Energy storage power station joint market optimal regulation and control method and system

Abstract

The invention provides an energy storage power station joint market optimal regulation and control method and system, and is applied to the field of power system optimal scheduling. The method comprises the steps of constructing a double-layer optimization model, merging a lower-layer model into an upper-layer model according to the Carlo demand-Kuhn-Tuck condition, carrying out linear processing to form a mixed integer linear programming model, solving and obtaining an optimal strategy, and regulating and controlling energy storage. The income is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system automation and power market transaction, and particularly relates to a joint market optimal regulation method and system of energy storage power station. BACKGROUND

[0002] With the increasing penetration of renewable energy sources such as wind power and photovoltaic power in the power grid, the volatility and uncertainty of the power system are increasing. Independent energy storage power stations, as a flexible regulation resource, play a role in stabilizing new energy fluctuations and ensuring the safe and stable operation of the power grid. To realize their commercial value, independent energy storage power stations need to actively participate in the power market and obtain revenue by arbitraging charging and discharging in different markets such as the energy market and the frequency regulation / reserve auxiliary service market.

[0003] Most of the existing declaration strategies take maximizing expected revenue as a single objective. However, the power market price, especially the spot market affected by new energy output, has a dramatic volatility and "fat tail" distribution characteristics, and extreme high / low price events occur from time to time. A strategy that only focuses on expected revenue may result in transaction losses for energy storage power stations in some extreme scenarios, threatening their assets. Existing risk management methods such as simple reserve are too rough to be quantified and balanced.

[0004] Most models simplify energy storage power stations as "price acceptors" of the market, that is, they assume that their own bidding behavior will not affect the final market clearing price. For large energy storage power stations or in local power grids, this assumption does not hold. The declaration behavior of energy storage actually changes the supply and demand balance of the market, thereby affecting the clearing price. Decision-making models that ignore this game relationship are short-sighted, and the strategies derived from them are often suboptimal in reality, and may even lead to decreased revenue due to incorrect market reaction prediction.

[0005] Although some advanced models that can describe the market game relationship have been proposed in the academic field, such as the Bilevel Optimization model, such models are usually NP-hard (Non-deterministic Polynomial-time hard) problems in mathematics, with non-linear and non-convex characteristics and extremely high computational complexity. Within the limited decision-making time required for day-ahead market declaration, general algorithms cannot obtain the global optimal solution. This makes it difficult for advanced theoretical models to be transformed into stable, efficient, and feasible practical tools in engineering.

[0006] Therefore, there is an urgent need for a technical solution that can accurately quantify and actively manage market price risk, scientifically depict the game relationship between energy storage and the market, make more forward-looking strategic decisions, and be efficient and feasible in calculation, and can be quickly and reliably deployed and executed in engineering practice.

[0007] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present application, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0008] Therefore, the present application provides a kind of energy storage power station joint market optimal regulation method and system, to solve the problem of risk management lack, short-sighted decision model and theory and application disconnection when energy storage power station participates in joint market regulation in prior art, by constructing the double-layer optimization model that can reflect market game relationship and linear processing, to obtain the optimal declaration strategy of risk and benefit balance and engineering practical value.

[0009] The present application provides an energy storage power station joint market optimal regulation method, comprising:

[0010] Based on the physical operation model of energy storage power station and the operation rules of power market, a double-layer optimization model comprising an upper optimization model and a lower optimization model is constructed, wherein the objective function of the upper optimization model is the risk-adjusted income of the energy storage power station, and the lower optimization model is a market clearing model;

[0011] The Karush-Kuhn-Tucker optimality condition of the lower optimization model is incorporated into the upper optimization model as an equivalent constraint to convert the double-layer optimization model into a single-layer optimization model;

[0012] The nonlinear product term generated by the complementary relaxation constraint of the Karush-Kuhn-Tucker optimality condition in the single-layer optimization model is linearized by using the big M method to form a mixed integer linear programming model;

[0013] The mixed integer linear programming model is solved to obtain the optimal declaration strategy of the energy storage power station, and control instructions are generated according to the optimal declaration strategy to regulate the physical charging and discharging process of the energy storage power station.

[0014] Further, the physical operation model of the energy storage power station at least includes a dynamic evolution equation for describing the change of the state of charge of the energy storage power station over time.

[0015] Further, the physical operation model of the energy storage power station further includes a power constraint equation for limiting the upper and lower limits of the charging and discharging power of the energy storage power station, and a binary variable constraint for representing the mutual exclusion of charging and discharging states.

[0016] Furthermore, the risk-adjusted return is determined by weighting the expected return of the energy storage power station with the conditional risk value, which characterizes its risk loss.

[0017] Furthermore, the weighted sum includes a risk aversion coefficient to characterize the risk preferences of energy storage operators.

[0018] Furthermore, the operating rules of the electricity market include at least a power flow constraint model of the power grid, and the power flow constraint model of the power grid is linearized using the power transmission distribution factor.

[0019] Furthermore, before incorporating the Caro-Kuhn-Tucker optimality condition as an equivalent constraint into the upper-level optimization model, it also includes obtaining a set of typical scenarios representing market uncertainty and constructing a two-level optimization model under the set of typical scenarios.

[0020] Furthermore, after linearization using the Big M method, the process also includes linearizing the bilinear term in the objective function of the upper-level optimization model, which is generated by the coupling of the bid price as a decision variable and the winning bid amount as an intermediate variable.

[0021] Furthermore, the linearization of the bilinear terms is achieved by segmenting the declared prices into discrete segments and introducing binary variables to represent the intervals in which the declared prices fall.

[0022] Furthermore, the optimal application strategy includes energy storage power stations applying for a combination of electricity volume and application price for the electricity market, frequency regulation ancillary service market, and standby ancillary service market.

[0023] Furthermore, the objective function of the lower-level optimization model is to minimize the total social cost of electricity purchase.

[0024] Furthermore, the Caro-Kuhn-Tucker optimality conditions include primal feasibility constraints, dual feasibility constraints, and complementary relaxation constraints.

[0025] Furthermore, the Big M method specifically transforms nonlinear product terms of the form A*λ=0 into linear constraints by introducing binary auxiliary variables δ and constant M: A≤M*δ and λ≤M*(1-δ).

[0026] Furthermore, the steps for solving the mixed-integer linear programming model are completed within a preset decision time limit using a commercial optimization solver.

[0027] Furthermore, the step of generating control commands based on the optimal reporting strategy also includes sending the control commands to the battery management system or energy management system of the energy storage power station.

[0028] This invention provides a joint market-optimized control system for energy storage power stations, comprising:

[0029] The model building module is used to construct a two-layer optimization model, which includes an upper-layer optimization model and a lower-layer optimization model, based on the physical operation model of the energy storage power station and the operation rules of the electricity market. The objective function of the upper-layer optimization model is the risk-adjusted return of the energy storage power station, and the lower-layer optimization model is a market clearing model.

[0030] The model transformation module, connected to the model building module, is used to incorporate the Caro-Kuhn-Tucker optimality conditions of the lower-level optimization model as equivalent constraints into the upper-level optimization model, so as to transform the two-level optimization model into a single-level optimization model.

[0031] The linearization module, connected to the model transformation module, is used to linearize the nonlinear product terms generated by the complementary relaxation constraints of the Caro-Kuhn-Tucker optimality conditions in the single-layer optimization model using the Big M method, so as to form a mixed-integer linear programming model.

[0032] The strategy solving and control module, connected to the linearization processing module, is used to solve the mixed-integer linear programming model to obtain the optimal application strategy for the energy storage power station, and generate control commands based on the optimal application strategy to send to the energy storage power station.

[0033] Furthermore, the linearization processing module is also configured as follows:

[0034] The bilinear term generated by the coupling of the bid price as a decision variable and the winning bid volume as an intermediate variable in the objective function of the upper-level optimization model is linearized.

[0035] Furthermore, the linearization module is specifically configured as follows when performing linearization on bilinear terms:

[0036] By segmenting and discretizing the declared prices and introducing binary variables to represent the intervals in which the declared prices fall, the bilinear terms can be linearized.

[0037] Furthermore, the model building module is specifically configured as follows when building the upper-level optimization model:

[0038] Risk-adjusted returns are determined by weighting and summing the expected returns of an energy storage power station with the conditional value of risk, which characterizes its risk losses.

[0039] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention.

[0040] The energy storage power station joint market optimal control method and system of the present invention has the following beneficial effects:

[0041] By constructing a two-layer optimization model, this invention can quantify market price risk, allowing operators to adjust their risk aversion coefficient, balance expected returns with extreme losses, and improve the robustness of asset operations. Through a master-slave game model, this invention predicts the impact of energy storage application behavior on market clearing, formulates more strategic bidding strategies, avoids the short-sightedness of the "price taker" model, and captures potential returns. This invention, through Caro-Kuhn-Tucker condition transformation and linearization techniques, transforms an NP-hard problem into an efficiently solvable mixed-integer linear programming model, solving the problem of theoretical limitations and making risk management and game theory decision-making practical in engineering. Attached Figure Description

[0042] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0043] Figure 1 This is a flowchart of an embodiment of the optimal regulation method for energy storage power stations in conjunction with the market.

[0044] Figure 2 This is a schematic diagram of the structure of an energy storage power station joint market optimal control system according to an embodiment of the present invention. Detailed Implementation

[0045] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0046] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0047] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined. Therefore, the actual execution order may change depending on the specific circumstances.

[0048] The role of energy storage power stations in electricity market regulation lies in optimizing their charging and discharging behavior to maximize economic benefits. The electricity market operates on the principle of supply and demand balance, with the market clearing price determined jointly by both supply and demand sides. The charging and discharging behavior of energy storage power stations directly impacts the supply and demand relationship in the electricity market, thus affecting the market clearing price. As market participants, energy storage power stations must fully consider the impact of their own behavior on the market when formulating their optimal regulation strategies. Mathematical optimization theory provides an effective tool for solving such problems. By establishing optimization models, the relationship between the charging and discharging decisions of energy storage power stations and market returns can be quantified. Two-level optimization models can be used to describe hierarchical decision problems, where the upper-level model represents the decisions of the energy storage power station, and the lower-level model represents the market clearing process. The Caro-Kuhn-Tucker (KKT) conditions are necessary conditions for solving constrained optimization problems, transforming the lower-level optimization problem into equivalent constraints. Mixed-integer linear programming (MILP) is a commonly used optimization model with mature solution algorithms. By transforming complex optimization problems into MILP models, commercial solvers can be used to efficiently obtain the global optimum. Therefore, by constructing a two-level optimization model to describe the game relationship between energy storage power stations and the market, and using KKT condition transformation and linearization techniques to transform the model into a MILP problem, an executable optimal control strategy can be provided for energy storage power stations.

[0049] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for optimal joint market control of energy storage power stations, which is implemented through the following steps:

[0050] Step S100: Construct a two-layer optimization model.

[0051] In this step, the first step is to establish a physical operation model of the energy storage power station. This model describes the energy conversion and constraints of the energy storage power station during charging and discharging. In one embodiment, the model includes the dynamic equation of the state of charge (SOC) of the energy storage power station, which describes how the SOC changes with time and charging / discharging power. For example, the SOC at time t can be expressed as the SOC at time t-1 plus the current charging amount minus the discharging amount. The charging / discharging efficiency can be set to 0.95. To further simulate the operational constraints of the energy storage power station, the model can also include upper and lower limits for charging and discharging power, which limit the maximum charging and discharging power of the energy storage power station. This charging / discharging power constraint can be achieved through inequalities, such as charging power less than or equal to the maximum charging power, and discharging power less than or equal to the maximum discharging power. Furthermore, mutual exclusion constraints for charging and discharging states are also incorporated into the model to prevent the energy storage power station from charging and discharging simultaneously. In one example, binary variables can be introduced to represent the charging and discharging states, and constraints can be added to ensure that only one state is 1. These physical constraints can be adjusted according to the specific model and operating characteristics of the energy storage power station, including but not limited to different types of energy storage systems such as lithium-ion batteries, lead-acid batteries, and flow batteries.

[0052] Secondly, a model of the electricity market's operating rules needs to be established. This model describes the supply and demand relationship and price formation mechanism of the electricity market. In one embodiment, the model can include a power flow constraint model of the power grid, which ensures the safe and stable operation of the grid. This power flow constraint model can be represented using a linearized power transmission distribution factor (PTDF) method. In another embodiment, the model can also include a unified clearing rule for the electricity market, which determines the winning bid volume and price for each market participant. For example, the market clearing rule could be a rule based on marginal cost pricing.

[0053] Based on the physical operation model of the aforementioned energy storage power station and the operational rule model of the electricity market, a two-layer optimization model is constructed, comprising an upper-layer optimization model and a lower-layer optimization model. The objective function of the upper-layer optimization model is the risk-adjusted return of the energy storage power station, while the lower-layer optimization model is a market clearing model. Risk-adjusted return can be quantified in various ways, such as expected return minus Value at Risk (VaR) or Conditional Value at Risk (CVaR). The objective of the market clearing model is to minimize the total social cost of electricity purchase while satisfying grid constraints. This total social cost of electricity purchase can be the sum of the cost of purchasing electricity from conventional generating units and the cost of purchasing electricity from energy storage power stations.

[0054] Step S200: The Caro-Kuhn-Tucker (KKT) optimality condition of the lower-level optimization model is used as an equivalent constraint and incorporated into the upper-level optimization model to transform the two-level optimization model into a single-level optimization model.

[0055] For the lower-level optimization model, its optimal solution must satisfy its KKT optimality conditions. KKT conditions include primal feasibility constraints, dual feasibility constraints, and complementary relaxation constraints. Primitive feasibility constraints are directly derived from the constraints in the lower-level model, such as system power balance constraints and transmission line power flow constraints. Dual feasibility constraints are related to the range of values ​​of the dual variables corresponding to each primal constraint in the lower-level model. Complementary relaxation constraints, for each inequality constraint, require that the product of the relaxation amount of that inequality constraint and its corresponding dual variable be zero.

[0056] These KKT conditions are added as equivalent constraints to the upper-level optimization model, thus transforming the original two-layer optimization model into a single-layer optimization model. For example, the equality constraints describing power balance and the constraints on line power flow are added as constraints to the upper-level model. Through this step, the upper-level and lower-level models, which originally needed to be solved separately, are integrated into a unified optimization problem.

[0057] Step S300: The Big M method is used to linearize the nonlinear product terms generated by the complementary relaxation constraints of the KKT optimality conditions in the single-layer optimization model to form a mixed-integer linear programming model.

[0058] The transformed single-layer model contains nonlinear terms, mainly stemming from the complementary relaxation constraints in the KKT optimality conditions. Complementary relaxation constraints are manifested as the product of two variables equaling zero. To address this issue, this step employs the Big M method for linearization.

[0059] For each complementary relaxation constraint Aλ = 0, a binary auxiliary variable δ (taking the value 0 or 1) and a sufficiently large positive constant M are introduced. The nonlinear constraints are replaced with the following two linear inequality constraints: A ≤ Mδ and λ ≤ M(1-δ). For example, if A represents the power flow of a certain line, and λ represents the dual variable corresponding to the power flow constraint of that line, then the above linearization method ensures that when δ is 0, the power flow A must be 0; and when δ is 1, the dual variable λ must be 0, thus guaranteeing that the complementary relaxation condition holds. This operation is performed on all complementary relaxation terms in the model.

[0060] Step S400: Solve the mixed-integer linear programming model to obtain the optimal application strategy for the energy storage power station, and generate control commands based on the optimal application strategy to regulate the physical charging and discharging process of the energy storage power station.

[0061] After the above steps, the entire model is transformed into a standard mixed-integer linear programming (MILP) model. This MILP model can be solved using commercial optimization solvers (such as Gurobi, CPLEX, and SCIP). These solvers integrate efficient branch-and-bound, cutting-plane, and other algorithms, enabling them to find the global optimum or near-optimal solution to this large-scale MILP problem within a preset computation time (e.g., 1 hour).

[0062] The solution yields a set of optimal decision variables, including the amount of electricity and price the energy storage power station should declare in the electricity market, frequency regulation market, and reserve market for each time period in the next 24 hours. This is the optimal declaration strategy. Next, the system translates this strategy into a specific, time-sequential sequence of control commands (e.g., charging at power P1 at time t1 and discharging at power P2 at time t2), and sends it to the energy management system (EMS) or battery management system (BMS) of the energy storage power station via a communication interface (such as Modbus, DNP3, or other industrial protocols) to guide and regulate its actual physical charging and discharging operations the following day.

[0063] In the above steps, the physical operation model of the energy storage power station, the power market operation rule model, KKT condition transformation, Big M method linearization, and MILP model solving techniques work together in a coordinated manner. By establishing an accurate two-level optimization model, the game relationship between the energy storage power station and the power market can be effectively simulated. Through KKT condition transformation, the complex two-level problem is transformed into a single-level problem, and the solvability of the model is guaranteed by Big M method linearization. Finally, by solving the MILP model, the optimal reporting strategy of the energy storage power station is obtained and transformed into control commands to achieve precise regulation of the energy storage power station. This coordinated approach solves the problems of lack of risk management, short-sighted decision-making models, and low computational efficiency in existing technologies, providing energy storage power stations with a day-ahead market reporting and physical regulation strategy that achieves an optimal balance between risk and return and has practical engineering value.

[0064] The method described in this embodiment can maximize the returns of energy storage power stations in the electricity market and effectively manage market risks while ensuring the safe operation of the power stations. By employing a two-level optimization model and linearization techniques, it can solve the difficult problems of complex optimization and achieve efficient engineering applications. This method can be adjusted according to different market rules and the characteristics of energy storage power stations, and has wide applicability.

[0065] In one specific implementation, based on the above embodiments, the dynamic evolution equation of the state of charge (SOC) of the energy storage power station is first defined more specifically. Specifically, this equation can be expressed in the following discrete-time form:

[0066] E(j,t+1)=E(j,t)+η ch (j,t)P ch (j,t)Δt-(1 / η ds (j,t))P ds (j,t)Δt-decay(j,t)Δt

[0067] in:

[0068] E(j,t) represents the state of charge of energy storage unit j at time t, usually expressed in megawatt-hours (MWh).

[0069] η ch (j,t) represents the charging efficiency of energy storage unit j at time t. Its value is affected by temperature, charging rate, and the degree of battery aging, and typically ranges from 0.9 to 0.98. In some embodiments, this charging efficiency can be modeled as a function of SOC, i.e., η. ch (j,t) = f(E(j,t)). For example, charging efficiency may decrease when the SOC approaches its upper limit.

[0070] P ch (j,t) represents the charging power of energy storage unit j at time t, in megawatts (MW). The charging power is limited by the rated capacity of the charger.

[0071] η ds (j,t) represents the discharge efficiency of energy storage unit j at time t. It is also affected by a variety of factors, and the value is usually between 0.9 and 0.98. It can also be modeled as a function of SOC.

[0072] P ds (j,t) represents the discharge power of energy storage unit j at time t, in megawatts (MW). The discharge power is limited by the rated capacity of the inverter.

[0073] Δt represents the time interval, which is usually 1 hour, but smaller time intervals, such as 15 minutes or 5 minutes, can also be used in high-resolution scheduling.

[0074] decay(j,t) represents the self-discharge rate of energy storage unit j at time t, describing the energy loss of the energy storage unit in the idle state, and is expressed in MW / h. The self-discharge rate is affected by ambient temperature and battery type, and is usually a small value, such as 0.001 MW / h. For some types of energy storage devices, such as flow batteries, the self-discharge rate may be relatively high.

[0075] Then, consider the upper and lower limits of the state of charge, i.e.:

[0076] E min (j)≤E(j,t)≤Emax (j)

[0077] Among them, E min (j) and E max (j) represent the minimum and maximum state of charge (SOC) of energy storage unit j, respectively. These values ​​depend on the battery's technical specifications and the operator's operating strategy. For example, to extend battery life, operators may limit the range of SOC usage and avoid deep charging and discharging.

[0078] Next, to ensure the accuracy of SOC dynamic evolution, a temperature model can be introduced to describe the impact of battery temperature on efficiency. Specifically, a thermodynamic model can be used to simulate the heat generation and conduction processes inside the battery, thereby predicting battery temperature. Battery temperature, in turn, affects charge / discharge efficiency and self-discharge rate.

[0079] Finally, this dynamic evolution equation can be used in the upper-level optimization model as a constraint to ensure that the operation of the energy storage power station conforms to physical laws. This equation ensures that the operation of the energy storage device at any given time is constrained by its physical capacity and efficiency, preventing overcharging or over-discharging.

[0080] By accurately describing the dynamic evolution of the state of charge of an energy storage power station, this embodiment can more accurately simulate the behavior of energy storage power stations in the electricity market, thereby optimizing the application strategy of energy storage power stations, improving their economic benefits, and extending their service life.

[0081] In one specific implementation, based on the above embodiments, the physical operation model of the energy storage power station also includes the following constraints to more comprehensively describe the operating characteristics of the energy storage power station.

[0082] First, the power constraint is defined as:

[0083]

[0084] Among them, P ch (j,t) and P ds (j,t) represent the charging and discharging power of energy storage station j at time t, respectively. and These are the maximum charging and discharging power of the energy storage power station j, respectively. These parameters depend on the physical capacity and technical specifications of the energy storage equipment. Their values ​​can be derived from the nameplate parameters or historical operating data of the energy storage equipment and can be adjusted according to the actual operating conditions.

[0085] As and Alternatively, a function related to ambient temperature can be used to dynamically adjust the power limit. For example, when the ambient temperature exceeds a certain threshold, the maximum charge and discharge power of the energy storage system will be reduced accordingly to prevent overheating risks. This function can be linear, piecewise linear, or nonlinear, and can be calibrated using experimental data or simulations. Furthermore, and It can also be a parameter that changes over time to adapt to different operating modes or aging states.

[0086] Specifically, to ensure that the model meets the physical constraint that energy storage power stations cannot charge and discharge simultaneously, binary variable constraints are introduced:

[0087] z ch (j,t)+z ds (j,t)≤1

[0088] Among them, z ch (j,t) and z dS (j,t) are binary variables, representing whether the energy storage power station is in a charging or discharging state at time t, respectively. ch When (j,t)=1, the energy storage power station is in a charging state; when z ds When (j,t)=1, the energy storage power station is in a discharging state; when z ch (j,t)=z ds When (j,t)=0, the energy storage power station neither charges nor discharges. This constraint ensures, through logical judgment, that the energy storage power station can only be in one state at any given time.

[0089] As an alternative to binary variables, other types of discrete variables can be used to represent the charging and discharging state, such as using integer variables to represent different operating modes (charging, discharging, standby, etc.). Alternatively, fuzzy logic variables can be used to represent the charging and discharging state, allowing for a certain degree of fuzziness and uncertainty.

[0090] Through the aforementioned power constraints and mutual exclusion constraints of charge and discharge states, the physical operation model of the energy storage power station can more accurately describe its operating characteristics, improving the accuracy and reliability of the model.

[0091] In one specific implementation, based on the above embodiments, the calculation method for risk-adjusted returns when constructing the upper-level optimization model is as follows: First, calculate the expected return of the energy storage power station. This expected return is the sum of the expected returns of the energy storage power station in different markets (such as the electricity market, frequency regulation market, and standby market). Specifically, for each market, the expected return is obtained by weighted averaging of returns under all possible scenarios, with the weight being the probability of the corresponding scenario occurring. Then, determine the Conditional Value at Risk (CVaR), which characterizes risk loss. CVaR quantifies the tail risk that the energy storage power station may face at a given confidence level α, i.e., the average loss exceeding a certain loss threshold. One method to determine CVaR is to use historical data or Monte Carlo simulations to generate a series of possible market price scenarios and calculate the return distribution of the energy storage power station under these scenarios. Then, set the confidence level α, for example, 95%, and calculate the average loss below the α quantile in the return distribution; this average loss is the CVaR. Next, the expected return and CVaR are weighted and summed to obtain the risk-adjusted return. The choice of weighting coefficient depends on the risk preference of the energy storage operator. For example, if an operator has a high degree of risk aversion, CVaR can be given a larger weight; conversely, if the operator is more focused on expected returns, CVaR can be given a smaller weight. The formula for calculating risk-adjusted return can be expressed as: Risk-adjusted return = Expected return - βCVaR, where β is the risk aversion coefficient. The risk aversion coefficient β typically ranges from 0 to 1, but can be adjusted according to actual needs. In other embodiments, alternative metrics to CVaR can be used to characterize risk losses, such as variance, semi-variance, or maximum drawdown. These metrics can reflect the risks faced by energy storage power stations from different perspectives and can be selected according to specific application scenarios and operator preferences. For example, variance measures the volatility of returns, semi-variance only considers the volatility of returns below the expected value, and maximum drawdown focuses on the maximum possible loss over a period of time.

[0092] By weighting the expected returns of an energy storage power station with the conditional value at risk (CVaR), which characterizes its risk losses, this embodiment can more comprehensively assess the profitability of the energy storage power station and fully consider risk factors when optimizing scheduling strategies, thereby improving the robustness and reliability of energy storage power station operation.

[0093] In one specific implementation, based on the above embodiments, a risk aversion coefficient β is introduced into the objective function of the upper-level optimization model to quantify the energy storage operator's risk preference. Specifically, the objective function is modified as follows:

[0094] maxF upper =E(Profit)-β*CVaR

[0095] Here, β is a non-negative real number whose value is preset by the energy storage operator based on its risk tolerance and business strategy.

[0096] The value of β can be adjusted according to the specific circumstances of the operator. For example, for risk-averse operators, the value of β can be set between 0.5 and 1.0. For risk-neutral operators, the value of β can be set to a value close to 0, such as 0.01. For risk-seeking operators, although this solution is mainly aimed at risk-averse scenarios, the value of β can also be set to a negative number, but this requires careful assessment of its potential risks.

[0097] More specifically, a state-owned power grid company with large-scale energy storage power stations has a lower risk tolerance, so β ​​can be set to 0.7, indicating that while pursuing expected returns, it pays more attention to controlling potential loss risks. Conversely, a private energy storage investment institution focusing on short-term market arbitrage has a higher risk tolerance, so β ​​can be set to 0.2 to pursue higher expected returns. In some embodiments, the risk aversion coefficient β can be time-varying, that is, dynamically adjusted according to market conditions or the operator's financial situation. For example, during periods of high market volatility, β can automatically increase to reduce risk exposure. This dynamic adjustment can be achieved through preset rules or machine learning algorithms.

[0098] The risk aversion coefficient β can be input into the system in various ways. For example, it can be manually input through a human-machine interface, or it can be automatically obtained from external data sources (such as the operator's financial statements or market risk assessment reports). Furthermore, in this embodiment, the risk aversion coefficient β can be simulated using a variable resistor composed of electronic components such as resistors and capacitors, and the value of this resistor is converted into a digital signal and input into the control system to achieve hardware adjustment of the risk aversion coefficient β. Moreover, this resistor can be a linear resistor or a non-linear resistor to provide more flexible risk preference settings.

[0099] In a specific example, an energy storage power station participates in both the day-ahead electricity market and the frequency regulation market. The station operator presets β = 0.6, representing a moderate level of risk aversion. After solving the optimization model, the resulting bidding strategy is more conservative compared to the strategy with β = 0. For example, it reduces the amount of discharge during high-price periods and increases the amount of charging during low-price periods, thereby reducing the risk of losses under extreme price fluctuations.

[0100] By introducing a risk aversion coefficient β, this embodiment enables the optimized scheduling strategy of energy storage power stations to better adapt to the operator's risk preferences, achieving a balance between expected returns and risks, and improving the robustness and security of asset operation.

[0101] In one specific implementation, based on the above embodiments, the operating rules of the electricity market at least include a power flow constraint model of the power grid. First, the network topology and line parameters of the power grid need to be modeled. Then, the power transmission distribution factor (PTDF) is used to approximate the linear relationship between power flow and node injection power. Specifically, for each line l in the power grid, its power transmission distribution factor (PTDF) is defined. l Let N be an N-dimensional vector, where N is the total number of nodes in the power grid. PTDF l The nth element represents the influence coefficient of unit power injected at node n on the active power flow of line l. PTDF can be obtained by simplifying the calculation of the grid admittance matrix. As an alternative, a linearized DC power flow model can be used to directly calculate the line power flow, or an AC power flow model can be used for nonlinear power flow calculation, but the computational complexity is high.

[0102] Then, the power flow constraints of line l are expressed as:

[0103] |PTDF l P injection |≤Flow max (l)

[0104] Among them, P injection It is an N-dimensional vector representing the power injected into each node (positive values ​​indicate injection, negative values ​​indicate outflow), Flow max (l) represents the maximum transmission capacity of line l. The absolute value operation can be linearized into two inequality constraints, namely PTDF. l P injection ≤Flow max (l) and -PTDF l P injection ≤Flow max (l).

[0105] Furthermore, considering the complexity of real-world power systems, this model can be extended to include multiple operating scenarios, such as different load levels or generator output combinations. Each scenario corresponds to a set of power flow constraints to ensure the safe operation of the system under various conditions. The model can also consider the N-1 safety criterion, which states that after any one line fails, the power flow of the remaining lines does not exceed its limit. This can be achieved by adding additional constraints, each corresponding to a possible line fault condition. Equivalent alternatives to this constraint include, but are not limited to, considering the Nk safety criterion, which states that after any k lines fail, the power flow of the remaining lines does not exceed its limit.

[0106] By incorporating the aforementioned power flow constraints into the lower-level market clearing model and using PTDF for linearization, it can be ensured that the market clearing results meet the requirements for safe operation of the power grid.

[0107] This embodiment introduces a power flow constraint model based on PTDF linearization to ensure the security of the electricity market clearing results and improve the reliability of energy storage power stations participating in market transactions.

[0108] In one specific implementation, based on the above embodiments, before constructing the two-layer optimization model, it is first necessary to obtain a set of typical scenarios representing market uncertainty. Specifically, this set of scenarios can be generated through methods such as historical market data analysis, Monte Carlo simulation, or expert experience. For example, historical data of the electricity market over the past five years can be analyzed to extract typical fluctuation patterns of factors such as renewable energy generation output, load demand, and fuel prices, forming a set of market states containing different scenarios. Alternatively, using the Monte Carlo simulation method, based on the statistical characteristics of historical data, a large number of possible market states can be randomly generated, and clustering algorithms can be used to extract the most representative scenarios. This set of scenarios needs to cover all possible market conditions, including normal conditions, extreme high / low price conditions, and various emergencies (e.g., unit failures, line trips, etc.). Then, when constructing the two-layer optimization model, an upper-layer optimization model and a lower-layer optimization model need to be constructed for each scenario in the scenario set, thus forming a set of two-layer optimization models. Different scenarios may cause changes in parameters such as load demand and renewable energy output in the lower-layer market clearing model, thereby affecting the market clearing results and the revenue of energy storage. The upper-level optimization model needs to formulate a comprehensive reporting strategy that maximizes risk-adjusted returns, taking into account all scenarios.

[0109] This implementation method, optimized in multiple typical scenarios and taking into account market uncertainties, can enhance the ability of energy storage power stations to cope with market risks and obtain a more robust application strategy.

[0110] In one specific implementation, based on the above embodiments, after linearizing the complementary relaxation constraints using the Big M method, the implementation further includes linearizing the bilinear term in the objective function of the upper-level optimization model. This bilinear term is generated by the coupling of the bid price of the energy storage power station and the winning bid volume.

[0111] First, a piecewise linearization method can be used to approximate the bid price. Specifically, the continuous bid price variable is discretized into several price intervals, for example, divided into ranges of 0-1000 yuan / MWh with an interval of 1 yuan / MWh. Then, binary variables are introduced to represent the specific price interval into which the bid price of the energy storage power station falls in each time period. For example, if divided into 10 price intervals, 10 binary variables need to be introduced, and the sum of these variables must be 1, indicating that only one price interval can be selected. Next, the bilinear term can be approximated as a piecewise linear function, that is, the center price of each price interval is multiplied by the corresponding winning bid volume. Another implementation is to use a piecewise linear function with an increasing slope to fit a nonlinear cost curve, such as a three-segment linear function. In this case, the cost is represented as the product of the piecewise linear function and the output variable, auxiliary variables are introduced to represent the segmentation points, and multiple linear equations are used to represent the piecewise function, thereby achieving linearization. In another implementation, the bilinear term can also be linearized using the McCormick Envelope method. Specifically, for the bilinear term w = xy, where x and y are the bid price and the winning bid volume, respectively, four new variables w are introduced. L ,w U ,x L ,x U ,y L ,y U Let w, x, and y represent the upper and lower bounds, respectively. The following linear inequality is used to constrain them: w ≥ x L y+xy L -x L y L w = x U y+xy U -x U y U w≥x U y+xy L -x U y L w = x L y+xy U -x L y U .

[0112] By linearizing the bilinear terms in the objective function, the solution efficiency of the model can be further improved and the computational complexity reduced while ensuring the accuracy of the solution. This makes it possible for large-scale energy storage power stations to participate in the joint optimization and regulation of the power market.

[0113] In one specific implementation, based on the above embodiments, the declared price variable of the energy storage power station is first discretized. Specifically, it is assumed that the reasonable range of the declared price is from 0 yuan per megawatt-hour to 2000 yuan per megawatt-hour. This range is evenly divided into 100 price intervals, each with a width of 20 yuan. The endpoint price of each interval is denoted as B. h , where h = 0, 1, ..., 100.

[0114] Then, for each energy storage unit j's declared price in each time period t, a set of binary variables u is introduced. j,t,h , where h ranges from 0 to 100. This binary variable indicates whether the bid price of energy storage unit j in time period t falls within the h-th price interval. To ensure that each bid price corresponds to only one price interval, a constraint is imposed: for each energy storage unit j and each time period t, all u j,t,h The sum of all must equal 1, that is

[0115] Next, the declared price is linearly represented using the aforementioned binary variables. Specifically, assuming the declared price lies within the h-th price interval, the declared price is approximately equal to the left endpoint price B of that interval. h-1 Therefore, the declared price b j,t It can be represented as This expression is a linear expression because it is simply a binary variable u. j,t,h With constant B h-1 The sum of the products of .

[0116] Specifically, if the midpoint price of the interval is used to approximate the declared price, then... Alternatively, a weighted average method can be used, for example, by introducing a continuous variable λ. j,t,h Let ∈[0,1], representing the relative position of the price within the interval h, and satisfying... but Communication between modules can be wired (e.g., via USB, Ethernet) or wireless (e.g., via Wi-Fi, Bluetooth, NFC). The processing unit can be implemented as a central processing unit (CPU), graphics processing unit (GPU), field-programmable gate array (FPGA), or application-specific integrated circuit (ASIC).

[0117] This embodiment achieves effective linearization of the bilinear terms related to the bid price of energy storage power stations by introducing a piecewise discretization method and binary variables, laying the foundation for solving the subsequent mixed integer linear programming model.

[0118] In one specific implementation, based on the above embodiments, the process for generating the optimal application strategy for energy storage power stations participating in multiple electricity markets is as follows:

[0119] First, for the electricity market, based on the predicted next-day market electricity price and the charging and discharging costs of the energy storage station, the charging and discharging revenue under different electricity price levels is calculated. Electricity price prediction can employ time series models (e.g., Autoregressive Moving Average (ARIMA)) or machine learning models (e.g., Support Vector Machine (SVM)). The bid price can be set slightly lower than the expected electricity sales price or slightly higher than the expected electricity purchase price to increase the probability of winning the bid. The bid electricity volume is constrained by the capacity and power of the energy storage station.

[0120] Specifically, for the frequency regulation ancillary services market, it is necessary to evaluate the performance indicators (e.g., regulation rate, regulation accuracy) of energy storage power stations providing frequency regulation services based on historical grid frequency fluctuation data. Then, according to market rules, the frequency regulation revenue at different performance levels is calculated. The declared capacity needs to be determined based on the actual adjustable range and regulation rate of the energy storage power station. The bid price for frequency regulation services can be adjusted based on historical winning bid prices and competitors' quotations.

[0121] Then, for the backup ancillary services market, the reliability of backup services provided by energy storage power stations needs to be assessed based on the grid's backup demand and the response speed of the energy storage power stations. The application for backup capacity needs to consider the available capacity and response time of the energy storage power stations. The application price for backup services can be adjusted based on historical winning bid prices and competitors' quotations.

[0122] Next, the aforementioned bidding strategies in the electricity market, frequency regulation market, and reserve market are combined as input variables for a mixed-integer linear programming model. By solving this model, an optimal bidding strategy that comprehensively considers the returns and risks of multiple markets is obtained. This optimal bidding strategy is specifically reflected in the following for each time period in the next 24 hours: the electricity volume and price bid by the energy storage power station for the electricity market, the capacity and price bid for the frequency regulation market, and the capacity and price bid for the reserve market.

[0123] Then, based on this optimal reporting strategy, corresponding control commands are generated and sent to the energy management system (EMS) of the energy storage power station to precisely regulate its actual physical charging and discharging process the following day. These control commands include: charging or discharging at a specific power within a specific time period; and responding to frequency deviations at a specific adjustment rate when the grid frequency deviates from the normal range. These power commands can be implemented using power electronic converters (e.g., bidirectional DC-DC converters, grid-connected inverters).

[0124] This implementation method enables joint and optimized scheduling of energy storage power stations in multiple electricity markets, maximizing their overall benefits and reducing market risks.

[0125] In one specific implementation, based on the above embodiments, the objective function of the lower-level optimization model is set to minimize the total social electricity purchase cost. Specifically, the total social electricity purchase cost includes the sum of the generation costs of all generating units (conventional units and energy storage power stations) participating in market transactions.

[0126] First, define set I as the set of all generator sets participating in market transactions, where i∈I represents an element in the set, i.e., a generator set. Define set T as the set of scheduling periods, where t∈T represents an element in the set, i.e., a scheduling period.

[0127] Specifically, the total social cost of electricity purchase can be expressed as:

[0128]

[0129] Where: C i (P i,t Let represent the power generation cost function of generator unit i during time period t. This function is related to the active power output P of the generator unit. i,t The function.

[0130] Then, the power generation cost function C i (P i,t () can be a linear function or a piecewise linear function.

[0131] Linear functions: C i (P i,t ) = a i P i,t Where a i The unit power generation cost of generator set i is expressed in yuan / megawatt-hour (¥ / MWh).

[0132] Piecewise linear function: The active power output range of the generator set is divided into multiple intervals, and the power generation cost function is linear within each interval. For example, the cost function for three intervals can be expressed as:

[0133] If P i,t ∈[0,P i,1 ], then C i (P i,t ) = a i,1 P i,t ;

[0134] If P i,t ∈(P i,1 ,P i,2 ], then C i (P i,t) = C i (P i,1 )+a i,2 (P i,t -P i,1 );

[0135] If P i,t ∈(P i,2 ,P i,max ], then C i (P i,t ) = C i (P i,2 )+a i,3 (P i,t -P i,2 ).

[0136] Among them, P i,1 and P i,2 It is the segmentation point, a i,11 a i,2 and a i,3 These represent the unit power generation cost for each segment. The power generation cost function can also use other forms such as quadratic functions; those skilled in the art can choose according to the specific circumstances.

[0137] Next, considering the safety constraints of power grid operation, the lower-level optimization model must satisfy system power balance constraints and transmission line power flow constraints. The system power balance constraint requires that the sum of the power generation of all generating units equals the total system load demand, expressed as: ∑ i∈I P i,t =D t D t This represents the total system load demand at time t. Power flow constraints on transmission lines must ensure that the transmission power of each line does not exceed its maximum transmission capacity. Using a DC power flow model, the power flow calculation formula for line l is: F l,t =B l (θ start,t -θ end,t ), where F k,t For the power flow of line l at time t, B l Let θ be the susceptance of line l. start,t and θ end,t These represent the phase angles of the nodes at both ends of line l. It should be noted that power flow calculations can use either a DC power flow model or an AC power flow model.

[0138] By minimizing the total social cost of electricity purchase while simultaneously satisfying grid security constraints, the lower-level optimization model can simulate the operation of the electricity market, determine the optimal output of each generator unit, and provide market price signals for the upper-level optimization model of the energy storage power station.

[0139] By adopting this embodiment, it is possible to ensure that the lower-level optimization model conducts market clearing with an economic orientation, and the resulting market price signal is closer to the real market supply and demand relationship, thereby improving the accuracy of the calculation of risk-adjusted returns of energy storage power stations.

[0140] In one specific implementation, based on the above embodiments, during the process of transforming the two-layer optimization model into a single-layer optimization model, the Caro-Kuhn-Tucker (KKT) optimality condition is expanded into three components, and constraints are applied to each component separately.

[0141] First, the original feasibility constraints, namely the basic equations describing the physical operation of the power system in the lower-level market clearing model, such as the nodal power balance equations (∑P gen +∑P discharge =∑P load +∑P charge ) and line power flow constraints (|PTDF(P injection -P withdrawal )|≤Flow max The parameters are completely replicated into the single-layer model to ensure that the single-layer model still satisfies the basic physical laws of the power system. These constraints directly limit the range of values ​​for variables such as generator output, energy storage charging and discharging power, and line power flow. Their specific mathematical expressions and parameter meanings have been described in detail in the above embodiments and will not be repeated here.

[0142] Next, dual feasibility constraints are the constraints that the dual variables corresponding to each primal constraint in the lower-level model must satisfy. For example, for the nodal power balance equation, its corresponding dual variable can be interpreted as the marginal electricity price of that node. Dual feasibility constraints are usually manifested as nonnegativity constraints on the dual variables, such as λ≥0, or constraints related to the coefficients of the objective function of the lower-level model. As an example, if the lower-level objective function includes minimizing the generation cost term c... g P g Then the dual feasibility constraint may manifest as the marginal electricity price being greater than or equal to the generation cost (λ≥c). g The specific form of these dual feasibility constraints depends on the type of constraints and the form of the objective function in the lower-level model. The physical meaning of the corresponding dual variables and the dual feasibility constraints that need to be satisfied for different types of constraints used in the lower-level model are well-known to those skilled in the art.

[0143] Then, complementary relaxation constraints are applied, meaning that for each inequality constraint in the lower-level model, the product of its relaxation variable and its corresponding dual variable is zero. Taking line power flow constraints as an example, if the power flow of line l is less than its upper limit Flow... max (l), then the corresponding dual variable λ l It must be zero; otherwise, if λ lIf the value is not zero, it means that line l is fully loaded, and its power flow is equal to Flow. max (l). Complementary slack constraints embody the complementary slackness in economics, meaning that resources, in optimal allocation, are either fully utilized or their price is zero. Complementary slack constraints are typically expressed as nonlinear equations, for example: (Flow max (l)-|PTDF(P injection -P withdrawal )|)λ l = 0. As in the above embodiment, this nonlinear equation can be linearized using the Big M method. Optionally, for power flow constraints, absolute value linearization can be used. For example, an auxiliary variable flow can be introduced. l,pos and flow l,neg , such that |PTDF(P injection -P withdrawal )|=flow l,pos +flow l,neg And add constraint PTDF(P) injection -P withdrawal ) = flow l,pos -flow l,neg flow l,pos ≥0, flow l,neg ≥0. This avoids the nonlinearity caused by absolute values.

[0144] By clearly defining the original feasibility constraints, dual feasibility constraints, and complementary relaxation constraints included in the KKT optimality conditions, and using them as constraints for the single-layer model, it can be ensured that the transformed single-layer model is mathematically equivalent to the original two-layer model, thus guaranteeing the accuracy and effectiveness of the solution results.

[0145] In one specific implementation, the Big M method is applied to the linearization process of the nonlinear product terms generated by the complementary relaxation constraints of the KKT optimality conditions in the above embodiments. First, all nonlinear terms of the form Aλ = 0 are identified in the single-layer model, where A represents the constraint relaxation amount and λ represents the corresponding dual variable, both of which are nonnegative. Specifically, consider an example scenario where A represents the relaxation variable of a power flow constraint for a transmission line, and λ represents the dual variable corresponding to that power flow constraint. The complementary relaxation condition requires that either the line power flow reaches its upper limit or the dual variable is zero.

[0146] Next, for each identified nonlinear term Aλ = 0, a binary auxiliary variable δ is introduced, with δ taking the value of 0 or 1. Simultaneously, a sufficiently large positive constant M is selected. The selection of M must ensure that, in the actual optimization problem, the maximum possible values ​​of both A and λ are less than M. The value of M can be estimated based on the physical parameters of the system (such as line capacity, generator output limit, etc.) or adjusted through multiple iterative tests. For example, if A represents the line power flow slack, and its maximum value will not exceed the line's maximum transmission capacity, then M can be set to a value slightly larger than that capacity.

[0147] Then, the nonlinear constraint Aλ = 0 is replaced with the following two linear inequality constraints: A ≤ Mδ and λ ≤ M(1-δ). For example, in the case of line power flow, these inequality constraints guarantee that if δ = 0, the line power flow slack A must be 0; if δ = 1, the dual variable λ must be 0. These linear constraints guarantee that in any case, at least one of A and λ is zero, thus being completely equivalent to the original complementary slack conditions. This operation is performed for all complementary slack terms in the model.

[0148] In the above process, the binary auxiliary variable δ can be implemented in various ways. For example, it can use the built-in binary variable type supported by the integer programming solver, or it can be simulated using other equivalent logical expressions. The value of the constant M, besides being estimated based on physical parameters, can also be determined using an adaptive adjustment method. That is, during the solution process, the size of M is dynamically adjusted according to the actual values ​​of the variables to improve the solution efficiency and accuracy of the model. The multiplication operations involved in the Big M method can be implemented using standard computer instructions or optimized mathematical library functions. This process can be implemented as a software module, firmware, or hardware circuit.

[0149] Through the above linearization process, the originally nonlinear complementary relaxation constraints are transformed into a set of linear constraints, thereby ensuring that the entire mathematical model can be transformed into a standard mixed integer linear programming (MILP) model, and can be solved efficiently by commercial solvers.

[0150] By introducing binary auxiliary variables and a sufficiently large constant M, and replacing nonlinear product terms with linear inequality constraints, this embodiment provides a concrete method for transforming nonlinear complementary relaxation constraints into linear constraints, thereby ensuring that the bi-level optimization model can be transformed into a mixed-integer linear programming model that can be solved efficiently.

[0151] In one specific implementation, the solution process for the mixed integer linear programming (MILP) model is refined based on the above embodiments.

[0152] First, choose a commercial optimization solver. As examples, Gurobi, CPLEX, XPRESS, and SCIP are available commercial optimization solvers. These solvers integrate branch and bound, cutting plane methods, and various heuristic algorithms.

[0153] Specifically, solver parameters can be configured to balance solution speed and solution quality. For example, the MIPGap parameter can be set to control the optimality gap, which defines the maximum percentage deviation between a feasible solution and the optimal solution. A smaller MIPGap value will increase solution time but will yield a solution closer to the global optimum.

[0154] Next, set the decision time limit. The decision time limit refers to the maximum time limit for the solver to run. If the global optimum is not found within the time limit, the solver will return the currently found feasible solution. This time limit can be adjusted according to the actual application scenario. For example, in day-ahead market reporting, decisions typically need to be completed within several hours, so the time limit can be set to 1 to 3 hours. For scenarios with higher real-time requirements, such as emergency power grid control, the time limit needs to be shortened to the minute level, such as 5 to 15 minutes.

[0155] Then, start the solver. The solver will read the model file and solve it according to the set parameters and time limit. During the solution process, the solver will output iteration information, including the objective function value, gap value, and number of nodes of the current optimal solution.

[0156] Finally, within the preset decision-making time limit, the solver returns the results, including the values ​​of each decision variable, such as the charging and discharging power of the energy storage power station, the declared electricity volume, and the declared price. If the solver finds the global optimal solution within the time limit, it returns the optimal solution; if it does not find the global optimal solution, it returns the current feasible solution and gives a corresponding status prompt, such as "Time limit reached".

[0157] This embodiment ensures the efficiency and controllability of the mixed-integer linear programming model solution process by explicitly specifying the commercial optimization solver and setting the decision time limit, enabling the energy storage power station to obtain a usable optimized scheduling strategy within a limited time.

[0158] In one specific implementation, based on the above embodiments, the process of generating control commands according to the optimal reporting strategy further includes: First, determining the communication protocol type between the control commands and the energy management system (EMS) or battery management system (BMS) of the energy storage power station. This communication protocol can be an industry standard protocol such as Modbus TCP, DNP3, or IEC 61850, or a proprietary protocol provided by the energy storage power station manufacturer. Then, according to the selected communication protocol, the control parameters such as charging / discharging power, voltage, and current derived from the optimal reporting strategy are encoded into data packets conforming to the protocol specifications. Specifically, if the Modbus TCP protocol is used, the control parameters are written to the corresponding register address; if the DNP3 protocol is used, the control parameters are written to the corresponding object. Next, the data packet containing the control parameters is sent to the EMS or BMS of the energy storage power station via a wired (e.g., Ethernet) or wireless (e.g., Wi-Fi, 4G / 5G) communication link. After receiving the data packet, the EMS or BMS decodes the data and converts it into physical control commands for the energy storage power station, such as controlling the switching state of the inverter or adjusting the charging / discharging current. Ultimately, the energy storage power station executes the corresponding charging and discharging operations based on the received control commands. As a fault-tolerant design for communication link interruptions or data transmission errors, the EMS or BMS can be configured to automatically switch to a predefined safe operating mode if no new control commands are received within a preset time, for example, stopping charging and discharging and maintaining the current state of charge.

[0159] By sending control commands directly to the EMS or BMS of the energy storage power station, this embodiment enables precise control of the energy storage power station, improves response speed, and reduces the need for manual intervention, thereby further enhancing the automation level and economic benefits of the energy storage power station's participation in power market regulation.

[0160] like Figure 2 As shown in the figure, this embodiment of the invention provides an optimal control system for energy storage power stations in conjunction with the market.

[0161] The system includes a model building module M100, which constructs a two-layer optimization model based on the physical operation model of the energy storage power station and the operating rules of the electricity market. This model comprises an upper-layer optimization model and a lower-layer optimization model. The physical operation model of the energy storage power station may include, but is not limited to, mathematical equations describing the charging and discharging efficiency, capacity limitations, and energy loss characteristics of the energy storage device. The operating rules of the electricity market, in one embodiment, may be the market rules for the day-ahead energy market, frequency regulation market, and reserve market, including market entry conditions, bidding methods, and clearing mechanisms. The objective function of the upper-layer optimization model is the risk-adjusted return of the energy storage power station. The risk-adjusted return can be quantified using Conditional Value at Risk (CVaR) or other risk metrics. The lower-layer optimization model is a market clearing model that simulates the supply and demand relationship in the electricity market to determine market prices and the winning bid volumes for each market participant, aiming to minimize the total social cost of electricity purchase. The model building module M100 can be implemented using programming languages ​​such as Python, MATLAB, or GAMS, or commercial optimization modeling software.

[0162] The model transformation module M200 is connected to the model building module M100. This module is used to incorporate the Caro-Kuhn-Tucker (KKT) optimality conditions of the lower-level optimization model as equivalent constraints into the upper-level optimization model, thereby transforming the two-level optimization model into a single-level optimization model. KKT conditions include primal feasibility conditions, dual feasibility conditions, and complementary relaxation conditions. The specific implementation of the model transformation module M200 can involve expressing the lower-level optimization problem and its KKT conditions in algebraic form, and then embedding these algebraic expressions into the upper-level optimization model using a programming language (e.g., Python). The connection between modules can be wired (e.g., via shared memory) or wireless (e.g., via TCP / IP protocol).

[0163] The linearization module M300 is connected to the model transformation module M200. This module is used to linearize the nonlinear product terms generated by the complementary relaxation constraints of the KKT optimality conditions in a single-layer optimization model using the Big M method, thus forming a mixed-integer linear programming model. The Big M method is a common technique for transforming nonlinear complementary relaxation conditions into linear constraints. Specifically, it introduces binary variables and a sufficiently large constant M, replacing the original nonlinear constraints with a set of linear inequality constraints. This linearization module M300 can be implemented using Python to call relevant mathematical optimization libraries (such as Pyomo).

[0164] The strategy solving and control module M400 is connected to the linearization processing module M300. This module solves the mixed-integer linear programming model to obtain the optimal application strategy for the energy storage power station and generates control commands based on the optimal application strategy to send to the energy storage power station. Solving the mixed-integer linear programming model can be achieved by calling commercial optimization solvers (such as Gurobi or CPLEX) or open-source solvers (such as SCIP). The optimal application strategy includes the energy storage power station's application volume and application price in various electricity markets. The control commands are timing signals used to control the charging and discharging behavior of the energy storage power station and can be sent to the energy management system (EMS) of the energy storage power station via industrial communication protocols such as Modbus or DNP3. The connection between modules can be wired (e.g., via serial port) or wireless (e.g., via Zigbee).

[0165] The various modules above work together to achieve optimal joint market regulation of energy storage power stations. Specifically, the model building module M100 is responsible for establishing a two-layer optimization model reflecting market game relations; the model transformation module M200 transforms the complex two-layer model into a single-layer model; the linearization module M300 eliminates nonlinear factors in the model, making it solvable efficiently; and the strategy solving and control module M400 solves the model and generates control commands, ultimately achieving optimal operation of the energy storage power station. This collaborative relationship solves technical problems in existing technologies such as lack of risk management, short-sighted decision-making models, and the disconnect between advanced theory and engineering application. It achieves a balance between risk and return, enhances the foresight of market decisions, and ensures the engineering practicality and computational efficiency of the solution.

[0166] The energy storage power station joint market optimal control system provided in this invention can quantify and manage market price risks, formulate more strategic pricing strategies, avoid the short-sightedness and bias of the traditional "price taker" model, solve the bottleneck of the inability of advanced theories to be applied in practice, and enable the routine and automated deployment of complex risk management and game decision-making in actual engineering projects. Thus, it can provide energy storage power station operators with an efficient and reliable decision support tool.

[0167] In one specific implementation, based on the above embodiments, the linearization processing module M300 is further configured to perform linearization processing on bilinear terms. This linearization processing targets the bilinear terms appearing in the objective function of the upper-level optimization model, which are generated by the coupling of the bid price and the winning bid quantity. These bilinear terms represent the revenue obtained by the energy storage power station participating in electricity market transactions, for example, the bid price multiplied by the winning bid quantity. For ease of processing, the linearization processing module M300 includes a price discretization unit, a binary variable introduction unit, and a linear constraint application unit.

[0168] The price discretization unit is configured to discretize a continuous bid price variable within its allowed bid price range. For example, this bid price range could be from the lowest to the highest price allowed by the power grid. The price discretization unit divides this range into multiple discontinuous price intervals, each with a defined upper and lower bound. Alternatively, the price discretization unit can employ a non-uniform discretization method, using denser discretization in areas of greater price volatility and sparser discretization in areas of less volatility.

[0169] The binary variable introduction unit, connected to the price discretization unit, is configured to introduce one binary variable for each price range. This binary variable indicates whether the bid price of the energy storage power station falls within the corresponding price range for that time period. The binary variable takes only two values: 0 or 1. To ensure that the bid price for each time period falls within only one price range, the binary variable introduction unit is also configured to impose a constraint such that the sum of these binary variables equals 1 across all price ranges.

[0170] The linear constraint application unit is connected to the binary variable introduction unit and is configured to transform bilinear terms into linear terms using the introduced binary and auxiliary variables. For example, a piecewise linearization method can be used to transform the bilinear term of the bid price multiplied by the winning bid volume into a series of linear constraints on the binary variable, the winning bid volume, and the auxiliary variable. Alternatively, the linear constraint application unit can also use the McCormick envelope method or other linearization techniques to linearize the bilinear terms.

[0171] With the above configuration, the linearization module M300 can linearize the bilinear terms in the objective function of the upper-level optimization model, thereby enabling the entire optimization model to be transformed into a mixed-integer linear programming (MILP) model and solved using existing commercial solvers.

[0172] This embodiment adds a linearization module M300 to linearize the bilinear terms in the objective function, thereby ensuring that the entire model can be transformed into a standard mixed-integer linear programming model, which is convenient for solving using a commercial optimization solver, thus improving computational efficiency and solution accuracy.

[0173] In one specific implementation, based on the above embodiments, the linearization processing module M300 includes a price discretization unit, a variable introduction unit, and a linearization constraint unit. The price discretization unit is configured to divide the continuous range of declared prices into multiple discrete price intervals. The number of intervals can be adjusted according to actual accuracy requirements, for example, into 10, 20, or more intervals. The variable introduction unit is configured to introduce a binary variable for each price interval, where a value of 1 indicates that the declared price of the energy storage power station falls within that interval, and a value of 0 indicates that it does not fall within that interval. The linearization constraint unit is configured to apply constraints to ensure that at any given time, only one interval's binary variable has a value of 1, meaning that the declared price of the energy storage power station can only fall within one price interval. As an alternative to the price discretization unit, a fuzzy logic method can also be used, treating the declared prices as fuzzy sets belonging to different price intervals, and using a membership function to characterize their degree of belonging. As an alternative to binary variables, multi-valued variables can be used, where each variable represents a price interval, and the value of the variable represents the confidence level of the energy storage power station's declared price within that interval. Communication between modules can be wired (e.g., via USB, Ethernet) or wireless (e.g., via Wi-Fi, Bluetooth, NFC).

[0174] This embodiment achieves linearization of the bilinear terms by segmenting and discretizing the declared prices and introducing binary variables, enabling the mixed-integer linear programming model to be solved efficiently.

[0175] In one specific implementation, based on the above embodiments, the model building module M100 is configured to, in the upper-level optimization model, weight and sum the expected return of the energy storage power station with the conditional value at risk (CVaR), which characterizes its risk loss, to determine the risk-adjusted return. Specifically, the calculation of expected return is based on the revenue forecasts of the energy storage power station participating in the electricity market, frequency regulation market, and reserve market. These forecasts consider factors such as historical market data, future load forecasts, and new energy power generation forecasts. The conditional value at risk (CVaR) is modeled using Monte Carlo simulation or other probabilistic forecasting methods to model the uncertainty of future market prices and the revenue of the energy storage power station, thereby quantifying the tail risk at a given confidence level. The weighting coefficients of the weighted sum include a risk aversion coefficient β, which can be set by the energy storage power station operator according to its own risk preferences. The value of β can be any real number between 0 and 1, or it can be obtained by looking up a pre-defined risk preference level table. The weighted sum can use linear weighting or non-linear weighting, such as exponentially transforming CVaR before weighting, to enhance the risk aversion effect.

[0176] In an alternative approach, Conditional Value at Risk (CVaR) can be replaced with other risk metrics, such as Value at Risk (VaR), Mean Absolute Deviation (MAD), or Semivariance. More complex functional relationships can also be introduced between expected return and risk metrics; for example, a utility function can be introduced to comprehensively consider both return and risk. This utility function could be an exponential utility function, a logarithmic utility function, or a power function.

[0177] Communication between modules can be wired (e.g., via USB, Ethernet) or wireless (e.g., via Wi-Fi, Bluetooth, NFC).

[0178] By weighting the expected returns and risk losses of energy storage power stations, the model building module M100 can more comprehensively assess the operational risks of energy storage power stations and provide a more accurate objective function for the subsequent formulation of the optimal application strategy.

[0179] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. An energy storage power station joint market optimal regulation method, characterized in that, The method comprises the following steps: a double-layer optimization model comprising an upper-layer optimization model and a lower-layer optimization model is constructed based on a physical operation model of an energy storage power station and operation rules of an electricity market, wherein an objective function of the upper-layer optimization model is a risk-adjusted income of the energy storage power station, and the lower-layer optimization model is a market clearing model; Karush-Kuhn-Tucker optimality conditions of the lower-layer optimization model are taken as equivalent constraints and incorporated into the upper-layer optimization model, so as to convert the double-layer optimization model into a single-layer optimization model; a nonlinear product term generated by complementary relaxation constraints of the Karush-Kuhn-Tucker optimality conditions in the single-layer optimization model is linearized by using a big M method, so as to form a mixed integer linear programming model; the mixed integer linear programming model is solved to obtain an optimal bidding strategy of the energy storage power station, and control instructions are generated according to the optimal bidding strategy to regulate and control a physical charging and discharging process of the energy storage power station.

2. The method of claim 1, wherein, The physical operation model of the energy storage power station at least comprises a dynamic evolution equation for describing a change of a state of charge of the energy storage power station over time.

3. The method of claim 2, wherein, The physical operation model of the energy storage power station further comprises a power constraint equation for limiting upper and lower limits of charging and discharging power of the energy storage power station, and a binary variable constraint for representing mutual exclusion of charging and discharging states.

4. The method of claim 1, wherein, The risk-adjusted income is determined by weighted summation of an expected income of the energy storage power station and a conditional value at risk representing a risk loss of the energy storage power station.

5. The method of claim 4, wherein, The weighted summation comprises a risk aversion coefficient for representing a risk preference of an energy storage operator.

6. The method of claim 1, wherein, The operation rules of the electricity market at least comprise a power flow constraint model of a power grid, and the power flow constraint model of the power grid is linearly represented by using a power transmission distribution factor.

7. The method of claim 1, wherein, Before the Karush-Kuhn-Tucker optimality conditions are incorporated into the upper-layer optimization model as the equivalent constraints, a typical scenario set representing market uncertainty is obtained, and the double-layer optimization model is constructed under the typical scenario set.

8. The method of claim 1, wherein, After the linearization by using the big M method, a bilinear term in the objective function of the upper-layer optimization model, which is generated by coupling of a bidding price as a decision variable and a winning electricity quantity as an intermediate variable, is linearized.

9. The method of claim 1, wherein, The optimal bidding strategy comprises combined bidding electricity quantity and bidding price of the energy storage power station for an electricity energy market, a frequency modulation auxiliary service market and a standby auxiliary service market.

10. An energy storage power station joint market optimal regulation system, characterized in that, The method comprises the following steps: a model construction module is configured to construct a double-layer optimization model comprising an upper-layer optimization model and a lower-layer optimization model based on a physical operation model of an energy storage power station and operation rules of an electricity market, wherein an objective function of the upper-layer optimization model is a risk-adjusted income of the energy storage power station, and the lower-layer optimization model is a market clearing model; a model conversion module connected with the model construction module is configured to take Karush-Kuhn-Tucker optimality conditions of the lower-layer optimization model as equivalent constraints and incorporate them into the upper-layer optimization model, so as to convert the double-layer optimization model into a single-layer optimization model; A linearization processing module, connected with the model conversion module, is configured to perform linearization processing on a nonlinear product term generated by a complementary relaxation constraint of the Karush-Kuhn-Tucker optimality condition in the single-layer optimization model by using a big M method, to form a mixed integer linear programming model; A strategy solving and control module, connected with the linearization processing module, is configured to solve the mixed integer linear programming model to obtain an optimal bidding strategy of the energy storage power station, and generate a control instruction according to the optimal bidding strategy to send to the energy storage power station.

Images