Optimized scheduling method and system for wind-solar-hydrogen storage micro-grid

Abstract

The invention discloses an optimal scheduling method and system for a wind-light-hydrogen storage micro-grid, and relates to the field of wind-light-hydrogen storage micro-grids, and the method comprises the steps: obtaining multi-source time sequence data, and carrying out the combined denoising and dynamic time alignment to generate a standardized input sequence; adopting a quantile regression model fused with a space-time attention mechanism to output a power prediction interval of wind and light output and load demand in a future time period; constructing a layered multi-objective optimization model; an improved adaptive particle swarm optimization algorithm is adopted to solve the layered multi-objective optimization model, and an equipment scheduling instruction set is generated; and dynamic correction is carried out, and micro-grid instruction distribution is carried out according to the equipment response priority. According to the method, multi-target conflicts such as power balance, equipment loss and energy efficiency are effectively balanced through multi-source data efficient preprocessing, space-time joint prediction interval generation and hierarchical multi-target optimization, and efficient and stable operation of the wind-light-hydrogen storage micro-grid is achieved.

Description

Technical Field

[0001] The present invention relates to the field of wind-solar-hydrogen storage microgrids, and more particularly to an optimization scheduling method and system for wind-solar-hydrogen storage microgrids. Background Art

[0002] As a key component of distributed energy systems, microgrids enable efficient clean energy utilization and regional energy autonomy by integrating wind and solar power generation, hydrogen energy storage, and batteries. In wind, solar, and hydrogen storage microgrids, the volatility of wind and solar power generation, the time-varying nature of load demand, and the complexity of multi-energy coupling make optimized scheduling a key technology for ensuring system reliability.

[0003] The sampling frequency and time series dynamic characteristics of multi-source data vary significantly. The existing technology lacks a joint denoising and dynamic time alignment mechanism, resulting in low input data quality and accumulated prediction model errors, which seriously affects the accuracy of subsequent scheduling decisions.

[0004] Traditional power forecasting models often rely on single-point time series analysis or spatial feature extraction, failing to fully capture the spatiotemporal coupling between wind and solar power output and load demand. Furthermore, existing models output multiple point forecasts and ignore uncertainty quantification, resulting in scheduling strategies that are unable to effectively address extreme fluctuations. Forecasting methods based on fixed confidence intervals struggle to adapt to the dynamic operating environment of microgrids, limiting the flexibility of risk-based decision-making.

[0005] Furthermore, existing microgrid scheduling optimization models often employ single-objective or simple weighted multi-objective approaches, making it difficult to balance conflicting objectives such as power balance, equipment loss, and energy efficiency. Furthermore, traditional particle swarm optimization algorithms suffer from fixed inertia weights and a single encoding scheme. They are prone to falling into local optima and have slow convergence, making them unable to meet the real-time requirements of multi-device coordinated scheduling under complex constraints.

[0006] Therefore, how to design an optimized scheduling method for wind-solar-hydrogen-storage microgrids that can improve data quality, quantify prediction uncertainty, and achieve efficient and reliable operation of wind-solar-hydrogen-storage microgrids is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0007] In view of this, the present invention provides an optimized scheduling method for a wind, solar, hydrogen and storage microgrid. By efficiently combining denoising and dynamic alignment of multi-source data, accurately capturing the spatiotemporal correlation of wind and solar output and load demand, and quickly finding the global optimal solution under complex constraints, the optimized scheduling algorithm can achieve fine control and optimized scheduling of each component in the wind, solar, hydrogen and storage microgrid.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] In a first aspect, the present invention provides a method for optimizing the scheduling of a wind-solar-hydrogen storage microgrid, comprising the following steps:

[0010] S1. Obtain multi-source time series data on wind and solar power output, load demand, and equipment status, and perform joint denoising and dynamic time alignment to generate a standardized input sequence;

[0011] S2. Based on the standardized input sequence, a quantile regression model integrating a spatiotemporal attention mechanism is used to output power forecast intervals for wind and solar power output and load demand in future time periods;

[0012] S3. Constructing a hierarchical multi-objective optimization model based on the power prediction interval;

[0013] S4. Using an improved adaptive particle swarm optimization algorithm, solving the hierarchical multi-objective optimization model to generate a device scheduling instruction set;

[0014] S5. Combined with the rolling optimization algorithm and feedback control mechanism, the device scheduling instruction set is dynamically modified, and the microgrid instructions are allocated according to the device response priority.

[0015] Furthermore, in S1, performing joint denoising includes:

[0016] The wavelet transform-empirical mode decomposition algorithm is used to decompose the multi-source time series data, and the intrinsic mode components with a signal-to-noise ratio greater than the preset signal-to-noise ratio threshold S are screened for reconstruction to obtain the denoised multi-source time series data X clean ;

[0017]

[0018] Where M represents the total number of eigenmode components, INF i represents the i-th eigenmode component, H() represents the step function, SNR i represents the signal-to-noise ratio of the i-th eigenmode component.

[0019] Furthermore, the S2 includes:

[0020] S21. Extracting temporal features h through long short-term memory networks t , and a one-dimensional convolutional network to extract spatial features f s , and fuse to generate attention weight Φ;

[0021] Φ=σ(w t h t +w s f s +b)

[0022] Among them, σ represents the sigmoid activation function, w t 、ws represents the weight matrix, b represents the bias term;

[0023] S22. Connect the attention weight Φ to the quantile regression layer to obtain the model output Q τ (P gen |X);

[0024] Q τ (P gen |X)=Φ·(w q ·[h t ;f s ])+b q

[0025] Among them, Q τ (P gen |X) represents the generated data P gen and the input feature X, the predicted value of the τth quantile, [h t ;f s ] represents the concatenation vector of LSTM temporal features and convolutional spatial features, w q 、b q Trainable parameters of the quantile regression layer;

[0026] S23, respectively train the quantile regression models with τ = 0.05 and τ = 0.95 to obtain the corresponding power prediction intervals;

[0027] [P low , P high ]=[Q 0.05 (P gen ), Q 0.95 (P gen )]

[0028] Among them, P low represents the lower bound of the prediction interval, P high represents the upper bound of the prediction interval, Q 0.05 (P gen ) represents the predicted value of τ=0.05 quantile, Q 0.95 (P gen ) represents the predicted value at τ = 0.95 quantile.

[0029] Furthermore, the objective function J of the hierarchical multi-objective optimization model S3 is expressed as:

[0030]

[0031] Among them, J1 represents the power balance term, J2 represents the equipment loss term, and J3 represents the renewable energy efficiency term. Indicates the corresponding weight coefficient.

[0032] Furthermore, the power balance term J1 is expressed as:

[0033]

[0034] Among them, P gen (t) = P wind (t)+P pv (t)+P FC (t)+P EC (t)+P bat (t), P wind (t) represents the wind power output at time t, P pv (t) represents the photovoltaic output at time t, P FC (t) represents the fuel cell power generation at time t, P EC (t) represents the hydrogen production power of the electrolyzer at time t, P bat (t) represents the battery charging and discharging power at time t.

[0035] Furthermore, the equipment loss term J2 is expressed as:

[0036]

[0037] Among them, P EC,rated Indicates the rated power of electrolyzer hydrogen production, SOC bat (t) represents the state of charge of the battery at time t, α represents the loss coefficient of the electrolyzer power deviation from the rated value, β represents the linear loss coefficient of the fuel cell power generation, and γ represents the penalty coefficient of the battery deviation from the median.

[0038] Furthermore, the renewable energy efficiency term J3 is expressed as:

[0039]

[0040] Among them, P load (t) represents the load demand power at time t.

[0041] Furthermore, the S4 includes:

[0042] S41, initialize the particle swarm, define the particle position vector as the power instruction D of the electrolyzer, fuel cell and battery, and randomly generate L particles to meet the device power constraint;

[0043] S42, dynamically adjust the algorithm parameters; wherein, the inertia weight ω t Using a nonlinear decreasing strategy, the learning factors C1 and C2 are adjusted based on the number of iterations;

[0044] S43, performing mixed variable optimization, including: retaining floating-point precision for continuous power variables, and mapping device start and stop instructions through binary coding;

[0045] S44. Calculate the objective function And add dynamic penalty terms to handle constraint violations;

[0046] S45 , after each round of iteration, the non-dominated solution set is screened, the optimal solution is selected by congestion sorting, and a scheduling instruction set D′ including power instructions and start / stop states within a preset time period is generated.

[0047] Furthermore, in S5, a proportional-integral controller is used to dynamically modify the device scheduling instruction set:

[0048]

[0049] Where ΔP adj Indicates the correction amount, K P Represents the proportional gain coefficient, K i represents the integral gain coefficient, h represents the rolling window length, ΔP load (t) represents the load power change at time t, ΔP load (tk) represents the load power change in the past k time steps.

[0050] In a second aspect, the present invention provides an optimization scheduling system for a wind-solar-hydrogen storage microgrid, comprising:

[0051] Data preprocessing module: used to obtain multi-source time series data of wind and solar power output, load demand, and equipment status, and perform joint denoising and dynamic time alignment to generate a standardized input sequence;

[0052] Power prediction interval generation module: used to output the power prediction interval of wind and solar power output and load demand in the future period based on the standardized input sequence and using a quantile regression model integrated with a spatiotemporal attention mechanism;

[0053] A hierarchical multi-objective optimization model construction module is used to construct a hierarchical multi-objective optimization model based on the power prediction interval;

[0054] Optimization solution module: used to solve the hierarchical multi-objective optimization model using an improved adaptive particle swarm optimization algorithm to generate a device scheduling instruction set;

[0055] Instruction allocation module: It is used to combine the rolling optimization algorithm and feedback control mechanism to dynamically modify the device scheduling instruction set and allocate microgrid instructions based on the device response priority.

[0056] It can be seen from the above technical solution that compared with the prior art, the technical solution of the present invention has the following advantages:

[0057] Beneficial effects:

[0058] 1. This solution uses a combined wavelet transform and empirical mode decomposition algorithm to decompose and reconstruct multi-source time series data. By screening intrinsic mode components with high signal-to-noise ratios, it achieves more refined noise removal. It effectively handles non-stationary noise in multi-source data such as wind and solar power output and load demand, addressing the problem of accumulated prediction errors caused by insufficient data preprocessing in existing technologies. This significantly improves the quality of input data and the robustness of subsequent models.

[0059] 2. By combining a long short-term memory (LSTM) network and a one-dimensional convolutional network (CNN), the model simultaneously extracts temporal and spatial features and utilizes a dynamic weighted attention mechanism to fuse them, enhancing its ability to capture the spatiotemporal correlation between wind and solar power output and load demand. Quantile regression is further introduced to generate power prediction intervals, providing not only point predictions but also quantifying uncertainty, making it more suitable for risk-based decision-making in dynamic microgrid scheduling.

[0060] 3. To address the challenge of multi-objective conflicts in microgrid scheduling, this solution constructs a hierarchical multi-objective optimization model, flexibly balancing the priorities of different objectives through weight coefficients. Furthermore, an improved adaptive particle swarm optimization algorithm employs dynamic inertia weights, hybrid variable encoding, and a non-dominated solution screening strategy. This addresses the issues of traditional algorithms prone to local optimality and slow convergence, achieving more efficient global optimization under complex constraints. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0062] Figure 1 A flow chart of an optimized scheduling method for a wind-solar-hydrogen storage microgrid provided in an embodiment of the present invention;

[0063] Figure 2 A schematic diagram of a process for obtaining a corresponding power prediction interval based on a quantile regression model provided by an embodiment of the present invention;

[0064] Figure 3 A schematic diagram of a process for generating a device scheduling instruction set based on an improved adaptive particle swarm optimization algorithm provided by an embodiment of the present invention;

[0065] Figure 4 A schematic diagram of the structure of an optimized scheduling system for a wind-solar-hydrogen storage microgrid provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0067] Example 1;

[0068] like Figure 1 As shown, this embodiment provides an optimization scheduling method for a wind-solar-hydrogen storage microgrid, including the following steps:

[0069] S1. Obtain multi-source time series data on wind and solar power output, load demand, and equipment status, and perform joint denoising and dynamic time alignment to generate a standardized input sequence;

[0070] S2. Based on the standardized input sequence, a quantile regression model integrating a spatiotemporal attention mechanism is used to output power forecast intervals for wind and solar power output and load demand in future time periods;

[0071] S3. Constructing a hierarchical multi-objective optimization model based on the power prediction interval;

[0072] S4. Using an improved adaptive particle swarm optimization algorithm, solving the hierarchical multi-objective optimization model to generate a device scheduling instruction set;

[0073] S5. Combined with the rolling optimization algorithm and feedback control mechanism, the device scheduling instruction set is dynamically modified, and the microgrid instructions are allocated according to the device response priority.

[0074] This method significantly improves the quality of input data and the robustness of subsequent models through efficient joint denoising and dynamic alignment of multi-source data; combined with quantile interval prediction integrating spatiotemporal attention mechanism, it enhances the ability to capture the spatiotemporal correlation between wind and solar power output and load demand, and provides a power prediction interval with quantified uncertainty; adopts hierarchical multi-objective optimization and improved adaptive particle swarm algorithm to achieve more efficient global optimization under complex constraints, and realizes efficient and stable operation of wind, solar, hydrogen and storage microgrids.

[0075] The following further describes each step in the above method in detail;

[0076] In this embodiment, S1, multi-source time series data of wind and solar power output, load demand, and device status are obtained, and joint denoising and dynamic time alignment are performed to generate a standardized input sequence;

[0077] Among them, joint denoising includes:

[0078] The wavelet transform-empirical mode decomposition algorithm is used to decompose the multi-source time series data, and the intrinsic mode components with a signal-to-noise ratio greater than the preset signal-to-noise ratio threshold S are screened for reconstruction to obtain the denoised multi-source time series data X clean ;

[0079]

[0080] Where M represents the total number of eigenmode components, INF i represents the i-th eigenmode component, H() represents the step function, SNR i represents the signal-to-noise ratio of the i-th eigenmode component.

[0081] Specifically, the wavelet transform excels at capturing high-frequency noise, while the EMD adaptively decomposes non-stationary signals. The combination of the two can effectively separate noise across different frequency bands while preserving key fluctuation characteristics. In practical applications, if a photovoltaic power station's output data generates high-frequency noise due to sudden weather changes, joint denoising can accurately identify and remove this interference while preserving the true output trend.

[0082] Furthermore, dynamic time alignment includes: using dynamic time warping algorithm to align the time axis of wind speed, light and load data, constraining the maximum deformation distance ∈ DTW ≤0.1T; where,∈ DTW represents the allowable time axis deformation distance, and T represents the length of the data sequence. This solves the problem of time sequence misalignment caused by differences in sampling frequency of multi-source data.

[0083] This step significantly improves the quality and consistency of multi-source data through combined denoising and dynamic alignment, laying a solid foundation for subsequent prediction and optimization.

[0084] In this embodiment S2, based on the standardized input sequence, a quantile regression model integrating a spatiotemporal attention mechanism is used to output the power prediction interval of wind and solar power output and load demand in the future period; Figure 2 As shown, it specifically includes:

[0085] S21. Extracting temporal features h through long short-term memory networks t , and a one-dimensional convolutional network to extract spatial features f s , and fuse to generate attention weight Φ;

[0086] Φ=σ(w t h t +w s f s +b)

[0087] Among them, σ represents the sigmoid activation function, w t 、ws represents the weight matrix, b represents the bias term;

[0088] S22. Connect the attention weight Φ to the quantile regression layer to obtain the model output Q τ (P gen |X);

[0089] Q τ (P gen |X)=Φ·(w q ·[h t ;f s ])+b q

[0090] Among them, Q τ (P gen |X) represents the generated data P gen and the input feature X, the predicted value of the τth quantile, [h t ;f s ] represents the concatenation vector of LSTM temporal features and convolutional spatial features, w q 、b q Trainable parameters of the quantile regression layer;

[0091] S23, respectively train the quantile regression models with τ = 0.05 and τ = 0.95 to obtain the corresponding power prediction intervals;

[0092] [P low , P high ]=[Q 0.05 (P gen ), Q 0.95 (P gen )]

[0093] Among them, P low represents the lower bound of the prediction interval, P high represents the upper bound of the prediction interval, Q 0.05 (P gen ) represents the predicted value of τ=0.05 quantile, Q 0.95 (P gen ) represents the predicted value at τ = 0.95 quantile.

[0094] This step uses a quantile regression model incorporating a spatiotemporal attention mechanism to predict wind and solar power output and load demand for future periods. This model combines the strengths of a long short-term memory (LSTM) network and a one-dimensional convolutional network (CNN). The LSTM extracts temporal features to capture temporal trends in wind and solar power output and load demand, while the CNN extracts spatial features to account for the interactions between different devices or regions. For example, when predicting wind power output, if there is a strong spatial correlation between adjacent wind turbines, the model automatically enhances the contribution of CNN features. Through the attention mechanism, the model can dynamically adjust the importance of different features, resulting in more accurate predictions of future power.

[0095] In addition, quantile regression is introduced to generate power prediction intervals of different quantiles, which not only provides point prediction values ​​of power but also quantifies the uncertainty of the prediction, providing more comprehensive information support for the dynamic scheduling of microgrids.

[0096] In this embodiment S3, a hierarchical multi-objective optimization model is constructed based on the power prediction interval. Specifically, the objective function J of the hierarchical multi-objective optimization model is expressed as:

[0097]

[0098] Among them, J1 represents the power balance term, J2 represents the equipment loss term, and J3 represents the renewable energy efficiency term. Indicates the corresponding weight coefficient.

[0099] Furthermore, the power balance term J1 is expressed as:

[0100]

[0101] Among them, P gen (t) = P wind (t)+P pv (t)+P FC (t)+P EC (t)+P bat (t), P wind (t) represents the wind power output at time t, P pv (t) represents the photovoltaic output at time t, P FC (t) represents the fuel cell power generation at time t, P EC (t) represents the hydrogen production power of the electrolyzer at time t, P bat (t) represents the battery charging and discharging power at time t.

[0102] Furthermore, the equipment loss term J2 is expressed as:

[0103]

[0104] Among them, P EC,rated Indicates the rated power of electrolyzer hydrogen production, SOC bat (t) represents the state of charge of the battery at time t, α represents the loss coefficient of the electrolyzer power deviation from the rated value, β represents the linear loss coefficient of the fuel cell power generation, and γ represents the penalty coefficient of the battery deviation from the median.

[0105] Differentiated loss assessment rules are designed for electrolyzers, fuel cells, and batteries, tailored to their operating characteristics. Electrolyzer losses increase significantly when deviating from rated power, so a quadratic penalty term constrains their operating range. Fuel cell losses are linearly related to their generated power, requiring optimization of output smoothness. Batteries, on the other hand, are penalized for state-of-charge deviations from the median to avoid life degradation caused by overcharging and discharging.

[0106] Furthermore, the renewable energy efficiency term J3 is expressed as:

[0107]

[0108] Among them, P load (t) represents the load demand power at time t.

[0109] Furthermore, the lower bound of the power interval P low As a hard constraint on wind and solar power output:

[0110] P wind (t)+P pv (t)≥P low

[0111] Ensure that wind and solar power output is not lower than the forecast lower limit to avoid overly conservative scheduling.

[0112] It quantifies the utilization rate of wind and solar power generation through renewable energy efficiency, promoting the maximum consumption of clean energy. At the same time, it sets a hard constraint on the lower limit of wind and solar output forecasts to ensure that actual output does not fall below the minimum expected value, thus avoiding energy waste caused by overly conservative scheduling.

[0113] In addition, the hierarchical multi-objective optimization model is also combined with weight coefficients Dynamic adjustment strategies include:

[0114] The initial value is set to

[0115] After each round of optimization, the weights are updated according to the rate of change of the objective function value:

[0116]

[0117] in, represents the new value of the i-th weight coefficient, Represents the old value of the i-th weight coefficient, ΔJ i Represents the rate of change of the i-th objective function.

[0118] The dynamic adjustment strategy for weight coefficients allows the system to flexibly adjust target priorities based on real-time demand. During peak load periods, power balance can be prioritized to reduce the risk of power outages; during equipment aging, loss reduction can be prioritized to extend equipment life. This dynamic balancing mechanism enables the model to adapt to complex and changing operating scenarios, avoiding the local optimality problem caused by single-objective optimization.

[0119] In this step, a hierarchical multi-objective optimization model integrates three key objectives: power balance, equipment losses, and renewable energy efficiency, to construct a microgrid scheduling framework that balances stability and sustainability. Its core lies in a dynamic weight adjustment strategy and differentiated loss modeling. The system can flexibly balance multi-objective conflicts under complex constraints while ensuring basic operational requirements through hard constraints. Combined with a feedback-driven optimization mechanism, the model not only adapts to real-time fluctuations but also optimizes equipment lifespan and energy utilization from a long-term perspective, providing efficient and reliable full-cycle scheduling support for wind, solar, and hydrogen storage microgrids.

[0120] In this embodiment S4, an improved adaptive particle swarm optimization algorithm is used to solve the hierarchical multi-objective optimization model and generate a device scheduling instruction set; Figure 3 As shown, it specifically includes:

[0121] S41, initialize the particle swarm, define the particle position vector as the power instruction D of the electrolyzer, fuel cell and battery, and randomly generate L particles to meet the device power constraint;

[0122] S42, dynamically adjust the algorithm parameters; wherein, the inertia weight ω t Using a nonlinear decreasing strategy, the learning factors C1 and C2 are adjusted based on the number of iterations;

[0123] S43, performing mixed variable optimization, including: retaining floating-point precision for continuous power variables, and mapping device start and stop instructions through binary coding;

[0124] S44. Calculate the objective function And add dynamic penalty terms to handle constraint violations;

[0125] S45 , after each round of iteration, the non-dominated solution set is screened, the optimal solution is selected by congestion sorting, and a scheduling instruction set D′ including power instructions and start / stop states within a preset time period is generated.

[0126] This step uses a dynamic inertia weight and learning factor adjustment strategy, maintaining high exploratory power (large inertia weight) in the early stages and enhancing local search (small inertia weight) in the later stages to avoid premature convergence of the algorithm. Furthermore, hybrid variable encoding optimizes the continuous power instructions and binary variables of the start and stop states of the electrolyzer and fuel cell. Specifically, during periods of low electricity prices, the electrolyzer is prioritized for hydrogen production, while its power is dynamically adjusted to the optimal value. At the same time, non-dominated solution screening is combined with congestion sorting to ensure that the solution set is evenly distributed and covers high-value areas. It can automatically select the solution with the lowest loss and highest efficiency from multiple feasible scheduling schemes.

[0127] It achieves efficient solution of hierarchical multi-objective optimization models, and thus obtains an instruction set that reasonably guides the operation and scheduling of various equipment in the microgrid.

[0128] In this embodiment S5, a rolling optimization algorithm and a feedback control mechanism are combined to dynamically modify the device scheduling instruction set and distribute microgrid instructions according to the device response priority.

[0129] Among them, the proportional-integral controller is used to dynamically correct the equipment scheduling instruction set:

[0130]

[0131] Where ΔP adj Indicates the correction amount, K P Represents the proportional gain coefficient, K i represents the integral gain coefficient, h represents the rolling window length, ΔP load (t) represents the load power change at time t, ΔP load (tk) represents the load power change in the past k time steps.

[0132] During microgrid operation, device response priorities must be dynamically adjusted based on their dynamic characteristics and real-time status. Specifically, when a sudden load increase creates a power shortfall, the system prioritizes discharging the fastest-responding battery and simultaneously activating the fuel cell to supplement it. If the shortfall continues to grow, the electrolyzer hydrogen production power is gradually increased to consume redundant wind and solar power output while reducing non-critical loads. Conversely, when wind and solar power output exceeds capacity, electrolyzer hydrogen production and storage are prioritized, and the fuel cell output is dynamically adjusted to the lowest range.

[0133] Furthermore, when the battery state of charge (SOC) falls below a safety threshold, the system automatically prioritizes hydrogen energy storage to ensure continuity of power supply to core loads. By monitoring device status and external demand in real time and dynamically assigning command priorities, efficient resource coordination and enhanced system resilience are achieved.

[0134] This example proposes an optimized scheduling method for a wind, solar, and hydrogen storage microgrid. This method improves data quality through joint denoising and dynamic alignment of multi-source time series data. It uses a quantile regression model incorporating a spatiotemporal attention mechanism to achieve accurate power forecasting. Furthermore, a hierarchical multi-objective optimization model is constructed to balance power balance, equipment losses, and renewable energy efficiency. Furthermore, by improving the adaptive particle swarm optimization algorithm to solve the model, a device scheduling instruction set is generated. Dynamically corrected using rolling optimization and feedback control mechanisms, the instructions are allocated based on device response priorities, ensuring efficient and stable operation of the microgrid in complex environments.

[0135] Example 2;

[0136] like Figure 4 As shown, this embodiment provides an optimization scheduling system for a wind-solar-hydrogen storage microgrid, including:

[0137] Data preprocessing module: used to obtain multi-source time series data of wind and solar power output, load demand, and equipment status, and perform joint denoising and dynamic time alignment to generate a standardized input sequence;

[0138] Power prediction interval generation module: used to output the power prediction interval of wind and solar power output and load demand in the future period based on the standardized input sequence and using a quantile regression model integrated with a spatiotemporal attention mechanism;

[0139] A hierarchical multi-objective optimization model construction module is used to construct a hierarchical multi-objective optimization model based on the power prediction interval;

[0140] Optimization solution module: used to solve the hierarchical multi-objective optimization model using an improved adaptive particle swarm optimization algorithm to generate a device scheduling instruction set;

[0141] Instruction allocation module: It is used to combine the rolling optimization algorithm and feedback control mechanism to dynamically modify the device scheduling instruction set and allocate microgrid instructions based on the device response priority.

[0142] The optimization scheduling system consists of five modules: data preprocessing, power prediction interval generation, hierarchical multi-objective optimization modeling, improved particle swarm algorithm solution, and dynamic correction and instruction allocation. It improves data quality through joint denoising and dynamic alignment, generates uncertainty-aware power predictions based on the spatiotemporal attention quantile regression model, balances multi-objective conflicts using a hierarchical optimization framework, and quickly solves the optimal scheduling scheme through an improved particle swarm algorithm. Finally, it combines rolling optimization with a feedback mechanism to dynamically adjust instruction priorities, achieving efficient coordinated scheduling and stable operation of wind, solar, hydrogen and storage microgrids in complex and fluctuating scenarios.

[0143] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. References to the same or similar parts between the various embodiments are sufficient. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For relevant parts, refer to the method description.

[0144] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An optimization scheduling method for a wind-solar-hydrogen storage microgrid, characterized in that: The following steps are involved: S1. Obtain multi-source time series data on wind and solar power output, load demand, and equipment status, and perform joint denoising and dynamic time alignment to generate a standardized input sequence; S2. Based on the standardized input sequence, a quantile regression model integrating a spatiotemporal attention mechanism is used to output power forecast intervals for wind and solar power output and load demand in future time periods; S3. Constructing a hierarchical multi-objective optimization model based on the power prediction interval; S4. Using an improved adaptive particle swarm optimization algorithm, solving the hierarchical multi-objective optimization model to generate a device scheduling instruction set; S5. Combined with the rolling optimization algorithm and feedback control mechanism, the device scheduling instruction set is dynamically modified, and the microgrid instructions are allocated according to the device response priority.

2. The optimization scheduling method of a wind-solar-hydrogen storage microgrid according to claim 1 is characterized in that: In S1, performing joint denoising includes: The wavelet transform-empirical mode decomposition algorithm is used to decompose the multi-source time series data, and the intrinsic mode components with a signal-to-noise ratio greater than the preset signal-to-noise ratio threshold S are screened for reconstruction to obtain the denoised multi-source time series data X clean ; Where M represents the total number of eigenmode components, INF i represents the i-th eigenmode component, H() represents the step function, SNR i represents the signal-to-noise ratio of the i-th eigenmode component.

3. The optimization scheduling method of a wind-solar-hydrogen storage microgrid according to claim 1 is characterized in that: Said S2 comprises: S21. Extracting temporal features h through long short-term memory networks t , and a one-dimensional convolutional network to extract spatial features f s , and fuse to generate attention weight Φ; Φ=σ(w t h t +w s f s +b) Among them, σ represents the sigmoid activation function, w t 、w s represents the weight matrix, b represents the bias term; S22. Connect the attention weight Φ to the quantile regression layer to obtain the model output Q τ (P gen |X); Q τ (P gen |X)=Φ·(w q ·[h t ;f s ])+b q Among them, Q τ (P gen |X) represents the generated data P gen and the input feature X, the predicted value of the τth quantile, [h t ;f s ] represents the concatenation vector of LSTM temporal features and convolutional spatial features, w q 、b q Trainable parameters of the quantile regression layer; S23, respectively train the quantile regression models with τ = 0.05 and τ = 0.95 to obtain the corresponding power prediction intervals; [P low ,P high ]=[Q 0.05 (P gen ),Q 0.95 (P gen )] Among them, P low represents the lower bound of the prediction interval, P high represents the upper bound of the prediction interval, Q 0.05 (P gen ) represents the predicted value of τ=0.05 quantile, Q 0.95 (P gen ) represents the predicted value at τ = 0.95 quantile.

4. The optimization scheduling method of a wind-solar-hydrogen storage microgrid according to claim 1 is characterized in that: The objective function J of the hierarchical multi-objective optimization model S3 is expressed as: Among them, J1 represents the power balance term, J2 represents the equipment loss term, and J3 represents the renewable energy efficiency term. Indicates the corresponding weight coefficient.

5. The optimization scheduling method of a wind-solar-hydrogen storage microgrid according to claim 4 is characterized in that: The power balance term J1 is expressed as: Among them, P gen (t) = P wind (t)+P pv (t)+P FC (t)+P EC (t)+P bat (t), P wind (t) represents the wind power output at time t, P pv (t) represents the photovoltaic output at time t, P FC (t) represents the fuel cell power generation at time t, P EC (t) represents the hydrogen production power of the electrolyzer at time t, P bat (t) represents the battery charging and discharging power at time t.

6. The optimization scheduling method of a wind-solar-hydrogen storage microgrid according to claim 4 is characterized in that: The equipment loss term J2 is expressed as: Among them, P EC,rated Indicates the rated power of electrolyzer hydrogen production, SOC bat (t) represents the state of charge of the battery at time t, α represents the loss coefficient of the electrolyzer power deviation from the rated value, β represents the linear loss coefficient of the fuel cell power generation, and γ represents the penalty coefficient of the battery deviation from the median.

7. The optimization scheduling method of a wind-solar-hydrogen storage microgrid according to claim 4 is characterized in that: The renewable energy efficiency term J3 is expressed as: Among them, P load (t) represents the load demand power at time t.

8. The optimization scheduling method of a wind-solar-hydrogen storage microgrid according to claim 1 is characterized in that: Said S4 comprises: S41, initialize the particle swarm, define the particle position vector as the power instruction D of the electrolyzer, fuel cell and battery, and randomly generate L particles to meet the device power constraint; S42, dynamically adjust the algorithm parameters; wherein, the inertia weight ω t Using a nonlinear decreasing strategy, the learning factors C1 and C2 are adjusted based on the number of iterations; S43, performing mixed variable optimization, including: retaining floating-point precision for continuous power variables, and mapping device start and stop instructions through binary coding; S44. Calculate the objective function And add dynamic penalty terms to handle constraint violations; S45 , after each round of iteration, the non-dominated solution set is screened, the optimal solution is selected by congestion sorting, and a scheduling instruction set D′ including power instructions and start / stop states within a preset time period is generated.

9. The optimization scheduling method of a wind-solar-hydrogen storage microgrid according to claim 1 is characterized in that: In S5, a proportional-integral controller is used to dynamically modify the device scheduling instruction set: Where ΔP adj Indicates the correction amount, K P Represents the proportional gain coefficient, K i represents the integral gain coefficient, h represents the rolling window length, ΔP load (t) represents the load power change at time t, ΔP load (tk) represents the load power change in the past k time steps.

10. An optimized dispatching system for a wind-solar-hydrogen storage microgrid, characterized in that: include: Data preprocessing module: used to obtain multi-source time series data of wind and solar power output, load demand, and equipment status, and perform joint denoising and dynamic time alignment to generate a standardized input sequence; Power prediction interval generation module: used to output the power prediction interval of wind and solar power output and load demand in the future period based on the standardized input sequence and using a quantile regression model integrated with a spatiotemporal attention mechanism; A hierarchical multi-objective optimization model construction module is used to construct a hierarchical multi-objective optimization model based on the power prediction interval; Optimization solution module: used to solve the hierarchical multi-objective optimization model using an improved adaptive particle swarm optimization algorithm to generate a device scheduling instruction set; Instruction allocation module: It is used to combine the rolling optimization algorithm and feedback control mechanism to dynamically modify the device scheduling instruction set and allocate microgrid instructions based on the device response priority.

Images