Multi-time scale optimization scheduling method for integrated energy system

Abstract

The invention provides a multi-time-scale optimal scheduling method for a comprehensive energy system, which aims to deal with the uncertainty of wind and light power generation, minimize the daily operation cost and ensure the supply and demand balance of multiple energy sources of electricity, heat, cold and gas. The method is based on chance constrained programming (CCP), a two-stage adaptive reserve optimization model (TSARO) is constructed, a quantile generalized addition model (QGAM) and a generalized autoregressive condition heterovariance model (GARCH) are utilized to quantify the nonlinear, time-varying and heterovariance characteristics of wind and light prediction errors, and the reserve capacity is optimized through an intraday dynamic adjustment mechanism. Simulation results show that compared with a deterministic optimization method, a random optimization method, a robust optimization method and a distributed robust optimization method, the method has the advantages that the total scheduling cost is remarkably reduced by about 11.18%, 5.33%, 30.39% and 23.92%, and the economical efficiency and the reliability are effectively balanced under different confidence levels. According to the method, through accurate uncertainty modeling and dynamic adjustment, a theoretical basis is provided for IES optimization scheduling in a high-permeability renewable energy source scene.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of integrated energy systems, and relates to a multi-time scale optimization scheduling method for an integrated energy system. BACKGROUND

[0002] With the growth of global energy demand and the highlighting of low-carbon trend, integrated energy system (IES) as a new energy supply and management mode gradually becomes an important path to promote energy transformation and realize sustainable development. IES integrates various energy forms and conversion technologies, relies on intelligent energy hubs to realize efficient coordination of energy networks, thereby improving overall energy utilization efficiency, meeting diversified user demands, and laying a foundation for building a clean and efficient energy ecosystem. However, with the increasing penetration of renewable energy, its intermittency and uncertainty bring new challenges to the stability and economy of IES operation. Therefore, how to optimize the operation strategy to ensure the reliability of energy supply while effectively responding to uncertainty fluctuations has become the current research focus.

[0003] In the process of energy system transformation towards high efficiency, the research on uncertain optimization scheduling of integrated energy system (IES) has become a key field. Robust optimization, distributed robust optimization, stochastic optimization and interval optimization methods are widely used in the treatment of renewable energy uncertainty. However, there are still some limitations in the modeling of wind and light uncertainty in existing research. The modeling dimension of dynamic uncertainty of wind and light resources is insufficient when using deterministic and robust optimization methods in the design and optimization of hybrid renewable energy systems, which is difficult to adapt to the real-time fluctuation scheduling scenario within a day. The two-stage scheduling method based on distributed robust optimization is limited in strategy regulation flexibility due to insufficient consideration of photovoltaic output uncertainty and insufficient quantification of scheduling risk. The method based on interval mathematics and static Monte Carlo simulation is not deep enough in analyzing the dynamic evolution law of wind and light output probability distribution, which affects the real-time fluctuation response ability. The fuzzy chance constraint programming depends on subjective parameters and lacks a rolling correction mechanism within a day, so its dynamic adaptability needs to be improved.

[0004] There is still optimization space in the existing research in terms of time scale dynamic optimization and real-time prediction error correction. The distributed collaborative strategy based on chance constrained programming has deficiencies in the intra-day dynamic process processing and short-term prediction error correction link, which restricts the real-time response efficiency of the system. The risk constraint multi-objective optimization method is difficult to cope with extreme volatility scenarios due to the narrow preset deviation range and the lack of intra-day dynamic adjustment mechanism. The model-free optimization strategy combines deep reinforcement learning and domain knowledge rules to realize adaptive scheduling of integrated energy systems in parks, optimize multi-dimensional supply and demand balance equations, and reduce high-grade energy waste through energy cascade utilization. However, its strategy is limited by the lack of dynamic adjustment capability of domain knowledge rules in multi-energy coupled scenarios, making it difficult to effectively realize adaptive correction of real-time prediction errors. The multi-unit collaborative optimization scheduling based on improved information gap decision theory targets high-altitude electric-thermal-oxygen integrated systems, optimizes uncertainty weights through entropy weight method and NSGA-II algorithm, quantifies wind and solar and electric / thermal / oxygen load fluctuation range, and realizes low-carbon scheduling and oxygen stable supply, but the single time period optimization mode limits the multi-time scale dynamic scheduling capability. The two-stage scheduling strategy based on stochastic model predictive control does not comprehensively consider photovoltaic output fluctuation, and the prediction error distribution update has time lag, making it difficult to effectively handle dynamic uncertainty in complex scenarios. SUMMARY

[0005] The purpose of the present application is to provide a multi-time scale optimization scheduling method for integrated energy systems, aiming to solve the above problems.

[0006] To solve the above technical problems, the present application provides a multi-time scale optimization scheduling method for integrated energy systems, characterized by the following contents:

[0007] S1, wind and solar power uncertainty estimation based on adaptive reserve optimization model and chance constrained programming

[0008] S1.1, day-ahead reserve capacity estimation of quantile generalized additive model and generalized autoregressive conditional heteroscedasticity model

[0009] S1.1.1, construction of quantile generalized additive model and generalized autoregressive conditional heteroscedasticity model

[0010] The quantile generalized additive model constructs a non-parametric quantile regression framework, embeds time period characteristics, captures error distribution nonlinearity and time-varying heterogeneity through spline function and factor function, adapts wind power asymmetric heavy-tailed and photovoltaic diurnal cycle modulation distribution characteristics, and the generalized autoregressive conditional heteroscedasticity model describes the heteroscedasticity of error sequence, extracts lag square term and previous variance memory term, captures wind power high-frequency fluctuation and photovoltaic inter-day period fluctuation characteristics, and in the collaborative framework, the quantile generalized additive model provides quantile boundary constraints, the generalized autoregressive conditional heteroscedasticity model generates fluctuation compensation term, realizes distribution adaptation and feature extraction;

[0011] S1.1.2. Day-ahead reserve capacity estimation based on dual-model coupling;

[0012] S1.2. Construction of intraday dynamic adjustment mechanism

[0013] The core task of the intraday phase is to ensure that the power system meets opportunity constraints under short-term uncertainties through a dynamic reserve capacity adjustment mechanism. This phase aims to dynamically adjust reserve capacity. The constructed intraday dynamic adjustment mechanism covers the trigger mechanism, adjustment amount calculation, and weight allocation, which complements the day-ahead reserve capacity planning in terms of time and space.

[0014] S2. Two-stage optimization scheduling

[0015] First, the day-ahead scheduling phase is executed. This phase uses a one-hour time resolution to develop a preliminary scheduling plan for the next 24 hours. The optimization process is based on historical forecast error data, the operating costs of various equipment, and day-ahead forecast data. The day-ahead scheduling model is used for calculations. The historical forecast error data is analyzed using a quantile generalized additive model and a generalized autoregressive conditional heteroskedasticity model to determine the day-ahead reserve capacity. The optimization results output includes the start / stop status and output plans of units such as the power grid, gas grid, wind power, photovoltaic power, combined heating and cooling power, gas boilers, power-to-gas, gas turbines, thermal and energy storage.

[0016] Subsequently, the framework enters the intraday rolling scheduling stage. The optimization window in this stage is 4 hours, the time resolution is 15 minutes, and rolling optimization is performed at intervals of 15 minutes. The optimization process is calculated through the rolling scheduling model based on real-time prediction data and errors, the actual scheduling results of the previous period, the day-ahead scheduling plan and short-term prediction data. The real-time prediction error is used to adjust the intraday rolling spare capacity. The optimization results generate a refined scheduling sequence for the next 4 hours, dynamically adjust the start and stop status and output of each unit, and realize precise regulation of system operation.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] First, this paper proposes a multi-timescale scheduling method for integrated energy systems based on chance-constrained programming and two-stage adaptive reserve optimization. This method accurately characterizes the nonlinear, time-varying, and heteroscedastic characteristics of wind and solar forecast errors through the QGAM and GARCH models, providing robust probabilistic constraints for day-ahead reserve capacity configuration. Simulations show that compared with deterministic, stochastic, robust, and distributed robust optimization methods, this method reduces day-ahead total costs by 15.81%, 6.21%, 32.53%, and 22.03%, respectively, and reduces intraday total costs by 6.63%, 4.51%, 28.36%, and 25.52%, significantly improving economic efficiency and renewable energy absorption capacity.

[0019] Secondly, to deal with short-term fluctuations, a dynamic adjustment mechanism is designed for intraday, using real-time error triggering and trend signal fusion strategy to dynamically optimize the reserve capacity with 15-minute resolution and 4-hour optimization window, ensuring coordination with day-ahead plan. The results analysis shows that the multi-device collaborative interaction can effectively smooth the output fluctuation of wind-solar power generation, while optimizing the peak-valley regulation. At the confidence level of 95%, 85%, and 75%, the method shows a small incremental cost to ensure the robustness of system reliability, which is suitable for high penetration rate of renewable energy scenarios.

[0020] Thirdly, through the modeling of multi-energy coupling of electricity, heat, cold, and gas, the method realizes collaborative optimization and enhances system flexibility. Compared with existing methods, the innovation lies in the refinement of uncertainty modeling, the flexibility of dynamic adjustment, and the systematicness of multi-energy collaboration. Simulation verifies the role of energy storage and intraday rolling optimization in improving real-time adaptability, providing a new path and engineering application value for building a clean and efficient energy ecosystem. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is the comprehensive energy system architecture diagram of the present application;

[0022] Figure 2 is the wind turbine output curve diagram of the dispatch day;

[0023] Figure 3 is the solar output curve diagram of the dispatch day;

[0024] Figure 4 is the load curve diagram;

[0025] Figure 5 is the energy storage SOC state diagram;

[0026] Figure 6 is the day-ahead dispatch electrical balance diagram;

[0027] Figure 7 is the day-ahead dispatch cold balance diagram;

[0028] Figure 8 is the day-ahead dispatch heat balance diagram;

[0029] Figure 9 is the day-ahead dispatch gas balance diagram;

[0030] Figure 10 is the intraday electrical balance diagram;

[0031] Figure 11 is the intraday cold balance diagram;

[0032] Figure 12 is the intraday heat balance diagram;

[0033] Figure 13 is the intraday gas balance diagram;

[0034] Figure 14 is a 95% confidence level backup plot;

[0035] Figure 15 is a 85% confidence level backup plot;

[0036] Figure 16 is a 75% confidence level backup. DETAILED DESCRIPTION

[0037] The comprehensive energy system multi-time scale optimal scheduling method of the present application is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present application will be more apparent from the following description. It should be noted that the drawings are in a very simplified form and are not drawn to precise scale, and are merely intended to facilitate the understanding of the embodiments of the present application. The same or similar reference numerals in the drawings represent the same or similar components.

[0038] Embodiment, a comprehensive energy system multi-time scale optimal scheduling method

[0039] 1. Comprehensive energy system architecture and modeling

[0040] 1.1 IES system architecture

[0041] The present application is directed to the IES containing multiple heterogeneous energy types of electricity, heat, cold, and gas, and the internal energy flow is as shown in Figure 1 Each type of energy subsystem in the IES includes energy supply, energy conversion, energy consumption, etc. The energy supply part includes the upper power grid, the upper gas grid, the wind turbine (WT), and the photovoltaic (PV). The energy conversion equipment includes the combined cooling, heating and power (CCHP), the gas boiler (GB), the power to gas (P2G), and the electric cooling (EC). The energy storage part includes the gas storage tank (GST), the thermal storage tank (TST), and the battery energy storage (BES).

[0042] 1.2 System device mathematical model

[0043] 1.2.1 Photovoltaic

[0044] The electrical output power of the PV system is mainly determined by the irradiance intensity on the surface of the PV unit and its physical properties, etc. The input-output relationship of the PV is shown in equation (1).

[0045]

[0046] wherein: Pv(t) is the electrical output power of PV at time t; f PV is the power derating factor of PV, which is used to represent the output power reduction caused by dust and aging, etc., and usually takes the value of 0.9; E PV is the peak capacity of PV; G T is the actual irradiance; G T,ST is the irradiance under standard test conditions, which usually takes the value of 1 kW / m 2 .

[0047] 1.2.2 Wind turbine

[0048] The electrical output power of WT mainly depends on the wind speed. The input-output relationship of WT is shown in equation (2).

[0049]

[0050] wherein: Pwt(t) is the electrical output power of WT at time t; P r is the rated power of WT; v r is the rated wind speed of WT, which takes the value of 13.5 m / s; v in is the cut-in wind speed of WT, which takes the value of 3 m / s; v out is the cut-out wind speed of WT, which takes the value of 25 m / s.

[0051] 2.2.3 Combined cooling, heating and power model

[0052] CCHP produces electrical energy, thermal energy and cooling energy by consuming natural gas. The input-output relationship of CCHP is shown in equation (3).

[0053]

[0054] wherein: Pcchp(t) is the electrical, thermal and cooling output power of CCHP device at time t; H is the natural gas consumption of CCHP device at time t; H gas is the low calorific value of natural gas; υ is the electrical-thermal conversion coefficient; is the upper and lower limit of input CCHP natural gas; η cchp,elec , η cchp,heat are the electrical and thermal efficiency of gas turbine, respectively; η cchp,cold is the cooling output efficiency of absorption refrigeration device, and the variable working condition characteristics of CCHP are ignored in the present application for simplifying the CCHP model.

[0055] 1.2.4 Electrical-to-gas model

[0056] P2G produces gas energy by consuming electricity energy, the input-output relationship is shown in equation (4).

[0057]

[0058] where: P2G(t) is the electricity power input to P2G at time t; Q2G(t) is the gas output of P2G at time t; η P2G η2G is the energy conversion efficiency of P2G; P2Gmin and P2Gmax are the lower and upper limits of the electricity power input to P2G, respectively.

[0059] 1.2.5 Electricity-to-Cooling Model

[0060] EC produces cooling energy by consuming electricity energy, the model is shown in equation (5).

[0061]

[0062] where: C2EC(t) is the cooling power output of EC at time t; P2EC(t) is the electricity power input to EC at time t; η ec,cold η2EC is the energy conversion efficiency of EC; P2ECmin and P2ECmax are the lower and upper limits of the electricity power input to EC, respectively.

[0063] 1.2.6 Energy Storage Model

[0064]

[0065] where: S2i(t) is the remaining capacity of the ith energy storage device at time t; η i η2i is the energy loss coefficient of the ith energy storage device; P2i(t) is the charging / discharging power of the ith energy storage device at time t; η i,char η2i i,dischar η2i is the charging / discharging efficiency of the ith energy storage device, S2imin and S2imax are the lower and upper limits of the energy storage of the ith energy storage device, respectively; S2i and S2i are the charging / discharging states of the ith energy storage device. S2i S2i

[0066] 2.2.7 Gas Boiler Model

[0067] GB produces heating energy by consuming natural gas energy, the model is shown in equation (7).

[0068]

[0069] where: Q2GB(t) is the natural gas input to GB at time t; is the heat energy output by GB at time t; η gb is the energy conversion efficiency of GB; are the upper and lower limits of the natural gas volume input to GB respectively.

[0070] 2 Uncertainty estimation of wind and solar power generation based on TSARO and chance-constrained programming

[0071] In the integrated energy system, the uncertainty of wind and solar power generation affects the stability and economy of the system. The configuration of reserve capacity is constrained by the nonlinear time-varying, heteroscedastic and asymmetric distribution characteristics of wind and solar forecast errors. The study proposed a two-stage adaptive reserve optimization model (TSARO) based on chance constrained programming (CCP). In the day-ahead stage, the quantile generalized additive model (QGAM) and the generalized autoregressive conditional heteroscedasticity model (GARCH) model are used to quantify the distribution and fluctuation of forecast errors, and the reserve capacity is configured based on the chance constraint theory; in the intraday stage, the dynamic triggering and adjustment model is used to achieve real-time optimization of the reserve capacity to ensure the short-term safe operation of the system. The core idea of ​​chance constrained programming is to convert uncertainty constraints into probabilistic form to ensure that the constraints are established with a specified probability. Its general mathematical expression is:

[0072] P(g i (x,ξ)≤0)≥1-α i (8)

[0073] Where: P(·) is the probability function, which measures the degree of satisfaction of the chance constraint; x is the decision variable vector; ξ is the random variable vector; g i (x,ξ) is the constraint function, α i In an integrated energy system, the opportunity constraint can be specifically described as a probabilistic constraint on power balance and reserve capacity requirements, which avoids the risk of power imbalance with a high confidence level:

[0074]

[0075] Where: is the load demand at time t; is the total output of each device in the system at time t; and is the actual output of wind power and photovoltaic power at time t; and are the uplink and downlink spare capacities at time t respectively; α u and α l To transform the above probabilistic constraints into a deterministic optimization problem, it is necessary to model the distribution characteristics of wind and solar forecast errors and combine the QGAM and GARCH methods in the TSARO model to achieve constraint conversion.

[0076] 2.1 Day-Ahead Reserve Capacity Estimation Methods Using QGAM and GARCH

[0077] 2.1.1 QGAM and GARC model construction

[0078] The reserve capacity estimation is achieved through the collaborative modeling of QGAM and GARCH to realize the joint estimation of the conditional quantile and time-varying volatility of wind and solar forecast errors. QGAM constructs a non-parametric quantile regression framework, embeds time period characteristics, captures the nonlinearity and time-varying heterogeneity of error distribution through spline functions and factor functions, and adapts to the asymmetric fat tail of wind power and the diurnal cycle modulation distribution characteristics of photovoltaic power. GARCH characterizes the heteroscedasticity of the error sequence, extracts the lagged square term and the previous variance memory term, and captures the high-frequency fluctuation characteristics of wind power and the daytime period fluctuation characteristics of photovoltaic power. Under the collaborative framework, QGAM provides quantile boundary constraints, and GARCH generates volatility compensation terms to achieve distribution adaptation and feature mining. The forecast error of wind power and photovoltaic power is defined as the difference between the actual output and the day-ahead forecast output:

[0079]

[0080] Where: y(t) is the actual output at time t; is the day-ahead output forecast at time t; e(t) is the forecast error sequence. The conditional quantile function of the error sequence is estimated using QGAM. For the quantile level t∈(0,1), the quantile function can be expressed as:

[0081]

[0082] Where: x(t) is a feature vector, which is composed of timestamp, periodic characteristics and predicted output value, and its specific form is determined according to the actual application scenario; f j (·,t) is a smooth basis function, and spline function is used for continuous features. The model parameters are optimized by minimizing the penalty quantile loss function. Under the chance constraint framework, it is assumed that the conditional cumulative distribution function of the error e(t) is F e(t)|x(t) (·), then the quantile function q t (x(t)) satisfies:

[0083] F e(t)|x(t) (q t (x(t)))=t (13)

[0084] The uplink and downlink reserve constraints combined with the opportunity constraints can be transformed into:

[0085]

[0086] Combining formula (13), we can get:

[0087]

[0088] Then we can further get:

[0089]

[0090] From equations (14)-(16), the lower bound of reserve capacity can be determined directly by quantile function. To capture the heteroscedasticity of error series, the conditional volatility is estimated by GARCH(1, 1) model, which can be expressed as:

[0091]

[0092] wherein: is the conditional variance at time t, and parameters are estimated by maximizing the log-likelihood function:

[0093]

[0094] From equation (18), ω, α and β can be determined, and satisfy ω>0, α≥0, β≥0 and α+β≥0. Then, using the estimated parameters, the future conditional variance is calculated recursively:

[0095]

[0096] The conditional standard deviation is:

[0097]

[0098] 2.1.2 Day-ahead reserve capacity estimation method based on dual model coupling

[0099] Combining equations (16) and (17), the wind power day-ahead up-bound reserve capacity and down-bound reserve capacity are:

[0100]

[0101] wherein: and are the quantile estimation values corresponding to the confidence level; k u (k u ≥1) is the conservative adjustment coefficient for enhancing the safety margin of reserve capacity; σ t is the volatility compensation term, which should compensate for the uncertainty in high volatility scenarios, determined by GARCH(1, 1) model; ζE[|e|] term is the absolute value of error expectation compensation term, which improves the robustness of negative reserve.

[0102] Considering the zero output characteristics of photovoltaic at night, the non-daytime period reserve capacity is set to zero, then the photovoltaic up-bound reserve capacity and down-bound reserve capacity are:

[0103]

[0104] where D is a set of daytime periods defined dynamically based on the time of light.

[0105] 2.2 Intra-day dynamic adjustment mechanism construction

[0106] The core task of the intra-day stage is to ensure that the power system meets the opportunity constraint condition in the short-term uncertainty environment through the dynamic reserve capacity adjustment mechanism. This stage aims to dynamically adjust the reserve capacity, and the constructed intra-day dynamic adjustment mechanism covers triggering mechanism, adjustment amount calculation, and weight allocation, etc. It is complementary to the day-ahead reserve capacity planning in time and space scale. The specific design is as follows:

[0107] 2.2.1 Trigger mechanism design

[0108] The trigger mechanism defines the trigger index I through a two-step method t , which realizes the accurate capture and anti-interference control of the adjustment signal. First, based on the real-time error e t , the trend signal s t and the dynamic thresholdò t , the candidate index I is constructed to preliminarily determine the adjustment direction:

[0109]

[0110] where: |(·) is the indicator function, which takes 1 when the condition is true, and 0 otherwise; A t is used to trigger the uplink reserve adjustment; B t is used to trigger the downlink reserve adjustment, and the specific model is:

[0111]

[0112] where: y t (e t ,s t ) is a fusion function that fuses the error and the trend signal at the previous time, and the expression is shown in equation (26); e t-1 is the real-time error at time t-1, which represents the power deviation at the previous time; s t is the trend signal, which reflects the short-term change trend of the error, and the expression is shown in equation (27);ò t is the dynamic threshold, which is adaptively adjusted according to the system state, and the expression is shown in equation (28).

[0113]

[0114] where:ò t / 2 is the trend item weight coefficient, which is dynamically adjusted according to the system volatility s tcontribution of the trend term. In high volatility scenarios, the influence of the trend term is weakened; in low volatility scenarios, the influence of the trend term is enhanced. t / ò t is the standardized trend signal, which is obtained by embeddingò t The fusion function realizes adaptive evaluation and balances the error and trend contributions.

[0115]

[0116] where m is the window length; n t is the local volatility of the historical data statistical range, defined as the mean square error of the error within the window relative to its mean value:

[0117]

[0118] where: is the mean value of the error within the window. To avoid unstable operation caused by frequent switching, a stability constraint condition is introduced:

[0119] C s = {λ t > λ0 ∧ |e t-1 |≤ 2ò t} (30)

[0120] where λ0 is the maximum switching frequency threshold, and the switching frequency λ t is defined as:

[0121] λ t = |(I t ≠ I t-1 )·|(I t ≠ 0)·|(I t-1 ≠ 0) (31)

[0122] where I t , I t-1 are the final trigger indicators at time t and t-1, used for switching frequency calculation. Finally, the trigger indicator I t is combined with the stability constraint optimization as shown in equation (32).

[0123]

[0124] Equation (32) balances the responsiveness and operational stability of the triggering mechanism by limiting high-frequency switching, providing a reliable basis for adjustment quantity calculation.

[0125] 2.2.2 Adjustment quantity calculation model

[0126] When the trigger indicator I tWhen ≠ 0, the adjustment amount calculation module is started to meet the opportunity constraint condition and ensure the reliability of power balance, and the expressions are shown in equations (33) and (34).

[0127]

[0128] In the formula, ω and ω are the uplink and downlink reserve adjustment amounts, respectively. The adjustment amount is calculated according to the direction of I t . For wind power, and the calculation model is equation (35).

[0129]

[0130] For photovoltaic, the uplink and downlink reserve adjustment amount calculation models are equations (36).

[0131]

[0132] In the formula, ω e is the direct error weight, reflecting the contribution of error to the adjustment amount; ω ν is the volatility weight, reflecting the contribution of local volatility to the adjustment amount; ω s is the trend weight, reflecting the contribution weight of short-term trend to the adjustment amount, ω e , ω ν , and ω s are shown in equation (39). γ w and γ p are the wind power and photovoltaic trend adjustment parameters, respectively. The adjustment amount is smoothed by functions φ + and φ - :

[0133]

[0134] The weights ω e , ω ν , and ω s are calculated by normalization to ensure the balance of each component of the adjustment amount, and are defined as:

[0135]

[0136] In the formula, ω , ω

[0137] , and ω are the weights of the direct error, volatility, and trend, respectively.

[0138] The weight models of Eqs. (38)-(39) dynamically adjust the importance of each component according to local volatility, ensuring that the adjustment quantity is adapted to the system state. The above method constructs a two-stage adaptive reserve optimization model (TSARO) coupled with a chance-constrained programming (CCP) framework, realizing the organic combination of wind and solar prediction error condition distribution modeling and reserve capacity optimization. The framework includes a joint modeling method for error condition quantile and volatility in the day-ahead stage, a dynamic trigger adjustment mechanism in the day-ahead stage, forming a multi-time scale optimization system, and achieving the trade-off between system operation safety and economy. By introducing a time period effectiveness judgment mechanism, the pertinence of reserve configuration of different energy types is improved.

[0139] 3 Two-stage optimization scheduling

[0140] In constructing the day-ahead and day-ahead optimization models, multiple factors in system operation need to be considered to ensure that the scheduling scheme is both economical and efficient and stable and reliable. Through day-ahead scheduling, a preliminary day-ahead economic operation plan is developed, and real-time correction of the plan is performed using day-ahead rolling scheduling to adapt to prediction errors and dynamic changes in system state, thereby improving the economy and reliability of system operation.

[0141] The two-stage optimization scheduling framework first performs the day-ahead scheduling stage, i.e., the first stage. This stage develops a preliminary scheduling plan for the next 24 hours with a time resolution of 1 hour. The optimization process is based on historical prediction error data, the operating costs of various devices, and day-ahead prediction data, and is calculated through a day-ahead scheduling model. The objective function is Eqs. (40), and the constraint conditions are Eqs. (1)-(7) and Eqs. (41)-(44). The historical prediction error data are used to determine the day-ahead reserve capacity through QGAM and GARCH models. The optimization results include the start-stop state and output plan of the grid, gas network, wind power, photovoltaic, combined cooling heating and power, gas boiler, electric-to-gas, gas turbine, thermal storage, and energy storage units. Subsequently, the framework enters the day-ahead rolling scheduling stage, i.e., the second stage. The optimization window is 4 hours, the time resolution is 15 minutes, and rolling optimization is performed at 15-minute intervals. The optimization process is based on real-time prediction data and errors, the actual scheduling results of the previous period, day-ahead scheduling plans, and short-term prediction data, and is calculated through a rolling scheduling model. The objective function is Eqs. (45), and the constraint conditions are Eqs. (1)-(7), Eqs. (41), Eqs. (43), Eqs. (44), and Eqs. (46). Real-time prediction errors are used to adjust the day-ahead rolling reserve capacity. The optimization results generate a refined scheduling sequence for the next 4 hours, dynamically adjusting the start-stop state and output of each unit to achieve accurate regulation of system operation.

[0142] 3.1 Day-ahead optimization scheduling

[0143] 3.1.1 Objective function

[0144] The day-ahead optimization model focuses on equipment operation cost, start-stop cost, standby capacity cost, and penalty for curtailment of wind and solar power. While meeting the load demand, it maximizes the use of renewable energy and reserves sufficient standby resources to cope with uncertainties. The day-ahead optimization objective function is shown in equation (40).

[0145]

[0146] In the formula: D run is the set of equipment considering operation cost; c d is the operation cost coefficient of equipment d; is the power output of equipment d at time t; D start is the set of equipment considering start-stop cost; is the start-stop cost coefficient of equipment d; is the start flag of equipment d at time t; c res is the cost coefficient of standby; c cur is the cost coefficient of wind and solar power curtailment penalty; and P s t are the available wind and photovoltaic power outputs at time t, and are the actual wind and photovoltaic power outputs at time t.

[0147] 3.1.2 Constraint conditions

[0148] 1) Power balance constraint

[0149] The balance of electricity, heat, cold, and gas energy flow in IES is shown in equation (41).

[0150]

[0151] In the formula: are the electric, gas, heat, and cold loads at time t, respectively; are the battery charging and discharging capacities at time t; are the gas tank charging and discharging capacities at time t; are the heat storage tank charging and discharging capacities at time t; are the upper-level electricity and gas purchase capacities at time t, respectively.

[0152] 2) Energy conversion equipment constraint

[0153]

[0154] In the formula: L d,max , L d,min are the maximum and minimum load rates of equipment d, respectively; respectively, are the upper and lower limits of the rated power of device d at time t; is a binary variable of the device operation state, device d is put into operation at and is taken out of operation at .

[0155] 3) Reserve constraint

[0156]

[0157] wherein: respectively, are the uplink and downlink reserve amounts provided by device d at time t; respectively, are the total uplink and downlink reserve amounts required by the system at time t, which are calculated by formula (16).

[0158] 4) Grid constraint

[0159] In order to ensure stable operation of the system, and due to the limitations of the transmission network, the interactive power of the IES with the upper-level power grid and the upper-level gas grid should be within a reasonable range:

[0160]

[0161] wherein: respectively, are the uplink and downlink flow limits of the IES purchasing electricity and gas from the upper-level power grid and the upper-level gas grid at time t.

[0162] 3.2 Intraday rolling optimization scheduling

[0163] 3.2.1 Objective function

[0164] The intraday rolling optimization model implements dynamic adjustment on the day-ahead scheduling plan in the real-time operation stage, and its operation mechanism is based on the rolling optimization strategy. In order to coordinate the fluctuation response in the short-term operation and the continuity of the day-ahead plan, by constructing an objective function containing the penalty term, the consistency with the day-ahead scheduling scheme is maximized while the short-term deviation is corrected. The intraday rolling optimization objective function is shown in formula (45).

[0165]

[0166] wherein: G is the set of devices that need to be penalized when intraday rolling optimization adjustment is performed; is the unit adjustment cost; is the device adjustment amount.

[0167] 3.2.2 Constraint conditions

[0168] The power balance constraints, energy conversion equipment constraints, backup constraints, and grid constraints for intraday rolling optimization are shown in Equations (41), (42), (43), and (44). Under the intraday rolling optimization framework, system scheduling needs to achieve a balance between economy and stability within each optimization window:

[0169]

[0170] Where Δt is the intraday scheduling time interval. This constraint ensures the continuity of the energy storage status, equipment operating status, and output power of continuously operating equipment between windows.

[0171] 4 Case Analysis

[0172] 4.1 Example Parameter Description

[0173] In order to prove the effectiveness of the optimization method proposed in this invention, Figure 1 The IES shown in the figure was used to conduct simulation experiments, where the energy parameters are shown in Table 1; the CCHP, GB, EC, P2G, PV, and WT equipment parameters are shown in Table 2, and the energy storage equipment parameters are shown in Table 3. In the comparative scheme designs: Scheme 1 implements a deterministic optimization strategy, with the reserve capacity set as 20% of the sum of the wind and solar power forecast values ​​for scheduling decisions; Scheme 2 constructs a two-stage stochastic optimization model with the goal of minimizing the expected operating cost while meeting the supply and demand balance constraints and reserve capacity requirements in all scenarios; Scheme 3 establishes a two-stage robust optimization model to deal with the uncertainty of wind power and photovoltaic output, defining the fluctuation range of renewable energy output through a box uncertainty set, and then implementing optimized scheduling; Scheme 4 adopts a two-stage distributed robust optimization method, constructing fuzzy sets of wind power and photovoltaic output based on statistical moments, and using this as the basis for scheduling planning; Scheme 5 implements the two-stage chance-constrained planning method proposed in this invention with a confidence level of 95%.

[0174] Table 1 Energy parameters

[0175]

[0176] Table 2 IES equipment parameters

[0177]

[0178] Table 3 Energy storage equipment parameters

[0179]

[0180] The day-ahead forecast, intraday forecast and actual wind turbine output curves on the dispatching day are as follows: Figure 2 、 Figure 3The day-ahead prediction, day-ahead prediction, and actual electric cold and hot gas load curve are shown as Figure 4

[0181] The five schemes are based on the same day-ahead prediction, day-ahead prediction, and actual wind and solar power generation curve and load curve to carry out two-stage optimization scheduling. The cost comparison of the optimization results of each scheme is shown in Table 4.

[0182] Table 4: Scheme cost comparison

[0183]

[0184] From Table 4, it can be seen that different optimization methods lead to significant differences in cost performance. The wind and light punishment cost of all schemes is zero, thanks to the efficient conversion of excess electricity by electric energy storage and electric to gas technology, effectively avoiding wind and light abandonment.

[0185] Scheme one adopts a deterministic optimization strategy, sets a fixed reserve capacity of 20%, ignores wind and light output uncertainty, and has the highest total scheduling cost of 2,001,600 yuan, including day-ahead cost of 99,100 yuan, operation cost of 90,930 yuan, and reserve cost of 8,170 yuan, which are all high.

[0186] Scheme two optimizes the expected cost based on a two-stage stochastic optimization model, and the total cost is reduced to 187,790 yuan, which is 6.17% lower than that of scheme one. However, due to the lack of fully depicting the time-varying and heteroscedastic characteristics of prediction errors, the adjustment cost is as high as 9,270 yuan, which is still high.

[0187] Scheme three adopts a robust optimization strategy, constructs a box-type uncertainty set, and pursues reliability in the worst case, resulting in a total cost of 255,390 yuan, which is 27.54% higher than that of scheme one, and the reserve cost increases to 15,440 yuan, which significantly damages the economy.

[0188] Scheme four balances conservatism and economy using a statistical moment fuzzy set, with a total cost of 233,700 yuan, which is lower than that of scheme three, but the fuzzy set modeling is difficult to adapt to dynamic error characteristics, and the adjustment cost is still as high as 11,560 yuan.

[0189] Scheme five proposes a two-stage adaptive reserve optimization model based on chance constrained programming (TSARO), which quantifies the wind and light error characteristics by combining QGAM and GARCH, and introduces a dynamic adjustment mechanism. The total cost is 177,800 yuan, which is 11.18%, 5.33%, 30.39%, and 23.92% lower than the previous four schemes, respectively. The day-ahead cost is 83,430 yuan, the reserve cost is only 5,350 yuan, and the day-ahead adjustment cost is 8,620 yuan, all of which are the lowest, reflecting the advantages of precise modeling and flexible scheduling.

[0190] ​The SOC curve shows that the actual fluctuation of the battery energy storage curve reflects the error response ability of the model, the smooth operation of the heat storage tank reflects the stability, and the curve change of the gas storage tank reflects the flexibility. Overall, scheme five balances reliability and economy through probability constraints, significantly improving the scheduling performance of the comprehensive energy system.

[0191] 4.2.1 Day-ahead scheduling results

[0192] The IES day-ahead optimal scheduling results are shown in Figure 6 , Figure 7 , Figure 8 and Figure 9 .

[0193] As can be seen from Figures 6 to 9 , in Figure 6 , the load curve shows a night low, morning rise, daytime stability, evening peak, and night fall. From 0 to 4, the fan and gas turbine provide stable power supply, with no significant grid interaction or battery activity. From 4 to 8, the load grows rapidly, and the battery discharges and the grid purchases power to supplement the deficiency. From 9 to 12, the photovoltaic output increases and reaches a peak at 12, and the total load is maximum at about 175 MW from 10 to 14. The fan and photovoltaic are the main supply, the gas turbine is flexibly adjusted, and the battery fills the gap. From 10 to 14, the photovoltaic surplus power is used for battery charging and electric-to-gas operation, and the gas balance is tightly coupled. From 14 to 20, the photovoltaic weakens, and the battery discharges and the gas turbine increases to cope with the secondary peak. From 22 to 24, the wind power supports a small amount of battery charging and electric-to-gas.

[0194] In Figure 7 , the cold load characteristics are night low, morning rapid rise, daytime high fluctuation, and afternoon sudden drop. The absorption chiller provides about 5 MW of base cold capacity throughout the day, using the combined heat and power waste heat. The electric chiller outputs about 7 to 9 MW from 8 to 14, meeting the peak cold load (about 14 MW), and the required power comes from photovoltaic, fan, or purchased power. From 14 to 16, the cold load drops to about 5 MW, and the electric chiller reduces the output.

[0195] In Figure 8 , the heat load peaks at 0 to 7 (about 35 MW), drops to a trough at 13 to 18, and rises to about 35 MW at 18 to 24. The combined heat and power unit is dynamically adjusted as the main heat source, and the gas-fired boiler further supplements the heat, and the heat storage device releases heat from 0 to 3, reducing the dependence on the gas-fired boiler.

[0196] In Figure 9 , the gas load is relatively stable, with troughs from 0 to 4 and 19 to 24, and remains basically unchanged (about 40 MW) during the rest of the period. External gas purchase is the main source, and electric-to-gas starts to supplement from 10, mainly using photovoltaic or wind power surplus gas. The gas storage tank releases energy from 5 to 7 to supplement the deficiency and adjust the peak demand.

[0197] From Figures 10 to 13 It can be seen that: Figure 10 The intra-day power balance is presented, with wind power and gas turbines supplying power at night, and a small amount of battery charging using surplus wind power. In the morning, photovoltaic power generation gradually increases, and battery discharge supplements the growing load. From 9:00 to 15:00, photovoltaic power generation dominates, and surplus power drives the operation of the electric-gas conversion equipment and battery charging, with the gas turbine output reduced to prioritize renewable energy consumption. From 15:00 to 21:00, photovoltaic power weakens, and battery discharge and gas turbine output increase to cope with the load peak. From 21:00 to 24:00, wind power and gas turbines are the main force, with a small amount of battery charging at night. Intra-day dispatching more accurately responds to actual fluctuations, and the behavior of electric-gas conversion and battery charging and discharging is more concentrated, improving system adaptability.

[0198] Figure 11 The intra-day cold energy balance is presented, with about 5 MW of base cold capacity provided by the cold absorption chiller all day, using combined heat and power waste heat. The electric chiller follows the load peak and fluctuation, reducing output during the load valley from 0:00 to 4:00 and from 16:00 to 24:00. The intra-day electric chiller output is more in line with actual load fluctuations, reducing energy waste.

[0199] Figure 12 The intra-day thermal energy balance is described, with combined heat and power units as the main heat source from 16:00 to 22:00, with gas boilers providing supplemental heat, and a small amount of thermal storage device heat released from 0:00 to 4:00 to cope with peak demand. The intra-day thermal balance dispatching is relatively stable, basically following the day-ahead plan.

[0200] Figure 13 The intra-day natural gas balance is presented, with gas turbines and gas boilers as the main natural gas consumption equipment, with reduced consumption from 10:00 to 14:00 due to sufficient photovoltaic power generation. External gas purchase is the main gas source, with electric-gas conversion equipment using surplus electricity to produce natural gas from 10:00 to 14:00 and subsequent periods, and the gas storage tank charging from 10:00 to 11:00 and from 21:00 to 22:00, and discharging from 5:00 to 9:00 to adjust peak demand. Intra-day electric-gas conversion and gas storage behavior is more concentrated, and gas discharge operation more accurately reduces peak demand. The natural gas balance is coupled with the electric-gas conversion and power balance, with the gas storage tank discharging optimizing external gas purchase demand and indirectly supporting the operation of the gas boiler in the thermal energy balance. Compared to day-ahead dispatching, intra-day dispatching optimizes the operation strategy of electric-gas conversion, battery, thermal storage device, and gas storage tank through more accurate prediction and dynamic adjustment, improving renewable energy consumption capacity and peak-valley regulation capacity.

[0201] 4.3 Analysis of reserve capacity under different confidence levels

[0202] To further evaluate the performance of the TSARO-CCP model under different risk preferences, Figure 14 , Figure 15 and Figure 16The optimization results of reserve capacity at 95%, 85% and 75% confidence levels are shown respectively. The comparative analysis of the three confidence levels aims to reveal the impact of risk tolerance on the trade-off between reserve capacity configuration, system reliability and economy. The cost comparison of different confidence levels is shown in Table 5. Higher confidence level emphasizes reliability, suitable for scenarios with strict requirements for power supply stability; lower confidence level pursues economy, suitable for systems with sufficient flexibility resources or lower reliability requirements. By comparing the performance of day-ahead scheduling, intra-day optimization and actual reserve calling under different confidence levels, the adaptability of TSARO-CCP model in balancing cost and safety can be comprehensively understood.

[0203] Table 5 Cost comparison of different confidence levels

[0204]

[0205] From Figure 14 , Figure 15 and Figure 16 it can be seen that day-ahead scheduling quantifies uncertainty with QGAM and GARCH models, and reserve capacity is significantly reduced when confidence level is lowered, such as the peak period from 10 to 14 o'clock, reserve capacity decreases from 30 MW at 95% confidence level to 24 MW at 85% confidence level and 22 MW at 75% confidence level, reflecting the strategy of accepting higher risk to reduce reserve cost. Cost analysis shows that under 85% confidence level, total scheduling cost, total day-ahead scheduling cost and reserve cost are reduced by 2.69%, 3.29% and 20.60% respectively compared with 95% confidence level; under 75% confidence level, the above indicators are reduced by 3.65%, 4.47% and 29.38% respectively, highlighting the optimization of economy while also accompanied by the increase of system risk.

[0206] 5 Conclusion

[0207] The application provides a multi-time scale optimal scheduling method for a comprehensive energy system based on an opportunity constraint programming and a two-stage adaptive reserve optimization model, and systematically studies the influence of wind and solar power generation uncertainty on system stability and economy. The QGAM and GARCH models are used to accurately depict the nonlinear, time-varying, heteroscedastic and asymmetric distribution characteristics of wind and solar prediction errors, thereby providing a robust probability constraint basis for day-ahead reserve capacity configuration. The designed intraday dynamic adjustment mechanism realizes dynamic optimization of reserve capacity through real-time error triggering, trend signal fusion and weight distribution strategy, and significantly enhances the response ability of the system to short-term fluctuations. Simulation analysis shows that compared with the deterministic optimization, stochastic optimization, robust optimization and distribution robust optimization methods, the total cost of the method is reduced by about 15.81%, 6.21%, 32.53% and 22.03% respectively, and the intraday total cost is reduced by about 6.63%, 4.51%, 28.36% and 25.52% respectively, which effectively improves the renewable energy consumption capacity and system operation flexibility. The analysis of reserve capacity at the confidence levels of 95%, 85% and 75% further verifies the adaptability of the method in various risk preference scenarios, especially at a high confidence level, which ensures the system reliability with a small cost increment, and provides a solid theoretical and practical basis for the optimal scheduling of comprehensive energy systems in high-penetration renewable energy scenarios. In addition, the comprehensive modeling of the multi-energy coupling relationship of electricity, heat, cold and gas reveals the key role of energy storage devices in dynamic scheduling. The state curve analysis shows that the battery, heat storage tank and gas storage tank can effectively smooth the wind and solar output fluctuation and optimize the peak-valley regulation. The 15-minute time resolution combined with the 4-hour optimization window of the intraday rolling optimization further improves the adaptability of the scheduling strategy to real-time conditions, ensuring high coordination with the day-ahead plan. Compared with existing methods, the innovation of the method lies in the refinement of uncertainty modeling, the flexibility of dynamic adjustment mechanism and the systematicness of multi-energy collaborative optimization, which provides a new technical path for building a clean and efficient energy ecosystem.

[0208] The above description is only a description of the preferred embodiments of the application, and does not limit the scope of the application in any way. Any modification or modification of the application by a person skilled in the art based on the above disclosure is within the scope of the claims.

Claims

1. A multi-time-scale optimization scheduling method for an integrated energy system, characterized in that: Includes the following: S1. Uncertainty estimation of wind and solar power generation based on adaptive reserve optimization model and chance-constrained programming S1.

1. Day-Ahead Reserve Capacity Estimation Using the Quantile Generalized Additive Model and the Generalized Autoregressive Conditional Heteroskedasticity Model S1.1.

1. Construction of Quantile Generalized Additive Model and Generalized Autoregressive Conditional Heteroskedasticity Model The quantile generalized additive model constructs a nonparametric quantile regression framework, embeds time period characteristics, and captures the nonlinearity and time-variability of the error distribution through spline functions and factor functions. It adapts to the asymmetric fat-tailed distribution characteristics of wind power and the diurnal cycle modulation distribution characteristics of photovoltaic power. The generalized autoregressive conditional heteroscedasticity model characterizes the heteroscedasticity of the error series, extracts the lagged square term and the previous variance memory term, and captures the high-frequency fluctuation characteristics of wind power and the daytime fluctuation characteristics of photovoltaic power. In the collaborative framework, the quantile generalized additive model provides quantile boundary constraints, and the generalized autoregressive conditional heteroscedasticity model generates volatility compensation terms to achieve distribution adaptation and feature mining. S1.1.

2. Day-ahead reserve capacity estimation based on dual-model coupling; S1.

2. Construction of intraday dynamic adjustment mechanism The core task of the intraday phase is to ensure that the power system meets opportunity constraints under short-term uncertainties through a dynamic reserve capacity adjustment mechanism. This phase aims to dynamically adjust reserve capacity. The constructed intraday dynamic adjustment mechanism covers the trigger mechanism, adjustment amount calculation, and weight allocation, which complements the day-ahead reserve capacity planning in terms of time and space. S2. Two-stage optimization scheduling First, the day-ahead scheduling phase is executed. This phase uses a one-hour time resolution to develop a preliminary scheduling plan for the next 24 hours. The optimization process is based on historical forecast error data, the operating costs of various equipment, and day-ahead forecast data. The day-ahead scheduling model is used for calculations. The historical forecast error data is analyzed using a quantile generalized additive model and a generalized autoregressive conditional heteroskedasticity model to determine the day-ahead reserve capacity. The optimization results output includes the start / stop status and output plans of units such as the power grid, gas grid, wind power, photovoltaic power, combined heating and cooling power, gas boilers, power-to-gas, gas turbines, thermal and energy storage. Subsequently, the framework enters the intraday rolling scheduling stage. The optimization window in this stage is 4 hours, the time resolution is 15 minutes, and rolling optimization is performed at intervals of 15 minutes. The optimization process is calculated through the rolling scheduling model based on real-time prediction data and errors, the actual scheduling results of the previous period, the day-ahead scheduling plan and short-term prediction data. The real-time prediction error is used to adjust the intraday rolling spare capacity. The optimization results generate a refined scheduling sequence for the next 4 hours, dynamically adjust the start and stop status and output of each unit, and realize precise regulation of system operation.

2. A multi-time-scale optimization scheduling method for an integrated energy system according to claim 1, characterized in that: In step SS1.1.1, the forecast error for wind power and photovoltaic power is defined as the difference between the actual output and the day-ahead forecast output: Where: y(t) is the actual output at time t; is the day-ahead output forecast at time t; e(t) is the forecast error sequence. The quantile generalized additive model is used to estimate the conditional quantile function of the error sequence. For the quantile level t∈(0,1), the quantile function can be expressed as: Where: x(t) is a feature vector, which is composed of timestamp, periodic characteristics and predicted output value, and its specific form is determined according to the actual application scenario; f j (·, t) is a smooth basis function, and spline function is used for continuous features. The model parameters are optimized by minimizing the penalty quantile loss function. Under the chance constraint framework, the conditional cumulative distribution function of the error e(t) is assumed to be F e(t)|x(t) (·), then the quantile function q t (x(t)) satisfies: F e(t)|x(t) (q t (x(t)))=t In order to capture the heteroskedasticity characteristics of the error series, the generalized autoregressive conditional heteroskedasticity model is used to estimate the conditional volatility. The model can be expressed as: Where: is the conditional variance at time t, and the parameters are estimated by optimizing the log-likelihood function using maximum likelihood estimation: Determine ω, α, and β from the above formula, and satisfy ω>0, α≥0, β≥0, and α+β≥0. Then use the estimated parameters to recursively calculate the future conditional variance: The conditional standard deviation is:

3. The multi-time-scale optimization scheduling method for an integrated energy system according to claim 1, characterized in that: In step S1.1.2, the wind power upstream reserve capacity and downlink spare capacity for: Where: and is the quantile estimate corresponding to the confidence level; k u (k u ≥1) is a conservative adjustment factor used to enhance the safety margin of spare capacity; σ t is a volatility compensation term to cope with the uncertainty in high volatility scenarios and is determined by the generalized autoregressive conditional heteroskedasticity model; the term ζE[|e|] is the expected absolute value error compensation term to improve the robustness of negative reserves; Considering the zero output characteristic of photovoltaic power generation at night, the reserve capacity during non-daytime periods is set to zero, and the photovoltaic upstream reserve capacity is and downlink spare capacity They are: Where D is the set of daytime periods dynamically defined based on the illumination time.

4. The multi-time-scale optimization scheduling method for an integrated energy system according to claim 1, characterized in that: In step S1.2, the specific design is as follows: S1.2.1 Trigger mechanism design The trigger mechanism defines the trigger indicator I through a two-step method t , to achieve accurate capture and anti-interference control of the adjustment signal, first based on the real-time error e t 、Trend Signals t and dynamic thresholds Constructing candidate indicators Preliminary judgment of adjustment direction: Where: I(·) is the indicator function, which takes 1 when the condition is met and 0 otherwise; A t Used to trigger uplink standby adjustment; B t Used to trigger downlink standby adjustment. The specific model is: Where: y t (e t ,s t ) is the fusion function, which integrates the error of the previous moment and the trend signal, e t-1 is the real-time error at time t-1, indicating the power deviation at the previous moment; s t It is a trend signal, reflecting the short-term trend of error; It is a dynamic threshold, which is adaptively adjusted according to the system status; Where: As the trend item weight coefficient, s is dynamically adjusted according to the system volatility. t The contribution of the trend term is weakened in high volatility scenarios; In low volatility scenarios, the impact of trend terms increases. To normalize the trend signal, by embedding The fusion function implements adaptive evaluation to balance the error and trend contributions; Where: m is the window length; n t The local volatility of the historical data statistics range is defined as the mean square error of the error within the window relative to its mean: Where: is the mean error within the window. To avoid unstable operation caused by frequent switching, a stability constraint is introduced: Where: λ0 is the maximum switching threshold, switching frequency λ t Defined as: λ t =|(I t ≠I t-1 )·|(I t ≠0)·|(I t-1 ≠0) Where: I t , I t-1 is the final trigger index at time t and t-1, which is used for switching frequency calculation. Finally, the trigger index I t Combined with the stability constraint optimization is: S1.2.2 Adjustment calculation model When the trigger indicator I t When ≠0, the adjustment amount calculation module is started to meet the opportunity constraint conditions and ensure the reliability of power balance. The expression is as follows: Where: and They are the uplink and downlink standby adjustment amounts, and the adjustment amount is based on I t Direction calculation, for wind power, and The calculation model is: For photovoltaics, the calculation model for upstream and downstream reserve adjustment is: Where, ω e is the direct error weight, reflecting the contribution of the error to the adjustment amount; ω ν is the volatility weight, reflecting the contribution of local fluctuations to the adjustment amount; ω s is the trend weight, reflecting the contribution weight of the short-term trend to the adjustment amount, γ w and γ p They are wind power and photovoltaic trend adjustment parameters respectively, and the adjustment amount is obtained through the function φ + and φ - To achieve smoothing: Weight ω e 、ω ν and ω s Through normalization calculation, the balance of each component of the adjustment is ensured, which is defined as: Where:

5. The multi-time-scale optimization scheduling method for an integrated energy system according to claim 4, characterized in that: Step S2 specifically includes: S2.1 Day-ahead Optimization Scheduling S2.1.1 Objective function The day-ahead optimization model focuses on equipment operating costs, startup and shutdown costs, backup capacity costs, and penalties for wind and solar curtailment. While meeting load demand, it maximizes renewable energy utilization and reserves sufficient backup resources to cope with uncertainty. The day-ahead optimization objective function is shown below: Where: D run A collection of equipment considering operating costs; c d is the operating cost coefficient of equipment d; is the power output of device d at time t; D start A collection of equipment that takes into account start-up and shutdown costs; is the start-stop cost coefficient of equipment d; is the start flag of device d at time t; c res is the cost coefficient of standby; c cur Cost coefficient for penalties for curtailing wind and solar power; and are the available wind power and photovoltaic power output at time t, and are the actual wind power and photovoltaic output at time t, respectively; S2.1.2 Constraints 1) Power balance constraints The energy flow balance of electricity, heat, cooling and gas in the integrated energy system is as follows: Where: are the electricity, gas, heating and cooling loads at time t respectively; The battery storage and power supply at time t; is the gas storage and supply volume of the gas tank at time t; The heat storage tank stores and supplies heat at time t; are the electricity and gas purchases of the superior at time t respectively; 2) Energy conversion equipment constraints Where: L d,max 、L d,min are the maximum load rate and minimum load rate of equipment d respectively; are the upper and lower limits of the rated power of device d at time t respectively; is a binary variable of the equipment operation status, and the equipment d is Put into operation at Exit the operation when 3) Backup constraints Where: are the uplink and downlink backup amounts provided by device d at time t respectively; are the total upstream and downstream reserves required by the system at time t respectively; 4) Mesh constraints To ensure stable system operation and due to the limitations of the transmission network, the interaction power between the integrated energy system and the upper-level power grid and gas grid should be within a reasonable range: Where: are the upper limits of electricity and gas purchased by the integrated energy system from the upper-level power grid and gas grid at time t, respectively; S2.2 Intraday rolling optimization scheduling S2.2.1 Objective function The intraday rolling optimization model dynamically adjusts the day-ahead scheduling plan during the real-time operation phase. Its operating mechanism is based on a rolling optimization strategy. To coordinate the fluctuation response in short-term operation with the continuity of the day-ahead plan, an objective function containing the penalty term is constructed. While achieving short-term deviation correction, it maintains consistency with the day-ahead scheduling plan to the greatest extent possible. The intraday rolling optimization objective function is shown in the following formula: Where: G is the set of devices that must be penalized during intraday rolling optimization adjustments; Adjust costs for units; Adjust the amount for the device; S2.2.2 Constraints In the intraday rolling optimization framework, system scheduling needs to achieve a balance between economy and stability within each optimization window: Where Δt is the intraday scheduling time interval. This constraint ensures the continuity of the energy storage status, equipment operating status, and output power of continuously operating equipment between windows.

Images