LINEAR LAV-BASED STATE ESTIMATION INTEGRATING HYBRID SCADA/PMU MEASUREMENTS

Reliable and secure grid operation depends on accurate monitoring of power system conditions, such as bus voltage magnitudes and phase angles. While Phasor Measurement Units (PMUs) give high-speed synchronized data at a much higher deployment cost, traditi onal Supervisory Control and Data Acquisition (SCADA) systems offer wide-area coverage but suffer from sluggish update rates and asynchronous measurements. In order to take use of both measuring methods' complimentary qualitiesThis project offers a framewo rk for state estimation based on Linear Least Absolute Value (LAV) that integrates hybrid SCADA/PMU observations. State estimates can be significantly distorted by conventional Weighted Least Squares (WLS) estimators, which are computationally efficient bu t extremely vulnerable to massive measurement errors and poor data. By minimizing the sum of absolute residuals, the Linear LAV estimator, on the other hand, provides better resilience against corrupted data and naturally suppresses the impact of outliers. Using ESP32 microcontrollers with voltage and current sensors to simulate widespread SCADA and PMU measurement nodes, the suggested system is realized as an inexpensive Internet of Things prototype. A Firebase Realtime Database, which acts as a cloud-base d communication bridge, receives sensor data over Wi-Fi. A Python backend uses the cvxpy package to solve the LAV optimization issue, get real-time data, and build the system matrix for a linearized DC power flow model. A Streamlit dashboard provides real-time visualization of estimated states, measurement residuals, and bad-data alarms. The Linear LAV estimator outperforms the traditional WLS technique by achieving accurate state estimation under both normal and noisy settings while successfully rejecting injected incorrect data, according to experimental results on a three-bus test network. The system demonstrates excellent communication reliability (98.9% success rate), low computational overhead (~215 ms per cycle), and effective bad-data identification (96% rate).