A New Depth Prediction Technique Based on Particle Swarm Optimization Algorithm with Multiple Main Controlling Factors Constraints
Zhimeng Gu, Haojie Liu, Min Gong, Yi Han · International Petroleum Technology Conference · 2025
Abstract In the exploration and development of certain oilfields, the problem of low accuracy in depth prediction often arises. The target strata have fast lateral velocity changes, low well control level, and many factors affecting velocity changes, such as structural compaction, velocity body anomalies, differences in seismic data (time and depth domain errors), and velocity field information. Conventional depth prediction methods often consider a single factor, leading to significant prediction errors and low accuracy, which cannot meet the requirements of fine development of oil fields. In order to improve the accuracy of depth prediction and make comprehensive use of various influencing factors, this paper proposes a depth prediction technology based on particle swarm optimization algorithm (PSO-BP), which is constrained by many main controlling factors, effectively solves the limitations of traditional methods, and is more in line with the essence of nonlinear changes of seismic wave velocity affected by many factors. Initially, the multifarious principal controlling factors that influence the lateral velocity variations are identified, encompassing changes in structural trends, shallow velocity anomalies, discrepancies among various seismic datasets, and velocity field information, with these influencing factors being quantified. Subsequently, by statistical analysis of the time-depth relationship of the target strata, a depth prediction background model constrained by structural trends is established using the Convolutional Neural Network algorithm. Thereafter, the relationships between various principal controlling factors and depth are analyzed. Employing the Particle Swarm Optimization algorithm, which adopts a global optimization scheme to circumvent the entrapment of prediction results in local optima, the search for optimized neural network weights and bias values is conducted, thereby enhancing the prediction accuracy of the Back Propagation neural network (BP). The multitude of principal controlling factors are integrated as influencing factors into the depth prediction background model constrained by structural trends, culminating in the establishment of a depth prediction model constrained by multiple principal controlling factors, achieving high-precision depth prediction. Upon analysis, it is posited that the principal controlling factors governing the lateral velocity variations in the B oil field of Bohai Sea include: ① lateral changes in the structural trend of the target strata; ② the localized development of low-velocity anomalies in shallow layers; ③ inconsistencies in the alignment between time-domain data and depth-domain data with actual drilling results at different structural positions, indicating that in some locations, time-domain data correlate more closely, while in others, depth-domain data exhibit better consistency; ④ velocity field data indicating lateral variations in velocity; and ⑤ the amalgamation of other influencing factors as a single principal controlling factor. Based on these five principal factors, a high-precision depth prediction model is constructed. As depicted in Figure 2, the error range for depth predictions made using three conventional methods based on different seismic data spans from -16.8m to 17m, with an error magnitude of 33.8m. In contrast, the depth prediction error range according to the methodology presented in this paper spans from -1.1m to 1.6m, with an error magnitude of 2.8m. This denotes a more than ten times increase in prediction accuracy compared to conventional methods, with a smaller and more stable error fluctuation range.