A data-driven stochastic-distributed bar low-carbon scheduling method for integrated energy systems in industrial parks

Abstract

The present invention discloses a random-distributed robust low-carbon scheduling method for a park integrated energy system based on data-driven. Taking into account the randomness of photovoltaic output and carbon emission benefits in the park, the scheduling cost of the park integrated energy system is minimized. The present invention first screens similar days for photovoltaic historical scenes according to the day-ahead photovoltaic forecast data, and classifies similar scenes through fuzzy C-means clustering. On this basis, a random-distributed robust two-layer scheduling model is constructed. The lower layer adopts a distributed robust optimization method to identify the probability distribution of the highest operating cost of scenes in each category, and generates representative scenes of each category; the upper layer adopts random planning to minimize the expected cost of representative scenes in each category, and obtains the optimal scheduling strategy for the park integrated energy system. The model reduces the conservatism of traditional robust optimization methods, improves the economy of system operation, and proposes a new method for scheduling the park integrated energy system taking into account the randomness of photovoltaic output and carbon emission benefits.

Description

Technical Field

[0001] The present invention belongs to the field of energy scheduling, and in particular to a random-distributed robust low-carbon scheduling method for a park integrated energy system based on data-driven. Background Art

[0002] The park's comprehensive energy system is a park-level energy management system that integrates multiple energy sources such as electricity, heat, and gas. Through various energy coupling equipment, such as cogeneration units, electric boilers, and P2G units, it achieves comprehensive and optimized utilization of multiple energy sources within the park, improves energy utilization efficiency within the park, and increases the proportion of new energy consumption. In the context of widespread access to distributed photovoltaics and the strategic goals of "carbon peak" and "carbon neutrality", the volatility and intermittency of photovoltaic output pose a huge threat to the reliable operation of the park's comprehensive energy system. How to achieve safe, economical, and green operation of the park's comprehensive energy system is a key issue currently faced. Summary of the invention

[0003] In order to solve the above problems, the present invention takes into account the randomness of photovoltaic output and carbon emission costs, and provides a data-driven random-distributed robust low-carbon scheduling method for the park integrated energy system, which reduces the conservatism of traditional robust optimization methods and improves the economy of system operation, thereby realizing safe, economical and green operation of the park integrated energy system.

[0004] The technical solution adopted by the present invention is as follows: a random-distributed robust low-carbon scheduling method for a park integrated energy system based on data-driven, comprising the following steps: a random-distributed robust low-carbon scheduling method for a park integrated energy system based on data-driven, comprising the following steps:

[0005] Step 1: Obtain comprehensive energy system parameters, historical photovoltaic output data and load demand data;

[0006] Step 2: Screen similar days based on PV forecast data and classify similar scenes through fuzzy C-means clustering;

[0007] Step 3: Based on similar photovoltaic output scenarios, a two-stage scheduling model of the park comprehensive energy system, day-ahead and day-intraday, is constructed. The model is a two-layer planning model.

[0008] Step 4: The upper model uses stochastic programming to minimize the expected cost of various representative scenarios to obtain the day-ahead scheduling strategy and scheduling cost C1;

[0009] Step 5: The lower-level model uses distributed robust optimization, brings in the day-ahead scheduling strategy, identifies the probability distribution of the highest operating cost in the most types of scenarios, generates representative scenarios of each type, and solves to obtain the intraday scheduling strategy and scheduling cost C2;

[0010] Step 6: Determine whether the model has converged, that is, whether |C2-C1| is less than the preset value. If not, return to step 4, return the representative scenario generated by the lower-level optimization to the upper-level random programming, and continue iterative optimization; if converged, output the optimal scheduling strategy for the park's integrated energy system.

[0011] Furthermore, in step 4, a day-ahead dispatch model for the integrated energy system of the park taking into account the carbon emission cost and the randomness of photovoltaic output is constructed to minimize the comprehensive operating cost shown in equation (1.1); the model constraints include the energy hub equipment operation constraints shown in equation (1.2) and the controllable load dispatch constraints shown in equation (1.3);

[0012]

[0013]

[0014]

[0015] In formula (1.1), TC is the total dispatch cost, f1 is the day-ahead dispatch cost, f2 is the intraday dispatch cost, and λ t and β t is the time-of-use electricity price of the upper-level power grid unit and the time-of-use gas price of the upper-level gas grid unit, c e is the carbon emission price per unit of electricity, c g is the unit carbon emission price of natural gas; P t,sub and Q t,g Respectively represent the real-time electricity and gas purchases from the superior power grid and gas grid; The cost coefficients for controllable hydrogen load, electricity load, and gas load are increased respectively; They are the cost coefficients for controllable hydrogen load, electrical load, and thermal load reductions, respectively; The power is increased for hydrogen load, electrical load and thermal load respectively; Respectively represent the power reduction of hydrogen load, electric load, and thermal load; in formula (1.2), P t,PV Represents the photovoltaic power generation in the park, P t,LG , P t,LE and P t,LH are the hydrogen load, electricity load and heat load requirements respectively; η P2G , η EB , η TF , η GB and describe the energy conversion efficiency of power-to-gas unit, electric boiler, transformer and gas boiler respectively; η ch and η ce are the heat conversion efficiency and gas conversion efficiency of the combined heat and power (CHP) unit, P t.e1 , P t,e2 and P t,e3are the power injected into P2G, EB and transformer respectively; Q t,g1 and Q t,g2 are the gas volumes injected into the CHP unit and GB unit respectively; and are the upper limits of electricity and gas purchase from the upper power grid and gas grid, respectively. In formula (1.3), They are the maximum values ​​of controllable hydrogen load, electric load and thermal load respectively; are the maximum values ​​under controllable hydrogen load, electrical load and thermal load respectively; P t,LG0 , P t,LE0 and P t,LH0 They are basic hydrogen load, electricity load and heat load demand; t,LG ,u t,LE ,u t,LH They represent the corresponding controllable load upward adjustment state variables when there is excess energy; v t,LG ,v t,LE ,v t,LH They represent the corresponding controllable load down-regulation state variables when there is energy shortage;

[0016] In formula (1.2), formulas 1-3 represent the conversion efficiency constraints of each device, formulas 4 and 5 represent the balance constraints of electricity and gas respectively, and formula 6 is the upper and lower limit constraints on the amount of electricity and gas purchased from the superior power grid. In formula (1.3), formulas 1-3 are the gas, electricity, and heat load response balance constraints; formula 4 is the response state constraint; and formulas 5-7 are the upper and lower limit constraints on load response.

[0017] Furthermore, in steps 4 and 5, the randomness of the photovoltaic output in the park is taken into account to construct a random + distributed robust two-layer scheduling model. The upper layer uses random planning to minimize the expected cost of the representative scenarios in each category. The representative scenarios of each category come from the distributed robust optimization of the lower layer. The upper layer random planning expression is shown in formula (1.4); the lower layer robust optimization identifies the probability distribution of the highest operating cost of the scenarios in each category based on the day-ahead scheduling strategy obtained by the upper layer random planning, obtains the intraday scheduling strategy, and generates representative scenarios. The lower layer distributed robust optimization form is shown in formula (1.5):

[0018] Upper level stochastic programming model:

[0019]

[0020] The lower layer is a robust optimization model:

[0021]

[0022] In the above formula, ρ k is the probability of the kth category, K is the maximum value of the scene category number; f 2,kis the intraday dispatch cost of the kth typical scenario; A, B, c represent various coefficients; x represents the purchased electricity and gas volume, that is, the day-ahead dispatch strategy; Represents the intermediate day-ahead dispatch variables of the park's integrated energy system, including P t.e1 , P t,e2 , P t,e3 , where superscript i is the i-th iteration and l is the number of iterations; is the expected PV output under the probability distribution with the highest operating cost of the kth category in the i-th iteration; is the intraday dispatch cost corresponding to the expected photovoltaic output scenario under the probability distribution with the highest operating cost of the kth category in the i-th iteration; Ω k is the probability distribution set with the highest running cost for the kth class; is the probability of the s-th scene sample of the k-th category; is the PV output of the kth category and the sth scenario; is the intraday dispatch cost corresponding to the s-th photovoltaic output scenario in the k-th category; q k,s is the probability of the sth scene sample of the kth category, which is a planned variable; F k The norm constraint of the uncertain probability distribution constructed in each class, y k,s is the intermediate intraday scheduling variable of the s-th scenario of the k-th category;

[0023] State variables in the upper stochastic programming model (1.4) To find the optimal solution for the lower-level planning model (1.5), split it into the following main and sub-problems and solve them iteratively:

[0024] Main question:

[0025]

[0026] Sub-questions:

[0027]

[0028] Where: x ＊ The optimal day-ahead scheduling strategy obtained for the main problem, where the superscript * represents the conversion of variables into constants; The result obtained is C1 in step 4; the sub-problem aims to solve the probability distribution of the highest operating cost corresponding to the day-ahead scheduling strategy of the main problem;

[0029] Since there is no coupling between the inner and outer optimizations of subproblem (1.7), the subproblems are further decomposed. Subproblem 1 is shown in formula (1.8), and the goal is to solve the optimal intraday scheduling cost for each similar scenario. Subproblem 2 is shown in formula (1.9), and the goal is to solve the probability distribution of the highest corresponding operating cost for each type of scenario.

[0030] Sub-question 1:

[0031]

[0032] Sub-question 2:

[0033]

[0034] In the above formula: N k is the number of scenes in each category; is θ1 a 1-norm constraint; θ ∞ is the ∞ norm constraint; is the initial probability of the s-th scenario in the k-th category of sub-problem 2; The intraday scheduling cost returned by subproblem 1 for subproblem 2;

[0035] The steps to solve the model are as follows:

[0036] Step 1) Initialization: Set the number of iterations l = 1, C1 = -∞, C2 = +∞,

[0037] Step 2) Solve the main problem (1.6) and obtain the current optimal day-ahead scheduling decision x ＊ , update the model target value

[0038] Step 3) Solve subproblems (1.8) and (1.9) to obtain the probability distribution with the highest running cost And add it to the main problem to update the model target value

[0039] Step 4) If |C2-C1|≥10 -3 , return to step 2), otherwise, the iteration ends and the optimal scheduling strategy of the park's integrated energy system is output. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of the method of the present invention;

[0041] Figure 2 This is the energy hub structure framework diagram. DETAILED DESCRIPTION

[0042] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention to further illustrate the present invention. Obviously, the described embodiments are only partial embodiments of the present invention and are not intended to limit the scope of implementation of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative work belong to the protection scope of the present invention.

[0043] like Figure 1As shown, the present invention proposes a data-driven random-distributed robust low-carbon scheduling method for a park integrated energy system, the method comprising the following steps:

[0044] Step 1: Obtain comprehensive energy system parameters, historical photovoltaic output data and load demand data;

[0045] Step 2: Screen similar days based on PV forecast data and classify similar scenes through fuzzy C-means clustering;

[0046] Step 3: Based on similar photovoltaic output scenarios, a two-stage scheduling model of the park comprehensive energy system, day-ahead and day-intraday, is constructed. The model is a two-layer planning model.

[0047] Step 4: The upper model uses stochastic programming to minimize the expected cost of various representative scenarios to obtain the day-ahead scheduling strategy and scheduling cost C1;

[0048] Step 5: The lower-level model uses distributed robust optimization, brings in the day-ahead scheduling strategy, identifies the probability distribution of the highest operating cost in the most types of scenarios, generates representative scenarios of each type, and solves to obtain the intraday scheduling strategy and scheduling cost C2;

[0049] Step 6: Determine whether the model has converged, that is, whether |C2-C1| is less than the preset value. If not, return to step 4, return the representative scenario generated by the lower-level optimization to the upper-level random programming, and continue iterative optimization; if converged, output the optimal scheduling strategy for the park's integrated energy system.

[0050] In step 4, an optimal day-ahead dispatch model for the integrated energy system of the park taking into account the carbon emission cost and the randomness of photovoltaic output is constructed to minimize the comprehensive operating cost shown in formula (1.1); the model constraints include the energy hub equipment operation constraints shown in formula (1.2) and the controllable load dispatch constraints shown in formula (1.3);

[0051]

[0052]

[0053]

[0054] In formula (1.1), TC is the total dispatch cost, f1 is the day-ahead dispatch cost, f2 is the intraday dispatch cost, and λ t and β t is the time-of-use electricity price of the upper-level power grid unit and the time-of-use gas price of the upper-level gas grid unit, c e is the carbon emission price per unit of electricity, c g is the unit carbon emission price of natural gas; P t,sub and Q t,g Respectively represent the real-time electricity and gas purchases from the superior power grid and gas grid; The cost coefficients for controllable hydrogen load, electricity load, and gas load are increased respectively; They are the cost coefficients for controllable hydrogen load, electrical load, and thermal load reductions, respectively; The power is increased for hydrogen load, electrical load and thermal load respectively; Respectively represent the power reduction of hydrogen load, electric load, and thermal load; in formula (1.2), P t,PV Represents the photovoltaic power generation in the park, P t,LG , P t,LE and P t,LH are the hydrogen load, electricity load and heat load requirements respectively; η P2G , η EB , η TF , η GB and describe the energy conversion efficiency of power-to-gas unit, electric boiler, transformer and gas boiler respectively; η ch and η ce are the heat conversion efficiency and gas conversion efficiency of the combined heat and power (CHP) unit, P t.e1 , P t,e2 and P t,e3 are the power injected into P2G, EB and transformer respectively; Q t,g1 and Q t,g2 are the gas volumes injected into the CHP unit and GB unit respectively; and are the upper limits of electricity and gas purchase from the upper power grid and gas grid, respectively. In formula (1.3), They are the maximum values ​​of controllable hydrogen load, electric load and thermal load respectively; are the maximum values ​​under controllable hydrogen load, electrical load and thermal load respectively; P t,LG0 , P t,LE0 and P t,LH0 They are basic hydrogen load, electricity load and heat load demand; t,LG ,u t,LE ,u t,LH They represent the corresponding controllable load upward adjustment state variables when there is excess energy; v t,LG ,v t,LE ,v t,LH They represent the corresponding controllable load down-regulation state variables when there is energy shortage;

[0055] In formula (1.2), formulas 1-3 represent the conversion efficiency constraints of each device, formulas 4 and 5 represent the balance constraints of electricity and gas respectively, and formula 6 is the upper and lower limit constraints on the amount of electricity and gas purchased from the superior power grid. In formula (1.3), formulas 1-3 are the gas, electricity, and heat load response balance constraints; formula 4 is the response state constraint; and formulas 5-7 are the upper and lower limit constraints on load response.

[0056] In steps 4 and 5, a random + robust double-layer scheduling model is constructed taking into account the randomness of the photovoltaic output in the park. The upper layer uses random planning to minimize the expected cost of the representative scenarios in each category. The representative scenarios of each category come from the robust optimization of the lower layer. The upper layer random planning expression is shown in formula (1.4); the lower layer robust optimization identifies the probability distribution of the highest operating cost of the scenarios in each category based on the day-ahead scheduling strategy obtained by the upper layer random planning, obtains the intraday scheduling strategy, and generates representative scenarios. The robust optimization form of the lower layer is shown in formula (1.5):

[0057] Upper level stochastic programming model:

[0058]

[0059] The lower layer is a robust optimization model:

[0060]

[0061] In the above formula, ρ k is the probability of the kth category, K is the maximum value of the scene category number; f 2,k is the intraday dispatch cost of the kth typical scenario; A, B, c represent various coefficients; x represents the purchased electricity and gas volume, that is, the day-ahead dispatch strategy; Represents the intermediate day-ahead dispatch variables of the park's integrated energy system, including P t.e1 , P t,e2 , P t,e3 , where superscript i is the i-th iteration and l is the number of iterations; is the expected PV output under the probability distribution with the highest operating cost of the kth category in the i-th iteration; is the intraday dispatch cost corresponding to the expected photovoltaic output scenario under the probability distribution with the highest operating cost of the kth category in the i-th iteration; Ω k is the probability distribution set with the highest running cost for the kth class; is the probability of the s-th scene sample of the k-th category; is the PV output of the kth category and the sth scenario; is the intraday dispatch cost corresponding to the s-th photovoltaic output scenario in the k-th category; q k,s is the probability of the sth scene sample of the kth category, which is a planned variable; F k The norm constraint of the uncertain probability distribution constructed in each class, y k,s is the intermediate intraday scheduling variable of the s-th scenario of the k-th category;

[0062] State variables in the upper stochastic programming model (1.4) To find the optimal solution for the lower-level planning model (1.5), split it into the following main and sub-problems and solve them iteratively:

[0063] Main question:

[0064]

[0065] Sub-questions:

[0066]

[0067] Where: x ＊ The optimal day-ahead scheduling strategy obtained for the main problem, where the superscript * represents the conversion of variables into constants; The result obtained is C1 in step 4; the sub-problem aims to solve the probability distribution of the highest operating cost corresponding to the day-ahead scheduling strategy of the main problem;

[0068] Since there is no coupling between the inner and outer optimizations of subproblem (1.7), the subproblems are further decomposed. Subproblem 1 is shown in formula (1.8), and the goal is to solve the optimal intraday scheduling cost for each similar scenario. Subproblem 2 is shown in formula (1.9), and the goal is to solve the probability distribution of the highest corresponding operating cost for each type of scenario.

[0069] Sub-question 1:

[0070]

[0071] Sub-question 2:

[0072]

[0073] In the above formula: N k is the number of scenes in each category; is θ1 a 1-norm constraint; θ ∞ is the ∞ norm constraint; is the initial probability of the s-th scenario in the k-th category of sub-problem 2; The intraday scheduling cost returned by subproblem 1 for subproblem 2;

[0074] The steps to solve the model are as follows:

[0075] Step 1) Initialization: Set the number of iterations l = 1, C1 = -∞, C2 = +∞,

[0076] Step 2) Solve the main problem (1.6) and obtain the current optimal day-ahead scheduling decision x ＊ , update the model target value

[0077] Step 3) Solve subproblems (1.8) and (1.9) to obtain the probability distribution with the highest running cost And add it to the main problem to update the model target value

[0078] Step 4) If |C2-C1|≥10-3 , return to step 2), otherwise, the iteration ends and the optimal scheduling strategy of the park's integrated energy system is output.

[0079] Example Analysis

[0080] According to the 13MW photovoltaic unit output data of a certain area in California, the United States in 2006 and the day-ahead photovoltaic forecast data provided by NREL, 117 similar photovoltaic output scenarios were screened and then converted into 0.5MW / PV output data for application in this embodiment; the time-of-use electricity price and load demand are shown in Table 1; the remaining parameters are shown in Table 2.

[0081] Table 1 Load demand and electricity price parameters

[0082]

[0083]

[0084] Table 2 Other system parameters

[0085]

[0086] In this embodiment, we compare the method proposed in the present invention with simple random programming and distributed robust optimization, and the comparison results are shown in Table 3. As can be seen from Table 3, compared with distributed robust optimization, the method proposed in this article is less conservative and more economical; compared with the random optimization method, the method proposed in this article is more conservative and the system operation is more reliable. Therefore, in the case of fluctuations in photovoltaic output, the present invention is conducive to achieving reliable and economical operation of the park integrated energy system.

[0087] Table 3 Comparison of costs of different methods

[0088]

Claims

1. A data-driven random-distributed robust low-carbon scheduling method for a park integrated energy system, characterized in that: The method comprises the following steps: Step 1: Obtain comprehensive energy system parameters, historical photovoltaic output data and load demand data; Step 2: Screen similar days based on PV forecast data and classify similar scenes through fuzzy C-means clustering; Step 3: Based on similar photovoltaic output scenarios, a two-stage scheduling model of the park comprehensive energy system, day-ahead and day-intraday, is constructed. The model is a two-layer planning model. Step 4: The upper model uses stochastic programming to minimize the expected cost of various representative scenarios to obtain the day-ahead scheduling strategy and scheduling cost C1; Step 5: The lower-level model uses distributed robust optimization, introduces the day-ahead scheduling strategy, identifies the probability distribution of the highest operating cost in various scenarios, generates representative scenarios of various types, and solves to obtain the intraday scheduling strategy and scheduling cost C2; Step 6: Determine whether the model has converged, that is, whether |C2-C1| is less than the preset value. If not, return to step 4, return the representative scenario generated by the lower-level optimization to the upper-level random programming, and continue iterative optimization; if converged, output the optimal scheduling strategy for the park's integrated energy system; In step 4, an optimal day-ahead dispatch model for the integrated energy system of the park taking into account the carbon emission cost and the randomness of photovoltaic output is constructed to minimize the comprehensive operating cost shown in formula (1.1); the model constraints include the energy hub equipment operation constraints shown in formula (1.2) and the controllable load dispatch constraints shown in formula (1.3); In formula (1.1), TC is the total dispatch cost, f1 is the day-ahead dispatch cost, f2 is the intraday dispatch cost, and λ t and β t is the time-of-use electricity price of the upper-level power grid unit and the time-of-use gas price of the upper-level gas grid unit, c e is the carbon emission price per unit of electricity, c g is the unit carbon emission price of natural gas; P t,sub and Q t,g Respectively represent the real-time electricity and gas purchases from the superior power grid and gas grid; The cost coefficients for controllable hydrogen load, electricity load, and gas load are increased respectively; They are the cost coefficients for controllable hydrogen load, electricity load, and gas load reductions, respectively; The power is increased for hydrogen load, electrical load and thermal load respectively; Respectively represent the power reduction of hydrogen load, electric load, and thermal load; in formula (1.2), P t,PV Represents the photovoltaic power generation in the park, P t,LG , P t,LE and P t,LH are the hydrogen load, electricity load and heat load requirements respectively; η P2G , η EB , η TF , η GB and describe the energy conversion efficiency of power-to-gas unit, electric boiler, transformer and gas boiler respectively; η ch and η ce are the heat conversion efficiency and gas conversion efficiency of the combined heat and power (CHP) unit, P t.e1 , P t,e2 and P t,e3 are the power injected into P2G, EB and transformer respectively; Q t,g1 and Q t,g2 are the gas volumes injected into the CHP unit and GB unit respectively; and are the upper limits of electricity and gas purchase from the upper power grid and gas grid, respectively. In formula (1.3), They are the maximum values ​​of controllable hydrogen load, electric load and thermal load respectively; are the maximum values ​​of controllable hydrogen load, electric load and thermal load respectively; P t,LG0 , P t,LE0 and P t,LH0 They are basic hydrogen load, electricity load and heat load demand; t,LG ,u t,LE ,u t,LH They represent the corresponding controllable load upward adjustment state variables when there is excess energy; v t,LG ,v t,LE ,v t,LH They represent the corresponding controllable load down-regulation state variables when there is energy shortage; In formula (1.2), formulas 1-3 represent the conversion efficiency constraints of each device, formulas 4 and 5 represent the balance constraints of electricity and gas respectively, and formula 6 is the upper and lower limit constraints of the amount of electricity and gas purchased from the superior power grid. In formula (1.3), formulas 1-3 are the gas, electricity, and heat load response balance constraints; formula 4 is the response state constraint; formulas 5-7 are the upper and lower limit constraints of load response; In steps 4 and 5, a random + robust double-layer scheduling model is constructed taking into account the randomness of the photovoltaic output in the park. The upper layer uses random planning to minimize the expected cost of the representative scenarios in each category. The representative scenarios of each category come from the robust optimization of the lower layer. The upper layer random planning expression is shown in formula (1.4); the lower layer robust optimization identifies the probability distribution of the highest operating cost of the scenarios in each category based on the day-ahead scheduling strategy obtained by the upper layer random planning, obtains the intraday scheduling strategy, and generates representative scenarios. The robust optimization form of the lower layer is shown in formula (1.5): Upper level stochastic programming model: The lower layer is a robust optimization model: In the above formula, ρ k is the probability of the kth category, K is the maximum value of the scene category number; f 2,k is the intraday dispatch cost of the kth typical scenario; A, B, c represent various coefficients; x represents the purchased electricity and gas volume, that is, the day-ahead dispatch strategy; Represents the intermediate day-ahead dispatch variables of the park’s integrated energy system, including P t.e1 , P t,e2 , P t,e3 , where superscript i is the i-th iteration and l is the number of iterations; is the expected PV output under the probability distribution with the highest operating cost of the kth category in the i-th iteration; is the intraday dispatch cost corresponding to the expected photovoltaic output scenario under the probability distribution with the highest operating cost of the kth category in the i-th iteration; Ω k is the probability distribution set with the highest running cost for the kth class; is the probability of the sth scene sample of the kth category; is the PV output of the kth category and the sth scenario; is the intraday dispatch cost corresponding to the s-th photovoltaic output scenario in the k-th category; q k,s is the probability of the sth scene sample of the kth category, which is a planned variable; F k The norm constraint of the uncertain probability distribution constructed in each class, y k,s is the intermediate intraday scheduling variable of the s-th scenario of the k-th category; State variables in the upper stochastic programming model (1.4) To find the optimal solution for the lower-level planning model (1.5), split it into the following main and sub-problems and solve them iteratively: Main question: Sub-questions: Where: x * The optimal day-ahead scheduling strategy obtained for the main problem, where the superscript * represents the conversion of variables into constants; The result obtained is C1 in step 4; the sub-problem aims to solve the probability distribution of the highest operating cost corresponding to the day-ahead scheduling strategy of the main problem; Since there is no coupling between the inner and outer optimizations of subproblem (1.7), the subproblems are further decomposed. Subproblem 1 is shown in formula (1.8), and the goal is to solve the optimal intraday scheduling cost for each similar scenario. Subproblem 2 is shown in formula (1.9), and the goal is to solve the probability distribution of the highest corresponding operating cost for each type of scenario. Sub-question 1: Sub-question 2: In the above formula: N k is the number of scenes in each category; is θ1 a 1-norm constraint; θ ∞ is the ∞ norm constraint; is the initial probability of the s-th scenario in the k-th category of sub-problem 2; The intraday scheduling cost returned by subproblem 1 for subproblem 2; The steps to solve the model are as follows: Step 1) Initialization: Set the number of iterations l = 1, C1 = -∞, C2 = +∞, Step 2) Solve the main problem (1.6) and obtain the current optimal day-ahead scheduling decision x * , update the model target value Step 3) Solve subproblems (1.8) and (1.9) to obtain the probability distribution with the highest running cost And add it to the main problem to update the model target value Step 4) If |C2-C1|≥10 -3 , return to step 2), otherwise, the iteration ends and the optimal scheduling strategy of the park's integrated energy system is output.

Images