A water-wind-solar short-term scheduling resilience improvement method considering uncertainty

Abstract

This invention discloses a method for enhancing the short-term resilience of hydro-wind-solar dispatch considering uncertainties, belonging to the field of power system operation and dispatch technology. The method includes: acquiring multi-source basic data; constructing stochastic scenarios using a differentiated generation model based on the physical and statistical characteristics of uncertainty elements to obtain a joint scenario set of hydro-wind-solar-load; using a scenario reduction algorithm based on Sinkhorn distance to representatively screen the joint scenario set to obtain a reduced scenario set; constructing a short-term stochastic dispatch model with the reduced scenario set as uncertainty input and minimizing system residual load fluctuations as the objective; solving the model to obtain the optimal hydro-wind-solar coordinated dispatch scheme; and executing system operation dispatch according to the scheme. This invention enhances the system's operational resilience under extreme scenarios, smooths residual load fluctuations, reduces peak-shaving pressure, and ensures the safe and stable operation of the system by jointly characterizing multi-source uncertainties and forming a forward-looking dispatch scheme.

Description

Technical Field

[0001] This invention belongs to the field of power system operation and dispatching technology, and in particular relates to a method for improving the short-term dispatching resilience of water-wind-solar power systems considering uncertainties. Background Technology

[0002] As the penetration rate of renewable energy sources such as wind and solar power in the power system continues to increase, the randomness and intermittency of their output pose significant challenges to the short-term peak shaving and safe operation of the power system. Hydropower, due to its rapid and flexible regulation capabilities, is often used in conjunction with wind and solar power to smooth out fluctuations and improve grid integration. However, in actual operation, hydro-wind-solar multi-energy complementary systems are simultaneously affected by multiple uncertainties, including the randomness of wind and solar power output, the uncertainty of runoff processes, and load fluctuations.

[0003] Existing hydro-wind-solar complementary dispatch methods still have significant shortcomings: First, for multiple uncertain sources such as wind power, solar power, runoff, and load, existing methods mostly adopt deterministic prediction or independent modeling, which makes it difficult to effectively characterize the joint statistical characteristics and spatiotemporal correlations among these uncertainties, resulting in deviations between the dispatch model input and the actual operating environment. Second, existing dispatch strategies generally lack forward-looking consideration of extreme operating scenarios, and the formulated schemes have poor adaptability when encountering actual extreme events, easily leading to problems such as water and electricity curtailment and peak-shaving failure. These shortcomings collectively result in insufficient system resilience under uncertain disturbances, i.e., large performance degradation and long recovery time, making it difficult to ensure the safe, reliable, and efficient operation of high-proportion renewable energy systems.

[0004] Therefore, there is an urgent need to develop a new short-term scheduling method that can deeply integrate multi-source uncertain information from wind, solar, hydro, and load, and is centered on improving the resilience of system operation, so as to achieve safe, efficient, and rapid recovery of the watershed water-wind-solar multi-energy complementary system under multiple disturbances. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a short-term water-wind-solar scheduling resilience enhancement method that considers uncertainties, thereby resolving the issues present in the prior art.

[0006] To achieve the above objectives, the present invention provides a method for improving the short-term scheduling resilience of water-wind-solar energy systems considering uncertainties, comprising: Multi-source basic data is acquired, and differentiated generative models are adopted according to the physical and statistical characteristics of different uncertainty factors. Random scenarios are constructed using the multi-source basic data as input to obtain a joint scenario set of water-wind-solar-load. The multi-source basic data includes wind power output data, photovoltaic power output data, runoff data, and system load data. A scene reduction algorithm based on Sinkhorn distance is used to perform representative screening of the water-wind-solar-load joint scene set to obtain a reduced scene set; Using the reduced scenario set as uncertainty input, and aiming to minimize the fluctuation of the system's remaining load within the scheduling period, a short-term stochastic scheduling model is constructed. Solve the short-term stochastic scheduling model to obtain the optimal water-wind-solar coordinated scheduling scheme under various uncertainty scenarios; The system operation scheduling is executed according to the optimal water-wind-solar coordinated scheduling scheme to improve the operational resilience of the short-term water-wind-solar scheduling.

[0007] Optionally, the process of obtaining a joint scene set of water-wind-solar-load includes: The wind power output data and the photovoltaic output data are serialized to obtain a historical wind and solar power combined output sequence; The historical combined wind and solar power output sequence is processed by a conditional variational autoencoder model to generate a random scene of wind and solar power output. The runoff data and the system load data are processed separately using time series decomposition methods to generate random runoff scenarios and random load scenarios. The random scenarios of wind and solar power output, the random scenarios of runoff, and the random scenarios of load are matched and combined to obtain an initial four-dimensional joint scenario set of water-wind-solar-load, and the quality is evaluated to obtain the joint scenario set of water-wind-solar-load.

[0008] Optionally, the conditional variational autoencoder model includes an encoder, a decoder, and a latent space; the conditional variational autoencoder model is trained by minimizing the comprehensive loss function through an optimizer; the comprehensive loss function includes reconstruction mean square error, KL divergence regularization term, energy conservation loss, output correlation loss, ramp rate constraint loss, and ramp insufficiency penalty term.

[0009] Optionally, the historical combined wind and solar power output sequence is processed using a conditional variational autoencoder model to generate a random scene of wind and solar power output, including: A conditional vector for embedding temporal features is constructed; the encoder concatenates the historical wind-solar power output sequence with the conditional vector and maps it to the latent space, outputting Gaussian distribution parameters; the Gaussian distribution parameters are processed using a reparameterization technique to sample latent variables from a standard normal distribution; the decoder processes the combination of the latent variables and the conditional vector to generate random wind-solar power output scenarios in batches; the generated random wind-solar power output scenarios are calibrated using a quantile mapping method to obtain random wind-solar power output scenarios that are consistent with the probability distribution of historical data.

[0010] Optionally, the process of generating stochastic runoff scenarios and stochastic load scenarios includes: Historical runoff time series and historical load time series are decomposed into trend components, seasonal components and residual components using a time series decomposition method; differential perturbation is applied to the residual components to obtain perturbed residual components; the trend components, the seasonal components and the perturbed residual components are superimposed and reconstructed to generate random runoff scenarios and random load scenarios.

[0011] Optionally, the process of obtaining a reduced scene set includes: The water-wind-solar-load joint scenario set is initially screened using a hierarchical clustering strategy to obtain an initial reduced scenario set; a cost matrix is ​​constructed between the water-wind-solar-load joint scenario set and the initial reduced scenario set, and the Sinkhorn distance is calculated by solving the entropy-regularized optimal transmission problem; the optimal transmission plan matrix is ​​calculated based on the dual variables obtained after the Sinkhorn iteration converges; and the probability distribution of the initial reduced scenario set is calculated based on the optimal transmission plan matrix to obtain the reduced scenario set.

[0012] Optionally, a short-term stochastic scheduling model is constructed, including: The water level process of the cascade reservoirs at different times during the scheduling period is set as the decision variable; An objective function is constructed with the goal of minimizing the fluctuation of the system's remaining load under all scenarios. The system's remaining load is determined by the combined output of electricity load and hydropower, wind power, and photovoltaic power. The constraints include reservoir water balance constraints, reservoir storage constraints, water level constraints at the beginning and end of the scheduling period, outflow constraints, hydropower station output constraints, and power grid constraints.

[0013] Optionally, the short-term stochastic scheduling model can be solved using an improved enzyme-based optimization algorithm. The solution process includes: Step 1: Generate multiple candidate solutions representing scheduling schemes by setting the population size, search space dimension, and maximum number of iterations, and randomly initializing them. Calculate the fitness value of each candidate solution. The candidate solution is the substrate position, and all substrate positions constitute the current solution space. Step 2: During the catalytic adaptation phase, two new candidate sites are generated for each substrate; Step 3: Select and update the two new candidate positions by introducing lens imaging reverse learning; Step 4: Explore the current solution space a second time by introducing the Levy flight strategy and update the global optimal solution; Repeat steps two through four until the maximum number of iterations is reached, and output the global optimal solution as the optimal water-wind-solar coordinated scheduling scheme.

[0014] Optionally, the process of generating two new candidate sites for each substrate includes: Based on the current substrate position and the current global optimal substrate position, the first candidate position is generated by combining the adaptive factor. Two different substrate locations are randomly selected to generate a difference vector. The difference vector and the current global optimal substrate location are then used to generate a second candidate location.

[0015] Optionally, the process of selecting and updating the two new candidate locations by introducing lens imaging reverse learning includes: Evaluate the fitness values ​​of the two new candidate positions and select the one with better fitness as the candidate solution to be updated; Whether lens imaging reverse learning is triggered is determined based on random probability. If triggered, a reverse candidate solution is generated based on the candidate solution to be updated. If the fitness value of the reverse candidate solution is better than that of the candidate solution to be updated, then the reverse candidate solution is used to replace the current substrate position.

[0016] Compared with the prior art, the present invention has the following advantages and technical effects: (1) By jointly characterizing the uncertainties of multiple sources such as wind power, photovoltaics, runoff and load, the scheduling model can be improved to adapt to the actual operating environment.

[0017] (2) In scenarios with multiple uncertainties, a forward-looking scheduling scheme is formed, which significantly improves the system's operational resilience under extreme conditions.

[0018] (3) It can effectively smooth the remaining load of the system, reduce peak shaving pressure, and reduce water or electricity waste.

[0019] (4) It is applicable to water-wind-solar multi-energy complementary systems at the watershed scale and has good engineering feasibility. Attached Figure Description

[0020] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a flowchart of step S1 in an embodiment of the present invention; Figure 3 This is a flowchart of step S2 in an embodiment of the present invention; Figure 4 This is a flowchart of step S3 in an embodiment of the present invention; Figure 5 This is a flowchart of step S4 in an embodiment of the present invention; Figure 6 This is a flowchart of step S5 in an embodiment of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0023] Example 1 like Figure 1 As shown, taking the upper reaches of the Yellow River as a case study, this embodiment provides a method for improving the resilience of short-term water-wind-solar scheduling considering uncertainties, including: Step S1: Collect and preprocess wind power output, photovoltaic power output, runoff data and system load data. Based on the physical and statistical characteristics of different uncertainty factors, use differentiated generative models to construct random scenarios and form a water-wind-solar-load joint scenario set. Step S2: Use a scene reduction algorithm based on Sinkhorn distance to perform representative screening of the water-wind-light-load joint scene set, and reduce the scene size while maintaining the joint statistical characteristics of multiple factors. Step S3: Input the above uncertainties and construct a short-term stochastic scheduling model with the goal of minimizing the fluctuation of the system's remaining load within the scheduling period; Step S4: The improved Enzyme Action Optimizer (IEAO) algorithm is used to solve the stochastic scheduling model to obtain the optimal water-wind-solar coordinated scheduling scheme under each uncertainty scenario. Step S5: Guide system operation based on optimization results, improve system performance retention and shorten recovery time under extreme operating scenarios, thereby enhancing system resilience.

[0024] like Figure 2 As shown, step S1 further includes: Step S11: For wind power and photovoltaic output that exhibit strong randomness, nonlinearity and complex spatiotemporal coupling relationship on a short time scale, construct a Conditional Variational Autoencoder (CVAE) model. Take the historical wind and solar power output sequence and conditional variables containing time and seasonal information as input, learn its joint probability distribution through an encoder-decoder structure, and generate a spatiotemporally correlated random scene of wind and solar power output. Step S12: For runoff and load processes that exhibit significant trend, seasonality, and periodicity characteristics, the time series decomposition (STL) method is used to model them, decomposing them into trend, seasonal, and residual components, and applying differential perturbations to the residual components to generate stochastic scenarios. Step S13: Match and combine the wind and solar power output scene generated in step S11 with the runoff and load scene generated in step S12 to form an initial four-dimensional joint scene set of water-wind-solar-load. Step S14: Conduct a quality assessment on the initial four-dimensional joint scene set of water-wind-light-load to verify whether the marginal distribution, autocorrelation characteristics and cross-correlation relationships of each element are consistent with historical data, and ensure that it can truly reflect the joint statistical characteristics and spatiotemporal correlation structure between multiple sources of uncertainty.

[0025] The generated scenarios show a high degree of consistency with historical data in terms of the correlation between wind and solar power output, runoff, and load, with an average absolute error of only 0.0331 for the correlation coefficient. The generated scenarios also exhibit high agreement with historical data in terms of the mean, standard deviation, and quantile characteristics of each factor, as detailed in Table 1. Table 1 Comparison of Statistical Characteristics of Various Elements in the Joint Scene Set

[0026] Step S11 further includes: Step S11a: Collect historical wind power and solar power combined output sequences ( T (For time series length), it is standardized to construct a condition vector. c It is used to embed temporal features, and its structure is as follows: ; In the formula, d It refers to the number of days in the year sequence; For seasonal heat codes; This indicates the weekend.

[0027] Step S11b: Construct the CVAE model, which consists of an encoder, a decoder, and a latent space. The encoder (Encoder...) The historical wind power and photovoltaic combined output sequence x With condition vector c The concatenation is mapped to the latent space, and the Gaussian distribution parameters are output: ; In the formula, and , respectively, are the mean and logarithmic variance of the latent variables; Encoder(⋅) is the encoder network.

[0028] Step S11c: Sample latent variables from the standard normal distribution using reparameterization techniques. z : ; In the formula, For random noise that follows a standard normal distribution; This indicates element-wise multiplication.

[0029] Step S11d: Sample the latent variables z Combined with condition vectors c Input the decoder network to generate random scenes of combined wind and light output in batches. : ; In the formula, Decoder(⋅) is the decoder network.

[0030] Step S11e: Train the model to learn the joint probability distribution of wind and solar power output. Employ a self-supervised learning strategy and use the Adam optimizer to minimize the comprehensive loss function. : ; In the formula, To reconstruct the mean square error; This is the KL divergence regularization term; , , These are losses due to energy conservation, output correlation, and ramp rate constraint, respectively. This is a penalty for insufficient hill climbing. ~ These are the corresponding weighting coefficients.

[0031] Step S11f: For the generated scene The distribution is calibrated using a quantile mapping method to ensure that it is consistent with the probability distribution of historical data.

[0032] Step S12 further includes: Step S12a: Use the STL method to process historical runoff and load time series. y t We decompose it into three independent additive components: ; In the formula, T t For trend components; S t This is a seasonal component; R t For the residual components.

[0033] Step S12b: Implement differentiated perturbation strategies for each decomposed component. The trend and seasonal components retain their original form without perturbation during scene generation. The residual components are generated using an enhanced bootstrap method, combined with the autoregressive process to maintain time correlation. Differentiated perturbations are implemented for different variable types. ; ; ; In the formula, The processed trend components; The processed seasonal components; These are the residual components after the disturbance; To enhance the bootstrap process, a first-order autoregressive process is used to generate residual sequences with similar autocorrelation. It is Gaussian noise. ,in The perturbation variance is set according to the variable type.

[0034] Step S12c, to , and Superimposed and reconstructed stochastic scenarios of runoff or load generation : .

[0035] like Figure 3 As shown, step S2 further includes: Step S21: Select an initial reduced scene set T from the original scene set S using a hierarchical k-means clustering strategy; Step S22: Construct the cost matrix C between the original scene set S and the initial reduced scene set T, and calculate the Sinkhorn distance; Step S23: After the Sinkhorn iteration converges, based on the final dual variables... u and v Calculate the optimal transmission plan matrix π: ; In the formula, K It is a Gibbs matrix; It is in diagonalized form.

[0036] Step S24: Based on the transmission plan matrix π Calculate the probability distribution of the reduced scene set S' : ; In the formula, n This represents the total number of scenes in the original scene set S; i For S, the first i One scenario, i =1,2,…, n ; m This represents the total number of scenes in the initial reduced scene set T; j For the first in T j One scenario, j =1,2,…, m .

[0037] Step S25: Compare the reduced scene set S' with the original scene set S in terms of statistical indicators such as mean and variance to verify the quality of the generated scene and ensure that it can be used for subsequent optimization and scheduling decisions.

[0038] To reduce the computational complexity of the stochastic scheduling model, the initial four-dimensional joint scene set containing 300 scenes (water, wind, light, and load) was reduced. As the scene reduction ratio increased, the computation time decreased significantly, while the increase in accuracy loss and distribution bias was relatively gradual. Specific data are shown in Table 2. Table 2 Performance metrics results at different scaling levels in different scenarios

[0039] Step S22 further includes: Step S22a: Construct the cost matrix C between the original scene set S and the reduced scene set T: ; ; In the formula, g The dimension of the feature; For the calculation based on characteristic variance, the first k The weights of each feature, where var k For the first k The variance of a feature across all scenarios; For S, the first i The first scenario k One eigenvalue; For the first in T j The first scenario k Each feature value.

[0040] Step S22b: Calculate the Sinkhorn distance by solving the entropy-regularized optimal transport problem: ; In the formula, For the feasible domain of the transmission plan, p , q These are the probability distributions for the original and reduced scenes, respectively; For regularization parameters; This is the entropy regularization term.

[0041] Step S22c: Initialize the Gibbs kernel matrix, then alternately update the dual variables until convergence: ; In the formula, h This represents the number of iterations.

[0042] Step S22d: Check the dual variables u and v If convergence has occurred, return to step S22b to continue iterating if convergence has occurred. If convergence has occurred, proceed to the next step.

[0043] like Figure 4 As shown, step S3 further includes: Step S31: Set the decision variables of the model as the water level process of the cascade reservoirs during each period of the scheduling period.

[0044] Step S32: Set the objective function of the model to minimize the fluctuation of the system's remaining load under all possible scenarios. The specific expression is as follows: ; ; In the formula, l Total number of scenes; For the scene The corresponding probability; For the scene exist t The power load at any given time; For the scene exist t Water, scenery, and landscape work together at all times; , , Scenes exist t The wind power output, solar power output, and hydropower output at any given time.

[0045] Step S33: Set the constraints of the model, including: reservoir water balance constraints, reservoir water storage constraints, water level constraints at the beginning and end of the scheduling period, outflow constraints, hydropower station output constraints, and power grid constraints.

[0046] like Figure 5 As shown, step S4 further includes: Step S41: Set basic parameters such as population size, search space dimension and maximum number of iterations, generate multiple candidate solutions through random initialization, these candidate solutions represent possible scheduling schemes, and calculate the fitness value of each candidate solution.

[0047] Step S42: In the catalytic adaptation phase, each substrate Two new candidate positions are generated.

[0048] Step S43: In the location selection and update stage, lens imaging reverse learning is introduced to update the location. The update path is selected by random probability to dynamically balance the global exploration and local development capabilities.

[0049] Step S44: After each generation update, introduce the Levy flight strategy to conduct a secondary exploration of the current solution space and check and update the global best solution.

[0050] Step S45: Repeat steps S42 to S44 until the maximum number of iterations is reached. T Finally, the globally optimal solution is output. .

[0051] Step S42 further includes: Step S42a: Calculate the first candidate position: ; In the formula, For the first t The generation i The substrate in the first j The first candidate position for the dimensional; For the first t The global optimal solution of generation -1; For the first t -1st generation i The substrate in the first j The current position of the dimension; t This represents the number of iterations. It is a random vector within the range [0,1]. As an adaptive factor, .

[0052] Step S42b: Calculate the second candidate position. The algorithm first randomly selects two different substrate positions. , Generate difference vector rUsing this difference vector and the current optimal substrate position, a second candidate position is formed: ; ; In the formula, N Population size; For the first t The generation i The substrate in the first j The second candidate position for wei; , for[ EC The random scaling factor within the range of [1], where EC This represents the enzyme concentration.

[0053] According to one aspect of this application, step S43 further comprises: Step S43a: Evaluate the fitness values ​​of the two candidate positions and select the better position as the update candidate. ; In the formula, To compare the better candidate solutions obtained from the two candidate positions; For position The corresponding fitness value; For position The corresponding fitness value.

[0054] Step S43b: Update the position using lens imaging reverse learning: ; ; In the formula, For the first t The generation i The substrate in the first j The final update position of the dimension; These are inverse candidate solutions generated through inverse learning of lens imaging; For position The corresponding fitness value; For position The corresponding fitness value; and represents the lower and upper bounds of the search space; rand is a uniformly distributed random number in the range [0,1]. The trigger probability threshold for lens imaging reverse learning; k This is the lens magnification factor.

[0055] Step S43c: If the candidate position has better fitness, replace the substrate position; otherwise, retain the substrate position.

[0056] Step S44 further includes: Step S44a: Introduce the Lévy flight strategy after each iteration: ; ; ; In the formula, For the first t +1st generation i The substrate in the first j The initial position of the dimension; This is the step size scaling factor; Let be the random step size of the Lévy flight, which follows a Lévy distribution; The Lévy index; It is a gamma function; As an intermediate variable; and They are respectively The standard deviation.

[0057] Step S44b: Check and update the global optimal solution: like ,but ; in, For position The corresponding fitness value; For the first t The global optimal fitness value of generation -1; For the first t The global optimal solution of the generation; For the first t The globally optimal fitness value of the generation. The above judgment logic is described as follows: If the current number is t The generation i The substrate in the first j Dimensional position The corresponding fitness value Less than the t -1 generation global optimal fitness value Then update the position to the first one. t Global optimal solution of the generation At the same time, update the fitness value at that position to the [number]th [position]. t Global optimal fitness value of generation Otherwise, retain the original global optimal solution and optimal fitness value unchanged.

[0058] like Figure 6 As shown, step S5 further includes: Step S51: Input the optimized scheduling scheme into the actual operating system and execute the water level and power generation scheduling for each time period according to the optimal scheme; specific data are shown in Table 3.

[0059] Table 3 Multidimensional resilience indicators under benchmark scenarios

[0060] Step S52: Design multiple extreme scenarios to conduct system resilience tests, including continuous no wind and no light, severe load fluctuations, and flood events; Three typical extreme operating scenarios were constructed to cover the main high-risk disturbance types that multi-energy complementary systems may face in short-term operation. Scenario A is a scenario of continuous low renewable energy output, simulating a situation where wind and solar power output is less than 5% of rated capacity for 6 consecutive hours; Scenario B is a scenario of severe load fluctuation, where the system load rapidly rises to 150% of the baseline level within 2 hours, and then drops sharply to 50% within 4 hours; Scenario C is a flood event, simulating a short-duration heavy rainfall flood event by amplifying the inflow runoff to 2.5 times, 3.5 times, and 4.0 times the baseline value, respectively.

[0061] Step S53: Evaluate the system performance under extreme scenarios, including residual load fluctuations, water wastage, performance retention, and recovery time; In the scenario of sustained low wind and solar power output (Scenario A), the model can reserve adjustment space for adverse scenarios in advance in the scheduling decision, improving the performance retention rate to 99.09%, and the system operation is basically unaffected by disturbances. In the scenario of severe load fluctuations (Scenario B), the model forms a more adaptive scheduling scheme through multi-scenario optimization, significantly reducing the standard deviation of the remaining load to 146,600 kW and shortening the recovery time to 1 hour, demonstrating stronger dynamic adjustment capabilities. In the scenario of a flood event (Scenario C), the constraint effect of runoff uncertainty on the system's peak-shaving capacity is particularly significant. By explicitly introducing runoff uncertainty into the decision, the model effectively coordinates power generation and flood discharge demands, reducing the water wastage from 85,494,400 m³. 3 Significantly reduced to 1.3618 million m³ 3 At the same time, it maintains a high level of system performance. Specific data is shown in Table 4.

[0062] Table 4 Comparison of multidimensional resilience indicators under three extreme scenarios

[0063] Step S54: Verify the effectiveness of the scheduling scheme in improving system resilience under extreme scenarios through comprehensive evaluation.

[0064] The model demonstrates significant advantages across most dimensions, particularly in terms of residual load standard deviation and recovery time. The results show that the scheduling strategy based on multi-scenario stochastic optimization can significantly improve the system's operational resilience under extreme disturbances. Its core mechanism lies in the fact that the stochastic scheduling model intrinsically incorporates multiple potential adverse scenarios into the decision-making process during the optimization phase, enabling the scheduling scheme to no longer be completely passively adjusted in the face of actual disturbances, but rather possessing a certain degree of forward-looking adaptability.

[0065] This invention provides a short-term water-wind-solar scheduling resilience enhancement system considering uncertainties, used to implement the method described above, including: A central processing unit; and a data storage device communicatively connected to the central processing unit; wherein the data storage device stores computer program instructions, which are read and executed by the central processing unit to perform the water-wind-solar short-term scheduling resilience enhancement method considering uncertainties as defined in steps S1 to S5 above.

[0066] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for improving the resilience of short-term water-wind-solar scheduling considering uncertainties, characterized in that, Includes the following steps: Multi-source basic data is acquired, and differentiated generative models are adopted according to the physical and statistical characteristics of different uncertainty factors. Random scenarios are constructed using the multi-source basic data as input to obtain a joint scenario set of water-wind-solar-load. The multi-source basic data includes wind power output data, photovoltaic power output data, runoff data, and system load data. A scene reduction algorithm based on Sinkhorn distance is used to perform representative screening of the water-wind-solar-load joint scene set to obtain a reduced scene set; Using the reduced scenario set as uncertainty input, and aiming to minimize the fluctuation of the system's remaining load within the scheduling period, a short-term stochastic scheduling model is constructed. Solve the short-term stochastic scheduling model to obtain the optimal water-wind-solar coordinated scheduling scheme under various uncertainty scenarios; The system operation scheduling is executed according to the optimal water-wind-solar coordinated scheduling scheme to improve the operational resilience of the short-term water-wind-solar scheduling.

2. The method for enhancing the resilience of short-term water-wind-solar scheduling considering uncertainties according to claim 1, characterized in that, The process of obtaining a combined scene set of water, wind, solar, and load includes: The wind power output data and the photovoltaic output data are serialized to obtain a historical wind and solar power combined output sequence; The historical combined wind and solar power output sequence is processed by a conditional variational autoencoder model to generate a random scene of wind and solar power output. The runoff data and the system load data are processed separately using time series decomposition methods to generate random runoff scenarios and random load scenarios. The random scenarios of wind and solar power output, the random scenarios of runoff, and the random scenarios of load are matched and combined to obtain an initial four-dimensional joint scenario set of water-wind-solar-load, and the quality is evaluated to obtain the joint scenario set of water-wind-solar-load.

3. The method for enhancing the resilience of short-term water-wind-solar scheduling considering uncertainties according to claim 2, characterized in that, The conditional variational autoencoder model includes an encoder, a decoder, and a latent space; the conditional variational autoencoder model is trained by minimizing the comprehensive loss function through an optimizer; the comprehensive loss function includes reconstruction mean square error, KL divergence regularization term, energy conservation loss, output correlation loss, ramp rate constraint loss, and ramp insufficiency penalty term.

4. The method for enhancing the resilience of short-term water-wind-solar scheduling considering uncertainties according to claim 3, characterized in that, The historical combined wind and solar power output sequence is processed using a conditional variational autoencoder model to generate random wind and solar power output scenarios, including: A conditional vector for embedding temporal features is constructed; the encoder concatenates the historical wind-solar power output sequence with the conditional vector and maps it to the latent space, outputting Gaussian distribution parameters; the Gaussian distribution parameters are processed using a reparameterization technique to sample latent variables from a standard normal distribution; the decoder processes the combination of the latent variables and the conditional vector to generate random wind-solar power output scenarios in batches; the generated random wind-solar power output scenarios are calibrated using a quantile mapping method to obtain random wind-solar power output scenarios that are consistent with the probability distribution of historical data.

5. The method for enhancing the resilience of short-term water-wind-solar scheduling considering uncertainties according to claim 2, characterized in that, The process of generating stochastic runoff scenarios and stochastic load scenarios includes: Historical runoff time series and historical load time series are decomposed into trend components, seasonal components and residual components using a time series decomposition method; differential perturbation is applied to the residual components to obtain perturbed residual components; the trend components, the seasonal components and the perturbed residual components are superimposed and reconstructed to generate random runoff scenarios and random load scenarios.

6. The method for enhancing the resilience of short-term water-wind-solar scheduling considering uncertainties according to claim 1, characterized in that, The process of obtaining a reduced scene set includes: The water-wind-solar-load joint scenario set is initially screened using a hierarchical clustering strategy to obtain an initial reduced scenario set; a cost matrix is ​​constructed between the water-wind-solar-load joint scenario set and the initial reduced scenario set, and the Sinkhorn distance is calculated by solving the entropy-regularized optimal transmission problem; the optimal transmission plan matrix is ​​calculated based on the dual variables obtained after the Sinkhorn iteration converges; and the probability distribution of the initial reduced scenario set is calculated based on the optimal transmission plan matrix to obtain the reduced scenario set.

7. The method for enhancing the resilience of short-term water-wind-solar scheduling considering uncertainties according to claim 1, characterized in that, Constructing a short-term stochastic scheduling model includes: The water level process of the cascade reservoirs at different times during the scheduling period is set as the decision variable; An objective function is constructed with the goal of minimizing the fluctuation of the system's remaining load under all scenarios. The system's remaining load is determined by the combined output of electricity load and hydropower, wind power, and photovoltaic power. The constraints include reservoir water balance constraints, reservoir storage constraints, water level constraints at the beginning and end of the scheduling period, outflow constraints, hydropower station output constraints, and power grid constraints.

8. The method for enhancing the resilience of short-term water-wind-solar scheduling considering uncertainties according to claim 1, characterized in that, The short-term stochastic scheduling model is solved using an improved enzyme-based optimization algorithm. The solution process includes: Step 1: Generate multiple candidate solutions representing scheduling schemes by setting the population size, search space dimension, and maximum number of iterations, and randomly initializing them. Calculate the fitness value of each candidate solution. The candidate solution is the substrate position, and all substrate positions constitute the current solution space. Step 2: During the catalytic adaptation phase, two new candidate sites are generated for each substrate; Step 3: Select and update the two new candidate positions by introducing lens imaging reverse learning; Step 4: Explore the current solution space a second time by introducing the Levy flight strategy and update the global optimal solution; Repeat steps two through four until the maximum number of iterations is reached, and output the global optimal solution as the optimal water-wind-solar coordinated scheduling scheme.

9. The method for enhancing the resilience of short-term water-wind-solar scheduling considering uncertainties according to claim 8, characterized in that, The process of generating two new candidate sites for each substrate includes: Based on the current substrate position and the current global optimal substrate position, the first candidate position is generated by combining the adaptive factor. Two different substrate locations are randomly selected to generate a difference vector. The difference vector and the current global optimal substrate location are then used to generate a second candidate location.

10. The method for enhancing the resilience of short-term water-wind-solar scheduling considering uncertainties according to claim 8, characterized in that, The process of selecting and updating the two new candidate positions by introducing lens imaging reverse learning includes: Evaluate the fitness values ​​of the two new candidate positions and select the one with better fitness as the candidate solution to be updated; Whether lens imaging reverse learning is triggered is determined based on random probability. If triggered, a reverse candidate solution is generated based on the candidate solution to be updated. If the fitness value of the reverse candidate solution is better than that of the candidate solution to be updated, then the reverse candidate solution is used to replace the current substrate position.

Images