Power system state estimation is vulnerable to stealthy false data injection attack (FDIA) that bypasses conventional bad data detectors. In this paper, an improved FDIA detection approach has been proposed using a phasor measurement unit (PMU) assisted linear power system state estimation scheme. The proposed detection approach tracks the changes of complex PMU measurements between the current time instant of the present-day and one step previous time instant of the previous day. This variation of complex PMU measurement is then compared with the variation of forecasted measurements. Manhattan distance has been applied to calculate the distance between the distribution of two different measurement variations. In the event of an FDIA, the Manhattan distance will increase significantly from normal conditions. The proposed approach has been validated on two IEEE benchmark test systems. The produced results clearly depict the efficacy of the proposed approach.