SNN Resistivity Recovery for Water-Flooded Layer Interpretation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional well logging interpretation models struggle to accurately interpret water-flooded layers due to their inability to adapt to reservoir heterogeneity and nonlinear characteristics during water flooding, leading to ambiguity in resistivity interpretation and low robustness to small sample and high noise data, especially in the presence of 'low-resistance oil layers' and 'high-resistance water layers'.
Innovation Solution
A system utilizing a spiking neural network (SNN) with a Leaky Integrate-and-Fire neuron model, combined with multi-modal well logging data preprocessing and dynamic feature extraction, to construct a resistivity recovering model that accurately captures spatiotemporal changes in resistivity, enhancing interpretation accuracy and reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional well logging interpretation models are used, then the interpretation process is simple, but the accuracy of resistivity interpretation deteriorates due to inability to adapt to reservoir heterogeneity and nonlinear characteristics
Solution Approach 1:
The patent applies dynamic neural network models that can adapt to changing reservoir conditions during water flooding. The model dynamically adjusts its parameters and structure to capture temporal evolution of resistivity, enabling accurate interpretation of dynamic water-flooded layer characteristics rather than relying on static traditional models
Solution Approach 2:
The patent transforms the interpretation approach by changing from fixed parameter models to adaptive parameter models. The neural network learns optimal parameters from training data and can adjust parameters dynamically to reflect actual reservoir conditions, including handling of low-resistance oil layers and high-resistance water layers
2Reliability
If static training mode is used, then the model training is fast, but the robustness to small sample and high noise data deteriorates
Solution Approach 1:
The patent implements comprehensive data preprocessing before model training, including noise filtering, outlier detection, and data normalization. This preliminary action cleans the data and prepares it optimally for training, enabling the model to achieve high robustness with smaller sample sizes and reduced training time
Solution Approach 2:
The patent employs iterative training with feedback mechanisms where the model performance is continuously evaluated and adjusted. The training process incorporates feedback from validation sets and uses techniques like early stopping and learning rate adjustment to optimize training efficiency and model robustness simultaneously
3Measurement precision
If traditional models are used for complex geological phenomena, then the model structure is simple, but the ability to capture temporal correlation characteristics deteriorates
Solution Approach 1:
The patent employs dynamic neural network architectures that explicitly model temporal correlations in resistivity data during water flooding. The model captures how resistivity evolves over time and space, enabling accurate identification of water-flooded layers and their dynamic characteristics rather than treating each measurement independently
Solution Approach 2:
The patent extends the interpretation from static single-point analysis to dynamic spatiotemporal analysis. By incorporating time as an additional dimension and using distributed neural network structures, the model captures the evolution of water saturation and resistivity patterns across multiple time steps and spatial locations
Data Source
AI summary
A system for intelligently recovering water-flooded layer of oil reservoir based on spiking neural network is provided. The system includes: a data acquiring module, a data preprocessing module, a model training module, a resistivity recovering module, and a water-flooded layer interpreting module. The data acquiring module is configured to acquire multi-modal data such as conventional and electrical well logging curves. The data preprocessing module is configured to perform range normalization and Z-Score standardization on the data. The model training module is configured to construct a resistivity recovering model based on a spiking neural network, and the resistivity recovering model is improved and verified by various manners. The resistivity recovering module is configured to recover an original resistivity. The water-flooded layer interpreting module is configured to calculate correlative parameters based on the original resistivity to determine a water-flooded layer and classify and interpret it.