Constructing a novel power system is an important measure to safeguard national energy security and achieve the goals of carbon peaking and carbon neutrality. However, with the substantial increase in uncertainties on both the power generation and load sides, the traditional balance mode where power generation follows load fluctuations can hardly meet the operational requirements of the novel power system. It is therefore imperative to fully tap into the flexibility potential of various adjustable resources on the load side, so as to address the dilemmas of insufficient regulation capacity and supply-demand imbalance in novel power system. To this end, this paper proposes a quantitative evaluation method for the adjustable potential of industrial users oriented to the novel power system. The improved complete ensemble empirical mode decomposition with adaptive noise is adopted to decompose industrial loads. And then, considering multi-dimensional influencing factors such as temperature, humidity, electricity price and order demand, a dataset for analyzing the adjustable potential of industrial users is constructed. The Newton-Raphson algorithm is used to optimize the parameters of the bidirectional long short-term memory network for each component, combined with the attention mechanism, thus realizing the quantitative evaluation of the adjustable potential of industrial users on the response day. A case study based on actual data verifies that the proposed method can overcome the problems of high discreteness and volatility of industrial user loads, and improve the accuracy of quantitative evaluation of adjustable potential.