Derivative works — US Patent US10430725B2

Defensive Disclosure Document

RE: Technical Disclosures and Derivative Works Related to the Art of US10430725B2
Publication Date: May 1, 2026
Status: Public Disclosure

This document serves as a defensive publication to establish prior art for a series of technical and operational variations related to the system and method described in U.S. Patent US10430725B2 ("Petroleum analytics learning machine system..."). The following disclosures are intended to place these concepts in the public domain, thereby rendering them obvious or non-novel for the purposes of future patent prosecution by third parties.

Axis 1: Material & Component Substitution

1.1. Quantum Annealing for Importance Weight Calculation

Enabling Description: The process of identifying "Importance Weights" is a high-dimensional optimization problem. This variation substitutes the classical machine learning ensemble (e.g., Support Vector Machines, Random Forests) with a quantum annealing processor. Attributes from the System Integration Database (SID) are mapped to a Quadratic Unconstrained Binary Optimization (QUBO) formulation. Each attribute's potential contribution to production is a variable, and inter-attribute correlations are couplings. A quantum annealer finds the QUBO's lowest energy state, which corresponds to the optimal set of "Importance Weights." The architecture comprises a hybrid classical-quantum stack where the SID feeds pre-processed data to a cloud-based quantum annealing service. The resulting weights are returned to the classical system for convolution and prescriptive output.

Diagram:

flowchart TD
    A[SID: Geological, Drilling, Frac Data] --> B{Data Pre-processing & Feature Engineering};
    B --> C[Map Features to QUBO Formulation];
    C --> D{{Quantum Annealing Processor}};
    D --> E[Lowest Energy State Solution];
    E --> F{Translate Solution to Importance Weights};
    F --> G[Classical PALM System];
    G --> H[Prescriptive Recommendations];

1.2. Federated Learning Architecture for Multi-Operator Data Collaboration

Enabling Description: This variation replaces the centralized System Integration Database (SID) with a federated learning architecture using a framework like TensorFlow Federated (TFF). This allows multiple operators to collaboratively train a global model without sharing proprietary raw data. Each operator maintains its local SID and trains a centrally-distributed base model. Only the updated model weights (gradients) are transmitted to a central server, which aggregates them using an algorithm like Federated Averaging. The improved global model, containing more robust "Importance Weights," is then redistributed to the participants, preserving data privacy while enhancing model accuracy.

Diagram:

sequenceDiagram
    participant Server
    participant Operator_A
    participant Operator_B
    participant Operator_C

    Server->>Operator_A: Distribute Global Model v1
    Server->>Operator_B: Distribute Global Model v1
    Server->>Operator_C: Distribute Global Model v1

    Operator_A->>Operator_A: Train Model on Local Data
    Operator_B->>Operator_B: Train Model on Local Data
    Operator_C->>Operator_C: Train Model on Local Data

    Operator_A-->>Server: Return updated weights_A
    Operator_B-->>Server: Return updated weights_B
    Operator_C-->>Server: Return updated weights_C

    Server->>Server: Aggregate weights (Federated Averaging)
    Server->>Server: Create Global Model v2

    Server->>Operator_A: Distribute Global Model v2
    Server->>Operator_B: Distribute Global Model v2
    Server->>Operator_C: Distribute Global Model v2

1.3. Neuromorphic Processing for Real-Time Edge Control

Enabling Description: To achieve low-latency "autopilot" control, this variation deploys neuromorphic computing hardware (e.g., Spiking Neural Network processors like Intel's Loihi) at the operational edge (e.g., on a drilling rig). The SNN processors are pre-trained to recognize specific patterns in high-frequency sensor data, such as acoustic signatures of drill bit wear or micro-seismic events during fracturing. The neuromorphic chip processes raw data streams and outputs low-dimensional, event-based signals corresponding to classified operational states. This allows the local controller to react in microseconds, with the "Importance Weights" from the patent encoded as synaptic weights within the SNN.

Diagram:

flowchart TD
    subgraph Edge Device
        A[High-Frequency Sensors] --> B(Neuromorphic SNN Processor);
        B --> C{Event-Based State Classification};
        C --> D[Local Control Unit];
    end
    D --> E[Actuator Control];
    B -- Synaptic Weights derived from --> F((Central PALM Model));
    D -- Logs data to --> F;

Axis 2: Operational Parameter Expansion

2.1. Nanofluidic Reservoir Analytics (Micro-Scale)

Enabling Description: The PALM system is applied to optimize laboratory "reservoir-on-a-chip" experiments. The "well" is a microfluidic chip with etched pore networks. "Production data" is sourced from high-speed microscopy and embedded nanosensors tracking fluid flow and pressure at the micron scale. The system analyzes thousands of experimental runs, using machine vision to classify flow patterns (unstructured data) and sensor readings (structured data). It calculates "Importance Weights" for parameters like pore throat geometry and fluid viscosity to predict hydrocarbon recovery, enabling high-throughput screening of Enhanced Oil Recovery (EOR) techniques.

Diagram:

flowchart TD
    A[Experiment Design Parameters] --> B(Microfluidic Chip Fabrication);
    B --> C{Reservoir-on-a-Chip Experiment};
    C --> D[High-Speed Microscopy & Nanosensors];
    D --> E{Data Processing & Feature Extraction};
    subgraph PALM Analytics
        E --> F[Calculate Importance Weights];
        F --> G[Predict EOR Efficacy];
    end
    G --> H[Rank EOR Techniques];

2.2. National-Level Energy Infrastructure Management (Macro-Scale)

Enabling Description: The PALM system is scaled to manage a nation's natural gas pipeline grid. "Wells" are major gas fields and storage facilities. The system ingests data from thousands of compressor stations, SCADA systems, weather forecasts, and economic indicators. The machine learning optimizer predicts regional demand, identifies network bottlenecks, and calculates "Importance Weights" for factors affecting grid stability (e.g., ambient temperature, market sentiment). Prescriptive recommendations manifest as autonomous control actions, such as rerouting gas flow, adjusting LNG terminal operations, and managing strategic reserves to prevent blackouts.

Diagram:

graph TD
    subgraph Data Inputs
        A[Gas Fields & Storage Data];
        B[National Pipeline SCADA];
        C[Weather Forecasts];
        D[Economic Indicators];
    end
    subgraph PALM for National Grid
        E{System Integration Database};
        F[ML Optimizer: Demand & Stability Prediction];
        G[Calculate Importance Weights for Grid Parameters];
        H[Prescriptive Control Engine];
    end
    Data Inputs --> E;
    E --> F --> G --> H;
    subgraph Autonomous Actions
        I[Pipeline Flow Rerouting];
        J[Compressor Station Adjustments];
        K[Strategic Reserve Dispatch];
    end
    H --> Autonomous Actions;

Axis 3: Cross-Domain Application

3.1. Pharmaceutical Batch Process Optimization

Enabling Description: The PALM framework is repurposed for optimizing biologic drug manufacturing in bioreactors. The "well" is a bioreactor batch. "Geological data" is the cell line's genetic profile. "Completion data" are the cell culture inoculation parameters. "Production data" is the real-time stream from Process Analytical Technology (PAT) sensors (e.g., Raman spectroscopy, dissolved oxygen). The system calculates "Importance Weights" for process parameters (e.g., nutrient feed strategy, temperature) that correlate with final product yield and purity, providing real-time adjustments to the bioreactor control system to steer the batch towards a desired metabolic state.

Diagram:

flowchart TD
    A[Historical Batch Data] --> B{PALM Training};
    B --> C[Importance Weights Model];
    subgraph Real-Time Batch
        D[Bioreactor PAT Sensors] --> E{Live Data Stream};
        E --> F[PALM Inference Engine];
        C --> F;
        F --> G[Prescriptive Control Actions];
        G --> H(Bioreactor Control System);
        H --> I((Bioreactor));
        I --> D;
    end

3.2. Precision Agriculture Yield Maximization

Enabling Description: The system is applied to a large-scale farm. A "well" is a field parcel. "Geological data" is soil composition and topography. "Fracturing data" is the irrigation and fertilization schedule. "Production data" comes from IoT soil sensors and drone-based multispectral imagery. The system integrates these sources to calculate "Importance Weights" for variables like soil moisture and nitrogen levels on crop yield. An "autopilot" function autonomously controls irrigation systems and triggers alerts for targeted resource application.

Diagram:

graph TD
    subgraph Farm Data Sources
        A[Soil Composition];
        B[Drone Multispectral Imagery];
        C[IoT Soil Sensors];
        D[Weather Data];
    end
    subgraph PALM for AgTech
        E[Integrated Farm Database];
        F[ML Model: Calculate Importance Weights for Yield];
        G[Prescriptive Recommendation Engine];
    end
    Farm Data Sources --> E --> F --> G;
    subgraph Autonomous Farm Operations
        H[Smart Irrigation System];
        I[Variable Rate Fertilizer Applicator];
        J[Pest Management Alerts];
    end
    G --> Autonomous Farm Operations;

3.3. Algorithmic Trading Strategy Optimization

Enabling Description: The PALM system is adapted to optimize high-frequency trading (HFT) strategies. A "well" is a trading algorithm. "Geological data" is historical market microstructure. "Production data" is the real-time market tick data and the algorithm's P&L. The unstructured data module performs sentiment analysis on news feeds. The system backtests parameter combinations, calculating "Importance Weights" for factors like order book imbalance and news sentiment. The prescriptive output dynamically adjusts the trading algorithm's parameters in real-time based on the system's classification of the market state.

Diagram:

sequenceDiagram
    participant Market
    participant NewsFeed
    participant PALM_HFT
    participant TradingAlgo

    Market->>PALM_HFT: Real-time Tick Data
    NewsFeed->>PALM_HFT: Real-time News/Sentiment
    PALM_HFT->>PALM_HFT: Analyze Data & Classify Market State
    PALM_HFT->>TradingAlgo: Prescribe Parameter Adjustment
    TradingAlgo->>Market: Execute Optimized Orders
    Market-->>TradingAlgo: Order Fill Confirmation
    TradingAlgo-->>PALM_HFT: Feedback P&L

Axis 4: Integration with Emerging Technologies

4.1. AI-Driven Generative Design of Frac Stages

Enabling Description: This variation integrates the PALM system with a Generative Adversarial Network (GAN). The PALM system's analysis of historical data and its "Importance Weights" are used to train the GAN's discriminator to distinguish between high- and low-producing fracture designs. The GAN's generator then creates novel frac stage designs (pump schedules, proppant volumes) that the discriminator classifies as "good." This creates an AI-driven design loop where the PALM system's predictive analytics guide a generative model to explore the design space and propose innovative, high-performance completion designs.

Diagram:

flowchart TD
    A[Historical Frac Data] --> B(PALM Analysis);
    B --> C[Train GAN Discriminator];
    D(GAN Generator) -- Creates Novel Designs --> E{Proposed Frac Design};
    E -- Is it "Good"? --> C;
    C -- Feedback --> D;
    C -- Validated High-Potential Designs --> F[Output for Field Use];

4.2. IoT-Enabled Digital Twin for Production Forecasting

Enabling Description: The PALM system is integrated with a network of Industrial IoT sensors (downhole gauges, acoustic sensors) that feed a physics-based digital twin of the well and reservoir. The PALM system's machine learning models continuously calibrate this digital twin, correcting discrepancies between the physics-based model and real-world performance. The calibrated twin provides a highly accurate, dynamic forecast of production. "Importance Weights" calculated by PALM identify which physical parameters in the twin (e.g., reservoir permeability) have the most uncertainty and require better sensor data for calibration.

Diagram:

graph TD
    A[IIoT Sensor Network] --> B{Real-Time Data Feed};
    B --> C(Physics-Based Digital Twin);
    C -- Prediction --> D{Compare};
    B -- Reality --> D;
    D -- Discrepancy --> E(PALM ML Calibration Engine);
    E -- Correction Factors --> C;
    C -- Calibrated Forecast --> F[Accurate Production Prediction];

4.3. Blockchain for Verifiable Carbon Sequestration Tracking

Enabling Description: The PALM system is adapted for Carbon Capture, Utilization, and Storage (CCUS) operations to optimize CO2 injection and storage. A private, permissioned blockchain (e.g., Hyperledger Fabric) creates an immutable record of the CCUS lifecycle. Each ton of injected CO2 is tokenized. Data from IoT sensors monitoring injection and potential leaks are written to the blockchain via oracles. The PALM system analyzes this data to predict long-term storage stability and autonomously adjusts injection parameters, while the blockchain provides a verifiable, auditable trail for carbon credits.

Diagram:

sequenceDiagram
    participant Operator
    participant PALM_CCUS
    participant IoT_Sensors
    participant Blockchain

    Operator->>PALM_CCUS: Define Injection Plan
    PALM_CCUS->>Operator: Provide Optimized Parameters
    IoT_Sensors->>Blockchain: Write Verified Injection Data via Oracle
    Blockchain->>PALM_CCUS: Read Immutable Operational Data
    PALM_CCUS->>PALM_CCUS: Analyze for Long-Term Stability
    PALM_CCUS->>Operator: Recommend Real-time Adjustments
    Blockchain-->>Operator: Provide Verifiable Audit Trail for Carbon Credits

Axis 5: The "Inverse" or Failure Mode

5.1. Graceful Degradation for Communications-Denied Environments

Enabling Description: For remote operations with intermittent connectivity, the edge device runs a compressed, lightweight version of the primary machine learning model (e.g., a quantized TensorFlow Lite model). When communication is lost, the system enters a "limited functionality" mode, using the local model to perform safety-critical functions like preventing drill string buckling based on a limited set of local sensor inputs. When communication is restored, the edge device syncs its logged data with the central SID and receives an updated lightweight model.

Diagram:

stateDiagram-v2
    [*] --> Connected
    Connected: Full PALM Functionality
    Connected: Syncing data with central SID
    Connected --> Disconnected: Communication Lost

    Disconnected: Limited Functionality Mode
    Disconnected: Operate on local, lightweight model
    Disconnected: Log all data and actions
    Disconnected --> Connected: Communication Restored

Combination Prior Art Scenarios with Open-Source Standards

Integration with Energistics PRODML: The system is disclosed wherein its System Integration Database (SID) is architected to natively ingest and export data using the Energistics PRODML open data exchange standard. The system specifically utilizes PRODML schemas for production data, including DtsInstalledSystem, WftRun, and FluidSample, enabling seamless interoperability with third-party production surveillance software and rendering a patent on such a specific integration obvious.
Integration with The Open Group OSDU™ Platform: The system is disclosed wherein the centralized SID is replaced with the Open Group's OSDU™ (Open Subsurface Data Universe) Platform. The PALM system's data processing and machine learning modules are architected as microservices that register with and query the OSDU platform via its standard APIs, using its defined schemas for subsurface data. This combination places in the public domain the concept of the patented system being built upon this specific open-source industry data platform.
Implementation with Apache Spark and MLflow: The system is disclosed wherein the Machine Learning Optimizer is specifically implemented using an open-source MLOps stack. Data processing and model training are performed on an Apache Spark cluster. The machine learning lifecycle is managed using the MLflow platform to track experiments, package the trained models (e.g., SVM, Random Forest), and deploy them as REST APIs for the prescriptive recommendation engine. This describes a specific, reproducible, open-source implementation of the patent's core machine learning process.