Automated data-driven optimization of small hydropower cascades

Dalane Kraft AS is a Norwegian medium-sized hydropower producer with 100 years of experience and a hydro portfolio of roughly 45MW installed capacity. Currently, the company owns and operates nine power plants in Rogaland County in Norway. It has several cascades in the range up to 10 MW.

Haukland is one of its cascades, a small hydropower (SHP) plant built in 1913, with the capacity of 4.9 MW located in south-western Norway, at the centre of Moi, Lund municipality. The power plant uses water from the Hauklandsbekken and Brekkebekken Rivers.

Despite its small capacity, it was a challenge to operate this plant. The site has a complex topology (Figure 3) and Dalane Kraft AS faced technical, environmental and other constraints.

The key challenges

• Multiple connected reservoirs
• Multiple gates used for flow control
• Different catchment areas with different inflow conditions and planning horizons
• Accurate short-term planning for the small basins, in particular Skårstemmevatn, is highly challenging since overflow not only causes spillage but also can lead to the flooding of a bridge located in the close surroundings
• Flexibility of upstream reservoirs is only usable at a short-term notice and only for a limited number of hours when hydrologically possible

{{challenges-section}}

Scope of the Project

1. Real-time automation: to coordinate a system of multiple reservoirs, gates and turbines in real time is key to avoid spill and imbalance costs;
2. Price-driven dispatch: to exploit the revenue opportunities created by market volatility while ensuring that all technical and environmental constraints are being respected.

As a prerequisite for the real-time capability, the availability of reliable data and robust & secure data communication were key. Over the last years, to achieve a fully automated production system, the company has developed KraftSCADA, a remote SCADA control solution (2018). The KraftSCADA solution is coupled with the optimization system, HYDROGRID Insight offers an innovative and dynamic data-driven approach that helps optimize hydropower plants of varying capacity in an automated way.

HYDROGRID Insight models Haukland’s complex cascade system as a digital copy that comprises Haukland’s multiple reservoirs, gates and turbines while taking into account its different flow times as well as its technical and environmental constraints. Haukland’s actual telemetry is transmitted with an hourly granularity over a standardized and secure RESTful API endpoint (a REST API is a medium for two computers to communicate over HTTP [Hypertext Transfer Protocol], in the same way clients and servers communicate), HYDROGRID Insight utilizes the telemetric data to calibrate their optimization and inflow model and calculates optimal plans for generation and gate opening on an hourly basis. The dispatch plan and gate instruction is read into Dalane Kraft’s KraftSCADA — to fully automate the steering of the power plant. Furthermore, HYDROGRID Insight calculates limit order bid files for the Day-Ahead Market, allowing Dalane Kraft to fully exploit the opportunities provided by power market volatility and, thus, maximize revenues of the Haukland cascade system.

Global hydropower industry context: a trend of digitalization

The intensified volatility of meteorological and market conditions has led to an increased demand for automated and real-time optimization solutions. Given the type of input — such as rainfall, power demand or market price data — optimization solutions face a complex set of non-linear problems. Additionally, various environmental and operational constraints require a complex planning process, resulting in a need for more resources, both regarding time and personnel. Consequently, the ability to react in real time is a major challenge for most hydropower producers. However, existing solutions for planning and optimization require high investments and manual resources.

Therefore, they are usually not profitable for many operators of small- and medium-scale hydropower plants, which excludes them from the possibility of optimized power plant control. The lack of possibilities for optimal plant control in real time leads to SHP producers not fully utilizing the available physical generation flexibility in their portfolio. New technical solutions are needed to make small- and medium-scale hydropower usable as “green battery” (storage power plants) as well as to promote further investments in renewable energy through higher profitability.

The international hydropower industry is currently going through a massive digitalization process, and such technologies as machine learning with its wide range of methodological solutions (e.g., deep learning using neural networks) can be used to optimize hydropower operations for all players in the market. The potential for automated and optimized processes spans the entire value chain, ranging from hydrological forecasting to automated dispatch power trading and predictive maintenance.

An optimized dispatch strategy leading to efficiency and profit gains can be achieved by solving the implicit non-linear problem of hydropower optimization, using mathematical models combined with optimization and machine learning approaches. The complexity of the optimization depends on the following parameters:

1. Number of components of the plant topology (gates/hatches, interconnected water bodies, turbines)
2. Characteristics of the connections between the individual plant’s components
3. Flow times and delays
4. Environmental restrictions
5. Technical restrictions
6. Catchment areas characteristics (size, topology, etc.)

The number of relevant parameters, combined with their possible range, results in a high degree of complexity for optimization solutions that are hard to solve by using cumbersome manual processes or individual statistical models for each underlying problem. New technological solutions support process efficiency, save time and can help operations teams to focus on other important tasks.

SHP plant operator/owner: Dalane Kraft  –  Technology Provider: HYDROGRID GmbH

Technical Characteristics

Key Technical Characteristics of Haukland Kraftverk

Real-time Gate Control

The modelling of the plant is done beforehand in a separate step and incorporates all relevant technical parameters of the actual plant as well as environmental restrictions. The precise modelling of gates is particularly relevant for the complex Haukland cascade, as gates play a significant role in balancing out the water flow in the system, avoiding spilling and imbalance costs. To optimize Haukland, one key aspect was the ability of HYDROGRID Insight to model level-dependent gate openings, resulting in a 3D gate mapping model. Furthermore, HYDROGRID Insight’s 3D gate mapping enables correlation to water level, flow rate and gate opening. Haukland’s gates can be automatically controlled by using Dalane Kraft’s KraftSCADA.

Consequently, KraftSCADA picks up both the optimal dispatch plan and gate opening plans via HYDROGRID Insight’s API endpoints to control all components of the Haukland cascade in a fully automated, yet optimal, manner.

Optimization methodology of HYDROGRID Insight - HIRO algorithm

State-of-the-art approaches to cascade optimization are based on stochastic dynamic programming models. These are taking the variations of all the input variables into account, providing probabilities for all scenarios as a result. However, this yields a dramatic increase in computational resources needed as the complexity of an asset’s topology grows. This increased calculation time renders the real-time reaction to changes in the market or hydrology nearly impossible.

To overcome this challenge, HYDROGRID Insight solves the cascade optimization problem using a more heuristic stochastic modelling approach based on proprietary algorithms. The in-house developed technology focuses on successfully reducing the calculation time. For example, the computing time of Haukland was reduced from 20 minutes to below 3 minutes. Furthermore, HYDROGRID Insight’s modelling approach is fully data-driven. The underlying plant model is continuously re-trained based on the actual plant telemetry and other external input data. Combined with proprietary solutions for inflow and price forecasts using machine learning, HYDROGRID Insight provides one integrated optimization solution for short-, medium- and long-term planning. Thereby, HYDROGRID Insight is capable of automatically identifying changes in hydrology or market conditions and automatically adjusts the entire planning horizon.

As a result, the data-driven approach of HYDROGRID Insight allows for high quality yet efficient optimization of multiple power plants. For Haukland, HYDROGRID Insight helped achieve a reduction in computing time by 65 per cent.

Power trading with HYDROGRID insight

Aside from the automated and optimal dispatch of the Haukland cascade, the other main goal of Dalane Kraft was to utilize the inherent cascade’s flexibility in order to maximize its financial results. One of the many advantages of real-time optimization systems is their ability to accurately forecast and react to changes in the power market and adjust the hydropower generation accordingly. Thereby, the nomination process can be shortened and changing from market order to limit order bids becomes possible.

Currently, Dalane Kraft is trading Haukland’s generation at Nord Pool AS Day-Ahead-Market. As its flagship feature, HYDROGRID Insight calculates an optimal Spot market position at any point in time. This position is either provided as a Market or Limit Order, depending on the needs of the customer. Historically, the Spot market nomination of Haukland’s production was executed based on a Market Order — an hourly generation plan that is immediately sold irrespective of the actual price in the market. To better exploit market volatility and to maximize Haukland’s revenues, Dalane Kraft decided to shift from a market order strategy to a limit order strategy, which is in turn provided by HYDROGRID Insight as a limit order bid. A limit order bid is a production matrix, providing the predicted generation on an hourly basis at different strike prices. Subsequently, the actual generation plan is derived by executing the bid against the actual realized prices in the Day-Ahead Market.

Power trading with HYDROGRID Insight

One of the main benefits of Dalane Kraft using HYDROGRID Insight was the reduction of manual effort within the organization’s operations team, in particular for the small reservoir Skårstemmevatn, which previously induced a high manual workload due to its inherent inflexibility and operational constraints.

In Figure 6, a typical reservoir development and dispatch situation of Skårstemmevatn is shown. With HYDROGRID Insight, the risk of overflow on 25 October 2022 was automatically identified and circumvented by optimizing the plant’s gate opening plan. This plan was, in turn, quickly picked up by DalaneKraft ’s KraftSCADA preventing reservoir spilling and thus avoiding imbalance costs. As a result, the full flexibility of the Haukland cascade system was exploited.

‍

A second main goal of Dalane Kraft using HYDROGRID Insight was maximizing revenue by exploiting the inherent flexibility of the Haukland cascade. Using HYDROGRID Insight, price-driven dispatch can be performed.

Figure 7 shows one example of the automatic price-driven dispatch calculated by HYDROGRID Insight for the Haukland cascade. Whenever the hydrological situation allows, the production is aligned with the power market prices.

Benefits

Key benefits of using HYDROGRID Insight technology for the operation of the Haukland SHP plant are described below.

Economic benefits
The data-driven approach of HYDROGRID Insight allows for high quality, yet efficient, optimization of multiple power plants with a low numerical effort and with minimum personnel expenditure. This brings considerable reduction of operation costs.
HYDROGRID Insight helps the operating company to fully control all data at all times. This enables Dalane Kraft to react extremely quickly to external changes and exploit the opportunities provided by power market volatility. Thanks to the HYDROGRID Insight, Dalane Kraft can execute the limit order strategy, minimize power trading risks and maximize revenues of the Haukland cascade system. Depending on the hydrological situation and the market price development, the financial outperformance ranges from 5 per cent to 12 per cent (equivalent to EUR 70,000–200,000).

HYDROGRID Insight is an affordable solution also for Small Hydro Power Plants of below 5MW, which do not require high investment and manual resources. Implementing this optimization solution, Dalane Kraft does not put its profitability at risk.

This optimization approach supports the market strengths and competitiveness of the operating company: Dalane Kraft can position itself as a top expert in operating SHP plants not only on the Norwegian market, but also on the international level, growing its portfolio with new clients.

Social benefits
HYDROGRID Insight helps reduce the manual effort within the organization’s operations team and free resources for other important tasks.

Environmental benefits
HYDROGRID Insight automatically ensures that all environmental restrictions of the project are followed.
Maximizing efficiency by optimizing water utilization (preventing reservoir spilling) and increasing the electricity production from renewable energy sources.

‍

Lessons learned and conclusion

Complex SHP cascades can considerably benefit from an innovative IT data-driven optimization approach. Technology such as HYDROGRID Insight can help achieve real-time automation and price-driven dispatch.

The potential for automated and optimized processes spans the entire value chain. In the HYDROGRID case study, these processes included hydrological forecasting, automated dispatch power trading and predictive maintenance.

New technical solutions are essential to make small- and medium-scale hydropower usable as “green battery” as well as to promote further investments in renewable energy through higher profitability.

Acknowledgement
The authors would like to acknowledge Dalane Kraft for their forward thinking and innovative attitude and for a fantastic collaboration.