General Overview & Capabilities
What do you do?
Gridraven is a grid software company. We predict line ratings using hyper local wind forecasts, it means how wind actually blows along power lines – how fast it blows and how trees, hills and buildings change its path. Our accurate wind forecasting helps grid operators see – days ahead – how much power every line will be able to handle without overheating.
Why are accurate line ratings important?
When transmission lines are congested, electricity prices spike. Power has to be rerouted or replaced with more expensive generation. If operators can safely move more electricity through existing lines, that reduces congestion and lowers costs. By forecasting wind more accurately, we help grid operators safely move more power through existing lines. That means less congestion, reduced risk of blackouts and lower power bills for ratepayers.
Before Gridraven, why have grid operators not accounted for wind cooling?
Historically, weather forecasting systems have not had sufficient accuracy and resolution. As a result, grid operators have applied conservative weather assumptions that are based on the worst-case conditions (low wind and hot days). While that keeps the system safe, real conditions can typically allow much more power flow than those assumptions.
How does the Claw system work?
Claw calculates ratings for each span individually based on hyper-local weather conditions. The rating for the line is determined by the minimum rating found amongst all the spans. We provide these rating forecasts to your SCADA or EMS, which can then be forwarded to or used by the system operator.
What data is Claw based on?
We use satellite imagery, LiDAR terrain models and thousands of weather stations to understand how wind actually moves along each individual power line. Unlike traditional forecasts that assume flat land and average wind speeds, Gridraven’s software takes into account how hills, trees and buildings change wind patterns. This gives grid operators a more accurate, line-by-line forecast of how much electricity each line can handle – without putting any kind of hardware on the wires.
What is the specific output provided to the grid operator?
Claw outputs the ratings of each line in amperes, or in MVA based on the nominal voltage. This is provided to the SCADA/EMS of the grid operator. Operators use these values in real-time monitoring and operational planning.
Does the system require physical sensors?
No, the solution is entirely software-based. It does not require any sensors to be installed on towers or wires, meaning that the system is immediately scalable to your full network and there is no hardware to calibrate or maintain in harsh conditions.
How do you validate accuracy without sensors?
We continually validate our weather forecasts against over 100,000 ground based weather stations. Further, we have done validations comparing our weather forecasts against measurements taken on high voltage lines in critical spans.
Methodology & Calculations
What is the forecast horizon and update frequency?
Ratings are provided for each line with hourly resolution and up to 10 days ahead. Forecasts are updated hourly.
0–5 Days (Dynamic): Gridraven's wind forecasts have sufficient certainty to provide dynamic capacity increases.
5–10 Days (Ambient): Wind forecast uncertainty becomes too high, so ratings revert to Ambient Ratings.
10+ Days (Seasonal): Beyond ten days, the ratings revert to standard Seasonal Ratings.
How volatile are the line ratings?
Dynamic ratings depend on large-scale wind patterns. The ratings will naturally rise and fall over the course of a few hours. This movement is clearly visible on DLR graphs.
In contrast, each forecast is stable. The predicted value for any specific hour does not "jump around" as real-time approaches. You can rely on the forecasted value to remain consistent.
What weather models do you use for ratings?
Claw is powered by a proprietary weather model built by Gridraven. The uniqueness of Gridraven’s weather model is twofold: first, it’s the highest resolution weather prediction out there, using meter-scale landscape data for forecasting weather parameters in each individual span. Second, the model outputs probability distributions for each variable, allowing the operator to set a desired safety level for the ratings, such as 98%.
How do you handle weather uncertainty?
Our key breakthrough is providing confidence ranges for weather parameters. For example, we can calculate the minimum wind speed with 99% certainty to ensure safety and reliability. Further, our models utilize satellite data, radar, and 1-meter resolution elevation models to understand the terrain, including ground objects like buildings and trees, ensuring reliability also in sheltered spans.
What methodology do you use for line ratings?
We primarily use CIGRE TB 601 or IEEE 738 methodology for both line rating and static rating calculations. The system can also switch to CIGRE TB 207 or custom methodologies if required.
Why do Dynamic Line Ratings (DLR) sometimes fall below the Static Rating (SR)?
This is a common occurrence because DLR calculations use more conservative assumptions regarding wind than the standard Static Rating to ensure safety. First, SR often assumes an optimal cooling angle of 90 degrees, while DLR uses a conservative minimum of 30 degrees (or 45 degrees fixed for Ambient Adjusted Ratings). Second, DLR applies a lower bound wind speed of 0.5 m/s, whereas SR often uses a fixed 0.61 m/s.
What are the default parameters for Static Rating (SR)?
Unless customized, the default settings are an ambient temperature of 35 degrees Celsius, a wind speed of 0.61 m/s, a wind attack angle of 90 degrees, and solar irradiance of 1033 W/m2.
How is conductor sag accounted for?
We account for sag by using the maximum allowed temperature for each span individually. We assume each span has a valid maximum temperature that avoids ground clearance violations. If LiDAR survey data is available, you can set individual temperature limits for specific spans to maximize capacity safely.
What statistical distribution is used to determine wind speed bounds?
We do not assume a normal (Gaussian) distribution or any other predefined shape for weather variables. Instead, our Bayesian non-parametric machine learning models allow the distribution to be formed mathematically based on raw data inputs for each specific site and timestep. This allows the model to accurately capture asymmetrical or skewed distributions typical of local conditions. Consequently, percentiles are derived directly from the model’s cumulative distribution rather than a classical standard deviation.
What does "95% confidence" mean mathematically?
For weather parameters like wind speed, a 95% confidence interval means the actual value is expected to fall between the 2.5th and 97.5th percentiles. Because our distributions are non-parametric and often asymmetrical, the distance from the median to these lower and upper bounds is rarely equal.
How is a two-sided weather forecast converted to a one-sided DLR value?
While weather inputs are processed with two-sided bounds, we treat the DLR calculation differently to prioritize operational safety. Since the upper bound of wind speed represents a "best-case" cooling scenario that could lead to unsafe overestimations, we resample the weather distributions into a one-sided DLR lower bound. When we refer to a 95% DLR (P95), there is a 95% probability that the actual capacity will be higher than that specific bound, providing a conservative "floor" for safe operations.
How are weather confidence intervals calculated compared to DLR percentiles?
We use two different mathematical approaches depending on the application:
- Weather Inputs (Two-Sided): For parameters like wind speed, a 95% confidence interval is two-sided, meaning the lower bound is the 2.5th percentile and the upper bound is the 97.5th percentile. In this case, there is a 97.5% probability that the actual wind speed will exceed the lower bound.
- DLR Ratings (One-Sided): For line capacity, we shift to a one-sided lower bound to prioritize safety. A DLR P95 rating means there is exactly a 95% probability that the actual line capacity is at least that value, with only a 5% statistical risk of it being lower.
Why doesn't Gridraven use standard deviation to define bounds?
Standard deviation is a metric used for normal (Gaussian) distributions. Because our Bayesian non-parametric models do not force data into a predefined Gaussian shape, we do not use standard deviation to define safety bounds. Instead, our models generate a unique probability density function for every timestep, and all percentiles are derived directly from that resulting distribution.
Does the model assume a specific distribution shape for weather variables?
No. Unlike traditional models that rely on predefined shapes, our machine learning models allow the distribution to be formed mathematically based on raw data inputs for each specific site and timestep. This allows the system to accurately handle asymmetrical, multi-modal, or skewed distributions based on real-world local conditions.
Implementation & Data Requirements
What data is needed to set up the system?
To configure the system, you are required to input: conductor type, conductor parameters and tower coordinates. Our system also makes use of high-resolution landscape data, which Gridraven obtains in the background. In most of Europe and the US, we rely on public LiDAR data with 1-meter resolution. If that is not available, we rely on satellite data.
Can this be used for distribution lines?
Yes. We have customers with lines ranging from 30 kV to 400 kV. The thermal balance calculations apply across this whole range. Our weather models are validated for accuracy at 5 meters from the ground to account for the lower ground clearance of distribution lines compared to transmission lines.
What is the forecast update frequency?
Forecasts are currently updated once per hour. While this is not currently user-configurable, higher frequencies may be possible in the future if weather data providers increase their update intervals.
API & Technical Integration
Can you support secure, on-premise installations?
Yes, we support on-premise installation. Setup usually requires VPN access.
How is the system integrated?
Claw can be hosted either on premises or in the Cloud. The integration is then customized to your SCADA / EMS. Options include for example IEC 104 for SCADA connection as a virtual RTU, or a REST API or Kafka for integrating to your EMS.
What authentication mechanism is used for APIs?
- Cloud: We use opaque API keys without an expiration date, manageable via the Claw admin panel.
- On-Premise: Authentication depends on your local technology (e.g., standard OIDC flow).
Are there API rate limits or bulk endpoints?
There are currently no rate limits. While there is no dedicated bulk endpoint yet, fetching each line with a separate API call is acceptable. To check for updates, polling a single line is sufficient as all lines update simultaneously.
What is the typical API response time?
The average response time is approximately 1 second.
How is the SFTP data delivery structured?
By default, our SFTP process pushes a single consolidated file containing all lines and all forecast steps. However, we can modify this process to push individual files per line if required by client’s specific integration needs.
How reliable is the SFTP process and are there SLAs?
The service is hosted on Google Cloud and its reliability matches the hosting, network, and Claw service uptime.
- Current Status: We currently offer this as a "best effort" transition solution for clients without a native API implementation; there is no default SLA for this specific integration.
- Future SLAs: We can establish specific SLAs based on client’s requirements. For added reliability, we recommend using previous forecast data as a fallback, which typically contains up to 3–4 days of hourly forecasts.
Are there limitations on SFTP transfer frequency?
There are no hard limitations. Both 1-hour and 20-minute update frequencies are supported. The limiting factor is primarily the REST API polling speed, as data aggregation and SFTP writing take negligible time.
Business & Safety
Why choose software over sensors?
Sensors only measure real-time conditions and cannot forecast, which is essential for day-ahead markets and operational planning. Furthermore, sensors only measure the specific span they are installed on, whereas our software covers 100% of spans to ensure the limiting span is always identified. Further, software scales up to cover all lines and ensures that congestion is reduced and not just pushed down to the next weakest line.
Is the system safe given that ratings change hour to hour?
Safety is our priority. We address the risk of sag and flashover by using conservative lower bound inputs, such as 99% confidence on wind speed and conservative wind attack angles, rather than average values. This ensures that even when ratings change, the operation remains within safe thermal limits.
Is it possible to view an evaluation package or trial?
Yes, we offer a trial version where you can add 1 line/10 weather stations and see the forecast. The free trial has a 30-day duration.