Comprehensive energy scheduling optimization method and system for green power direct connection park

Abstract

The invention relates to the technical field of energy scheduling, and provides a comprehensive energy scheduling optimization method and system for a green power direct connection park, which can accurately describe the synchronous fluctuation characteristics of source load uncertainty and reduce the amplification effect of prediction deviation by constructing a hybrid prediction model, performing combined modeling and predicting park load and upstream green power output, and can improve the scheduling efficiency of the green power direct connection park. And the response capability to the dynamic change of the source load is improved. Meanwhile, for the problem that a traditional optimization algorithm is prone to falling into local optimum in a high-dimensional nonlinear system, a wind-sensitive golden-spot butterfly optimization algorithm is introduced, higher global search and convergence performance is achieved, and a scheduling solution with better quality and smaller fluctuation can be obtained under multiple constraints and multiple targets; and in combination with an energy efficiency-economic equilibrium corridor decision-making mechanism, an optimal compromise strategy giving consideration to both operation stability and economy is screened out from the candidate schemes, so that the scheduling efficiency, economy and performability of the integrated energy system of the green power direct connection park are improved.

Description

A comprehensive energy dispatch optimization method and system for green electricity direct-connection parks Technical Field

[0001] This invention relates to the field of energy dispatching technology, specifically to a comprehensive energy dispatching optimization method and system for green electricity direct-connection parks. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In recent years, with the rapid development of renewable energy technologies such as distributed wind power and photovoltaics, industrial park energy systems have gradually evolved towards cleaner, more localized, and intelligent directions. Among these, the green electricity direct connection model, as an important form of new integrated energy systems, has received increasing attention. This model achieves local production and consumption of green electricity by physically or virtually coupling the park's load side with nearby renewable energy power plants (such as wind farms and photovoltaic arrays), bypassing the intermediate links of the traditional power grid. Compared with traditional energy supply methods, green electricity direct connection has multiple advantages: first, it reduces energy loss and dispatch lag during transmission and distribution, improving energy utilization efficiency; second, it enhances the local absorption capacity of renewable energy, reducing external dependence and carbon emission risks; and third, the system response is more flexible, helping to achieve dynamic adaptation of the load side to green energy fluctuations. This approach is widely considered one of the important paths to achieving zero-carbon parks and a high proportion of green electricity utilization.

[0004] Although the direct green power connection model structurally reduces the grid connection, the existing deployment and operation schemes of integrated energy systems in industrial parks still follow the design approach of traditional grid-connected energy systems, failing to match the operational characteristics of the direct connection architecture. The main core technical problems are as follows: First, existing deployment schemes fail to effectively address the direct coupling problem of uncertainties between the source and load sides under direct green power connection conditions. In the direct green power connection model, random fluctuations in wind and solar power output act directly on the park's load system without grid buffering, while the park's load itself exhibits significant dynamic changes. Existing schemes typically model and predict the generation and load sides separately, lacking a unified characterization and coordinated control mechanism for the "synchronous superposition and amplification effect of source-load uncertainties." This leads to amplified prediction deviations in actual operation, making it difficult to maintain system power balance and affecting energy supply reliability.

[0005] Second, existing optimization algorithms struggle to achieve globally optimal scheduling results that balance economy and stability in high-dimensional, nonlinear, and highly uncertain integrated energy systems. Faced with scheduling models involving multiple energy sources, load types, and operational constraints, traditional optimization methods are prone to getting trapped in local optima, and their convergence speed and solution stability are difficult to guarantee when wind and solar power output changes rapidly. Furthermore, existing methods often prioritize economic efficiency, neglecting the smoothness of the scheduling process and equipment operational constraints, leading to frequent and drastic power adjustments. This not only affects system stability but also weakens the long-term economic viability and feasibility of the overall scheduling scheme. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a comprehensive energy dispatch optimization method and system for green energy direct-connection parks. It integrates load forecasting, green energy output forecasting, and the Wind Sensitive King Butterfly Optimization (WGSBA) algorithm to jointly optimize the comprehensive energy system. This effectively solves the source-load matching problem caused by wind and solar fluctuations and load uncertainties, and avoids the problems of traditional optimization methods easily getting trapped in local optima and having poor dispatch economy and stability.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: One or more embodiments provide a comprehensive energy dispatch optimization method for green electricity direct-connection parks, including the following steps: acquiring historical operating data of the park and upstream power plants used for direct power supply; using the acquired historical operating data, predicting the park load and green electricity supply of upstream power plants in future periods through a constructed CNN-BiLSTM-Attention hybrid prediction model, obtaining predicted load and green electricity output curves; establishing mathematical models for key equipment and operating constraints of the comprehensive energy system, obtaining an operation optimization model of the comprehensive energy system; in the operation optimization model, establishing a multi-objective function with the optimization objectives of minimizing annual total operating cost, minimizing equipment power fluctuations, and maximizing the overall energy efficiency of the system; based on the obtained predicted load and green electricity output curves, using the wind-sensitive kingpin optimization algorithm to solve for a set of candidate dispatch schemes; using the energy efficiency-economic equilibrium corridor decision method to screen the dispatch schemes in the generated set of candidate dispatch schemes, determining the optimal compromise dispatch scheme.

[0008] One or more embodiments provide an integrated energy dispatch optimization system for green energy direct-connection industrial parks, comprising: a data acquisition module configured to acquire historical operating data of the industrial park and upstream power plants used for direct power supply; a prediction module configured to use the acquired historical operating data to predict the load of the industrial park and the green energy supply of upstream power plants in future periods through a constructed CNN-BiLSTM-Attention hybrid prediction model, obtaining predicted load and green energy output curves; a model building module configured to establish mathematical models for key equipment and operating constraints of the integrated energy system, obtaining an operating optimization model of the integrated energy system; an optimization module configured to establish a multi-objective function in the operating optimization model with the optimization objectives of minimizing annual total operating cost, minimizing equipment power fluctuation, and maximizing the overall energy efficiency of the system, and to solve for a set of candidate dispatch schemes using the wind-sensitive golden butterfly optimization algorithm based on the obtained predicted load and green energy output curves; and a decision module configured to use the energy efficiency-economic equilibrium corridor decision method to screen the generated set of candidate dispatch schemes and determine the optimal compromise dispatch scheme.

[0009] One or more embodiments provide an integrated energy dispatch optimization system for green electricity direct connection parks, including: an upstream wind and solar power generation system that achieves green electricity direct connection with the park through a set line; and an energy management system configured to execute the above-mentioned integrated energy dispatch optimization method for green electricity direct connection parks, output the optimal dispatch scheme, and dispatch the various equipment units of the integrated energy system.

[0010] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, it completes the steps in the above-described integrated energy dispatch optimization method for green electricity direct connection parks.

[0011] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention effectively solves the power balance problem caused by the direct coupling of uncertainties between the source and load sides in the green electricity direct connection mode. By constructing a CNN-BiLSTM-Attention hybrid prediction model, jointly modeling and predicting the park load and upstream green electricity output, it can accurately characterize the synchronous fluctuation characteristics of source-load uncertainty, reduce the amplification effect of prediction bias, improve the response capability to dynamic changes in source and load, and enhance the stability and robustness of the energy supply system. Simultaneously, addressing the problem that traditional optimization algorithms are prone to getting trapped in local optima in high-dimensional nonlinear systems, this method introduces the wind-sensitive monarch butterfly optimization algorithm, which has stronger global search and convergence performance and can obtain higher-quality, less volatile scheduling solutions under multiple constraints and objectives. Furthermore, combined with the energy efficiency-economic equilibrium corridor decision mechanism, it selects the optimal trade-off strategy that balances operational stability and economy from candidate solutions, fundamentally improving the scheduling efficiency, economy, and feasibility of the integrated energy system of the green electricity direct connection park.

[0012] The advantages of the present invention, as well as its additional advantages, will be described in detail in the following specific embodiments. Attached Figure Description

[0013] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0014] Figure 1 is a flowchart of the integrated energy dispatch optimization method of Embodiment 1 of the present invention; Figure 2 is a flowchart of the prediction based on CNN-BiLSTM-Attention hybrid prediction of Embodiment 1 of the present invention; Figure 3 is a flowchart of the energy efficiency-economic equilibrium corridor decision method of Embodiment 1 of the present invention; Figure 4 is an example system block diagram of the integrated energy dispatch optimization system of Embodiment 2 of the present invention. Detailed Implementation

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0016] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0017] It should be noted that the terminology used herein is for describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features within those embodiments can be combined with each other. The embodiments will now be described in detail with reference to the accompanying drawings.

[0018] Example 1: In one or more of the technical solutions disclosed in the embodiments, as shown in Figures 1 to 3, a comprehensive energy dispatch optimization method for green electricity direct-connection parks includes the following steps: S1, acquiring historical operating data of the park and upstream power plants used for direct power supply; S2, using the acquired historical operating data, predicting the park load and green electricity supply of upstream power plants for future periods through a constructed CNN-BiLSTM-Attention hybrid prediction model, obtaining predicted load and green electricity output curves; S3, establishing mathematical models for key equipment and operating constraints of the comprehensive energy system, obtaining an operation optimization model for the comprehensive energy system; S4, in the operation optimization model, minimizing the total annual operating cost, minimizing equipment power fluctuations, and maximizing the overall system energy efficiency are prioritized. To achieve the desired outcome, a multi-objective function is established. Based on the predicted load and green power output curves obtained in S2, the Wind-Sensitive Golden Butterfly Optimization Algorithm (WGSBA) is used to solve for a set of candidate scheduling schemes. In S5, the generated candidate scheduling scheme set is screened using the energy efficiency-economic equilibrium corridor decision method to determine the optimal compromise scheduling scheme. This method first collects historical operating data of the park's load and the upstream wind farms or photovoltaic power stations directly supplying power as basic data input. Then, a hybrid prediction model combining convolutional neural networks (CNN), bidirectional long short-term memory networks (BiLSTM), and attention mechanisms is used to extract time-series features and key influencing factors, jointly predicting load demand and green power output for several future periods to obtain dynamic prediction curves. Next, based on the composition structure of the integrated energy system, a mathematical model of physical constraints and operational boundaries is constructed for key components including energy storage systems, combined cooling, heating, and power (CCHP) equipment, and energy loads, forming a complete definition of the optimization problem. In this model, long-term operating costs, equipment regulation stability, and system energy efficiency are considered to construct a multi-objective optimization function. The wind-sensitive monarch butterfly optimization algorithm, which possesses global search capabilities and adaptability to wind and solar power output, is used to obtain a set of candidate scheduling solutions that meet the constraints. Finally, by introducing an "energy efficiency-economic equilibrium corridor" decision-making mechanism, the trade-off performance of each candidate scheme in terms of energy efficiency and economy is comprehensively evaluated, and the optimal scheduling strategy is selected to guide actual operation.

[0019] This method effectively solves the power balance problem caused by the direct coupling of uncertainties between the source and load sides in the green electricity direct connection mode. By constructing a CNN-BiLSTM-Attention hybrid prediction model, it jointly models and predicts the park load and upstream green electricity output, accurately characterizing the synchronous fluctuation characteristics of source-load uncertainty, reducing the amplification effect of prediction bias, improving the responsiveness to dynamic changes in source and load, and enhancing the stability and robustness of the energy supply system. Simultaneously, addressing the problem that traditional optimization algorithms are prone to getting trapped in local optima in high-dimensional nonlinear systems, this method introduces the wind-sensitive monarch butterfly optimization algorithm, which has stronger global search and convergence performance and can obtain higher-quality and less volatile scheduling solutions under multiple constraints and objectives. Combined with the energy efficiency-economic equilibrium corridor decision mechanism, it selects the optimal trade-off strategy that balances operational stability and economy from candidate solutions, fundamentally improving the scheduling efficiency, economy, and feasibility of the integrated energy system of the green electricity direct connection park.

[0020] Specifically, the energy efficiency-economic equilibrium corridor decision-making method involves setting a minimum acceptable comprehensive energy efficiency value to exclude schemes that fail to meet energy efficiency standards, selecting cost-acceptable schemes within the economic tolerance range, and determining the optimal compromise scheduling scheme. Step S5 outputs the optimal compromise scheduling scheme, including the real-time power allocation of each device and the values ​​of the three objective functions corresponding to the scheme: annual total operating cost, device power fluctuation index, and system comprehensive energy efficiency value. At the same time, it outputs the state of charge change curve of the energy storage system and the power interaction sequence with the grid.

[0021] In step S1, the park in this embodiment can be an industrial park. Green electricity refers to electricity provided directly by local power plants without going through the power grid system. Specifically, historical operating data includes historical operating data of the industrial park and upstream power plants. The upstream power plants are power plants that directly supply electricity to the industrial park and can be any type of power plant, such as photovoltaic power generation, wind power generation, gas power generation, etc. Optionally, the historical operating data of the industrial park can include historical hourly electricity load, heat load, and cold load data, as well as meteorological characteristic parameters, production intensity, and personnel activity coefficients corresponding to the time sequence. It also includes the operating parameters of key equipment in the park, covering the capacity and power of electrochemical energy storage, ice storage, and high-temperature hot water thermal storage systems. Optionally, the historical operating data of the upstream power plants can include historical power generation data and meteorological characteristic parameters corresponding to the time sequence, i.e., the green electricity supplied. Meteorological characteristic parameters include ambient temperature, wind speed, and solar radiation intensity. Furthermore, it can also include market and policy parameter data, such as electricity price data, including the electricity sales price of the upstream green electricity grid and the specific price range of the tiered carbon trading mechanism.

[0022] In step S2, the park load includes electrical load, thermal load, and cooling load. In some embodiments, the CNN-BiLSTM-Attention hybrid prediction model is constructed by sequentially connecting a convolutional neural network (CNN), a bidirectional long short-term memory network (BiLSTM), and an attention mechanism layer. The CNN is used to perform convolution operations on the input historical operating data (multidimensional time series data) to extract local features of the data in the time dimension. The BiLSTM is used to learn the local features output by the CNN based on time-series dependency learning to capture forward and backward long-term time-series dependencies, obtaining the hidden state vector. Specifically, the high-level feature sequence extracted by the CNN layer is input into the BiLSTM. The BiLSTM learns the long-term dependencies of the time series from both forward and backward directions, which can remember past trends and capture future contextual information, thereby gaining a more comprehensive understanding of the dynamic changes in load and power generation.

[0023] The attention layer introduces an attention mechanism to calculate features at different time points with different weights and perform weighted fusion on the hidden state vector output by the Bidirectional Long Short-Term Memory Network (BiLSTM).

[0024] The CNN-BiLSTM-Attention hybrid prediction model in this embodiment introduces an attention mechanism, assigning different weights to features at different times, enabling the model to automatically focus on the most critical historical moments for the prediction target, effectively improving the prediction accuracy of key fluctuations.

[0025] The model inputs include historical load data, historical power generation data, and meteorological data obtained in step S1, and the outputs are the predicted values ​​of park load and green electricity for a specific future period.

[0026] In step S2, the acquired historical operating data is used to predict the load of the park and the green electricity supply of the upstream power station in the future period through the constructed CNN-BiLSTM-Attention hybrid prediction model, and the method for obtaining the predicted load and green electricity output curve includes the following steps: S21, data preprocessing: the historical operating data of the park and the upstream power station obtained in step S1 is preprocessed to obtain the input sequence; the hourly meteorological and environmental data of the park and the upstream power station obtained in S1, including ambient temperature, wind speed and solar radiation amplitude, as well as the production intensity and personnel activity coefficient of the park, are preprocessed by cleaning, normalization and other operations.

[0027] CNN-BiLSTM-Attention Model Input and Output: Model Input: The preprocessed multivariate time series data from S21 is used as the model input. For a prediction time, the input is a matrix composed of historical data from the past L time steps. , each of which It is a feature vector that includes the temperature, irradiance, and wind speed at that moment; the historical load value, production intensity coefficient, and personnel activity coefficient of the park; and the historical power generation output value of the upstream power station, such as the wind and solar power generation value.

[0028] Model output: The output data is the predicted value for the next T time steps.

[0029] The load forecast output is: ;in, These are the predicted values ​​of the park's electrical load, thermal load, and cooling load at time t, respectively.

[0030] The predicted output for wind and solar power generation is: ;in, These are the predicted values ​​of photovoltaic and wind power supply from the upstream power station at time t, respectively.

[0031] S22. CNN Feature Extraction: The input sequence is fed into the convolutional neural network of the CNN-BiLSTM-Attention model for convolutional processing to extract local short-term dependencies and spatial features from the input sequence, obtaining local features in the time dimension. Specifically, the input sequence X is processed through a one-dimensional convolutional layer, with the following formula: Among them, W c and b c Here, k represents the weights and biases of the convolution kernel, and k is the kernel size. This represents the convolution operation, where σ is the ReLU activation function, and C... t Let be the feature map obtained at time t.

[0032] The convolution operation in this step generates a series of high-level feature maps through multiple convolution kernels, capturing local patterns such as "increased solar power output due to enhanced afternoon sunlight" or "concentrated start-up of electrical load during working hours".

[0033] S23. BiLSTM Temporal Dependency Learning: The local features extracted in step S22 are input into a bidirectional long short-term memory network to learn long-term temporal dependencies in both forward and backward directions, outputting a hidden state vector. Specifically, BiLSTM consists of a forward LSTM and a backward LSTM. For a time step t: ;in, These represent the forward and backward hidden state vectors at time t, respectively; These represent the computational processes of forward and backward long short-term memory networks, respectively.

[0034] Finally, the hidden state vector at this time step is formed by concatenating the forward and backward hidden states: This structure allows the model to both remember "the peak load at the same time in the past few days" and perceive "the future impact of the upcoming temperature drop on the heat load".

[0035] S24. Attention Mechanism Weighting: Based on the attention mechanism, the weights of features at different time points are calculated, and the hidden state vectors output by the Bidirectional Long Short-Term Memory Network (BiLSTM) are weighted and fused to obtain the context vector.

[0036] To emphasize the varying importance of different historical moments in predicting future moments, an attention mechanism is introduced to compute the hidden state at each historical moment. Attention weights The calculation formula is: ;in, This represents the hidden state of historical time i at the current decoding time t. The importance of the information contained, i.e., the attention score; This is the weight vector; and This is the weight matrix; Used to introduce nonlinearity; This represents the normalized attention weights assigned to the information at the current decoding time t, where the sum of the weights for all historical times i is 1.

[0037] The context vector is obtained after weighted summation: ;in, The context vector generated at the current decoding time t represents all historical hidden states. The weighted average, where the weights are the attention weights. .

[0038] This mechanism enables the model to adaptively focus on key information points. For example, when predicting the evening peak load, it automatically assigns higher weights to "the load level at noon on the same day" and "the load at the same time on the previous day", thereby reducing the weight of irrelevant information and effectively improving prediction accuracy.

[0039] S25, Output Layer: The context vector contextL from the last time step is linearly transformed through a fully connected layer to obtain the final prediction result. Where Wo and bo are the weights and biases of the output layer.

[0040] Through the above steps, the hybrid model achieves accurate quantification of uncertainties on both the source and load sides, and its high-precision prediction data provides crucial forward-looking information for subsequent optimized scheduling.

[0041] S3. Establish an integrated energy system operation model, specifically establishing mathematical models for the key equipment and operational constraints in the integrated energy system. Specifically, key equipment may include air source heat pumps, electric air conditioners, electrochemical energy storage systems, ice storage systems, and high-temperature hot water thermal storage systems. The equipment models include mathematical models for air source heat pumps, electric air conditioners, electrochemical energy storage systems, ice storage systems, and high-temperature hot water thermal storage systems. The electrochemical energy storage system model, ice storage system model, and high-temperature hot water thermal storage system model all include state equations, power constraints, and mutual exclusion constraints for charging and discharging energy.

[0042] Specifically, the operational constraints of integrated energy dispatch optimization include energy balance constraints, equipment capacity constraints, equipment operation constraints, energy storage system constraints, and grid interaction constraints. Energy balance constraints include power balance constraints, thermal balance constraints, and cooling balance constraints. Equipment operation constraints include equipment capacity constraints and equipment ramp-up constraints. Energy storage system constraints include energy storage capacity constraints, charging and discharging power constraints, and charging and discharging mutual exclusion constraints. Grid interaction constraints limit the scope of real-time power purchase with the upstream green power grid.

[0043] The mathematical models established for key equipment and operational constraints in the integrated energy system are explained as follows: (1) Air source heat pump model: Air source heat pumps can work in both heating and cooling modes, and their performance is characterized by the coefficient of performance (COP).

[0044] The thermal power in heating mode is expressed as: The cooling power in cooling mode is expressed as: ;in, and The output power of the air source heat pump during time period t is the heat power and cooling power, in kW. The electrical power consumed by the air source heat pump during time period t is expressed in kW. and denoted by and respectively, these represent the coefficients of performance of the air source heat pump in heating and cooling modes, and are functions of the ambient temperature Tamb(t).

[0045] (2) Electric air conditioner model: As a supplementary and backup energy supply device for the system, the electric air conditioner model is similar to that of the air source heat pump.

[0046] The thermal power in heating mode is expressed as: The cooling power in cooling mode is expressed as: ;in, and These represent the thermal power and cooling power output of the air conditioner during time period t, respectively, in kW; Let t be the electrical power consumed by the air conditioner during the time period, in kW; and These represent the coefficients of performance (COPs) of the air conditioner in heating and cooling modes, respectively.

[0047] (3) Electrochemical energy storage model: The state equation of the electrochemical energy storage system is expressed as: In electrochemical energy storage systems, power and mutual exclusion constraints are expressed as: ;in, The state of charge of the electrochemical energy storage system during time period t represents the proportion of current charge to total capacity. The self-discharge rate of the battery represents the proportion of energy lost due to self-discharge during each scheduling period. and These are the battery's charging efficiency and discharging power, respectively, and 0 < ≤1, 0< ≤1; These represent the charging and discharging power of the battery during time period t. In this embodiment, it is assumed that the battery is not allowed to charge and discharge simultaneously during the same time period. The unit is kW. The scheduling time interval is in hours; The rated capacity of the battery is expressed in kWh. and This is the charge / discharge status flag of the battery during time period t, which is a binary variable (0 or 1), where 1 corresponds to being in the charge / discharge state; These represent the maximum permissible charge and discharge power of the battery, in kW; and These are the minimum permissible state of charge (lower limit for safe operation) and the maximum permissible state of charge (rated capacity) of the battery, respectively.

[0048] (4) Ice storage system model; The state equation of the ice storage system is: The power and mutual exclusion constraints in an ice storage system are expressed as follows: ;in, The state of the ice storage system during time period t represents the current storage capacity as a percentage of the total capacity. The cooling loss rate of an ice storage system represents the proportion of cooling lost due to natural melting and heat exchange during each scheduling period. and These are the storage efficiency and release efficiency of the ice storage system, respectively, and 0 < ≤1, 0< ≤1; These represent the electrical power consumed by the chiller in the ice-making mode of the ice storage system during time period t, and the cooling power released by the ice storage system during ice melting, in kW. The scheduling time interval is in hours; This refers to the rated cold storage capacity of the ice storage system, expressed in kWh. and t represents the ice-making and cooling-releasing states of the ice storage system during time period t. It is a binary variable (0 or 1), where 1 indicates that the system is in the ice-making / cooling-releasing state. These are the maximum electrical power of the chiller in the ice storage system and the maximum cooling capacity of the ice storage system, respectively, in kW; and These represent the minimum and maximum allowable cold storage capacity states of the ice storage system, respectively.

[0049] (5) High-temperature hot water thermal storage system model; State equation of the high-temperature hot water thermal storage system: The power and mutual exclusion constraints of the high-temperature hot water storage system are as follows: ;in, The heat storage status of the high-temperature hot water thermal storage system during time period t represents the proportion of the current heat storage to the total capacity. The heat loss rate of a thermal storage system represents the proportion of heat lost due to heat exchange during each scheduling period. and These are the heat storage efficiency and heat release efficiency of the thermal storage system, respectively, and 0 < ≤1, 0< ≤1; These represent the electrical power consumed by the electric heater of the thermal storage system in thermal storage mode and the thermal power released by the thermal storage system, respectively, in kW. The scheduling time interval is in hours; This refers to the rated thermal storage capacity of the thermal storage system, expressed in kWh. and t represents the heat storage and heat release states of the thermal storage system during time period t, a binary variable (0 or 1), where 1 corresponds to the state of heat storage / heat release; These are the maximum electrical power and maximum heat release power of the electric heater in the thermal storage system, respectively, in kW; and These represent the minimum and maximum allowable heat storage states of the thermal storage system, respectively.

[0050] Furthermore, the constraints on the operation of the integrated energy system include: (1) energy balance constraints, including power balance constraints, thermal balance constraints and cooling balance constraints; the power balance constraints are: ;in, The power input during time period t via the green electricity direct connection line is in kW. and , respectively, represent the discharge power of the electrochemical energy storage system during time period t, in kW; Let t be the electrical load of the park during time period t, in kW; The total electrical power consumed by the air source heat pump during time period t, in kW; Let t be the total electrical power consumed by the air conditioner during time period t, in kW; These represent the electrical power consumed by the chiller in the ice storage system during ice-making mode and the electrical power consumed by the electric heater in the high-temperature hot water storage system during heat storage mode, respectively, in kW.

[0051] The thermal equilibrium constraint is: ;in, and Let t be the thermal power output of the air source heat pump and electric air conditioner during time period t, in kW; and The heat charging power and heat dissipation power of the high-temperature hot water storage system during time period t are given in kW. Let t be the heat load of the park during time period t, in kW.

[0052] The cold force balance constraint is: ;in, and These represent the cooling power output (kW) of the air source heat pump and the electric air conditioner during time period t, respectively. and These are the cooling storage power and cooling release power of the ice storage system during time period t, respectively, in kW; Let t be the cooling load of the park during time period t, in kW.

[0053] (2) Equipment operation constraints, including upper and lower limits of equipment output, equipment ramp-up constraints, and grid interaction constraints; used to characterize the maximum rate of increase or decrease of the equipment's output power between two adjacent scheduling periods, reflecting the inertia of the equipment's regulation capability; the upper and lower limits of equipment output are: ;in, This represents the actual operating power of each device during time period t, expressed in kW. and These are the minimum output (lower limit for on / off) and rated output of each device, respectively, in kW; optionally, the applicable devices for the equipment operation constraints may include air source heat pumps, electric air conditioners, etc.

[0054] Equipment ramping constraints: ;in, This represents the maximum allowable power increase for each device per unit time, expressed as kW / Δt. This represents the maximum allowable power reduction of each device per unit time, expressed in kW / Δt.

[0055] Optionally, the applicable equipment for the ramp constraint may include adjustable power equipment such as air source heat pumps and electric air conditioners.

[0056] Power grid interaction constraints: ;in, This indicates the maximum power that can be purchased from upstream power plants during the current time period, in kW; This indicates the current power supply gap (in kW) in the park's system.

[0057] In this embodiment, for the direct connection mode, the focus is on strengthening the power balance constraints and grid interaction constraints. A specific grid interaction constraint is applied to one of the decision variables: green electricity purchase. The green electricity purchase range is obtained based on the upstream green electricity supply predicted in the current time period and the electricity storage obtained from the previous time period through rolling optimization. The purchase volume for the current hour should be within this range to meet the optimized scheduling requirements and achieve the comprehensive goals of economy, stable power supply, and maximum energy efficiency. Step S4 involves setting the objective function and solving the multi-objective optimization problem based on the Wind-Sensitive Golden Butterfly Optimization Algorithm (WGSBA). In the operational optimization model, the optimization objectives are minimizing the annual total operating cost, minimizing equipment power fluctuations, and maximizing the overall system energy efficiency. The multi-objective function is established as follows: Annual total operating cost (F1) is expressed as follows: Among them, grid interaction costs It is expressed as follows: ;in, For the power purchase capacity, The electricity purchase price.

[0058] Equipment maintenance costs It is expressed as follows: Where D represents the set of devices, Let d be the unit power operation and maintenance cost coefficient of equipment d. Let d be the operating power of device d during time period t.

[0059] Equipment depreciation cost It is expressed as follows: S represents the collection of key equipment (air source heat pumps, energy storage equipment, etc.). Let be the power fluctuation depreciation cost coefficient for equipment s.

[0060] Carbon trading costs This represents the net cost (with negative returns) incurred by the system due to carbon emissions from purchasing allowances or selling surplus allowances in the carbon trading market. This model introduces a tiered carbon trading mechanism, calculated as follows: In the formula: The actual total carbon emissions of the system during the scheduling period, expressed in tons (t). The total amount of free carbon allowances (tons) allocated to the park by the government, in tons; For the step carbon valence function, yuan / t where This represents the difference between the quota and the actual emissions. The function is defined as follows: In the formula, The base price for carbon trading is yuan / ton; Price growth rate; The tiered carbon trading range is t.

[0061] Equipment power fluctuation (F2) is expressed as: S represents a collection of key equipment, including air source heat pumps, energy storage devices, etc. Let s be the output power of device s during time period t, in kW; Let s be the rated power of the device, in kW.

[0062] System overall energy efficiency (F3) represents the ratio of the total effective energy output by the integrated energy system during the dispatch cycle to the total primary energy consumed, i.e., the direct green electricity input. The formula is: The numerator of the system's comprehensive energy efficiency formula is the total effective energy output by the system during the scheduling cycle (meeting electrical, thermal, and cooling loads); the denominator is the total primary energy consumed by the system during the scheduling cycle, i.e., the green electricity direct-connection input power.

[0063] This embodiment constructs a multi-objective function comprising three objective functions: The first objective function is the total annual cost of the integrated energy system (F1), addressing the cost-effectiveness of direct connection to green electricity. The total annual cost includes the investment cost of each piece of equipment, the cost of purchasing green electricity directly connected to the system, operation and maintenance costs, and the system's energy storage, cold storage, and heat storage costs. The second objective function is the equipment power fluctuation of the integrated energy system (F2), specifically designed to address the impact losses caused by wind and solar power fluctuations through direct connection channels on critical equipment. It extends equipment lifespan by measuring and optimizing the operational stability of critical equipment. This objective function primarily quantifies the rate of change of the charging and discharging power of the electrochemical energy storage system in adjacent time periods, as well as the fluctuation amplitude of the output power of major conversion equipment such as air source heat pumps. Its value is the ratio of the power change of all relevant equipment to the rated power within the scheduling cycle. The third objective function is the comprehensive energy efficiency of the integrated energy system (F3), ensuring high-quality energy utilization while reducing emissions at the source. This function calculates the ratio of the system's total output energy to its total input energy, i.e., the ratio of the system's total output load to its total power consumption. Optimizing this objective aims to force the model to prioritize efficient conversion technologies, thereby reducing energy demand at its source and achieving a higher level of energy conservation.

[0064] In this embodiment, the Wind-Sensitive Monarch Butterfly Optimization Algorithm (WGSBA) is employed to solve this multi-objective optimization problem. This algorithm simulates the intelligent behavior of monarch butterflies sensing wind speed and dynamically adjusting their flight patterns in variable wind fields to address uncertainties on both the source and load sides of the system. To optimize effectively, the decision variables of the Wind-Sensitive Monarch Butterfly Optimization Algorithm are set as key control parameters of each energy flow in the integrated energy system during each scheduling period. Specifically, these include: the power purchased from the grid, the charging and discharging power of the energy storage / thermal storage / cold storage modules, and the electrical power allocated to the air source heat pump and electric air conditioner for heating / cooling. Through the coordinated optimization of these variables, the algorithm can automatically find a scheduling scheme that achieves the optimal balance of the three objective functions while satisfying all operational constraints. This achieves a comprehensive improvement in economy, energy supply stability, and high energy efficiency under a zero-carbon background.

[0065] The existing standard Hybrid Butterfly Optimization Algorithm (HBFA) suffers from a fixed strategy switching probability, making it unsuitable for adapting to the dual uncertainties of power system operation (e.g., wind power, solar power, and load fluctuations). In green power direct connection scenarios, the system operation risk is high and prediction accuracy is limited. The original algorithm cannot flexibly cope with the dynamic and complex "wind field" environment, resulting in poor robustness and unstable performance. This embodiment proposes a Wind-Sensitive Monarch Butterfly Optimization Algorithm (WGSBA), inspired by the flight behavior of monarch butterflies in natural wind. It introduces a prediction uncertainty measurement mechanism to construct a "wind condition" evaluation index; it changes the fixed switching probability of the butterfly optimization algorithm to a switching probability that dynamically adjusts with prediction uncertainty, forming a "wind-sensitive" adaptive mechanism; and it links the optimization process with system risk by changing the strategy weights of global exploration and local development. The Wind-Sensitive Monarch Butterfly Optimization Algorithm (WGSBA) proposed in this embodiment achieves dynamic linkage between optimization strategy and system operation risk by introducing an adaptive mechanism driven by prediction uncertainty, enabling the search behavior to be intelligently adjusted according to the actual "wind condition". WGSBA outperforms in convergence speed, solution quality, and robustness, and is particularly suitable for scheduling optimization tasks in high-volatility green power direct connection scenarios. While improving system stability and operational economy, it also better meets the current actual needs of new energy consumption and intelligent scheduling, and has good engineering application prospects.

[0066] Specifically, in step S4, based on the predicted load and green power output curves obtained in S2, the Wind-Sensitive Monarch Butterfly Optimization Algorithm (WGSBA) is used to solve the multi-objective function to obtain a set of candidate scheduling schemes. During the solution process, the source load uncertainty data predicted by the CNN-BiLSTM-Attention model is used to generate dynamic switching probabilities, dynamically adjust the allocation of global exploration and local development strategies to determine the direction of population individual mutation updates, iteratively update the population, and finally generate a set of candidate scheduling schemes.

[0067] In the above implementation, a prediction uncertainty quantification factor is introduced to dynamically adjust the allocation of strategies for global exploration and local development. When prediction uncertainty is high (similar to "strong wind disturbance"), exploration is enhanced to broadly search for robust solutions; when prediction certainty is high (similar to "gentle wind and stable conditions"), development is strengthened to finely optimize. This "wind-sensitive" adaptive mechanism achieves intelligent coordination between algorithmic behavior and the operational risks of the physical system. By dynamically switching probabilities to guide the direction of individual mutations and step size updates, it drives the iterative evolution of the population, ultimately forming a set of scheduling schemes that satisfy operational constraints and have excellent performance.

[0068] Based on the predicted load and green power output curves obtained in S2, the process of solving for the candidate scheduling scheme set using the Wind-Sensitive Golden Butterfly Optimization Algorithm (WGSBA) includes the following steps: S41, Initialize algorithm parameters by randomly generating them within the solution space consisting of all decision variables (power of each device in each scheduling period). Initialize the population with a set of individuals; specifically, set the parameters required for running the Wind-Sensitive Monarch Butterfly Optimization Algorithm (WGSBA), including: Population size, i.e., the number of individual butterflies.

[0069] MaxIter: Maximum number of iterations.

[0070] λ: Inertial displacement constant, derived from SOA.

[0071] Mortality rate: Used to eliminate the worst-performing individuals.

[0072] Within the solution space comprised of all decision variables (the power of each device in each scheduling period), random generation is performed. One initial individual. Each individual It represents a complete system scheduling scheme, including all decision variables.

[0073] Furthermore, the feasibility of the obtained initial individuals is verified to check whether they meet the operational constraints in step S3, including: power balance constraints, equipment capacity constraints, ramp-up constraints, and energy storage SOC constraints.

[0074] For feasible individuals, calculate the values ​​of three objective functions: total annual operating cost, equipment power fluctuation, and overall system energy efficiency.

[0075] S42. Individual Evaluation and Selection of High-Quality Solutions: Each individual in the population is decoded into a specific equipment power scheduling sequence. After verifying feasibility, a multi-objective function value is calculated. Individuals that meet the requirements are selected based on the objective function values ​​for subsequent iterations. Each individual is decoded into a specific equipment power scheduling sequence. After verifying feasibility, three objective function values ​​(F1, F2, F3) are calculated. Subsequently, a pre-screening logic for the energy efficiency-economic equilibrium corridor is introduced: a temporary, relatively lenient energy efficiency threshold is set. and economic tolerance coefficient , will simultaneously satisfy and Individuals that meet the criteria are marked as "high-quality individuals" and given priority in subsequent population updates.

[0076] The purpose of this step is not to make a final decision, but to guide the algorithm's search direction and avoid wasting computational resources in low-quality solution domains.

[0077] S43, WGSBA core iterative optimization: The population is iteratively updated. In each iteration, dynamic switching probabilities are generated based on the predicted uncertainty to guide individuals to choose global exploration or local development for position updates, and mutation is completed by combining direction vectors and adaptive step size. Subsequently, the individuals are constrained and repaired, inferior solutions are eliminated and new solutions are added, and finally merged to form a new generation of population.

[0078] Specifically, in each iteration, the following operations are performed on each individual in the population: S431, calculate the direction vector of the individual toward the global optimum for individual movement control. ; Calculate individuals Towards the global optimum in the current population (i.e., a representative solution in the first level of the non-dominated sorting) ) direction vector : S432. Based on the individual's position, calculate the adaptive step size for the individual to move towards the target direction. ; Calculate the step length based on the individual's position This step size is related to the distance between the individual and the optimal solution, in order to achieve a fine-grained search in the near term and a coarse-grained search in the far term; ;in, This represents the pollination probability. Let be the inertial displacement constant. , These represent the positions of the current and previous generation individuals, respectively.

[0079] S433. Calculate the dynamic switching probability based on the source-load uncertainty data predicted by the CNN-BiLSTM-Attention model. Based on dynamic switching probability Direction vector and adaptive step size Update the individual data; further, based on the dynamic switching probability. The process of updating an individual includes the following: 1) Generating a random number r∈[0,1]; 2) If The butterfly optimization algorithm is used for global exploration, controlling individuals to move towards the current global optimum, and combining this with the direction vector. With adaptive step size Explore and update individuals; the individual update formula is: ;3) If It executes a self-organizing algorithm for local development, controlling individuals to perform random searches within a local area, and combining the direction and step size direction vectors. With adaptive step size To develop and update individuals, the update formula for an individual is: ;in, This refers to the fragrance concentration defined in BOA, used to guide individuals towards better solutions. I is the individual's fitness value. and To adjust the parameters; and is a randomly selected individual from the same population; e is a random number in the interval [0,1], used to dynamically balance the contributions of the two search strategies. For the individual before the update; For the updated individual; The global optimal solution; BOA refers to the Butterfly Optimization Algorithm; SOA stands for Self-organizing Algorithm, which is a heuristic optimization algorithm framework.

[0080] S434. Constraint Processing and Feasibility Repair: Constraint verification is performed on the newly generated individual after the update. If it violates the operational constraints of the integrated energy system, a repair operator is used to adjust the new individual, resulting in a constrained-repaired individual. Specifically, for the newly generated individual after the update... Constraint verification is performed. If it violates any of the operational constraints of the integrated energy system defined in S3, a repair operator is used to adjust it to a solution that satisfies the feasible region. S435, Population update and elite retention: By calculating the Euclidean distance between each individual and the current global optimal solution, a set number of individuals are eliminated to remove inferior solutions, and an equal number of new individuals are randomly generated in the solution space of the feasible region to maintain population diversity. Then, the parent individuals, the position-updated and constraint-repaired offspring individuals, and the newly generated random individuals are merged to form the updated new generation of population.

[0081] Eliminate the worst individual: Calculate the sum of all individuals and the current global optimal solution. The Euclidean distance is calculated as follows: Remove the m% of individuals with the furthest distance to eliminate the worst-performing solution; the Euclidean distance is calculated as follows: .

[0082] Adding new individuals: An equal number of new individuals are randomly generated within the feasible region of the solution space to maintain population diversity and exploration capability. The formula for randomly generating new individuals is: Where LB and UB are the lower and upper bound vectors of the decision variables, respectively, defined by the specific optimization problem; rand is a random number uniformly distributed in the interval [0,1].

[0083] The parent population, along with the offspring of mutated individuals after position updates, the population of repaired individuals generated after constraint repair, and newly generated random individuals, are merged to form the updated new population. The standard Hybrid Butterfly Optimization (HBFA) algorithm uses a fixed switching probability p to control the balance between global exploration and local exploitation, which is like a butterfly flying in a constant wind speed—a rigid strategy. To adapt to the highly fluctuating environment of green power direct-connection systems, WGSBA proposes a wind-sensitive adaptive switching probability mechanism, improving the fixed probability p into a dynamic switching probability that changes dynamically with the level of system uncertainty. This mechanism simulates the behavior of monarch butterflies in sensing wind strength and adjusting their flight attitude. Its dynamic probability is calculated in real time based on the uncertainty quantification results output by the previous prediction model. Its core is: when the prediction uncertainty is high, it increases the global exploration tendency to cope with unknown risks; when the prediction certainty is high, it enhances the local development capability to improve economic benefits.

[0084] In step S433, the method for calculating the dynamic switching probability based on the source-load uncertainty data predicted by the CNN-BiLSTM-Attention model includes the following steps: S4331. For the current scheduling period, predict the load and output based on the CNN-BiLSTM-Attention model, calculate the uncertainty, and normalize the uncertainty based on the uncertainty and the installed capacity and maximum historical load of the corresponding equipment. First, construct a metric model that can comprehensively quantify the uncertainty of photovoltaic, wind power output, and park load prediction. Specifically, the larger the value, the more complex and difficult to predict the "wind conditions" (i.e., source-load fluctuation) of the operating environment in time period t. From the CNN-BiLSTM-Attention model, the prediction variables for the future scheduling period t can be obtained, and then the uncertainty estimate is calculated, i.e., the uncertainty includes: Uncertainty component of renewable energy output forecast for time period t; : Uncertainty component of the park load forecast for time period t.

[0085] The specific calculation method for the uncertainty component is as follows: In the formula: Let be the predicted renewable energy output at time t, in kW. This can be any form of new energy source, such as wind power (with an uncertainty component of ) or photovoltaic power (with an uncertainty component of ). wait; The average renewable energy output, in kW, is the historical average for the same period (e.g., the same time in the past 30 days). and These represent the maximum and minimum renewable energy output for the same historical period, in kW; Let be the predicted load value at time t, in kW; The average load forecast for the same period in history (e.g., the same time in the past 30 days), in kW; and These represent the maximum and minimum predicted load values ​​for the same historical period, in kW. During the calculation of the uncertainty component, if the numerator and denominator of any term in the calculation formula are zero, it indicates that there were no fluctuations in the historical period, and the calculation is no longer performed using the formula; the dynamic switching probability is then used. If only the denominator is zero, then replace the denominator with a certain value, such as 1% of the installed capacity of renewable energy.

[0086] This formula is a general formula expression. The specific calculation requires calculating the uncertainty components of the upstream photovoltaic and wind power forecast output and the uncertainty components of the downstream cooling, heating and power load forecast according to the formula respectively.

[0087] S4332. Weight the normalized uncertainties to obtain the comprehensive forecast uncertainty measure. This embodiment defines an uncertainty measure. The weighted sum of the three normalized components mentioned above:

[0088] Among them, the weighting coefficient satisfy The weight values ​​can be determined based on the degree of impact of historical prediction errors on the energy balance of each part. It is a dimensionless quantity. The larger its value, the more uncertain the prediction of the operating environment in time period t is. , These represent the uncertainty components corresponding to photovoltaic and wind power, respectively. The range is between 0 and 1: when Current forecasts are very close to historical averages, with a high degree of certainty; when The prediction deviation is approximately half of the normal fluctuation range, which falls under the category of moderate uncertainty; when The prediction deviation exceeds the historical normal fluctuation range, indicating a high degree of uncertainty. S4333: A dynamic switching probability generation function is constructed by monotonically increasing the dynamic switching probability following the uncertainty measure, and the dynamic switching probability is calculated based on the uncertainty measure. This is based on a comprehensive prediction uncertainty measure. Dynamic switching probability It is generated by the following monotonically increasing function, i.e., the dynamic switching probability generation function: In the formula: Let be the dynamic switching probability used by the algorithm in scheduling period t. ,in Here, represents the lower and upper bounds of the switching probability, and represents a preset constant, which typically satisfies . k is a sensitivity coefficient that controls the steepness of the switching; k > 0, and a value of 5 is recommended. It is used for adjustment. right Sensitivity to change; , represents the midpoint of the switching; e is the natural constant.

[0089] when The prediction is very reliable. Focusing on localized development; when Prediction reliability is moderate. Equilibrium strategy; when Low prediction reliability It focuses on global exploration and aims to search for robust scheduling schemes that can cope with various potential risks in a broad solution space.

[0090] The dynamic switching probability generation function is the mathematical core of the "wind-sensitive" behavior. It enables the algorithm to intelligently switch between robust cruising (global exploration) and efficient honey gathering (local development) based on the strength of the "wind conditions" (uncertainty), just like a monarch butterfly.

[0091] The specific process of integrating dynamic switching probabilities in the WGSBA algorithm flow is as follows: In each iteration of the algorithm, for the current optimized future scheduling period t: calculate dynamic parameters: first calculate the comprehensive prediction uncertainty of the current period. Then calculate the dynamic switching probability. .

[0092] Strategy selection: For each individual in the population, generate a random number r uniformly distributed in the interval [0,1]. The decision logic is updated as follows: If... If so, a global exploration strategy will be executed for that individual.

[0093] like If so, a local development strategy will be implemented for that individual.

[0094] This mechanism ensures that the behavior of the optimization algorithm is wind-sensitive and adaptive, rather than fixed. It makes dynamic and adaptive adjustments based on a forward-looking assessment of future operational risks, fundamentally improving the adaptability, robustness, and biomimetic intelligence of the optimized scheduling scheme in the face of dual uncertainties of source and load.

[0095] S44. Optimization Result Output and Solution Set Generation: Determine if the iteration termination condition is met. If not, proceed to the next iteration until the termination condition is met, and output the candidate scheduling solution set. Further, perform algorithm termination judgment. The algorithm terminates iteration when any of the following termination conditions are met: 1) Reaching the set maximum number of iterations MaxIter; 2) The improvement magnitude of the optimal solution set for K consecutive generations is less than a set threshold ε. Specifically, the improvement magnitude of the solution set performance can be used as the change magnitude of the multi-objective function. For each generation of optimal solution set, calculate the rate of change of its mean or minimum value on the three objective functions as the improvement magnitude of the optimal solution set; 3) The algorithm running time reaches a preset upper limit. After the algorithm terminates, output all high-quality solutions obtained during the optimization process, forming a candidate scheduling solution set. This set has three characteristics: diversity, quality, and feasibility. Further, perform performance index calculation and solution feasibility verification on the obtained candidate scheduling solution set to obtain the final candidate scheduling solution set. Performance index calculation: Specifically, calculate the precise values ​​of the three objective functions for each solution; Solution feasibility verification: Perform final technical feasibility confirmation for each solution. This involves verifying whether the equipment power scheduling sequence strictly meets all the operational constraints in the integrated energy system operation optimization model established in step S3, including energy balance constraints, equipment capacity and ramping constraints, energy storage system constraints, and grid interaction constraints.

[0096] Step S5 aims to select a single scheduling instruction that achieves the best engineering balance among economy (F1), stability (F2), and energy efficiency (F3) from the large number of candidate schemes generated by the WGSBA algorithm. Specifically, a minimum acceptable value for comprehensive energy efficiency is set, and all scheduling schemes with comprehensive energy efficiency lower than this value are excluded. The minimum annual total cost among the candidate schemes is calculated, and the economic tolerance upper limit is set as the minimum cost plus a certain percentage. Among the schemes that simultaneously meet the energy efficiency and economic requirements, the scheme with the smallest equipment power fluctuation is selected as the optimal compromise scheduling scheme.

[0097] In step S5, the generated candidate scheduling scheme set is screened using the energy efficiency-economic equilibrium corridor decision method to determine the optimal compromise scheduling scheme. This includes the following steps: S51. For the scheduling schemes in the candidate scheduling scheme set obtained by the WGSBA algorithm, calculate the multi-objective function values, including annual total operating cost (F1), equipment power fluctuation (F2), and overall system energy efficiency (F3). The non-dominated solution set, which includes multiple operating strategies, output by the WGSBA algorithm, is then used as the candidate scheduling scheme set. Each scheme in this set represents a feasible operating strategy that achieves different trade-offs among the three objectives.

[0098] S52. Based on the operational requirements of the park or integrated energy system, set a minimum acceptable comprehensive energy efficiency value. Dispatch schemes with energy efficiency below the minimum acceptable comprehensive energy efficiency (CAE) value are eliminated, resulting in qualified energy efficiency schemes. Specifically, a minimum acceptable CAE value is set based on the park's zero-carbon operation requirements. All comprehensive energy efficiency The proposed alternatives were initially ruled out to ensure that the selected alternatives meet the basic requirements for efficient energy use.

[0099] S53. In an energy-efficient scheduling scheme, calculate the upper limit of cost based on the minimum annual total operating cost (F1). Eliminate those with operating costs exceeding the cost ceiling The scheduling scheme is optimized to obtain a high-quality compromise solution, which serves as the energy efficiency-economic equilibrium scheduling scheme. Among the schemes that meet the energy efficiency threshold, the lowest annual total operating cost (F1) is identified. Set a tolerable cost increase percentage α, and calculate the cost ceiling using the following formula: Where α is the percentage increase in costs acceptable to the decision-maker.

[0100] In this embodiment, candidate solutions that simultaneously meet the following conditions are considered as high-quality compromise solutions and constitute a "corridor set". Solutions in this "corridor set" are considered to be high-quality solutions that achieve a reasonable balance between energy efficiency and economy.

[0101] Will simultaneously satisfy and The candidate solutions are considered as high-quality compromise solutions and defined within the "energy efficiency-economic equilibrium corridor," forming a "corridor set." Solutions within this "corridor set" are considered high-quality solutions that achieve a reasonable balance between energy efficiency and economy. S54: For the energy efficiency-economic equilibrium scheduling schemes, calculate the equipment power fluctuation value for each scheduling scheme. The scheduling scheme with the smallest power fluctuation (i.e., the most stable operation) is selected as the final optimal compromise scheduling scheme; The solution with the smallest value, that is, the most stable and equipment-friendly solution, is taken as the final "optimal compromise scheduling solution". .

[0102] Where argmin represents the parameter that minimizes the subsequent function value; j∈Corridor indicates that scheme j belongs to the set of candidate schemes within the "energy efficiency-economic equilibrium corridor".

[0103] The output optimal scheduling scheme includes: the hourly real-time power allocation of each device, the state of charge change curve of the energy storage system, the system operating cost, the power fluctuation of the device, and the three key indicators of comprehensive energy efficiency, along with the power purchase sequence of the power grid. Through the above implementation method, the green electricity direct connection integrated energy system can realize direct optimized scheduling from the source of green energy to the end load, forming a complete technical closed loop of green electricity direct connection - prediction and early warning - multi-objective optimization - intelligent decision-making.

[0104] This embodiment addresses the application scenario of direct green power connection between the park and adjacent power stations, and proposes a full-link optimized scheduling method that effectively overcomes key bottlenecks such as multi-level transmission and distribution losses, scheduling delays, and insufficient prediction accuracy in traditional grid-connected power supply modes. By constructing a dedicated direct connection channel between the park and the power station, point-to-point green electricity transmission was achieved, improving energy utilization efficiency. Addressing the characteristics of strong fluctuations and high prediction difficulty on both the source and load sides of the direct connection channel, a CNN-BiLSTM-Attention hybrid model was innovatively introduced to improve the accuracy of multi-source prediction. In terms of optimization objective construction, the traditional model focusing solely on economic efficiency was broken through, and for the first time, equipment power fluctuations were incorporated into the objective function system, enhancing the stability of system operation. Regarding the optimization algorithm design, based on the wind-sensitive monarch butterfly optimization algorithm, a prediction uncertainty quantification factor was introduced, improving the original fixed switching probability mechanism into dynamic probability control. This enabled the algorithm to adaptively search in high-dimensional uncertain environments, significantly improving the robustness and solution efficiency. Simultaneously, an energy efficiency-economic equilibrium corridor decision mechanism was proposed, constructing a multi-objective trade-off framework based on energy efficiency thresholds and economic tolerance ranges. This achieves optimal coordination between ensuring full green electricity consumption and balancing operating costs, equipment lifespan, and system energy efficiency, demonstrating good engineering adaptability and promotional value, and providing systematic technical support for the intelligent and low-carbon operation of zero-carbon parks.

[0105] Example 2 is based on Example 1. This example provides a comprehensive energy dispatch optimization system for green electricity direct connection parks, as shown in Figure 4. It includes: an upstream power station power generation system, which achieves green electricity direct connection with the park through a set line; and an energy management system, which is configured to execute the comprehensive energy dispatch optimization method for green electricity direct connection parks described in Example 1, output the optimal dispatch scheme, and dispatch each equipment unit of the comprehensive energy system.

[0106] Specifically, the park's integrated energy system includes: energy production and conversion units, including air source heat pump units and electric air conditioning systems; multi-element energy storage units, including electrochemical energy storage systems, ice storage systems, and high-temperature hot water thermal storage systems; and a real-time monitoring interface that displays the operating status of each device, optimization results, and the values ​​of the three objective functions.

[0107] Furthermore, the electrochemical energy storage system, ice storage system, and high-temperature hot water thermal storage system are all equipped with: a status monitoring unit to collect the state of charge data of the energy storage device in real time; a power control unit to control the charging and discharging power according to the scheduling instructions and ensure that charging and discharging are mutually exclusive; and a loss compensation unit to dynamically correct the energy storage status based on the self-discharge rate.

[0108] Example 3, based on Example 1, provides a comprehensive energy dispatch optimization system for green electricity direct-connection industrial parks, comprising: a data acquisition module configured to acquire historical operating data of the industrial park and upstream power plants used for direct power supply; a prediction module configured to use the acquired historical operating data to predict the load of the industrial park and the green electricity supply of upstream power plants in future periods through a constructed CNN-BiLSTM-Attention hybrid prediction model, obtaining predicted load and green electricity output curves; a model building module configured to establish mathematical models for key equipment and operating constraints of the comprehensive energy system, obtaining an operation optimization model of the comprehensive energy system; an optimization module configured to establish a multi-objective function in the operation optimization model with the optimization objectives of minimizing annual total operating cost, minimizing equipment power fluctuations, and maximizing the overall energy efficiency of the system, and to solve for a set of candidate dispatch schemes using the wind-sensitive kingpin optimization algorithm based on the obtained predicted load and green electricity output curves; and a decision module configured to use the energy efficiency-economic equilibrium corridor decision method to screen the generated set of candidate dispatch schemes and determine the optimal compromise dispatch scheme.

[0109] It should be noted that each module in this embodiment corresponds one-to-one with each step in embodiment 1, and their specific implementation process is the same, so it will not be repeated here.

[0110] Example 4 is based on Example 1. This example provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps in the integrated energy dispatch optimization method for green electricity direct connection parks described in Example 1.

[0111] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0112] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A comprehensive energy dispatch optimization method for green electricity direct-connection industrial parks, characterized in that, The process includes the following steps: acquiring historical operating data of the industrial park and upstream power plants used for direct power supply; using the acquired historical operating data, predicting the future load of the industrial park and the green electricity supply of upstream power plants through a constructed CNN-BiLSTM-Attention hybrid prediction model to obtain predicted load and green electricity output curves; establishing mathematical models for key equipment and operational constraints of the integrated energy system to obtain an operational optimization model for the integrated energy system; in the operational optimization model, establishing a multi-objective function with the optimization objectives of minimizing annual total operating cost, minimizing equipment power fluctuations, and maximizing overall system energy efficiency; and using the wind-sensitive kingpin optimization algorithm to solve for a set of candidate scheduling schemes based on the obtained predicted load and green electricity output curves; and using the energy efficiency-economic equilibrium corridor decision method to screen the generated set of candidate scheduling schemes and determine the optimal compromise scheduling scheme.

2. The comprehensive energy dispatch optimization method for green electricity direct-connection parks as described in claim 1, characterized in that: The method of using historical operational data to predict the load of the park and the green electricity supply of upstream power plants in the future through a CNN-BiLSTM-Attention hybrid prediction model to obtain the predicted load and green electricity output curves includes the following steps: preprocessing the acquired historical operational data to obtain an input sequence; inputting the input sequence into the convolutional neural network of the CNN-BiLSTM-Attention model to perform convolution operations, extracting local short-term dependencies and spatial features in the input sequence, and obtaining local features in the time dimension; The extracted local features are input into a bidirectional long short-term memory network to learn long-term temporal dependencies in the forward and backward directions, and output a hidden state vector. Based on the attention mechanism, the weights of features at different time steps are calculated, and the hidden state vectors output by the bidirectional long short-term memory network are weighted and fused to obtain a context vector. The context vector of the last time step is linearly transformed through a fully connected layer to obtain the final prediction result.

3. The comprehensive energy dispatch optimization method for green electricity direct-connection parks as described in claim 1, characterized in that: Mathematical models are established for key equipment and operational constraints in integrated energy systems, including mathematical models for air source heat pumps, electric air conditioners, electrochemical energy storage systems, ice storage systems, and high-temperature hot water thermal storage systems. The operational constraints for integrated energy dispatch optimization include energy balance constraints, equipment capacity constraints, equipment operation constraints, energy storage system constraints, and grid interaction constraints.

4. The comprehensive energy dispatch optimization method for green electricity direct-connection parks as described in claim 1, characterized in that: Based on the obtained predicted load and green power output curves, the process of using the wind-sensitive golden butterfly optimization algorithm to solve for the candidate scheduling scheme set includes the following steps: initializing algorithm parameters, and randomly generating parameters within the solution space consisting of all decision variables, i.e., the power of each device in each scheduling period. An initial population is constructed using an initial individual; each individual in the population is decoded into a specific device power scheduling sequence, and after verifying its feasibility, a multi-objective function value is calculated. Individuals that meet the requirements are selected based on the objective function value for subsequent iterations. The population is iteratively updated. In each iteration, dynamic switching probabilities are generated based on the predicted uncertainty to guide individuals to choose global exploration or local development for position updates, and mutation is completed by combining direction vectors and adaptive step size. Subsequently, constraints are repaired for individuals, inferior solutions are eliminated and new solutions are added, and finally merged to form a new generation of population. It is determined whether the iteration termination condition is met. If not, the next iteration is carried out until the termination condition is met, and a set of candidate scheduling schemes is output.

5. The integrated energy dispatch optimization method for green electricity direct-connection parks as described in claim 4, characterized in that: In each iteration, perform the following operation for each individual in the population: compute the direction vector of the individual toward the global optimum used for individual movement control. ;Calculate the adaptive step size for the individual to move towards the target direction based on its position. Based on the source-load uncertainty data predicted by the CNN-BiLSTM-Attention model, the dynamic switching probability is calculated. Based on dynamic switching probability Direction vector and adaptive step size The individual is updated; the new individual generated after the update is constrained and verified. If the operational constraints are violated, the new individual is adjusted using the repair operator to obtain the constrained and repaired individual; by calculating the Euclidean distance between each individual and the current global optimal solution, a set number of individuals are eliminated to remove inferior solutions, and an equal number of new individuals are randomly generated in the solution space of the feasible region to maintain population diversity; then the parent individuals, the position-updated and constrained and repaired offspring individuals, and the newly generated random individuals are merged to form the updated new generation of population.

6. The comprehensive energy dispatch optimization method for green electricity direct-connection parks as described in claim 1, characterized in that: A method for calculating dynamic switching probability based on source-load uncertainty data predicted by a CNN-BiLSTM-Attention model includes the following steps: For the current scheduling period, predict the load and output based on the CNN-BiLSTM-Attention model, calculate the uncertainty, normalize the uncertainty based on the uncertainty and the installed capacity and maximum historical load of the corresponding equipment, and then weight the normalized uncertainty to obtain a comprehensive prediction uncertainty measure. A dynamic switching probability generation function is constructed by using the dynamic switching probability to follow the uncertainty measure monotonically increasing, and the dynamic switching probability is calculated based on the uncertainty measure.

7. The comprehensive energy dispatch optimization method for green electricity direct-connection parks as described in claim 1, characterized in that: Based on dynamic switching probability The process of updating an individual includes the following: generating a random number r∈[0,1]; if The butterfly optimization algorithm is used for global exploration, controlling individuals to move towards the current global optimum, and combining this with the direction vector. With adaptive step size Explore and update the individual; if It executes a self-organizing algorithm for local development, controlling individuals to perform random searches within a local area, and combining the direction and step size direction vectors. With adaptive step size Develop and update individuals.

8. A comprehensive energy dispatching and optimization system for green electricity direct-connection industrial parks, characterized in that, include: The data acquisition module is configured to acquire historical operating data of the park and the upstream power plants used for direct power supply; The prediction module is configured to use the acquired historical operating data to predict the load of the park and the green electricity supply of the upstream power station in the future through the constructed CNN-BiLSTM-Attention hybrid prediction model, and obtain the predicted load and green electricity output curves; the model building module is configured to build mathematical models for the key equipment and operating constraints of the integrated energy system, and obtain the operation optimization model of the integrated energy system. The optimization module is configured to minimize the total annual operating cost, minimize equipment power fluctuations, and maximize the overall system energy efficiency in the running optimization model. It establishes a multi-objective function and uses the wind-sensitive golden butterfly optimization algorithm to solve for a set of candidate scheduling schemes based on the obtained predicted load and green electricity output curves. The decision-making module is configured to use the energy efficiency-economic equilibrium corridor decision method to screen the generated set of candidate scheduling schemes and determine the optimal compromise scheduling scheme.

9. A comprehensive energy dispatching and optimization system for green electricity direct-connection industrial parks, characterized in that: include: The upstream wind and solar power generation system achieves direct green electricity connection with the park through a set line; the energy management system is configured to execute the comprehensive energy dispatch optimization method for green electricity direct connection parks as described in any one of claims 1-7, output the optimal dispatch scheme, and dispatch each equipment unit of the comprehensive energy system.

10. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, it completes the steps in the integrated energy dispatch optimization method for green electricity direct connection parks as described in any one of claims 1-7.

Images