Improved Deep Learning Model Based on Self-Paced Learning for Multiscale Short-Term Electricity Load Forecasting

Electricity loads are basic and important information for power generation facilities and traders, especially in terms of production plans, daily operations, unit commitments, and economic dispatches. Short-term load forecasting (STLF), which predicts power loads for a few days, plays a vital role in the reliable, safe, and efficient operation of a power system. Currently, two main challenges are faced by existing STLF prediction models. The first involves how to fuse multiscale electricity load data to obtain a high-performance model and remove data noise after integration. The second involves how to improve the local optimal solution despite the sample quality problem. To address the above issues, this paper proposes a multiscale electricity load data fusion- and STLF-based short time series prediction model built on a sparse deep autoencoder and self-paced learning (SPL). A sparse deep autoencoder was used to solve the multiscale data fusion problem with data noise. Furthermore, SPL was utilized to solve the local optimal solution problem. The experimental results showed that our model was better than the existing STLF prediction models by more than 15.89% in terms of the mean squared error (MSE) indicator.

Intelligent alarm data noise reduction for power systems based on singular value decomposition

The noise types of power system intelligent alarm data are complex. When reducing the intelligent alarm data, the profile noise statistics of the noise data are large, resulting in the actual noise reduction value is too small. To solve this problem, a power system intelligent alarm data noise reduction method based on singular value decomposition is designed. The selected normalized decomposition matrix iteratively processes the original matrix, the singular value decomposes the power system alarm data, sets an estimation quantity within the paradigm of the alarm data, controls the noise profile noise statistics, characterizes the noise alarm data structure, uses the SC algorithm to process the cluster basis vectors in the noise data structure, and constructs a repeated iterative convergence process to realize intelligent data noise reduction processing. The original alarm data within a known power system is used as test data, the power system alarm window is set, and the power system alarm data singular values are circled. The data mining-based alarm data noise reduction method, the regularized filter-based alarm data noise reduction method and the designed data noise reduction method are applied to the noise reduction process, and the results show that the designed data noise reduction method has the largest noise value and the best noise reduction effect.

MODEL-BASED CORRECTION FOR DIP AND SHOULDER BED EFFECTS ON LWD PROPAGATION DIELECTRIC CONSTANT AND RESISTIVITY LOGS

Electromagnetic propagation logging has been primarily used to measure formation resistivity. A sensitivity study shows that the formation dielectric constant becomes detectable when it is larger than 10 for the typical LWD 2 MHz propagation measurements. At this frequency, in some field cases, the dielectric constant can be tens to hundreds, making it measurable directly from propagation data. Factors causing this high dielectric constant include connate water volume, the interfacial polarization due to clays, and the Maxwell-Wagner effect as a result of coexistence of conducting and insulating materials. Current methods for determining dielectric constant are based on a homogeneous assumption for the formation regardless of its actual complexity. These methods give reasonable results in thick beds (larger than 10 ft) and low-resistivity-contrast (less than 5) formations. In thinner beds with larger resistivity contrast, both resistivity and dielectric constant logs can be adversely affected by the strong shoulder bed effect. The dip effect in dipping formations can only exacerbate the situation. Like resistivity, dielectric constant can also be anisotropic, but the anisotropy will be ignored here. These effects must be corrected to mitigate undesirable results in the quantitative use of resistivity and dielectric constant logs. The goal of this paper is to address this problem by incorporating the layered structure in the formation model to correct for the shoulder bed and dip effects on resistivity and dielectric constant logs. To account for these effects, we first take advantage of our previous work (Wang et al., 2019) on induction dielectric processing for its fast convergence, robustness to data noise, and weak dependence on initial formation model. The regularization popular for feature selection problems in supervised learning is then added to the existing method. An attractive feature of this regularization is its unique noise-suppression capability while being able to preserve bed boundaries on resistivity and dielectric constant logs. Numerical experiments with synthetic data demonstrate the clear benefits of the new processing in comparison to current methods that have been popular for dielectric constant processing. Field testing also confirms that this data processing is superior to the current methods as far as dip and shoulder bed effects are concerned. Both synthetic and field results indicate that this advanced data processing should be run preferentially as long as the relative dip is not extremely high (less than 70 deg). Some questions with practical importance are addressed in detail that provides insight into the characteristics and performance of the processing. These questions include the effects of data noise, inaccurate dip input, and drilling fluid invasion. In addition, the depth of investigation and vertical resolution are studied for resistivity and dielectric constant to enable a quantitative comparison with other logs. The limitations of the processing and guidelines are also discussed before field applications.

PRESTACK SEISMIC INVERSION WITH STRUCTURAL CONSTRAINTS

Pre-stack seismic inversion usually suffers from the lower signal-to-noise ratio, which could result in unstable inversion results. The conventional multi-trace lateral constrained inversion blurs the steeply dipping layers, whereas the simple structural constrained inversion is affected by noise. To solve this issue, an inversion method with multiple constraints is proposed, which include 1) A local smoothing operator is used to suppress the inversion anomalies caused by data noise, 2) a difference operator is used to protect the stratum boundary, 3) a structural dipping constraint is used to enhance the characterization of the possible dipping stratum. The multi-constraint inversion method suppresses the inversion anomalies caused by data noise without blurring the stratum boundary. The effects of different constraints in the inversion process and the influence of noise on the inversion results are analyzed. In multi-constraint inversion, the regularization coefficient of each constraint operator is dynamically changed, thereby controlling the significance of each regularization term in the inversion. The proposed algorithm is tested on synthetic and field data, which demonstrates its effectiveness and improved accuracy on the inversion results.