A multisource heterogeneous data fusion method based on time alignment and federated learning for the dual carbon digital intelligence monitoring center

Due to the complex composition of the "Dual Carbon" digital intelligence monitoring data, the error in using it to analyze carbon emission status is large. Therefore, a multi-source heterogeneous "Dual Carbon" digital intelligence monitoring center based on time registration and federated learning is proposed. First, in terms of "double carbon" historical data accumulation, the composition of static basic data and carbon emission historical data is analyzed; in terms of "double carbon" real-time monitoring data, the composition of IoT sensing data and dynamic business data is analyzed. After adding timestamps to each data type, align the data to a unified timeline, build the "Double Carbon" digital intelligence monitoring center multi-source heterogeneous data high-order tensor in the federated learning framework, and perform Tucker decomposition and iterate to outputs the converged quadratic linear correlation function. In the test results, the method has the smallest difference between the analysis results of carbon emissions in region A and the actual situation. The corresponding error is always within 0.5*107 t, and the maximum error is only 0.46*107 t.