The implementation of non-intrusive load monitoring for industrial users contributes to power demand-side management and enhances energy utilization efficiency. To address the scarcity of sample data in industrial scenarios and the limitation of existing methods that don’t account for industrial production process constraints

a non-intrusive industrial load decomposition method is proposed that considers the load power characteristics and timing correlation. Firstly

based on the time-varying characteristics of load operating power

the load is categorized into three types: switching load

multi-state load

and continuously varying load. Secondly

since the first two types of loads have stable power states

the integer programming method is used to establish the load power model by considering both active and reactive power characteristics. Thirdly

by extracting the actual production cycle of the enterprise and considering that the power of continuously varying loads has a continuous consumption range

the matrix factorization method is used to establish a load power model. Finally

considering the effects of base vector grouping and the constraints of load timing correlation under production process constraints

load decomposition is achieved by combining and solving the two models. The algorithm is validated using public datasets and measured private datasets

and the results indicate that the proposed method has higher accuracy and better engineering practicality compared to existing methods.