses the charge depleting (CD) - charge sustaining (CS) control strategy, which forces the battery energy to be utilized in priority. However this control approach does not guarantee optimal performance.;Dynamic programming (DP), an offline optimization tool, guarantees global optimal solution to a constrained cost function. Equivalent consumption minimization strategy (ECMS) is a real-time control strategy, which minimizes an equivalent cost function in form of J = m˙fuel + lambda * SO˙C where m˙fuel is fuel consumption rate, lambda is equivalent factor and SO˙C is battery state of charge (SOC) variation rate. For PHEVs lambda is sensitive to trip length since the energy stored in the battery should be used more aggressively on shorter trips, and vice versa.;In this thesis, we propose a novel control method for PHEVs that combines DP and ECMS to find the optimal lambda value given remaining distance dR and current SOC value. Using DP results, lambda is determined; ECMS then calculates the engine and motor operating points based on its cost function. DP results are used as follows: given an optimal trajectory for a known drive cycle and the change in SOC when covering the remaining distance dR, lambda i s obtained at that instance by linear regression of the DP results. Therefore, a look-up table of lambda (referred as lambda map) can be generated with respect to d R and SOC for a given drive cycle. With a lambda map available, and trip length and target SOC known, ECMS can be implemented. Results show that the equivalent cost has an average of 5% gap to DP results over twelve standard drive cycles. This benchmarking exercise showed us that given a known driving pattern, the estimate of lambda was near-optimal.;Next, we generated a universal map by merging the twelve lambda maps into one. We find an increased 14% gap to DP results on average over the twelve drive cycles, but a 30% improvement over CD-CS strategy. For further verification, we simulate 4719 real-world drive cycles, proving controller robustness with an average 9.3% cost reduction compared to CD-CS strategy.;To further improve the performance of the proposed controller, we classify the lambda maps into three patterns (city, suburban, and highway), and introduce an offset in the instantaneous value of lambda based on the driving style. Fuzzy logic controllers are designed for driving pattern and driving style recognition. The twelve standard drive cycles are divided into two groups: six for training and six for test. Simulations show only a 5% gap to DP. Further test on real-world drive cycles shows an average 8% gap to DP results compared to 14% gap by ECMS with the universal lambda map and 35% gap by CD-CS strategy.;This DP-ECMS combined optimization method is not limited to vehicle fuel economy optimization. Other factors such as emissions, can be considered in the cost function as well. More generally, for any optimization problem with multiple resources to meet a demand, a weighting factor can be assigned to each resource in the cost function and optimal weighting factors can be found using DP results.