The surface dynamometer diagram of a rod-pumped well is obtained by load sensor, from which oil well working conditions can be analyzed, oil well failures can be diagnosed, and oil production can be calculated. However, such problems as high cost, low popularity and susceptibility to drift and distortion existing with the way to obtain a dynamometer diagram have restricted the development of oil well digital management. Electric parameters are the most basic ones for oil well operation, which are characterized by high popularity, low acquisition cost and stable data. Electric parameters are applied to the dynamometer diagram conversion, analysis and metering, which can take place of the load sensor and realize low-cost and high-efficiency digital management of oil wells. In most cases, the application of electric parameters inversing dynamometer diagram is based on the torque factor method. The torque factor is zero when the polished rod is at the top and bottom dead centers. When this factor is used as a divisor, the calculated load will have no convergence. In this paper, an electric power curve inversion dynamometer diagram approach using big data technologies is proposed: Hadoop technology and Spark technology are utilized to establish a big data platform for electric power curve conversion dynamometer diagram, which collects more than 60,000 site power curve-dynamometer diagram sample data to set up an deep learning sample database. 144 points are selected from the power curve as eigenvalues. After filling the dynamometer diagram, 256 pixels are selected as eigenvalues. So, the eigenvalues are sufficient enough. Deep learning technologies such as Restricted Boltzmann Machine, Sparse AutoEncoder and Softmax Mapping are applied to training sample database, finding out the inter-relation between power curve and dynamometer diagram and obtaining the "power curve-dynamometer diagram conversion model" to realize power curve inversing dynamometer diagram. This technology has been applied in Jilin Oilfield for 350 wells. According to the comparison of a dynamometer diagram converted from power curve with one actually measured, the diagnosis accuracy of working conditions can reach 95.4%, the conformance rate of maximum load is 95%, and the conformance rate of minimum load is 93%. It is shown by field application that using big data technologies the deep learning model is accurate, the algorithm is highly stable, the inversion result is highly precise and the conformance rate is high, which is of great significance to the improvement of digital management level as well as the reduction of production cost. This technology has a promising prospect for popularization.