Extreme high temperature scene multi-energy complementary optimization scheduling method considering uncertainty

Abstract

The invention discloses an extreme high-temperature scene multi-energy complementary optimization scheduling method considering uncertainty, which mainly comprises the following steps of: 1, fusing bidirectional time convolution, a bidirectional long-short-term memory network, an attention mechanism and a quantile regression forest, realizing high-precision prediction and uncertainty modeling of wind speed, solar irradiation and load, and constructing a typical day scene set; 2, constructing a two-stage scheduling model fusing epsilon-constraint multi-objective optimization and opportunity constraint mixed integer programming, optimizing and adjusting margin day ahead, and rolling and correcting a scheduling path within the day; 3, three physical correction mechanisms of wind power air density correction, photovoltaic temperature response and hydroelectric evaporation-water level coupling are provided for extreme high-temperature disturbance; and 4, a prediction-optimization-feedback-correction closed-loop process is integrated and constructed, and the stability and response toughness of the system under extreme climate are improved. The method has the beneficial effect that the prediction precision, the scheduling flexibility and the operation toughness of the system under the extreme climate are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of short-term optimal dispatching of power systems, and relates to a multi-energy complementary optimal dispatching method for extreme high temperature scenarios taking uncertainty into account. Background Art

[0002] With the continuous improvement of clean energy penetration, high-proportion clean energy systems face key problems such as insufficient scheduling flexibility and weak system resilience when facing extreme climate events. Although a large number of studies have focused on wind and solar power output prediction and scheduling optimization, the current technical system still has the following deficiencies: (1) Most studies are still based on point predictions, lacking confidence intervals or probability distribution information, making it difficult to provide a credible risk boundary for scheduling optimization; (2) Faced with the deep coupling of source-load bidirectional uncertainty, traditional static scheduling strategies and deterministic operation frameworks are unable to cope with the increasingly complex operating environment, and there is an urgent need to build an uncertainty perception mechanism that integrates multi-energy complementarity, supports dynamic regulation, and covers multiple time scales; (3) Current models generally ignore the physical impact mechanism of extreme high temperatures on wind, solar, and hydropower output, and lack temperature response modeling methods that can be applied to actual scheduling optimization. (4) When responding to extreme climate shocks, there is a lack of an integrated scheduling framework, and it is impossible to achieve a high degree of coordination between prediction, scheduling, and feedback control; Summary of the Invention

[0003] The purpose of the present invention is to provide a multi-energy complementary optimization scheduling method for extreme high temperature scenarios taking into account uncertainty, so as to achieve multi-energy coordination, supply and demand balance and stable operation of the power grid in high temperature environments.

[0004] The technical solution adopted by the present invention is a multi-energy complementary optimization scheduling method for extreme high temperature scenarios taking into account uncertainty, which is specifically implemented according to the following steps:

[0005] Step 1: Fusing a bidirectional temporal convolutional neural network-bidirectional long short-term memory network-attention mechanism (BiTCN-BiLSTM-AM) with a quantile regression forest (QRF) jointly extracts the spatiotemporal characteristics of wind speed, solar irradiation, and load time series, achieving high-precision point forecasts and 95% confidence interval modeling. A typical daily scenario set is constructed based on this fusion model.

[0006] Step 2: A two-stage scheduling model was constructed that integrates ε-constrained multi-objective optimization and chance-constrained mixed integer programming. This model optimizes margins and absorption rates at the hourly level, and makes rolling adjustments to scheduling paths at a 15-minute granularity within the day, enabling efficient response and dynamic adjustment to uncertain scenarios.

[0007] Step 3, a temperature coupling correction mechanism for wind, light and water output under extreme high temperature scenarios is proposed: wind power introduces air density-wind speed linkage correction, photovoltaic establishes component temperature response model, and hydroelectricity constructs evaporation-water level-output coupling adjustment structure. This mechanism integrates three types of physical correction models in a unified framework, realizes dynamic adjustment of power prediction value;

[0008] Step 4, integrate prediction modeling, high temperature correction and dispatching optimization module, and build a closed-loop control process of "prediction-optimization-feedback-correction". Realize the whole process linkage of dispatching chain, and improve the stability and response resilience of the system under extreme high temperature scenarios.

[0009] The features of the present application are also:

[0010] Step 1 is implemented according to the following implementation:

[0011] In the prediction of wind speed, solar radiation and load of the receiving end power grid, the present application proposes BiTCN-BiLSTM-AM model combined with QRF to deal with uncertainty problems. Among them, BiTCN is used to extract long time dependence and capture the fluctuation characteristics of time series, and BiLSTM further uses bidirectional propagation to capture the relationship between time steps before and after.

[0012] In BiTCN, assume that the input time series data is X={x1,x2,...,x T}, the input x t at each time is processed through convolution layer and activation function, and the feature sequence output after multiple convolution layers is output. For the convolution operation of the lth layer, the output can be expressed as:

[0013]

[0014] Among them: is the lth layer convolution kernel, is the lth layer reverse convolution kernel, b (l) is the bias, sigma(·) is the activation function, k is the size of the convolution kernel, and t is the current time step.

[0015] In the stacking structure of BiTCN, with the increase of the number of layers, the output features will be gradually abstracted, and higher order time dependence will be captured. Finally, the output sequence Z t of the BiTCN layer is sent to the BiLSTM layer.

[0016] The BiLSTM structure combines forward and reverse LSTM, and learns the bidirectional dependence of data through forward propagation and backward propagation , and finally outputs:

[0017]

[0018] On this basis, the QRF is introduced to map the point prediction result Y into a probability density function f(Y), which characterizes the prediction distribution characteristics of wind speed, solar radiation intensity and load variables. Its training process relies on the structure of random forest, and the results of multiple decision trees are integrated to generate the final distribution estimate. Specifically, it is expressed as:

[0019] f(Y t )=P(Y t ∣X t ) (3)

[0020] Where: X t is the input feature vector, and P(Y t ) is the probability distribution of the predicted value estimated based on the QRF model. QRF optimizes the uncertainty estimate through tree structure, so that it can output the prediction probability density function (PDF) of each time step, further providing sample distribution for the simulation module.

[0021] Step 2 is implemented as follows:

[0022] To improve the scheduling performance of the wind-solar-water storage system in different operating periods, the invention constructs a double-time-scale optimization framework with day-ahead scheduling as the core of robustness and intra-day scheduling as the core of flexibility. Among them, the day-ahead scheduling model takes 1 hour as the time granularity, makes 24-hour prospective prediction and resource coordination, while the intra-day scheduling model adopts 15-minute rolling optimization method and continuously updates the scheduling scheme based on the latest load and new energy output prediction. Under this optimization framework, the objective function is designed respectively by stage.

[0023] (a). Day-ahead risk scheduling

[0024] The day-ahead scheduling stage is based on prediction information, focusing on developing a highly executable and resource-efficient scheduling scheme. According to the operating characteristics of the wind-solar-water storage system, the following objective function is set:

[0025] (1) Maximum water reservoir regulation capacity:

[0026]

[0027] Where: SOC(t) represents the state of the water power adjustable resource (such as water level, water volume or equivalent energy) at time t, is the desired water storage capacity of the terminal period.

[0028] (2) Maximum renewable energy utilization rate:

[0029]

[0030] Where: The wind power and photovoltaic power effectively absorbed by the system at time t, respectively.

[0031] (3) Maximizing system output stability:

[0032]

[0033] P gen (t) is the total power generation of the system at time t.

[0034] (b). Intra-day opportunity constraint scheduling

[0035] Intra-day scheduling mainly relies on rolling prediction correction to achieve rapid response to actual load and new energy uncertainty disturbance. The objective function design focuses on short-term deviation control and real-time adjustment capability:

[0036] (1) Minimizing system operation deviation:

[0037]

[0038] T' is the intra-day rolling optimization period, P load (t) is the load demand considering prediction error.

[0039] (2) Maximizing flexible resource adjustment responsiveness:

[0040]

[0041] P ESS (t) is the net output of the energy storage system.

[0042] (3) Minimizing curtailment and load loss:

[0043]

[0044] The first term is the load loss, and the second term is the curtailment.

[0045] (c). Energy storage system output constraint

[0046] (1) State of charge (SOC) capacity constraint:

[0047]

[0048] SOC ESS (t) is the state of charge of the energy storage system at time t.

[0049] (2) Power balance constraint:

[0050] The difference between daily charging and discharging should not exceed the set threshold:

[0051] P gen(t)+P dis (t) = P load (t)+P chg (t) (11)

[0052] where: P dis (t), P ch (t) are the discharging and charging power of the energy storage, respectively.

[0053] (3) The chance constraint of uncertainty processing:

[0054] Pr(∣P gen (t)-P load (t)∣≤δ)≥1-∈ (12)

[0055] where: ∈ represents the maximum constraint violation probability allowed by the system, and the present application sets ∈ = 0.05, indicating that the constraint should be satisfied at a confidence level of 95%.

[0056] Step 3 is specifically implemented as follows:

[0057] (a). Photovoltaic power output

[0058] To improve the accuracy of photovoltaic modeling under extreme high temperature, the present application proposes a dynamic photovoltaic power estimation method considering temperature influence. The method introduces a component working temperature (T c ) estimation mechanism, which more accurately describes the coupling effect of irradiation and environmental temperature on photovoltaic performance than traditional linear correction. The specific formula is as follows:

[0059]

[0060] where: T a is the environmental temperature, G is the current irradiance, and NOCT is the nominal operating temperature of the photovoltaic module.

[0061] Subsequently, the temperature coefficient γ (value -0.004℃-1) is used to correct the temperature of the unit component power:

[0062] P output = P module ·(1+γ(T c -T stc ))(14)

[0063] where: T stc is the temperature under standard test conditions, and P module is the rated power of each photovoltaic module. When T c increases, (T c -T stc ) in the formula will be positive, resulting in a decrease in P output .

[0064] (b) Wind power output

[0065] This invention introduces a temperature correction mechanism for wind speed and air density, comprehensively assesses the impact of high temperature on wind power output, and performs dynamic correction calculations on wind power accordingly. The specific steps are as follows:

[0066] Assume that at standard temperature (15℃, T stc =298.15K) The air density is 1.225kg / m 3 The change in air density according to temperature is calculated using the following formula:

[0067]

[0068] Where ρ0 is the air density at standard temperature, and T is the ambient temperature (unit: °C), which is converted to Kelvin temperature for calculation.

[0069] Based on this, the present invention considers introducing a wind speed adjustment factor for modeling in high temperature extreme weather scenarios (T≥35°C), and its expression is as follows:

[0070] v adjusted =v base ·(1-α(T-35)), T≥35℃(16)

[0071] Among them, v base is a hypothetical baseline wind speed sequence, and T is the current ambient temperature (unit: °C). α is the temperature sensitivity coefficient, with a recommended range of 0.02 to 0.036, representing a 2% to 3.6% decrease in wind speed for every 1°C increase in temperature. If T is < 35°C, no wind speed adjustment is performed.

[0072] (c) Hydropower output

[0073] The output of a hydropower unit is affected by the reservoir water level, inflow, and ambient temperature. Extremely high temperatures can exacerbate water evaporation, leading to a drop in water levels and affecting the unit's power generation capacity. Its output can be calculated using the following formula:

[0074] P hydro (t) = η·g·Q(t)·H(t)(17)

[0075] Where: P hydro (t) is the output of the hydropower unit at time t, η is the efficiency of the hydropower unit, g is the acceleration of gravity, Q(t) is the flow rate at time t, and H(t) is the water head at time t.

[0076] To more accurately estimate hydropower output under high temperature conditions, it is necessary to consider the impact of evaporation on water levels. The change in reservoir water level is determined by precipitation, evaporation, and inflow, and can be expressed by the following formula:

[0077] AH(t) = R(t) - E(t) - Q(t) (18)

[0078] wherein: AH(t) is the change of water level of the reservoir at time t; R(t) is the precipitation at time t; E(t) is the evaporation at time t; Q(t) is the water flow into the reservoir.

[0079] Step 4 is specifically implemented as follows:

[0080] The application constructs a short-term optimal scheduling framework for extreme climate conditions, which is composed of three links of uncertainty modeling, day-ahead multi-objective scheduling and intra-day rolling optimization, and runs through system prediction, modeling, optimization and real-time adjustment.

[0081] In the uncertainty analysis stage, the BiTCN-BiLSTM-AM is fused to extract deep spatio-temporal features from multi-source time series data such as wind speed, solar radiation and load, and output high-precision point prediction; further, the QRF is introduced to construct a 95% confidence interval, and a typical uncertainty scenario is generated to provide an input boundary for scheduling modeling.

[0082] In the day-ahead optimization stage, based on the above typical scenario and dispatchable resources, a three-objective optimization model containing "maximum reservoir regulation margin, maximum renewable absorption rate and minimum system volatility" is constructed, and the epsilon-constraint method is used to scan the Pareto frontier to extract a balanced scheduling scheme between the objectives as the preliminary operation path of the system.

[0083] In the intra-day rolling optimization stage, the prediction information is dynamically updated at a 15-minute step, and the unit boundary is real-time corrected to consider the extreme high temperature inhibition effect; under the rolling time window mechanism, an opportunity constraint mixed integer programming model (MIP) is constructed, and each round executes the first period scheduling instruction to realize the coordinated optimization of supply-demand deviation, flexible responsiveness and curtailment / loss of load, and ensure the stable, flexible and resilient operation of the system under high confidence. BRIEF DESCRIPTION OF DRAWINGS

[0084] Figure 1 is the BiTCN-BiLSTM-AM model architecture and flowchart in the extreme high temperature scenario multi-energy complementary optimal scheduling method considering uncertainty of the application;

[0085] Figure 2 is the multi-energy complementary system operation framework considering uncertainty in the extreme high temperature scenario multi-energy complementary optimal scheduling method considering uncertainty of the application;

[0086] Figure 3 is the research data set graph in the extreme high temperature scenario multi-energy complementary optimal scheduling method considering uncertainty of the application: (a) wind speed, (b) solar irradiance, (c) power load;

[0087] Figure 4 is a typical day Pareto solution set and optimal trade-off solution graph in a multi-energy complementary optimization scheduling method considering uncertainty in an extremely high temperature scenario of the application;

[0088] Figure 5 is a power output process graph of day-ahead scheduling plan in a multi-energy complementary optimization scheduling method considering uncertainty in an extremely high temperature scenario of the application;

[0089] Figure 6 is a superimposed graph of each energy power output of the intra-day scheduling plan in a multi-energy complementary optimization scheduling method considering uncertainty in an extremely high temperature scenario of the application;

[0090] Figure 7 is a system total output and power generation plan deviation graph of the intra-day scheduling plan in a multi-energy complementary optimization scheduling method considering uncertainty in an extremely high temperature scenario of the application. DETAILED DESCRIPTION

[0091] The application will be described in detail below in combination with the drawings and specific embodiments.

[0092] The application is a multi-energy complementary optimization scheduling method considering uncertainty in an extremely high temperature scenario, which is specifically implemented according to the following steps:

[0093] Step 1 is specifically implemented according to the following:

[0094] In the prediction of wind speed, solar radiation and load of the receiving end power grid, the application proposes a BiTCN-BiLSTM-AM model combined with QRF to handle uncertainty problems. Among them, BiTCN is used to extract long-time dependence and capture the fluctuation characteristics of time series, and BiLSTM further utilizes bidirectional propagation to capture the relationship between previous and subsequent time steps. The structure of the proposed model is as shown in Figure 1 .

[0095] In BiTCN, it is assumed that the input time series data is X={x1,x2,...,x T}, the input x t at each time is processed through a convolution layer and an activation function, and a feature sequence output by multiple convolution layers. For the convolution operation of the lth layer, the output can be expressed as:

[0096]

[0097] wherein: is the lth layer convolution kernel, is the lth layer reverse convolution kernel, b (l) is the bias, and σ(·) is the activation function, k is the size of the convolution kernel, and t is the current time step.

[0098] In the stacked structure of BiTCN, as the number of layers increases, the output features will gradually abstract, capturing higher-order temporal dependencies. Finally, the output sequence Z t of the BiTCN layer is sent to the BiLSTM layer.

[0099] The BiLSTM structure combines forward and reverse LSTMs, which learn the bidirectional dependencies of the data through forward propagation and backward propagation , and finally output:

[0100]

[0101] On this basis, QRF is introduced to map the point prediction result Y to the probability density function f(Y), which describes the prediction distribution characteristics of wind speed, solar radiation intensity and load variables. Its training process relies on the random forest structure, and the results of multiple decision trees are integrated to generate the final distribution estimate. Specifically, it is expressed as:

[0102] f(Y t )=P(Y t ∣X t ) (3)

[0103] Where: X t is the input feature vector, and P(Y t ) is the probability distribution of the predicted value estimated based on the QRF model. QRF optimizes uncertainty estimation through tree structure, so it can output the prediction probability density function (PDF) at each time step, further providing sample distribution for the simulation module.

[0104] Step 2 is implemented as follows:

[0105] To improve the scheduling performance of the wind-solar-water storage system at different operating periods, the invention constructs a double-time-scale optimization framework with day-ahead scheduling as the core of robustness and intra-day scheduling as the core of flexibility. Among them, the day-ahead scheduling model takes 1 hour as the time granularity, and makes 24-hour forward-looking prediction and resource coordination, while the intra-day scheduling model adopts 15-minute rolling optimization method, and continuously updates the scheduling scheme based on the latest load and new energy output prediction. Under this optimization framework, the objective function is designed by stage.

[0106] (a). Day-ahead risk scheduling

[0107] The day-ahead scheduling stage is based on prediction information, focusing on developing a highly executable and resource-efficient scheduling scheme. According to the operating characteristics of the wind-solar-water storage system, the following objective function is set:

[0108] (1) Maximum reservoir regulation capacity:

[0109]

[0110] Where: SOC(t) represents the state of hydropower adjustable resources (such as water level, water volume or equivalent energy) at time t, The expected amount of water left in the reservoir at the end of the period.

[0111] (2) Maximizing the utilization of renewable energy:

[0112]

[0113] in: are the wind power and photovoltaic power outputs effectively absorbed by the system at time t.

[0114] (3) Maximizing system output stability:

[0115]

[0116] Where: P gen (t) is the total power generation of the system at time t.

[0117] (b) Intraday opportunity-constrained scheduling

[0118] Intraday scheduling mainly relies on rolling forecast correction to achieve rapid response to actual load and new energy uncertainty disturbances. The objective function design focuses on short-term deviation control and real-time adjustment capabilities:

[0119] (1) Minimize system operation deviation:

[0120]

[0121] Where: T' is the rolling optimization period within the day, P load (t) is the load demand after considering the prediction error.

[0122] (2) Maximizing responsiveness of flexible resource adjustments:

[0123]

[0124] Where: P ESS (t) is the net output of the energy storage system.

[0125] (3) Minimizing power abandonment and load loss:

[0126]

[0127] Among them: the first item is the load loss amount, and the second item is the power abandonment amount.

[0128] (c) Energy storage system output constraints

[0129] (1) State of Charge (SOC) capacity constraint:

[0130]

[0131] where: SOC ESS (t) is the state of charge of the energy storage system at time t.

[0132] (2) Power balance constraint:

[0133] The difference between daily charge and discharge should not exceed a set threshold:

[0134] P gen (t) + P dis (t) = P load (t) + P chg (t) (11)

[0135] where: P dis (t), P ch (t) are the discharging and charging power of the energy storage, respectively.

[0136] (3) Chance constraint for uncertainty handling:

[0137] Pr(∣P gen (t) - P load (t)∣≤δ) ≥ 1-ε (12)

[0138] where: ε represents the maximum constraint violation probability allowed by the system, and the present invention sets ε = 0.05, indicating that the constraint should be satisfied at a confidence level of 95%.

[0139] Step 3 is implemented according to the following:

[0140] (a). Photovoltaic power output

[0141] To improve the accuracy of photovoltaic modeling under extreme high temperature, the present invention proposes a dynamic photovoltaic power estimation method considering temperature influence. This method introduces a component operating temperature (T c ) estimation mechanism, which more accurately describes the coupling effect of irradiance and ambient temperature on photovoltaic performance than traditional linear correction. The specific formula is as follows:

[0142]

[0143] where: T a is the ambient temperature, G is the current irradiance, and NOCT is the nominal operating temperature of the photovoltaic module.

[0144] Subsequently, the temperature coefficient γ (value -0.004℃-1) is used to correct the temperature of the unit component power:

[0145] Poutput = P module ·(1 + γ(T c - T stc ))(14)

[0146] where: T stc is the temperature under standard test conditions, P module is the rated power of each photovoltaic module. When T c increases, (T c - T stc ) in the formula will be positive, resulting in P output decreases.

[0147] (b). Wind power output

[0148] The present application introduces a correction mechanism for temperature, wind speed and air density, comprehensively evaluates the impact of high temperature on wind power output, and accordingly carries out dynamic correction calculation of wind power. The specific steps are as follows:

[0149] Assume that the air density is 1.225 kg / m 3 at the standard temperature (15℃, i.e. T stc = 298.15K). The change of air density is calculated according to temperature by the following formula:

[0150]

[0151] wherein p0 is the air density at the standard temperature, T is the ambient temperature (unit: ℃), and is converted to Kelvin temperature for calculation.

[0152] Based on this, the present application considers to introduce wind speed adjustment factor for modeling in high temperature extreme weather scenario (T≥35℃), and its expression is as follows:

[0153] v adjusted = v base ·(1 - α(T - 35)), T≥35℃(16)

[0154] wherein v base is the assumed reference wind speed sequence, T is the current ambient temperature (unit: ℃). α is the temperature sensitivity coefficient, and the recommended value range is 0.02-0.036, representing 2%-3.6% decrease of wind speed per 1℃ increase. If T<35℃, no wind speed adjustment is made.

[0155] (c). Hydropower output

[0156] The output of hydropower unit is jointly affected by reservoir water level, inflow and ambient temperature. Extreme high temperature weather will exacerbate water evaporation, and then cause water level to drop, affecting the generating capacity of the unit. Its output can be calculated according to the following formula:

[0157] P hydro (t) = η · g · Q(t) · H(t) (17)

[0158] where P hydro (t) is the output of the hydroelectric unit at time t, η is the efficiency of the hydroelectric unit, g is the acceleration of gravity, Q(t) is the flow at time t, and H(t) is the water head at time t.

[0159] To more accurately estimate the output of the hydroelectric power under high temperature conditions, the effect of evaporation on the water level needs to be considered. The change in the water level of the reservoir is determined by precipitation, evaporation and inflow, which can be expressed by the following formula:

[0160] ΔH(t) = R(t) - E(t) - Q(t) (18)

[0161] where ΔH(t) is the change in the water level of the reservoir at time t, R(t) is the precipitation at time t, E(t) is the evaporation at time t, and Q(t) is the flow into the reservoir.

[0162] Step 4 is specifically implemented as follows:

[0163] The present application constructs a short-term optimization scheduling framework for extreme weather conditions. As shown in Figure 2 , it is composed of three links of uncertainty modeling, day-ahead multi-objective scheduling and intra-day rolling optimization, and runs through system prediction, modeling, optimization and real-time adjustment.

[0164] In the uncertainty analysis stage, BiTCN-BiLSTM-AM is used to extract deep spatio-temporal features from multi-source time series data such as wind speed, solar radiation and load, and output high-precision point prediction; QRF is further introduced to construct a 95% confidence interval, generating typical uncertainty scenarios to provide input boundaries for scheduling modeling.

[0165] In the day-ahead optimization stage, based on the above typical scenarios and dispatchable resources, a three-objective optimization model containing "maximum reservoir regulation margin, maximum renewable absorption rate, and minimum system volatility" is constructed, and the ε-constraint method is used to scan the Pareto frontier to extract a balanced scheduling scheme between the objectives as the initial operation path of the system.

[0166] In the intra-day rolling optimization stage, the prediction information is dynamically updated at a step of 15 minutes, and the unit boundary is real-time corrected to consider the extreme high temperature suppression effect; under the rolling time window mechanism, an opportunity constraint mixed integer programming model (MIP) is constructed, which executes the first period scheduling instruction in each round to realize the coordinated optimization of supply-demand deviation, flexible responsiveness and curtailment / loss of load, ensuring the stable, flexible and resilient operation of the system under high confidence.

[0167] Embodiment

[0168] The present application takes the 9 units of 6930 MW in the Sichuan side of the left bank of the Jinsha River Lugu Hydropower Station as the core, constructs a wind-solar-water-storage complementary system, and simulates its joint dispatching performance under the background of extreme high temperature in summer (31 days) in Sichuan Province in 2022. The wind, light prediction variables and power grid load data are divided into training set (first 26 days) and test set (last 5 days), and the energy storage system with 30% of the installed capacity is configured to form a flexible resource pool in the dispatching. The data set characteristics and division are as follows Figure 3 , and various energy parameters are shown in Table 1.

[0169] In the aspect of prediction modeling, BiTCN-BiLSTM-AM-QRF is used for probabilistic modeling of wind speed, irradiance and load, and the prediction period of solar irradiance is limited to 7:00-19:00 to avoid night-time errors. Based on the probability density function of QRF output, 500 typical future day scenarios are generated, and 50 representative scenarios are selected through a fast backward reduction algorithm.

[0170] Table 1 Various energy-related parameters

[0171]

[0172] The dispatching model adopts a double-time-scale structure:

[0173] Day-ahead stage: A multi-objective optimization model is constructed with the main objective of maximizing the water power regulation margin, and the renewable energy utilization rate ≥ε1∈[0.85, 0.95] and system output fluctuation ≤ε2∈[800000, 1200000] are set as constraints. The ε-constraint method is used to scan the non-inferior solution set, and the optimal compromise solution is selected by the minimum polarization distance criterion. The calculation results are shown in Figure 4 .

[0174] Subsequently, the corresponding power generation time series is further back-calculated based on this compromise solution as the benchmark, and the dispatching operation trajectory is plotted. The output results of the typical day under the recommended dispatching path in the day-ahead stage are shown in Figure 5 , the new absorption rate is 91.89%, the water consumption is 257 million m 3 , and the fluctuation is 904,280.41 MW. From the time series characteristics, wind, light and water support the load change, although there is some fluctuation at noon and in the evening, the overall operation trajectory is smooth, and the system fluctuation is effectively suppressed.

[0175] Intra-day stage: The opportunity constraint rolling optimization mechanism is introduced to dynamically correct the prediction and update the power generation boundary at 15-minute intervals, and each rolling optimization covers the next 4 hours to perform the first period dispatching. The optimization objectives are minimum supply-demand deviation, minimum curtailment / loss of load, and maximum responsiveness to ensure the operation safety under 95% confidence level. Figure 6 and Figure 7The results show that the dispatching strategy achieves 100% new energy absorption rate, no power loss and no load loss, the total power generation is 143,912.18 MWh, and the system has excellent response and regulation ability.

Claims

1. A multi-energy complementary optimization scheduling method for extreme high temperature scenarios taking into account uncertainty, characterized by: Please follow the steps below to implement it: Step 1: Fusing a bidirectional temporal convolutional neural network-bidirectional long short-term memory network-attention mechanism (BiTCN-BiLSTM-AM) with a quantile regression forest (QRF) jointly extracts the spatiotemporal characteristics of wind speed, solar irradiation, and load time series, achieving high-precision point forecasts and 95% confidence interval modeling. A typical daily scenario set is constructed based on this fusion model. Step 2: A two-stage scheduling model was constructed that integrates ε-constrained multi-objective optimization and chance-constrained mixed integer programming. This model optimizes margins and absorption rates at the hourly level, and makes rolling adjustments to scheduling paths at a 15-minute granularity within the day, enabling efficient response and dynamic adjustment to uncertain scenarios. Step 3: A temperature-coupled correction mechanism for wind, solar, and hydropower output in extreme high-temperature scenarios was proposed: wind power incorporates an air density-wind speed linkage correction, photovoltaic power establishes a module temperature response model, and hydropower constructs an evaporation-water level-output coupled regulation structure. This mechanism integrates three types of physical correction models within a unified framework to achieve dynamic adjustment of power forecasts. Step 4: Integrate predictive modeling, high-temperature correction, and scheduling optimization modules to build a closed-loop control process of "prediction-optimization-feedback-correction." This enables full-process linkage in the scheduling chain, improving the system's stability and responsiveness in extreme high-temperature scenarios.

2. The multi-energy complementary optimization scheduling method for extreme high temperature scenarios taking uncertainty into account according to claim 1 is characterized in that: The step 1 is specifically implemented as follows: In the forecasting of wind speed, solar irradiation, and receiving grid load, this paper proposes a BiTCN-BiLSTM-AM model combined with QRF to handle uncertainty. BiTCN is used to extract long-term dependencies and capture the volatility of time series, while BiLSTM further utilizes bidirectional propagation to capture the relationship between previous and next time steps. In BiTCN, it is assumed that the input time series data is X = {x1,x2,...,x T }, the input x at each moment t After processing through convolutional layers and activation functions, the output is a feature sequence that has passed through multiple convolutional layers. For the convolution operation of layer l, the output is It can be expressed as: in: is the l-th layer convolution kernel, is the reverse convolution kernel of layer l, b (l) is the bias, σ(·) is the activation function, k is the size of the convolution kernel, and t is the current time step. In the stacked structure of BiTCN, as the number of layers increases, the output features will gradually abstract and capture higher-order temporal dependencies. Ultimately, the output sequence Z of the BiTCN layer is t is fed into the BiLSTM layer. The BiLSTM structure combines forward and backward LSTM, and and backpropagation To learn the bidirectional dependency of the data, the final output is: On this basis, we introduce QRF to map the point prediction result Y into a probability density function f(Y), which characterizes the predicted distribution characteristics of wind speed, solar radiation intensity, and load variables. Its training process relies on a random forest structure, integrating the results of multiple decision trees to generate the final distribution estimate. Specifically, it is expressed as: f(Y t )=P(Y t ∣X t ) (3) Where: X t is the input feature vector, P(Y t ) is the probability distribution of the predicted value estimated by the QRF model. QRF optimizes the uncertainty estimate through a tree structure, thereby outputting the predicted probability density function (PDF) at each time step, further providing the sample distribution for the simulation module.

3. The multi-energy complementary optimization scheduling method for extreme high temperature scenarios taking uncertainty into account according to claim 1 is characterized in that: The step 2 is specifically implemented as follows: Step 2 is implemented as follows: To improve the dispatch performance of wind, solar, and hydropower storage systems across different operating periods, this paper constructs a dual-time-scale optimization framework, centered around day-ahead scheduling for robustness and intraday scheduling for flexibility. The day-ahead scheduling model uses a one-hour time granularity to perform 24-hour forward-looking forecasts and resource coordination, while the intraday scheduling model employs a 15-minute rolling optimization approach and continuously updates the dispatch plan based on the latest load and renewable energy output forecasts. Within this optimization framework, the objective function is designed for each stage. (a) Day-ahead risk scheduling The day-ahead dispatch phase is based on forecast information, with the focus on developing a dispatch plan that is highly executable and has high resource utilization efficiency. Based on the operational characteristics of the wind, solar, and hydropower storage systems, the following objective function is set: (1) Maximizing reservoir regulation capacity: Where: SOC(t) represents the state of hydropower adjustable resources (such as water level, water volume or equivalent energy) at time t, The expected amount of water left in the reservoir at the end of the period. (2) Maximizing the utilization of renewable energy: in: are the wind power and photovoltaic power outputs effectively absorbed by the system at time t. (3) Maximizing system output stability: Where: P gen (t) is the total power generated by the system at time t. (b) Intraday opportunity-constrained scheduling Intraday scheduling mainly relies on rolling forecast correction to achieve rapid response to actual load and new energy uncertainty disturbances. The objective function design focuses on short-term deviation control and real-time adjustment capabilities: (1) Minimize system operation deviation: Where: T' is the rolling optimization period within the day, P load (t) is the load demand after considering the prediction error. (2) Maximizing responsiveness of flexible resource adjustments: Where: P ESS (t) is the net output of the energy storage system. (3) Minimizing power abandonment and load loss: Among them: the first item is the load loss amount, and the second item is the power abandonment amount. (c) Energy storage system output constraints (1) State of Charge (SOC) capacity constraints: Among them: SOC ESS (t) is the state of charge of the energy storage system at time t. (2) Power balance constraints: The difference between daily charging and discharging must not exceed a set threshold: P gen (t)+P dis (t)=P load (t)+P chg (t) (11) Where: P dis (t), P ch (t) are the discharging and charging powers of the energy storage, respectively. (3) Opportunity constraints for uncertainty handling: Pr(∣P gen (t)-P load (t)∣≤δ)≥1-∈ (12) Wherein: ∈ represents the maximum constraint violation probability allowed by the system. The present invention sets ∈=0.05 to indicate that the constraint should be satisfied at a confidence level of 95%.

4. The multi-energy complementary optimization scheduling method for extreme high temperature scenarios taking uncertainty into account according to claim 1 is characterized in that: The step 3 is specifically implemented as follows: Step 3 is implemented as follows: (a) Photovoltaic power output In order to improve the accuracy of photovoltaic modeling under extreme high temperatures, this paper proposes a dynamic photovoltaic power estimation method that takes into account the temperature effect. This method introduces the module operating temperature (T c ) estimation mechanism, which more accurately describes the coupled effects of irradiation and ambient temperature on photovoltaic performance than traditional linear correction. The specific formula is as follows: Where: T a is the ambient temperature, G is the current irradiance, and NOCT is the nominal operating temperature of the PV module. Then, the temperature coefficient γ (value -0.004℃ -1 ) Temperature correction for unit component power: P output =P module ·(1+γ(T c -T stc )) (14) Where: T stc is the temperature under standard test conditions, P module is the rated power of each photovoltaic module. c When the increase, the formula (T c -T stc ) will be positive, resulting in P output decline. (b) Wind power output This invention introduces a temperature correction mechanism for wind speed and air density, comprehensively assesses the impact of high temperature on wind power output, and performs dynamic correction calculations on wind power accordingly. The specific steps are as follows: Assume that at standard temperature (15℃, T stc =298.15K) The air density is 1.225kg / m 3 The change in air density according to temperature is calculated using the following formula: Where ρ0 is the air density at standard temperature, and T is the ambient temperature (unit: °C), which is converted to Kelvin temperature for calculation. Based on this, the present invention considers introducing a wind speed adjustment factor for modeling in high temperature extreme weather scenarios (T≥35°C), and its expression is as follows: v adjusted =v base ·(1-α(T-35)),T≥35℃ (16) Among them, v base is a hypothetical baseline wind speed sequence, and T is the current ambient temperature (unit: °C). α is the temperature sensitivity coefficient, with a recommended range of 0.02 to 0.036, representing a 2% to 3.6% decrease in wind speed for every 1°C increase in temperature. If T is < 35°C, no wind speed adjustment is performed. (c) Hydropower output The output of a hydropower unit is affected by the reservoir water level, inflow, and ambient temperature. Extremely high temperatures can exacerbate water evaporation, leading to a drop in water levels and affecting the unit's power generation capacity. Its output can be calculated using the following formula: P hydro (t)=η·g·Q(t)·H(t) (17) Where: P hydro (t) is the output of the hydropower unit at time t, η is the efficiency of the hydropower unit, g is the acceleration of gravity, Q(t) is the flow rate at time t, and H(t) is the water head at time t. To more accurately estimate hydropower output under high temperature scenarios, it is necessary to consider the impact of evaporation on water levels. The change in reservoir water level is determined by precipitation, evaporation, and inflow, and can be expressed by the following formula: ΔH(t)=R(t)-E(t)-Q(t) (18) Where: ΔH(t) is the change in reservoir water level at time t; R(t) is the precipitation at time t; E(t) is the evaporation at time t; Q(t) is the water flow into the reservoir.

5. The multi-energy complementary optimization scheduling method for extreme high temperature scenarios taking uncertainty into account according to claim 1 is characterized in that: The step 4 is specifically implemented as follows: Step 4 is implemented as follows: The present invention constructs a short-term optimization scheduling framework for extreme climate conditions, which is composed of three collaborative links: uncertainty modeling, day-ahead multi-objective scheduling, and intraday rolling optimization, and runs through system prediction, modeling, optimization and real-time adjustment. In the uncertainty analysis stage, BiTCN-BiLSTM-AM is integrated to extract deep spatiotemporal features from multi-source time series data such as wind speed, solar irradiation, and load, and output high-precision point predictions; QRF is further introduced to construct 95% confidence intervals and generate typical uncertainty scenarios, providing input boundaries for scheduling modeling. During the day-ahead optimization phase, based on the aforementioned typical scenarios and dispatchable resources, a three-objective optimization model was constructed, encompassing "maximum reservoir regulation margin, maximum renewable energy absorption rate, and minimum system volatility." The ε-constraint method was then used to systematically scan the Pareto front and extract a compromise dispatch solution that balanced these objectives, which served as the system's initial operational path. During the intraday rolling optimization phase, forecast information is dynamically updated with a 15-minute step size, and unit boundaries are corrected in real time to take into account the extreme high temperature suppression effect; Under the rolling time window mechanism, a chance-constrained mixed integer programming model (MIP) is constructed to execute the first-period scheduling instructions in each round to achieve coordinated optimization of supply-demand deviations, flexible responsiveness, and power curtailment / load loss, ensuring that the system maintains stable, flexible, and resilient operation with high confidence.

Images