A multi-objective optimization scheduling method for new energy power system considering wind-solar correlation and prediction error

Abstract

This invention relates to a multi-objective optimal scheduling method for new energy power systems that considers wind-solar correlation and prediction errors, belonging to the field of power system optimal scheduling technology. It includes: First, preprocessing the raw data based on adaptive noise complete ensemble empirical mode decomposition technology; second, establishing a wind-solar output prediction model collaboratively constructed by a convolutional neural network and a bidirectional long short-term memory network, and adaptively optimizing the prediction model parameters based on the chaotic sparrow algorithm to extract prediction error features; third, fusing the error modes and original features to establish a wind-solar output coupling cross-prediction mechanism to reuse prediction errors; finally, importing the prediction results into a multi-objective collaborative optimal scheduling model for new energy power systems for day-ahead decision optimization. Compared with independent prediction models, the model proposed in this invention has higher prediction accuracy and a lower total system operating cost, effectively balancing the economic benefits of new energy power system operation with the capacity for new energy absorption.

Description

Technical Field

[0001] This invention relates to a multi-objective optimal scheduling method for new energy power systems that considers the correlation between wind and solar power and prediction errors, and belongs to the field of power system optimal scheduling technology. Background Technology

[0002] New energy power systems (NEPS) are accelerating the transformation of the energy structure. Their significant differences from traditional power systems lie in the significantly increased penetration rate of new energy installations, the strong random fluctuations in power output, and increased forecast uncertainty, posing a severe challenge to the safe and stable operation of the power grid. Currently, NEPS development has broken through the boundaries of single systems, achieving energy synergy and intelligent linkage of multiple types of equipment through the construction of multi-energy network interconnection and data sharing mechanisms, thereby greatly improving the capacity for renewable energy absorption. Furthermore, in recent years, many countries have successively issued relevant documents, pointing out the advantages and necessity of interconnection and data sharing among different energy sources. my country's renewable energy plan also requires improving the accuracy of power forecasting and points out the need to promote the energy complementarity and joint dispatch of renewable energy sources such as hydropower, wind power, and photovoltaics. Therefore, exploring the feasibility of wind and solar joint forecasting and improving forecast accuracy are of great significance for improving the stability of NEPS and enhancing energy utilization efficiency.

[0003] Regarding the improvement of new energy prediction accuracy, Yang Jiajun et al. constructed a heterogeneous graph structure, integrating the power output of multiple new energy sources such as wind and solar power, as well as their spatial and topological information, and used graph neural networks for feature learning to achieve accurate prediction of new energy power. However, early studies used static network architectures for prediction modeling, whose inherent properties limited prediction accuracy and generalization ability. Jing Huitian et al. constructed a multimodal coupled new energy power prediction framework by integrating data reduction algorithms and convolutional neural networks, effectively overcoming the representation bottleneck of traditional single algorithms. Faced with the complexity of multi-source heterogeneous prediction data, simply relying on intelligent algorithm optimization is difficult to effectively overcome the prediction performance degradation caused by defects in the quality of the original data. In terms of decomposition technology, Yang Weixi et al. mainly explored its depth optimization and performance improvement; for example, Zhu Liuzhu et al. comprehensively considered the correlation between source and load, and used wavelet packet decomposition and recurrent neural networks for short-term load prediction; Wang Zhenhao et al. introduced the ensemble empirical mode decomposition (EEMD) method, which made progress in alleviating mode aliasing, but still faces new challenges such as low computational efficiency and poor reconstruction fidelity. In recent years, iterative utilization strategies of prediction residuals have gradually become an emerging research direction in the field of prediction modeling. Error correction methods have been applied to improve the phase synchronization of predictions, thereby improving the lag problem; Wang Fuzhong et al. constructed a residual compensation model and used a multi-order prediction reconstruction strategy to generate integrated prediction output; Liu Junchao et al. achieved synergistic optimization of system operation benefits and risk avoidance by introducing prediction error distribution characteristics; Zhu Jizhong et al. introduced prediction residual information feedback into a multi-objective collaborative scheduling model to correct the stage deviation of wind and solar power combined output; Luo Qing et al., from the perspective of scenario division, deeply analyzed the coupling characteristics of wind and solar power output prediction errors, pointing out that reasonable utilization of residual information can effectively reduce the volatility of wind and solar power output and improve the system load tracking capability.

[0004] In terms of multi-energy synergy and complementarity, Yao Wenliang et al. adopted scenario generation to quantify the random fluctuation characteristics of new energy sources and established a source-load multi-energy flow synergy optimization model for integrated energy systems; Chen Qichao et al. proposed a multi-energy complementary operation framework based on a robust optimization model with embedded new energy output randomness to address the economic dispatch problem of integrated energy systems; in order to achieve risk avoidance, Wang Z et al. combined conditional value at risk (CVaR) risk cost with reserve capacity cost to construct a risk-avoidance-type optimization dispatch method that considers energy storage synergy response.

[0005] While current wind / solar power prediction technologies for single scenarios are relatively mature, the coupling effect of wind-solar hybrid systems, the strong randomness of distributed power sources in NEPS (Non-Powered Power Utilization) systems, and the uncontrollability of prediction error propagation paths collectively lead to deterioration in prediction accuracy, making it difficult for current optimized scheduling schemes to balance economy and completeness. Based on these issues, a multi-objective collaborative optimization scheduling model for NEPS that considers wind-solar output correlation prediction is proposed. First, mode decoupling of the original sequence is achieved through complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), supplemented by approximate entropy and correlation criteria to select the prediction feature set. Second, a CNN-BiLSTM prediction framework is constructed, combining a convolutional neural network (CNN) optimized by the chaos sparrow search algorithm (CSSA) with a bidirectional long short-term memory (BiLSTM) network. Based on this, a wind-solar output coupling cross-prediction mechanism is proposed. Unlike traditional single-power source error feedback, this invention utilizes the strong correlation between wind and solar resources in meteorological factors, taking wind power prediction error as an incremental feature of photovoltaic prediction, establishing a cross-mapping relationship between "error and feature", realizing the effective reuse of error information and secondary prediction of new energy output; finally, based on the day-ahead optimized scheduling scenario, a multi-dimensional performance evaluation of wind and solar joint output prediction is carried out to verify the effectiveness of the proposed model. Summary of the Invention

[0006] To address the problems in existing technologies, this invention provides a multi-objective optimization scheduling method for new energy power systems that considers the correlation between wind and solar power output and prediction errors. This method aims to solve the technical challenge of declining reliability in system scheduling decisions due to the correlation and prediction uncertainties in existing methods regarding wind and solar power output. Compared to independent prediction models, the proposed model achieves higher prediction accuracy and reduces the total system operating cost, effectively balancing the economic benefits of new energy power system operation with the capacity for new energy absorption.

[0007] The technical solution of this invention is: a multi-objective optimal scheduling method for new energy power systems considering wind-solar correlation and prediction errors, the method comprising:

[0008] Step 1: Preprocess the raw data based on the adaptive noise complete set empirical mode decomposition technique;

[0009] Step 2: Establish a wind and solar power output prediction model that is jointly constructed by a convolutional neural network and a bidirectional long short-term memory network, and adaptively optimize the prediction model parameters based on the chaotic sparrow algorithm to extract prediction error features.

[0010] Step 3: Integrate error modes and original feature quantities to establish a wind and solar power output coupling cross-prediction mechanism to realize the reuse of prediction errors;

[0011] Step 4: Import the prediction results into the multi-objective collaborative optimization scheduling model of the new energy power system for day-ahead decision optimization.

[0012] Furthermore, Step 1 includes:

[0013] Gaussian white noise with zero mean n g (t) Repeat the superposition g times to the original sequence x g Above, construct the set of signals to be decomposed, x. g (t), as shown in Equation (1); use internal EMD to perform intrinsic mode decomposition on each signal to be decomposed, integrate all decomposition results and obtain the mean value to obtain IMF; use the remaining residual signal as a new input sequence and repeat until the signal meets the monotonicity termination condition;

[0014] (1)

[0015] Where ε is The weighting coefficients are denoted by g; g represents the sample value.

[0016] Furthermore, Step 2 includes:

[0017] CSSA is used to automatically find and optimize key hyperparameters in CNN and BiLSTM models;

[0018] The BILSTM network model plays a role in temporal prediction in the joint model. Its architecture consists of bidirectional LSTM coupling. By fusing forward and backward temporal features, it effectively solves the problem of missing inverse information in traditional LSTM. This model is sensitive to parameter configuration. Among them, the three general parameters of hidden layer dimension, upper limit of training period, and initial value of learning rate dominate the model's convergence characteristics. To address issues such as prediction performance degradation and convergence lag, CSSA is used to implement parameter-oriented optimization, which simultaneously improves the model's robustness and computational efficiency.

[0019] Furthermore, Step 3 includes:

[0020] By integrating multi-source correlation features into the NEPS short-term joint prediction framework for wind and solar power, the barrier between independent wind and solar power prediction is broken, and an error feedback mechanism is established. Specifically, it consists of four parts: wind and solar power correlation analysis, input feature set CEEMDAN decoupling, CSSA parameter optimization, and CNN-BILSTM model two-layer prediction.

[0021] Furthermore, Step 4 includes:

[0022] Step 4.1, System Parameter Settings:

[0023] By taking the predicted wind and solar power output values ​​that take into account the correlation between wind and solar power as input, a comprehensive multi-objective function that can adaptively balance is established to achieve day-ahead optimal scheduling of NEPS.

[0024] Step 4.2, Establishing the objective function:

[0025] Step 4.2.1, System operating costs:

[0026] (2)

[0027] (3)

[0028] in, This represents the overall operating cost of NEPS at time t; Do not consider the total transaction cost and natural gas purchase cost at time t; The operating costs of the battery, P2G, and GT at time t; Let t be the cost of purchasing electricity from the main grid and the revenue from selling electricity. Let GT represent the natural gas consumption and the user's natural gas demand at time t, respectively. For P2G and GT, the operation and maintenance coefficients are used.

[0029] Step 4.2.2, System environment governance costs:

[0030] (4)

[0031] (5)

[0032] in, Let t be the environmental remediation cost of NEPS at time t; These represent the environmental governance costs of the main network, P2G, and GT at time t, respectively. Let and n represent the treatment cost and total number of types of pollutant x, respectively. Purchase electricity from the main network at time t; These are the emission coefficients of pollutant x in the main grid, P2G, and GT, respectively.

[0033] Step 4.2.3, Minimum Load Loss and CVaR Cost:

[0034] Uncertainty in new energy output leads to a contraction in system regulation margin, inducing the risk of load shedding and wind / solar curtailment; optimizing the allocation of reserve resources can effectively cope with unexpected disturbances such as sudden shutdowns, grid failures, and sudden changes in environmental parameters; the cost of reserve capacity is shown in equation (6):

[0035] (6)

[0036] Among them, a t Let λ be the wind power confidence level at time t; T be the system operating period; and λ be the penalty coefficient. Total cost for backup.

[0037] When the output of new energy sources exceeds the boundary of its confidence interval, the system will face the risk of wind / solar curtailment or insufficient power supply. Based on this dual risk scenario, a risk-cost mapping mechanism is constructed using the CVaR model to achieve an economic quantification of operational risks.

[0038] Step 4.3: Membership function processing;

[0039] To overcome the surge in computational complexity caused by simultaneous optimization of multiple objective functions, a membership-weighted aggregation mechanism is adopted to realize the mathematical transformation from multiple objectives to a single objective, effectively reducing the search dimension of the Pareto solution set and accelerating the optimization process;

[0040] Step 4.4 System constraints, including grid constraints, WT constraints, P2G constraints, and GT constraints;

[0041] Step 4.5: Solve using the genetic particle swarm optimization algorithm.

[0042] Furthermore, Step 4.4 includes:

[0043] Step 4.4.1, Power Grid Constraints:

[0044] (7)

[0045] in, This represents the electrical power value of the NEPS interacting with the main grid at time t. When the value is greater than 0, it indicates that electricity is purchased from the main grid. A value less than 0 indicates that electricity is sold to the main grid. These represent the upper and lower limits of the interaction between NEPS and the main grid power; These represent the wind and solar power absorption capacity at time t; These represent the power generation and power consumption of the P2G at time t, respectively. This represents the output value of GT at time t.

[0046] Step 4.4.2, WT Constraint:

[0047] (8)

[0048] Step 4.4.3, P2G Constraints:

[0049] (9)

[0050] Step 4.4.4, GT Constraints:

[0051] (10)

[0052] in, This represents the predicted maximum output of WT at time t; , These represent the wind speed values ​​when WT enters / exits the main grid, respectively. Represents the rated wind speed; Δt represents the rated output power of WT; Δt represents the time step. Represents natural gas production. ; The unit representing the higher heating value of methane is 39 MJ / m³ under standard conditions. ; Represents the P2G electro-gas conversion efficiency; Represents the gas-to-electricity conversion coefficient; This represents the maximum electrical power of the P2G; The gas consumption of GT within the time period t; Represents the lower heating value of natural gas; Represents GT power generation efficiency; These represent the upper and lower limits of GT output, respectively. These represent the upper and lower limits of the GT gradeability, respectively.

[0053] The present invention also provides a multi-objective optimal scheduling method for a new energy power system that considers wind-solar correlation and prediction error. The system includes a module for executing the aforementioned multi-objective optimal scheduling method for a new energy power system that considers wind-solar correlation and prediction error.

[0054] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned multi-objective optimization scheduling method for a new energy power system that considers wind and solar correlation and prediction errors.

[0055] The present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the multi-objective optimization scheduling method for a new energy power system that considers wind and solar correlation and prediction errors.

[0056] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the multi-objective optimization scheduling method for new energy power systems that considers wind and solar correlation and prediction errors.

[0057] The beneficial effects of this invention are:

[0058] (1) In the collaborative prediction of wind-solar hybrid power generation, this invention proposes a wind-solar coupling cross-prediction mechanism based on error reuse. By coupling the spatiotemporal correlation characteristics among multiple influencing factors, complementary information in the wind-solar prediction error is mined and cross-compensated, effectively capturing the fluctuation characteristics that are difficult to represent with single data, and significantly improving the accuracy of joint wind-solar prediction. Compared with independent prediction models, this multi-source collaborative prediction model has better prediction performance;

[0059] (2) In the prediction model construction part of this invention, compared with the traditional subjective adjustment and PSO parameter optimization methods, CSSA is used to adaptively optimize the hyperparameters of BiLSTM, which can effectively enhance the model's adaptive capability. This intelligent optimization strategy not only significantly improves the efficiency of parameter optimization, but also effectively avoids the problem of prediction performance degradation caused by parameter mismatch;

[0060] (3) Compared with wind and solar independent forecasting, the present invention uses the new energy output forecast value that takes into account the correlation between wind and solar to optimize the scheduling of the NEPS system, which has lower risk reserve cost and system operation and maintenance cost, and has better overall economic efficiency. It is conducive to the consumption of new energy and meets the requirements of green and low carbon. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the NEPS prediction model that considers the correlation between wind and solar power in this invention.

[0062] Figure 2 This is a schematic diagram of the prediction process of the joint prediction model in this invention;

[0063] Figure 3 This is a schematic diagram of the GAPSO process in this invention;

[0064] Figure 4 This is a schematic diagram illustrating the correlation analysis of input variables for wind power prediction in this invention;

[0065] Figure 5 This is a schematic diagram of the correlation analysis of input variables for photovoltaic power prediction in this invention;

[0066] Figure 6 This is a schematic diagram of the wind / solar power prediction results in this invention;

[0067] Figure 7 This is a schematic diagram comparing wind power generation prediction in this invention;

[0068] Figure 8 This is a schematic diagram comparing the photovoltaic power generation prediction in this invention;

[0069] Figure 9 This is a schematic diagram showing the output of each unit in the system under different scenarios in this invention;

[0070] Figure 10 This is a schematic diagram of the electro-pneumatic network coupling system in this invention. Detailed Implementation

[0071] Example 1: As Figures 1-10 As shown, a multi-objective optimization scheduling method for a new energy power system considering wind-solar correlation and prediction error is characterized in that: the method includes:

[0072] Step 1: Preprocess the raw data based on the adaptive noise complete set empirical mode decomposition technique;

[0073] Step 2: Establish a wind and solar power output prediction model that is jointly constructed by a convolutional neural network and a bidirectional long short-term memory network, and adaptively optimize the prediction model parameters based on the chaotic sparrow algorithm to extract prediction error features.

[0074] Step 3: Integrate error modes and original feature quantities to establish a wind and solar power output coupling cross-prediction mechanism to realize the reuse of prediction errors;

[0075] Step 4: Import the prediction results into the multi-objective collaborative optimization scheduling model of the new energy power system for day-ahead decision optimization.

[0076] Furthermore, Step 1 includes:

[0077] Due to the strong randomness and complex signal characteristics of wind and solar power and meteorological data, data decoupling is necessary to improve prediction accuracy. Traditional EMD suffers from mode aliasing, leading to distortion of the physical meaning of intrinsic mode functions; while EEMD alleviates mode aliasing, it still has limitations such as low computational efficiency and large reconstruction residuals. In contrast, CEEMDAN significantly optimizes the decomposition effect of nonlinear / non-stationary time series through an adaptive noise injection mechanism. Its calculation steps are as follows:

[0078] Gaussian white noise with zero mean n g (t) Repeat the superposition g times to the original sequence x g Above, construct the set of signals to be decomposed, x. g(t), as shown in Equation (1); use internal EMD to perform intrinsic mode decomposition on each signal to be decomposed, integrate all decomposition results and obtain the mean value to obtain IMF; use the remaining residual signal as a new input sequence and repeat until the signal meets the monotonicity termination condition;

[0079] (1)

[0080] Where ε is The weighting coefficients; g is the sample value;

[0081] Furthermore, Step 2 includes:

[0082] The Chaotic Sparrow Algorithm (CSSA) network model plays a role in hyperparameter adaptive optimization in joint models. Traditional Sparrow Algorithm (SSA) uses random initialization to generate initial solutions, which leads to a low degree of uniformity in the initial population distribution, limiting the algorithm's optimization ability and making it susceptible to local optima. Chaotic mapping, with its randomness, ergodicity, and regularity, allows the SSA algorithm to explore the search space more dynamically and globally, greatly improving its susceptibility to local optima and low convergence accuracy. Furthermore, using CSSA to automatically find and optimize key hyperparameters in CNN and BiLSTM models avoids the time-consuming and unstable manual parameter tuning process, significantly reducing the influence of subjective human factors on the model.

[0083] CNN network models play a role in feature extraction in joint models. They are often used for high-dimensional mapping processing when external data is imported. They automatically extract short-term, local dependencies and hidden patterns in the input data through convolutional kernels. The model adopts a sparse connection structure, which reduces parameter complexity and maintains the consistency and stability of the feature extraction process through a weight sharing mechanism, thereby significantly improving the efficiency and accuracy of feature data extraction.

[0084] The BILSTM network model plays a role in temporal prediction within the joint model. Its architecture consists of bidirectional LSTM coupling, effectively addressing the problem of missing inverse information in traditional LSTMs by fusing forward and reverse temporal features, thus significantly improving the prediction performance of sequence data. This model is highly sensitive to parameter configuration, with the hidden layer dimension, upper limit of training epochs, and initial learning rate being the three general parameters that dominate the model's convergence characteristics. To address issues such as prediction performance degradation and convergence lag, CSSA is used for parameter-oriented optimization, simultaneously improving model robustness and computational efficiency.

[0085] Furthermore, Step 3 includes:

[0086] Since meteorological factors such as wind speed, irradiance, and temperature have a combined impact on wind and solar power output, the prediction error of one of them often contains common meteorological disturbance information that has not been fully extracted by the model. Existing prediction models are insufficient in capturing the spatiotemporal correlation of wind and solar power in New Energy Power Systems (NEPS), and their error correction mechanisms lack adaptability. This invention constructs a short-term joint prediction framework for NEPS wind and solar power that integrates multi-source correlation features, breaking down the barriers between independent wind and solar prediction and establishing an error feedback mechanism. Specifically, it consists of four parts: wind and solar power correlation analysis, input feature set CEEMDAN decoupling, CSSA parameter optimization, and a two-layer prediction using a CNN-BILSTM model. The specific structure is as follows: Figure 1 As shown, the prediction process steps are as follows: Figure 2 As shown.

[0087] Furthermore, Step 4 includes:

[0088] Step 4.1, System Parameter Settings:

[0089] Using the predicted wind and solar power output, which considers the correlation between wind and solar power, as input, a comprehensive multi-objective function capable of adaptive balance is established to achieve day-ahead optimal scheduling of NEPS (Non-Electric Power Utilization System). Electro-pneumatic coupled network systems, such as... Figure 10 As shown in the figure. A simplified equivalent summary of the park system is provided based on the input-output characteristics of each component, output upper and lower limits constraints, and ramping constraints. The main consideration is the conversion and utilization of energy flow in the power grid and gas grid power flow components, thereby simplifying the system and calculations. CvaR cost represents the cost of quantifying the uncertainty risk of wind / solar output, with wind / solar confidence fixed at 95%. The key output units of NEPS include the gas turbine (GT), wind turbine (WT), photovoltaic (PV), power-to-gas (P2G) equipment, and battery (BESS), and their specific parameters are shown in Table 1-3.

[0090] Table 1 shows the unit parameters.

[0091] Table 2 shows the pollutant parameters.

[0092] Table 3 shows the energy storage parameters.

[0093] Step 4.2, Establishing the objective function:

[0094] Step 4.2.1, System operating costs:

[0095] (2)

[0096] (3)

[0097] in, This represents the overall operating cost of NEPS at time t; Do not consider the total transaction cost and natural gas purchase cost at time t; The operating costs of the battery, P2G, and GT at time t; Let t be the cost of purchasing electricity from the main grid and the revenue from selling electricity. Let GT represent the natural gas consumption and the user's natural gas demand at time t, respectively. For P2G and GT, the operation and maintenance coefficients are used.

[0098] Step 4.2.2, System environment governance costs:

[0099] (4)

[0100] (5)

[0101] in, Let t be the environmental remediation cost of NEPS at time t; These represent the environmental governance costs of the main network, P2G, and GT at time t, respectively. Let and n represent the treatment cost and total number of types of pollutant x, respectively. Purchase electricity from the main network at time t; These are the emission coefficients of pollutant x in the main grid, P2G, and GT, respectively.

[0102] Step 4.2.3, Minimum Load Loss and CVaR Cost:

[0103] Uncertainty in renewable energy output leads to a contraction in system regulation margin, inducing the risk of load shedding and wind / solar curtailment. Optimizing the allocation of reserve resources can effectively address unexpected disturbances such as sudden outages, grid failures, and sudden changes in environmental parameters. The cost of reserve capacity is shown in equation (6):

[0104] (6)

[0105] Among them, a t Let λ be the wind power confidence level at time t; T be the system operating period; and λ be the penalty coefficient. Total cost for backup.

[0106] When the output of new energy sources exceeds the boundary of its confidence interval, the system will face the risk of wind / solar curtailment or insufficient power supply. Based on this dual-risk scenario, a risk-cost mapping mechanism is constructed using a CVaR model to achieve a quantitative representation of the economic risks of operation. Taking wind power as an example, the CVaR cost is:

[0107] (7)

[0108] In the formula: , These represent the risk costs at time t for exceeding and falling below the confidence range, respectively. Represents the number of Wts; , These represent the wind power prediction error and the predicted output value at time t, respectively. , These represent the upper and lower limits of the system's ability to absorb wind power at time t, respectively. This represents the error probability density function.

[0109] The minimum load loss is:

[0110] (8)

[0111] in, Let t be the NEPS load loss at time t.

[0112] Step 4.3: Membership Function Processing

[0113] To overcome the surge in computational complexity caused by simultaneous optimization of multiple objective functions, a membership-weighted aggregation mechanism is adopted to realize the mathematical transformation from multiple objectives to a single objective, effectively reducing the search dimension of the Pareto solution set and accelerating the optimization process.

[0114] Step 4.4, System Constraints

[0115] Step 4.4.1, Power Grid Constraints:

[0116] (9)

[0117] in, This represents the electrical power value of the NEPS interacting with the main grid at time t. When the value is greater than 0, it indicates that electricity is purchased from the main grid. A value less than 0 indicates that electricity is sold to the main grid. These represent the upper and lower limits of the interaction between NEPS and the main grid power; These represent the wind and solar power absorption capacity at time t; These represent the power generation and power consumption of the P2G at time t, respectively. This represents the output value of GT at time t.

[0118] Step 4.4.2, WT Constraint:

[0119] (10)

[0120] Step 4.4.3, P2G Constraints:

[0121] (13)

[0122] Step 4.4.4, GT Constraints:

[0123] (14)

[0124] in, This represents the predicted maximum output of WT at time t; , These represent the wind speed values ​​when WT enters / exits the main grid, respectively. Represents the rated wind speed; Δt represents the rated output power of WT; Δt represents the time step. Represents natural gas production. ; The unit representing the higher heating value of methane is 39 MJ / m³ under standard conditions. ; Represents the P2G electro-gas conversion efficiency; Represents the gas-to-electricity conversion coefficient; This represents the maximum electrical power of the P2G; The gas consumption of GT within the time period t; Represents the lower heating value of natural gas; Represents GT power generation efficiency;

[0125] These represent the upper and lower limits of GT output, respectively. These represent the upper and lower limits of the GT gradeability, respectively.

[0126] Step 4.5, Solving the target algorithm:

[0127] The optimization scheduling model of this invention features multi-objective coupling, nonlinear constraints, and a complex search space. Traditional particle swarm optimization (PSO) algorithms are prone to getting trapped in local optima in complex scenarios, while genetic algorithms (GA), although possessing strong global search capabilities, have slow convergence speeds and suffer from premature convergence. In contrast, the genetic particle swarm optimization algorithm (GAPSO) combines the global search advantages of GA with the fast convergence advantages of PSO, improving iteration efficiency and search stability while maintaining population diversity. Its optimization accuracy and convergence speed are superior to both single algorithms, making it more suitable for the optimization scheduling model of this invention. The specific flowchart of GAPSO is shown below. Figure 3 As shown.

[0128] The present invention also provides a multi-objective optimal scheduling method for a new energy power system that considers wind-solar correlation and prediction error. The system includes a module for executing the aforementioned multi-objective optimal scheduling method for a new energy power system that considers wind-solar correlation and prediction error.

[0129] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned multi-objective optimization scheduling method for a new energy power system that considers wind and solar correlation and prediction errors.

[0130] The present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the multi-objective optimization scheduling method for a new energy power system that considers wind and solar correlation and prediction errors.

[0131] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the multi-objective optimization scheduling method for new energy power systems that considers wind and solar correlation and prediction errors.

[0132] The present invention also includes Step 5: Case analysis and verification.

[0133] The specific steps of Step 5 are as follows:

[0134] Step 5.1, Verification of the short-term joint wind and solar forecasting model:

[0135] Since the proposed model comprises two parts—wind-solar correlation prediction and day-ahead optimal scheduling—the wind-solar correlation prediction provides data for day-ahead optimal scheduling. Therefore, when conducting case studies, the accuracy of the model's predictions should be verified first, followed by the rationality of the model's scheduling.

[0136] The proposed prediction model is based on hourly wind / solar power and meteorological data from an actual renewable energy power station in East China. The data collection period was from September 1, 2021 to August 31, 2022, with a sampling interval of 1 hour. The window length was set to 30 days to predict the 31st day, with a step size of 1 day.

[0137] Step 5.1.1, Screening of the Scenery-Related Feature Set:

[0138] The Pearson correlation coefficient is used to determine the degree of association between power and each influencing factor. The correlation results of the sample are as follows: Figure 4 and Figure 5 As shown.

[0139] Table 4 shows the Pearson correlation classification.

[0140] like Figure 4-5As shown, the correlation strength of each feature variable with wind power / solar power output varies significantly. Given that multiple weakly correlated features can easily cause instability in the prediction sequence and lead to accuracy degradation, based on the threshold standard mentioned by Li Li et al., a correlation coefficient of 0.2 is used as the dividing threshold between strong and weak correlations. Only feature variables with correlation coefficients exceeding this threshold are retained to construct the input feature set of the prediction model.

[0141] Step 5.1.2, Analysis of the effect of the joint prediction model:

[0142] To verify the effectiveness of the proposed model, three other models—CEEMDAN-SSA-CNN-BILSTM (hereinafter referred to as CE-S-CN-B), CEEMDAN-CSSA-BILSTM (hereinafter referred to as CE-CS-B), and CSSA-CNN-BILSTM (hereinafter referred to as CS-CN-B)—were selected for comparison. The comparative analysis of the prediction performance of each model is as follows: Figure 6 As shown, since the photovoltaic output value is greatly affected by the weather, in order to reflect the adaptability of the proposed model, data from one day each of sunny, cloudy, and rainy days were selected for comparison.

[0143] Depend on Figure 6 It can be seen that the CE-S-CN-B model has a relatively poor fitting effect. This is because the SSA algorithm often gets trapped in local optima when optimizing the BiLSTM hyperparameters, leading to improper parameter settings in the subsequent prediction model network, reduced model prediction accuracy, and thus limiting model performance. Without the CEEMDAN data processing stage, the prediction curve of the CS-CN-B model fluctuates wildly, and the error is further amplified during periods of significant fluctuation in the true values, resulting in a significant deviation from the measured sequence. In contrast, the proposed method first uses CEEMDAN to smooth the sequence, and then uses the CSSA algorithm to optimize the CNN-BiLSTM hyperparameters, avoiding the local optima that SSA parameter optimization is prone to, as well as the subjective uncertainty brought about by manual parameter tuning. Furthermore, even in scenarios with drastic fluctuations such as rainy weather, the model's estimation of peak points still closely matches the measured values, verifying the model's stability under different conditions.

[0144] Further Figure 4 To objectively verify the effectiveness of the model, quantitative analysis of the indicators was conducted. Based on national standards, a comprehensive evaluation system was used to assess the experimental results. The comparison results of each model are shown in Tables 5 and 6.

[0145] Table 5 shows the predicted wind power output.

[0146] Table 6 shows the photovoltaic power generation prediction results.

[0147] Based on the evaluation results in Tables 5 and 6, the CE-CS-CN-B composite prediction architecture performs better in both wind and solar power prediction tasks, exhibiting smaller prediction errors, higher prediction accuracy, and a higher pass rate. The photovoltaic power prediction accuracy achieved a 100% pass rate, fully validating the rationality and effectiveness of the CE-CS-CN-B composite prediction architecture.

[0148] Step 5.1.3 Analysis of the two-layer prediction effect including wind / solar power prediction errors:

[0149] The predictive performance of the model is evaluated by setting up comparative experiments, and the key is to assess the effectiveness of the predictive error characteristics. Figure 7 , Figure 8 The prediction curves obtained with and without the prediction error feature set are shown in Table 7, and the error comparison results further confirm the above judgment. Among them, the photovoltaic output prediction selected one day's data each for sunny, cloudy, and rainy days for comparison.

[0150] Table 7 shows the wind / solar power prediction results.

[0151] The comparison of the above charts shows that:

[0152] 1) As shown in Table 7, under consistent meteorological conditions, the RMSE index of the model in predicting photovoltaic / wind power with cross-prediction errors of wind and solar power is reduced by 5.20% and 6.24% respectively compared with the baseline scenario without errors. Furthermore, the photovoltaic power prediction pass rate reaches 100%, confirming that the model has good prediction stability and accuracy. This also demonstrates that using the prediction errors between wind and solar power as feature inputs can effectively improve the overall accuracy of wind and solar power prediction.

[0153] 2) Combining Table 7 with Figure 7 , Figure 8 The prediction curves show a significant coupling correlation between wind power and photovoltaic power prediction errors. Incorporating the photovoltaic prediction deviation into the feature space effectively improves the accuracy of wind power prediction; conversely, incorporating the photovoltaic prediction deviation into the feature space also effectively improves the accuracy of wind power prediction, verifying the synergistic enhancement effect of multi-energy prediction.

[0154] The reason for the above phenomenon is that by incorporating the model's own prediction bias into the feature set input, the model can be driven to achieve defect-oriented iterative optimization, enhance dynamic adaptability, and ultimately form a positive cycle mechanism of "dynamic error feedback - internal parameter adjustment - prediction accuracy improvement", thereby effectively improving the overall accuracy of wind / solar power prediction.

[0155] Step 5.2, Day-ahead optimization scheduling of NEPS including wind and solar correlation:

[0156] To highlight the effectiveness of the proposed model in terms of economic efficiency and renewable energy consumption, this invention sets up two scenarios for comparative analysis:

[0157] Scenario 1: NEPS day-ahead optimization scheduling under wind and solar forecasting alone.

[0158] Scenario 2: Day-ahead optimization scheduling of NEPS considering joint forecasting of wind and solar correlation.

[0159] Table 8 compares the daytime total output fluctuations of key generating units under different scenarios.

[0160] To demonstrate more intuitively Figure 9 The output of each unit in the system under different scenarios is further analyzed. To reflect the volatility of its output curve, variance, standard deviation, and range are used to describe it. The results are shown in Table A2 of Appendix A. The optimization rate of volatility index of scenario 2 relative to scenario 1 is shown in Table 9. The specific target costs and output of important units are shown in Tables 10 and 11.

[0161] Table 9 shows the optimization rate of the volatility index.

[0162] Table 10 compares the day-ahead scheduling costs under different scenarios.

[0163] Table 11 compares the total daytime output of key generating units under different scenarios.

[0164] An analysis of the above charts reveals the following:

[0165] (1) As can be seen from Table 8, the optimization rates of the two scenarios are positive and negative, but overall the volatility index in scenario 2 is better than that in scenario 1. Furthermore, as shown in Table 9, the power volatility of BESS, PV, Grad, and MT in scenario 2 is better than that in scenario 1, especially the power interaction curve with the main grid and the volatility index of photovoltaic power are significantly better than the corresponding indicators in scenario 1. This indicates that the park in scenario 2 can have a more stable trend when exchanging power with the main grid, reducing the impact effect on the main grid when connected to the grid.

[0166] (2) As shown in Table 10, under the day-ahead scheduling scenario of wind and solar joint forecasting, the cost of each sub-objective and the overall scheduling cost are lower than the day-ahead scheduling cost of wind and solar forecasting alone. Among them, the system day-ahead operation and maintenance cost and system risk cost are reduced by 30.35% and 18.76% respectively, while the environmental protection cost is similar, resulting in a significant optimization of the total cost. The specific analysis is as follows:

[0167] 1) When wind and solar correlation is considered for joint forecasting, the forecast results are more accurate and closer to the actual power output than when wind and solar are forecasted independently. Therefore, the uncertainty risk domain is smaller when CVaR risk is quantified, and the reserve capacity set for this uncertainty risk is also reduced accordingly. Thus, the risk cost in Table 9 is less.

[0168] 2) When the uncertainty risk domain of renewable energy output is smaller, the theoretical absorption capacity of renewable energy by the system also increases accordingly. Furthermore, compared to the unit price of electricity purchased from the main grid, the unit price of renewable energy power is relatively lower. Therefore, the system will prioritize the use of renewable energy output that can be absorbed. With the total load remaining constant, the increase in renewable energy absorption leads to a corresponding decrease in electricity purchased from the main grid and a corresponding increase in electricity purchased from the main grid, thus reducing the overall system operation and maintenance costs. As shown in Table 11, the wind and solar absorption capacity in Scenario 2 is 39.96MW, which is 2.23MW more than in Scenario 1; the electricity sold to the main grid increases by 2.29MW.

[0169] Based on the content described in Section 1.3, day-ahead forecasts were performed for wind and solar power output. The forecast results were then used for day-ahead optimization scheduling in Scenario 1 and Scenario 2. To reduce the randomness of the optimization algorithm, each algorithm was run independently 10 times, and the average value was taken to obtain the day-ahead output of each operating unit as follows: Figure 9 As shown.

[0170] In summary, this invention proposes a multi-objective collaborative optimization scheduling model for NEPS that considers the correlation between wind and solar power output prediction. Taking into account the multi-factor coupling effect and error correlation in wind and solar power output prediction, a short-term power combination prediction framework for NEPS is constructed to achieve collaborative forward-looking prediction of wind and solar power generation capabilities. The prediction results are then used for day-ahead optimization scheduling of NEPS.

[0171] This invention relates to a multi-objective optimal scheduling method for new energy power systems that considers wind-solar correlation and prediction errors, belonging to the field of power system optimal scheduling technology. It includes: First, preprocessing the raw data based on adaptive noisy complete ensemble empirical mode decomposition (ERD); second, establishing a wind-solar output prediction model collaboratively constructed by a convolutional neural network and a bidirectional long short-term memory network, and adaptively optimizing the prediction model parameters based on the chaotic sparrow algorithm to extract prediction error features; third, fusing the error modes and original features to establish a wind-solar output coupling cross-prediction mechanism to reuse prediction errors; finally, importing the prediction results into a multi-objective collaborative optimal scheduling model for new energy power systems for day-ahead decision optimization. Compared with independent prediction models, the proposed model has higher prediction accuracy and a 28% year-on-year reduction in total system operating costs, effectively balancing the economic benefits of new energy power system operation with the capacity for new energy absorption.

[0172] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A multi-objective optimal scheduling method for new energy power systems considering wind-solar correlation and prediction errors, characterized in that: The method includes: Step 1: Preprocess the raw data based on the adaptive noise complete set empirical mode decomposition technique; Step 2: Establish a wind and solar power output prediction model that is jointly constructed by a convolutional neural network and a bidirectional long short-term memory network, and adaptively optimize the prediction model parameters based on the chaotic sparrow algorithm to extract prediction error features. Step 3: Integrate error modes and original feature quantities to establish a wind and solar power output coupling cross-prediction mechanism to realize the reuse of prediction errors; Step 4: Import the prediction results into the multi-objective collaborative optimization scheduling model of the new energy power system for day-ahead decision optimization.

2. The multi-objective optimal scheduling method for new energy power systems considering wind-solar correlation and prediction error as described in claim 1, characterized in that: Step 1 includes: Gaussian white noise with zero mean n g (t) Repeat the superposition g times to the original sequence x g Above, construct the set of signals to be decomposed, x. g (t), as shown in Equation (1); use internal EMD to perform intrinsic mode decomposition on each signal to be decomposed, integrate all decomposition results and obtain the mean value to obtain IMF; use the remaining residual signal as a new input sequence and repeat until the signal meets the monotonicity termination condition; (1); where ε is The weighting coefficients are denoted by g; g represents the sample value.

3. The multi-objective optimal scheduling method for new energy power systems considering wind-solar correlation and prediction error as described in claim 1, characterized in that: Step 2 includes: CSSA is used to automatically find and optimize key hyperparameters in CNN and BiLSTM models; The BILSTM network model plays a role in temporal prediction in the joint model. Its architecture consists of bidirectional LSTM coupling. By fusing forward and backward temporal features, it effectively solves the problem of missing inverse information in traditional LSTM. This model is sensitive to parameter configuration. Among them, the three general parameters of hidden layer dimension, upper limit of training period, and initial value of learning rate dominate the convergence characteristics of the model. To address the problems of prediction performance degradation and convergence lag, CSSA is used to carry out parameter-oriented optimization, which simultaneously improves the robustness and computational efficiency of the model.

4. The multi-objective optimal scheduling method for a new energy power system considering wind-solar correlation and prediction error as described in claim 1, characterized in that: Step 3 includes: By integrating multi-source correlation features into the NEPS short-term joint prediction framework for wind and solar power, the barrier between independent wind and solar power prediction is broken, and an error feedback mechanism is established. Specifically, it consists of four parts: wind and solar power correlation analysis, input feature set CEEMDAN decoupling, CSSA parameter optimization, and CNN-BILSTM model two-layer prediction.

5. A multi-objective optimization scheduling method for new energy power systems considering wind-solar correlation and prediction error, as described in claim 4, is characterized in that: Step 4 includes: Step 4.1, System Parameter Settings: By taking the predicted wind and solar power output values ​​that take into account the correlation between wind and solar power as input, a comprehensive multi-objective function that can adaptively balance is established to achieve day-ahead optimal scheduling of NEPS. Step 4.2, Establishing the objective function: Step 4.2.1, System operating costs: (2) (3) in, This represents the overall operating cost of NEPS at time t; Do not consider the total transaction cost and natural gas purchase cost at time t; The operating costs of the battery, P2G, and GT at time t; Let t be the cost of purchasing electricity from the main grid and the revenue from selling electricity. Let GT represent the natural gas consumption and the user's natural gas demand at time t, respectively. For P2G and GT, the operation and maintenance coefficients are used. Step 4.2.2, System environment governance costs: (4) (5) in, Let t be the environmental remediation cost of NEPS at time t; These represent the environmental governance costs of the main network, P2G, and GT at time t, respectively. Let and n represent the treatment cost and total number of types of pollutant x, respectively. Purchase electricity from the main network at time t; These are the emission coefficients of pollutant x in the main grid, P2G, and GT, respectively. Step 4.2.3, Minimum Load Loss and CVaR Cost: Uncertainty in new energy output leads to a contraction in system regulation margin, inducing the risk of load shedding and wind / solar curtailment; optimizing the allocation of reserve resources can effectively cope with unexpected disturbances such as sudden shutdowns, grid failures, and sudden changes in environmental parameters; the cost of reserve capacity is shown in equation (6): (6) Among them, a t Let λ be the wind power confidence level at time t; T be the system operating period; and λ be the penalty coefficient. Total cost of backup; When the output of new energy sources exceeds the boundary of its confidence interval, the system will face the risk of wind / solar curtailment or insufficient power supply. Based on this dual risk scenario, a risk-cost mapping mechanism is constructed using the CVaR model to achieve an economic quantification of operational risks. Step 4.3: Membership function processing; To overcome the surge in computational complexity caused by simultaneous optimization of multiple objective functions, a membership-weighted aggregation mechanism is adopted to realize the mathematical transformation from multiple objectives to a single objective, effectively reducing the search dimension of the Pareto solution set and accelerating the optimization process; Step 4.4 System constraints, including grid constraints, WT constraints, P2G constraints, and GT constraints; Step 4.5: Solve using the genetic particle swarm optimization algorithm.

6. A multi-objective optimal scheduling method for a new energy power system considering wind-solar correlation and prediction error, as described in claim 5, is characterized in that: Step 4.4 includes: Step 4.4.1, Power Grid Constraints: (7) in, This represents the electrical power value of the NEPS interacting with the main grid at time t. When the value is greater than 0, it indicates that electricity is purchased from the main grid. A value less than 0 indicates that electricity is sold to the main grid. These represent the upper and lower limits of the interaction between NEPS and the main grid power; These represent the wind and solar power absorption capacity at time t; These represent the power generation and power consumption of the P2G at time t, respectively. This represents the output value of GT at time t; Step 4.4.2, WT Constraint: (8) Step 4.4.3, P2G Constraints: (9) Step 4.4.4, GT Constraints: (10) in, This represents the predicted maximum output of WT at time t; , These represent the wind speed values ​​when WT enters / exits the main grid, respectively. Represents the rated wind speed; Δt represents the rated output power of WT; Δt represents the time step. Represents natural gas production. ; The unit representing the higher heating value of methane is 39 MJ / m³ under standard conditions. ; Represents the P2G electro-gas conversion efficiency; Represents the gas-to-electricity conversion coefficient; This represents the maximum electrical power of the P2G; The gas consumption of GT within the time period t; Represents the lower heating value of natural gas; Represents GT power generation efficiency; These represent the upper and lower limits of GT output, respectively. These represent the upper and lower limits of the GT gradeability, respectively.

7. A multi-objective optimal scheduling method for a new energy power system considering wind-solar correlation and prediction errors, characterized in that, The system includes a module for executing a multi-objective optimization scheduling method for a new energy power system that considers wind-solar correlation and prediction error as described in any one of claims 1 to 6.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a multi-objective optimization scheduling method for new energy power systems that considers wind and solar correlation and prediction errors, as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a multi-objective optimization scheduling method for new energy power systems that considers wind and solar correlation and prediction errors, as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements a multi-objective optimization scheduling method for new energy power systems that considers wind and solar correlation and prediction errors, as described in any one of claims 1 to 6.

Images