White Paper

Veritone Energy Solutions:
CDI For Grid Optimization and Resilience

Delivering real-time dynamic grid modeling and control for predictable, cost-effective and resilient energy dispatch

With Veritone Energy Solutions, you can optimize, synchronize and intelligently control the energy grid, using predictive AI to make clean energy more predictable and efficient, more cost-effective, and more safe and resilient.

Use cases include:

  • Near real-time optimal economic dispatch
  • Real-time demand response
  • Volt/VAR optimization
  • Microgrid energy management control and resiliency
  • Solar smoothing
  • Near real-time energy arbitrage

Veritone’s patented CDI (Cooperative Distributed Inferencing) technology forms the backbone of Veritone Energy Solutions, delivering real-time dynamic modeling and control that ensures predictable energy distribution and resilience across the grid. CDI self-learns and adapts to ensure all energy devices in a microgrid, such as solar and battery power, deliver optimal energy at peak demand times and continue to operate autonomously if isolated from the main grid due to extreme weather or natural disaster.

Veritone Energy For Grid Resilience

Traditional energy planning strategies can no longer ensure uninterrupted power service.

Extreme weather, natural disasters, power fluctuations and unpredictable green energy sources translate to more spinning reserves and higher energy costs.

Veritone Energy Solutions and CDI optimize smart grid energy distribution by continuously knowing how much of what type of energy to deliver where ensuring grid resilience in the face of the unexpected.

In an emergency, Veritone’s real-time AI-based grid learning optimally routes energy to critical loads from neighboring states, from another microgrid, or from behind-the-meter battery storage.

Distributed AI agents ensure optimal economic dispatch of energy between multiple Distributed Energy Resources (DERs) allowing for autonomous, continuous operation when portions of the grid fail. Batteries, solar inverters, and wind turbines can be synchronized within a microgrid and across grids, enabling islanding and autonomous grid management.

Ongoing grid resilience can be achieved by using Veritone to fuse together data from real-time weather and load forecasting, economics, rules, and real-time grid learning to deliver optimal grid management and control, whether between grids or autonomously in an islanding configuration. With AI controlling the grid, everyone has the power they need, and blackouts can be prevented.

Veritone Energy Solutions Components

Veritone’s patented AI-based energy solutions consist of three subsystems. CDI plays within these subsystems in various ways, covered later in this paper.

Forecaster – a distributed forecaster service that generates predictions of the state of the smart grid devices.

Optimizer – a real-time distributed agent that learns, optimizes and tunes models of smart grid components, generates desired behavior directions and provides synchronization of smart grid components.

Controller – a bank of edge controllers that implements the desired behavior as a function of the predictive state of the smart grid.

The Arbitrage solution brings these three subsystems together to deliver predictive energy buy, sell and dispatch capabilities.

Veritone aiWARE, the leading cognitive operating system, provides deployment, integration, data services, and weather services to these subsystems.

How Veritone Energy Works

The Veritone Forecaster, Optimizer, and Arbitrage solutions leverage CDI to perform their functions. CDI works hand-in-hand with the Veritone Controller solution, which is a dynamic, incremental feedback control system that obtains sensor-based environmental information and produces actions to control a network of electric microgrid devices including distributed loads, solar panels, batteries, and utilities.

How Energy Works

CDI Overview

CDI is the grid modeling and learning core of Veritone Energy Solutions. CDI leverages forecast data and rules to build and continually update device state models, which are then used to intelligently control edge devices. Figure 1 shows a high-level diagram of CDI and a distributed edge control system.

CDI Overview

FIGURE 1
A high-level diagram of CDI and a distributed edge control system.

Why CDI Is Unique – Real-Time Dynamic Modeling

Veritone’s patented CDI (Cooperative Distributive Inferencing) technology uses advanced optimization and modeling techniques applied to distributed dynamic systems with sensor data to dramatically improve the performance of complex grid operations in real-time. CDI makes real-time dynamic modeling reality in clean energy and IoT applications as well as thousands of others across industries.

Adaptive Models

CDI uses Veritone’s patented Tomograph, which continuously constructs a control model of a complex system in real-time as the system evolves with changing conditions. Other solutions on the market simply learn static model parameters, while Veritone applies dynamic, adaptive learning to models via model Hamiltonian. With CDI, the system is always operating at peak performance as each device’s edge controller gets its instructions from the most accurate model possible at any given time considering changing environmental conditions.
The Veritone Tomograph builds dynamic control models from sensory data, based on rules that govern the system. Models are summarized by a system Hamiltonian, which is computed in a distributed manner across an agent-based network of grid devices. In this way, models apply to devices and are also synchronized across multiple devices on the grid via an aggregated Hamiltonian approach.

Distributed Agents

CDI uses a distributed agent-based approach that reduces overall processing requirements while drastically improving latency. Systems and devices including batteries, solar inverters, and wind turbines use models that are synchronized both with each other and the grid, allowing for autonomous energy grid management via Veritone’s predictive intelligent controllers, and resulting in estimated energy efficiency improvements in the range of 15-25%.

Integrated Solution

Unlike competitive energy offerings that focus on a single aspect of energy management, such as:

  • Forecasting (price, demand, generation)
  • Optimal economic dispatch of energy from multiple distributed energy resources (DER’s)
  • Real-time optimal battery control
  • High-frequency energy arbitrage (buy/sell) for the wholesale energy markets

Veritone’s CDI-based energy solutions provide all needed capabilities in a single integrated system, fusing together real-time forecasting, economics, rules, and real-time learning for device and network model building/updating to deliver autonomous energy grid management and control.

CDI Applications

Benefits of CDI’s real-time dynamic modeling span any device under control. For example, fast and accurate dynamic modeling of batteries increases battery utilization, reduces the risk of a thermal event, extends battery life, and improves the economics. To build an accurate model of a battery, Veritone uses its patented Model Realization System (MRS) which injects a micro-volt pulse into the battery and then records its response.

For electric car batteries, dynamic, real-time modeling and optimization result in improved range, battery longevity, and reduced thermal event risk. Complex energy dispatch challenges for EV charging stations can also be solved using Veritone’s unique CDI-based energy solutions.

CDI & Edge Controller Interaction

A more detailed view of the CDI and Edge Control system is shown in Figure 2 and lists the components of the CDI Agents and Edge Controllers. These components are described in more detail in the sections below.

CDI uses hard and soft rules for edge controller components, along with a mean field-based dynamic synchronization of the network, to perform a real-time construction and updating of device models under control via a tomograph. CDI then generates a tracking signal representing the most optimal model at any point in time. That model combines dynamic, optimal demand satisfaction with rules describing device longevity, operational limits and other device characteristics.

This CDI tracking signal is sent to the edge controller, which generates an implementable signal that controls a physical energy device.

Distributed CDI agents are dynamically synchronized through the mean approach with a blackboard architecture that collects and transfers information and ensures optimization and synchronization of energy across a distributed energy network.

CDI & EDGE CONTROLLER INTERACTION

FIGURE 2
Key energy solution components for optimal control of a device at the edge. A CDI agent may control a single or several edge controllers.

Key Technical Components

The Veritone Energy CDI and edge control system combines data acquired in its ML-based knowledge base, distributed CDI agents, edge controllers and sensors to control components in the network for optimal energy dispatch.

Key technical components of the overall Veritone Energy CDI and edge control system include:

1. Rule translator:

Rule Translator
Converts domain rules, including absolute, hard and soft rules, into potential equations and inequalities used by CDI agents.

2. CDI agent:

CDI Agent

  • Rule potential: Local representation (rule-potentials) of controlled components associated with an edge controller
  • Tomograph: Learns and updates a model of the device under control, based on a data Hamiltonian with initial and terminal conditions for states, co-states and controls, for a CDI agent by probing with active sensor controls.
  • Optimizer: Continuously generates a policy, using a sliding window, for a dynamic system that incorporates rule-potentials to satisfy operational and behavioral goals.
  • Mean field: Synchronizes and optimizes distributed CDI agents by projecting states of the entire network onto local CDI agents.

3. Edge controller:

Edge Controller

  • State estimator: Estimates the state of the system, including discrete and continuous modalities, using an incremental state estimator. Also uses a parameter adaptive engine (PAE) to learn parameters.
  • Incremental controller: Determines the optimal device control signal that tracks an input signal (for example, power rate or power) using a proportional linear-quadratic (LQ) tracker, a PID LQ tracker, or a chattering optimization algorithm.
  • Emergency control system: A controller that can operate independently in an emergency situation when there is no connection to the network and can reconcile with the network when connectivity is restored.

4. Blackboard: The blackboard provides flow of information, including state and synchronization data, between agents.

5. Forecaster: Uses historical and current sensor data to forecast future energy factors, such as demand and irradiation, to be used in the edge controller. Leverages multiple trainable models including a parameter adaptive engine (PAE).

Ready to leverage Veritone Energy Solutions for improved grid performance and resilience? Contact Veritone today.

References

1. Wolf Kohn, Zelda B. Zabinsky, and Anil Nerode, A Micro-Grid Distributed Intelligent Control and Management System, IEEE Transactions on Smart Grid, Vol. 6, No. 6, November 2015.
2. Cooperative Distributed Inferencer (CDI) is discussed in “xPatterns CDI Architectures 112015”