Adaptive Robust Optimization for the Security Constrained Unit Commitment Problem
Summary (3 min read)
Introduction
- The objective of the UC problem is to find a unit commitment schedule that minimizes the commitment and dispatch costs of meeting the forecast system load, taking into account various physical, inter-temporal constraints for generating resources, transmission, and system reliability requirements.
- This framework could also be integrated into the robust optimization formulation proposed below.
- The method of uncertainty quantification (UQ) proposed in [11] can be integrated into the robust optimization UC model, where the UQ module updates the uncertainty model as more information is obtained in time.
II. DETERMINISTIC SCUC PROBLEM
- The deterministic SCUC problem is extensively studied in the power system literature (e.g., [24], [25]).
- The binary variable is a vector of commitment related decisions including the on/off and start-up/shut-down status of each generation unit for each time interval of the commitment period, usually 24 h in an ISO setting.
- By convention, generation, reserve, and flow take positive sign, whereas load takes negative sign.
- Constraint (3) couples the commitment and dispatch decisions, including minimum and maximum generation capacity constraints.
III. TWO-STAGE ADAPTIVE ROBUST SCUC FORMULATION
- The authors first discuss the uncertainty set, which is a key building block of the robust model.
- Then, the authors introduce the two-stage adaptive robust SCUC formulation and provide a detailed explanation.
- Furthermore, in their formulation the optimal second-stage decision is a function of the uncertain net injection , therefore,fully adaptive to any realization of the uncertainty.
- Notice that the worst case dispatch cost has a max-min form, where determines the economic dispatch cost for a fixed commitment and net injection , which is then maximized over the uncertainty set .
- Due to the bilinear structure of the objective function, the optimal solution of problem (9) is an extreme point of the polyhedron , and similarly the optimal solution is an extreme point of .
IV. SOLUTION METHOD TO SOLVE THE ADAPTIVE ROBUST MODEL
- As analyzed in the previous section, the adaptive robust formulation (7) is a two-stage problem.
- The second-stage is to find the worst-case dispatch cost under a fixed commitment solution.
- Naturally, the authors will have a two-level algorithm.
- The outer level employs a Benders decomposition (BD) type cutting plane algorithm to obtain using the information (i.e., cuts) generated from the inner level, which approximately solves the bilinear optimization problem (9).
A. Outer Level: Benders Decomposition Algorithm
- The Benders decomposition algorithm is described below.
- To speed up the convergence of the above BD algorithm, the authors find it helpful to add dispatch constraints to the BD master problem (11) at certain iteration when or has improved slowly.
- Choose an inner level convergence tolerance level .
- Since the uncertainty set is assumed to be polyhedral, is a linear program.
- The next theorem shows that the Benders cuts generated by the inner level are valid cuts.
V. COMPUTATIONAL EXPERIMENTS
- The authors present a computational study to evaluate the performance of the adaptive robust approach and the reserve adjustment approach.
- If the authors relax the BD convergence tolerance to and set the MIP gap of the BD master problem to be , the average computation time to solve the robust UC problem significantly decreases to 1.46 h with an average of 0.17% increase in terms of the worst-case total cost.
- The computational results for normally and uniformly distributed loads are similar in illustrating a) and b).
- A. Cost Efficiency and the Choice of the Budget Level Table I reports the average dispatch costs and total costs of AdptRob and ResAdj solutions for normally distributed load when the uncertainty budget varies from 0 to .
- Therefore, a proper level of uncertainty budget in the uncertainty set (5) should be chosen as .
B. Reliability of Dispatch Operation
- The adaptive robust approach also greatly reduces the volatility of the real-time dispatch costs.
- Recall that the dispatch cost is the sum of the production cost and penalty cost.
- The system operator has to take expensive emergency actions such as dispatching fast-start units or load-shedding to maintain system reliability.
- In particular, for different levels of uncertainty budget, Tables I and II characterize the economic efficiency obtained by the robust solutions, whereas Table III shows the risk of the robust solutions in terms of the standard deviations of the cost.
- Using these tables, a proper tradeoff can be made by decision makers.
C. Robustness Against Load Distributions
- In practice, it is not easy to accurately identify the probability distribution of the load uncertainty for each node, especially in a large-scale power system.
- Thus, it is important for a UC solution to have stable economic and operational performance over different distributions of the uncertain load.
- The absolute difference between the two curves is between $6.32 k and $15.80 k for the entire range of .
- The authors also study the effect of the load distribution on the standard deviation of dispatch costs.
- As shown in the table, the relative change of the std is around 18.8% for the AdptRob approach, and is around 59.6% for the ResAdj approach, which is more than three times higher.
D. Cases for Higher Level of Load Variation
- The authors also test the performance of the robust adaptive UCmodel at a higher level of load variation, namely .
- Again, the authors can see that the ResAdj approach has extremely high std on dispatch costs at high reserve adjustment levels due to high levels of violations.
- Comparing to the previous results of times higher std (see Table III), this demonstrates that the AdptRob approach has an even more significant reduction in cost volatility at higher level of demand variation.
- Table VIII lists the penalty costs of the two approaches, where the penalty costs of the AdptRob approach are on average less than 0.06% of the dispatch costs, while the numbers are more than two orders of magnitude higher for the ResAdj approach (from 5.14% to 13.98% for , and as high as 61.74% for ).
- To be compact, the authors omit the plots and tables, but summarize the key statistics.
VI. CONCLUSION AND DISCUSSION
- The adaptive robust model and its solution technique presented in this paper provide a novel and practical approach to handle uncertainties in the unit commitment process.
- The second stage dispatch solution is adaptive to not only the demand uncertainty, but also generator outage uncertainty.
- How to build the associated uncertainty set and how to deal with the more complicated robust counterpart are important questions in this direction.
- Equation (20) is the transmission line constraint for the th contingency where transmission line is tripped.
- Equation (21) is the constraint that the sum of the production output and the reserve should be within the upper and lower bounds for each generator.
Did you find this useful? Give us your feedback
Citations
1,010 citations
Cites background or methods from "Adaptive Robust Optimization for th..."
...This is the first presentation of this cutting plane algorithm with the column-and-constraint generation strategy in a general setup and the first systematic comparison of its computational performance with the Benders-dual cutting plane method....
[...]
...These algorithms gradually construct the value function of the first-stage decisions using dual solutions of the second-stage decision problems [18, 19, 10, 14, 12]....
[...]
...However, two-stage RO models are very difficult to compute....
[...]
742 citations
718 citations
Cites background from "Adaptive Robust Optimization for th..."
...Without DSM, robust scheduling problems with penalty-based costs for uncertain supply and demand have been investigated in [6]....
[...]
519 citations
Cites background or methods from "Adaptive Robust Optimization for th..."
...In the power system literature, RUCmodels have been used to address uncertainties mainly from nodal net electricity injection [29], wind power availability [28], [35], power systems component contingencies [32]–[34], and demand-side management [87]....
[...]
...Outer approximation algorithms have been used to solve this bilinear program when is assumed to be uncertain within a polyhedral set [29]....
[...]
...contingency models [32], [33]; the total allowed number of wind output cases can be varied [35]; the linear coefficients of the polyhedral sets also can be varied, such as in [28] and [29]....
[...]
...For wind output uncertainty, polyhedral or ellipsoidal constraints can be used instead [29]....
[...]
470 citations
Cites background from "Adaptive Robust Optimization for th..."
...There have been many publications that show the value of RO in many fields of application including finance (Lobo 2000), energy (Bertsimas et al. 2013b, Babonneau et al. 2010), supply chain (Ben-Tal et al. 2005, Lim 2013), healthcare (Fredriksson et al. 2011), engineering (Ben-Tal and Nemirovski…...
[...]
...Bertsimas et al. (2013a) show how to construct uncertainty sets based on historical data and statistical tests....
[...]
References
[...]
139,059 citations
"Adaptive Robust Optimization for th..." refers background in this paper
...The deterministic SCUC problem is extensively studied in the power system literature [22], [23]....
[...]
3,364 citations
3,359 citations
2,780 citations
"Adaptive Robust Optimization for th..." refers methods in this paper
...A nonconvex cost function can also be approximated by a nonconvex piecewise linear function with binary variable techniques as shown in [27]....
[...]
2,501 citations
"Adaptive Robust Optimization for th..." refers background in this paper
...Robust optimization has recently gained substantial popularity as a modeling framework for optimization under parameter uncertainty, led by the work in [12]–[18]....
[...]
Related Papers (5)
Frequently Asked Questions (13)
Q2. What future works have the authors mentioned in the paper "Adaptive robust optimization for the security constrained unit commitment problem" ?
Many interesting directions are open for future research. For example, it would be interesting to study re-commitment that is adaptive to load forecast. It would be interesting to study the extent that the volatility in the energy price is reduced. The authors can easily adjust the parameters such as in the uncertainty set and re-run their model for future re-commitment when a better estimation of uncertainty is available.
Q3. What is the purpose of the proposed robust model?
The proposed robust model includes basic reserve requirement in order to cover generator contingencies, since generator contingency is not considered in the uncertainty model.
Q4. What is the variable production cost in a market setting?
The variable production cost, or the supply curve in a market setting, is an increasing convex piece-wise linear function of the production output .
Q5. What are the types of reserve that are available to the system operator?
Other types of reserves exist, such as regulation service (automatic generation control) which responds to frequency changes in the system second by second, and supplement reserve.
Q6. What is the framework of the proposed solution methodology?
The framework of the proposed solution methodology, especially the outer approximation technique to solve the second-stage problem, is not restricted to the budgeted uncertainty set and can be applied to general polyhedral uncertainty sets.
Q7. What is the effect of the adaptive robust approach on the system?
In conclusion, the low volatility of the dispatch cost and the zero penalty cost of the adaptive robust approach demonstratesits operational effectiveness in reducing costly emergency actions and improving system reliability.
Q8. What is the way to solve the unit commitment problem?
In the current practice, the SFT runs iteratively with the unit commitment procedure by gradually adding violated transmission security constraints to the economic dispatch problem.
Q9. What is the reserve capacity of a generator?
The reserve capacity will be available to the system operator in the real-time operation to prepare for unexpected loss of generators or other system disruptions.
Q10. What is the cost of a system operator's emergency actions?
The system operator has to take expensive emergency actions such as dispatching fast-start units or load-shedding to maintain system reliability.
Q11. what is the set of reserve products needed to satisfy the reserve requirement?
The set of reserve products needed to satisfy reserve requirement :, , .• : Reserve capacity of generator , requirement , time .• : System reserve requirement of , time .
Q12. How can a nonconvex cost function be approximated?
A nonconvex cost function can also be approximated by a nonconvex piecewise linear function with binary variable techniques as shown in [27].
Q13. What is the uncertainty set of nodal net injection at each time period in the planning ?
The authors consider the following uncertainty set of nodal net injection at each time period in the planning horizon :(5)where is the set of nodes that have uncertain injections, is the number of such nodes, is the vector of uncertain net injections at time , is the nominal value of the net injection of node at time , is the deviation from the nominal net injection value of node at time , the interval is the range of the uncertain , and the inequality in (5) controls the total deviation of all injections from their nominal values at time .