Low-carbon park flexible resource optimization scheduling method based on improved parrot optimization algorithm

Abstract

The application discloses a low-carbon park flexible resource optimization scheduling method based on an improved parrot optimization algorithm and belongs to the technical field of energy optimization scheduling. The method comprises the following steps: defining a fitness evaluation function based on a multi-objective optimization model and generating initial candidate solution individuals; constructing a corresponding reverse candidate solution individual for each initial candidate solution individual, simultaneously constructing a candidate solution individual set by combining the initial candidate solution individual and the reverse candidate solution individual, and constructing an initial population based on the candidate solution individual set; iteratively optimizing the initial population based on the improved parrot optimization algorithm; when the iterative optimization has reached a preset termination condition, outputting an optimal solution and decoding the optimal solution into standardized scheduling instructions of each flexible resource in each scheduling period, so as to realize collaborative optimization scheduling of the low-carbon park.

Description

Technical Field

[0001] This application relates to the field of energy optimization scheduling technology, and more specifically, to a flexible resource optimization scheduling method for low-carbon industrial parks based on an improved parrot optimization algorithm. Background Technology

[0002] With the development of low-carbon industries and the improvement of the intelligence level of industrial park energy systems, low-carbon parks integrating various types of flexible resources such as photovoltaics, wind power, energy storage, and controllable loads are gradually becoming an important form of energy management. Optimizing the scheduling of various flexible resources within the park helps to improve the capacity for renewable energy absorption and reduce operating costs and carbon emission levels.

[0003] However, the low-carbon park scheduling problem typically features multiple objectives (economic efficiency, environmental protection, etc.), multiple constraints (power balance, equipment operation constraints, etc.), and strong nonlinearity and uncertainty, significantly increasing the difficulty of optimization. Existing intelligent optimization methods based on genetic algorithms, particle swarm optimization, etc., generally suffer from slow convergence speed, susceptibility to local optima, and difficulty in balancing global search and local exploitation capabilities in practical applications, making it difficult to meet the needs of efficient solutions in complex scheduling scenarios.

[0004] In summary, how to provide an optimization scheduling method based on the improved parrot optimization algorithm that can achieve efficient optimization under complex constraints, take into account both global search and local development capabilities, and improve the flexible resource scheduling effect of low-carbon parks has become an urgent technical problem to be solved. Summary of the Invention

[0005] To overcome a series of shortcomings in existing technologies, the purpose of this application is to provide a flexible resource optimization scheduling method for low-carbon industrial parks based on an improved parrot optimization algorithm, comprising the following steps: A fitness evaluation function is defined based on a multi-objective optimization model, and initial candidate solutions are generated. For each initial candidate solution individual, a corresponding reverse candidate solution individual is constructed. At the same time, the initial candidate solution individual and the reverse candidate solution individual constitute a candidate solution individual set, and an initial population is constructed based on the candidate solution individual set. The initial population was iteratively optimized based on the improved parrot optimization algorithm; When the iterative optimization has reached the preset termination condition, the optimal solution is output, and the optimal solution is decoded into standardized scheduling instructions for each flexible resource in each scheduling period, so as to realize the collaborative optimization scheduling of the low-carbon park.

[0006] In some embodiments, the multi-objective optimization model includes: An economic objective function is used to minimize the total operating cost of the park during the scheduling cycle. The total operating cost includes the cost of purchased electricity, gas, equipment depreciation, and operation and maintenance, and deducts the revenue from interruptible load compensation and surplus electricity fed into the grid. An environmental objective function is used to minimize the total carbon emissions of the park, which includes indirect carbon emissions from the grid side corresponding to purchased electricity and direct carbon emissions from gas consumption within the park. The renewable energy consumption rate objective function is used to maximize the utilization ratio of renewable energy and improve the local consumption level of clean energy by minimizing the amount of wind and solar power curtailment. The load fluctuation mitigation objective function is used to minimize the inter-period fluctuation amplitude of the power interaction between the park and the power grid, so as to reduce the impact on the external power grid. The weighted coefficient method is used to linearly combine multiple objective functions into a single objective optimization problem. The weight coefficients are set by the park's operators based on actual preferences and policy guidance. Alternatively, the constraint method can be used to transform some objectives into constraints.

[0007] In some embodiments, the method for generating initial candidate solutions is as follows: Control parameters and initial conditions are set based on a chaotic mapping model to generate an initial chaotic sequence; Iterative calculations are performed on the chaotic mapping model to obtain a sequence of chaotic variables whose values ​​are located within a preset interval; The chaotic variable sequence is mapped to the range of values ​​of each decision variable in the optimization problem to obtain the corresponding initial values ​​of the decision variables; For different types of decision variables, initial values ​​of the decision variables are constrained to obtain the variable values. The values ​​of each decision variable are combined to obtain the initial candidate solutions.

[0008] In some embodiments, the method for constructing a corresponding reverse candidate solution for each initial candidate solution is as follows: For each decision variable in the initial candidate solution, based on its value range and variable type, construct the inverse value symmetric about the center point or boundary within the domain of the corresponding variable; Based on the combined values ​​of the inverse variables corresponding to each decision variable, an inverse candidate solution individual corresponding to each initial candidate solution individual is constructed.

[0009] In some embodiments, the initial population is constructed as follows: Statistical analysis of historical optimal solutions yields the statistical characteristics of each decision variable; Based on statistical characteristics, the candidate solutions in the candidate solution individual set are guided and corrected. Constraint violation degree is calculated for individual candidate solutions after guided correction, and fitness is evaluated based on a multi-objective optimization model; The candidate solutions are ranked according to the fitness assessment results, and the candidate solutions whose fitness meets the preset screening conditions are selected according to the preset population size to construct the initial population.

[0010] In some embodiments, the method for iteratively optimizing the initial population based on the improved parrot optimization algorithm is as follows: Select the individual with the best fitness from the initial population as the global best individual; The initial population is divided into neighborhoods according to a preset ring topology, and the optimal individual in the neighborhood is determined within the corresponding neighborhood of each individual. For each individual to be updated, a random number is generated and compared with a preset following probability threshold. If the random number is less than the following probability threshold, the individual to be updated follows the global optimal individual; otherwise, it follows the neighborhood optimal individual. Based on the current position of the individual to be updated, the position of its following objects, and the preset step size factor, the position of the individual to be updated is updated, and a random perturbation term is introduced during the update process; Boundary constraints are applied to the updated individual positions to ensure that the values ​​of each decision variable are within the preset domain. Repeat the above steps until the preset termination condition is met, and output the optimized result.

[0011] In some embodiments, the method for optimal decoding into standardized scheduling instructions for each flexible resource in each scheduling period is as follows: The optimal solution is analyzed according to the preset decision variable coding rules to obtain the power variables and operating status variables of each flexible resource in each scheduling period. Based on power variables and operating status variables, scheduling information for each flexible resource is generated. Among them, the energy storage system corresponds to the charging and discharging power of each time period, the controllable load equipment corresponds to the adjustable power or transfer power of each time period, the photovoltaic power generation system and wind turbine generator set correspond to the scheduling output value of each time period, and the start-stop status of each equipment corresponds to the 0-1 operating status command. The scheduling information of each flexible resource is organized according to the scheduling time period sequence to generate a power sequence and a state sequence with the scheduling time period as the unit; The power sequence and state sequence are subjected to a feasibility check to determine whether they meet the physical constraints of the equipment. The physical constraints include at least the power ramp-up constraint, the minimum start-up time constraint, and the minimum downtime constraint. When a constraint violation is detected, the power sequence is smoothed or the state sequence is time-adjusted to meet the constraint conditions. After passing the feasibility verification, standardized scheduling instructions are generated. The standardized scheduling instruction file includes at least equipment identification information, scheduling time period information, power setting value, and operating status instructions.

[0012] In some embodiments, the flexible resource optimization scheduling method for low-carbon industrial parks further includes: During the iterative optimization process, each candidate solution is updated in position by following the current global best solution or a good neighboring solution, and the search space is expanded under the action of random perturbation. During the iterative optimization process, the exploration intensity parameter and the randomness parameter are dynamically adjusted based on the population dispersion index and the optimization progress index to achieve dynamic coordination between global search and local development capabilities.

[0013] Compared with the prior art, this application has the following beneficial effects: This application introduces an improved parrot optimization algorithm that integrates multiple strategies, including chaotic mapping and reverse learning initialization, population stratification and differentiated evolution, elite-guided mutation, and adaptive parameter adjustment, to achieve dynamic coordination between global search and local development in multi-objective scheduling of low-carbon parks. This significantly improves the convergence efficiency and solution quality of flexible resource collaborative optimization under complex constraints. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the flexible resource optimization and scheduling method for low-carbon industrial parks based on the improved parrot optimization algorithm disclosed in this application. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some embodiments of this invention, but not all embodiments.

[0016] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] The embodiments and directional terms described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0018] like Figure 1 As shown, the flexible resource optimization scheduling method for low-carbon industrial parks based on the improved parrot optimization algorithm includes the following steps: Based on the flexible resource allocation information and operational constraints of low-carbon industrial parks, and combined with the load forecast data, renewable energy output forecast data and electricity price information of the current scheduling cycle, a multi-objective optimization model for flexible resource collaborative scheduling is constructed. A fitness evaluation function is defined based on a multi-objective optimization model, and an initial candidate solution is generated using a chaotic mapping method. Based on the reverse learning strategy, a corresponding reverse candidate solution individual is constructed for each initial candidate solution individual. At the same time, the initial candidate solution individual and the reverse candidate solution individual are combined to form a candidate solution individual set, and an initial population is constructed based on the candidate solution individual set. The initial population was iteratively optimized based on the improved parrot optimization algorithm; During the iterative optimization process, each candidate solution is updated in position by following the current global best solution or a good neighboring solution, and the search space is expanded under the action of random perturbation. During the iterative optimization process, the exploration intensity parameter and the randomness parameter are dynamically adjusted based on the population dispersion index and the optimization progress index to achieve dynamic coordination between global search and local development capabilities. During the iterative optimization process, the initial population is divided into high-performance subgroups, medium-performance subgroups, and low-performance subgroups based on individual performance indicators. Differentiated strategies are adopted for position updates for different subgroups, and the hierarchical results are dynamically adjusted according to the optimization effect of each update strategy. When the preset triggering conditions are met, candidate individuals that meet the preset elite determination conditions are selected from the high-performance subgroup based on the fitness evaluation results, and they are constructed into an elite individual set. Based on the elite individual set, guided mutation operations are performed on non-elite individuals to enhance local search capabilities. If the iterative optimization does not reach the preset termination condition, the iterative optimization will continue until the termination condition is reached. If the iterative optimization has reached the preset termination condition, the optimal solution is output, and the optimal solution is decoded into standardized scheduling instructions for each flexible resource in each scheduling period. The standardized scheduling instructions are then sent to the corresponding flexible resource equipment to achieve collaborative optimization scheduling of the low-carbon park.

[0019] The proposed method for optimizing the scheduling of flexible resources in low-carbon industrial parks constructs a multi-objective optimization model that integrates load forecasting, renewable energy output forecasting, and time-of-use pricing information. This model enables collaborative scheduling decision-making for various flexible resources within the park. By introducing chaotic mapping and reverse learning strategies to generate a high-quality initial population, the uniformity and diversity of candidate solution distribution are improved, enhancing global search capabilities from the outset. Furthermore, the Parrot Optimization algorithm is combined with a joint feedback mechanism between population dispersion and optimization progress to dynamically adjust exploration intensity and randomness parameters, achieving an adaptive balance between global search and local development capabilities. Additionally, a population hierarchical and differentiated update strategy based on individual performance, along with a mutation mechanism guided by elite individuals, effectively improves the algorithm's convergence speed and optimization accuracy. This results in efficient collaborative scheduling of various flexible resources within low-carbon industrial parks across multiple time periods, significantly reducing overall operating costs and carbon emission levels, while improving stability and economic efficiency.

[0020] In some embodiments, the flexible resource allocation information includes: The capacity parameters, power boundary parameters, charge / discharge efficiency coefficients, and cycle life degradation characteristics of the various types of energy storage systems deployed in the park; The rated power range, allowable adjustment window, load transfer time constraints, and user comfort sensitivity coefficient of controllable load equipment; The number of electric vehicle charging pile clusters connected, the charging power range of a single pile, the statistical characteristics of vehicle arrival and departure times on a typical day, and the boundary constraints of battery state of charge. The installed capacity of photovoltaic power generation systems, the rated capacity of inverters, the power factor adjustment range and the active and reactive power coordination control capabilities, as well as the rated power of wind turbine generators, the cut-in and cut-out wind speed thresholds, the output characteristic curves and the power ramp-up constraints per unit time. The heat-to-power ratio parameters, minimum technical output, start-up and shutdown time constants, and coal consumption rate curves for power supply and heating under different operating conditions of the combined heat and power units in the industrial park.

[0021] For example, in the actual operation scenario of Low-Carbon Industrial Park A, the park is equipped with two sets of electrochemical energy storage systems, each with a capacity of 5MWh, a power boundary parameter of ±2.5MW, and charge / discharge efficiency coefficients of 0.92 and 0.90, respectively. The cycle life degradation characteristic parameter is characterized by a 1.5% capacity reduction after every 500 equivalent charge / discharge cycles. Controllable load equipment includes a central air conditioning system and some process heating equipment, with a rated power range of 0.5MW to 3MW. The allowable adjustment window is from 08:00 to 22:00 daily, with a load transfer time constraint of no more than 15 minutes. The user comfort sensitivity coefficient is set between 0.3 and 0.8 based on the indoor temperature deviation. A total of 120 DC fast charging piles are deployed for electric vehicles, with a single pile charging power range of 30kW to 120kW. On typical days, vehicle arrival times are concentrated between 08:00 and 22:00. The operating hours are 10:00 and 17:00-20:00, with departure times concentrated between 09:00-11:00 and 18:00-21:00. The battery state of charge boundary is limited to 20% to 95%. The photovoltaic power generation system has an installed capacity of 8MW, an inverter rated capacity of 7.5MW, and a power factor adjustment range of 0.95 (lagging) to 0.95 (leading), possessing a certain active and reactive power coordination control capability. The wind turbine generators have a total rated power of 6MW, a cut-in wind speed of 3m / s, and a cut-out wind speed of 25m / s. The output characteristic curve is represented by a piecewise linear fitting method, and the power ramp-up per minute does not exceed 10% of the rated power. The park's cogeneration unit is equipped with one 2MW unit with a heat-to-power ratio parameter of 1.2, a minimum technical output of 0.4MW, and a start-stop time constant of 45 minutes. The coal consumption rate curves for power supply and heating under different operating conditions exhibit a non-linear relationship.

[0022] The aforementioned flexible resource allocation information in this application, by uniformly characterizing the operating boundary and efficiency degradation characteristics of the energy storage system, controllable load adjustment constraints, and user comfort parameters, ensures that various adjustable resources have a quantifiable scheduling basis. On this basis, combined with the charging behavior characteristics and state of charge constraints of electric vehicles, as well as the output characteristics and adjustment capabilities of photovoltaic and wind power, collaborative modeling of multi-source uncertain resources is achieved. Furthermore, the thermoelectric coupling operation parameters of cogeneration units are introduced to achieve a unified expression of multi-energy constraints of electricity and heat, thereby constructing a complete and accurate flexible resource parameter system.

[0023] In some embodiments, the operational constraints include: The power balance equation constraint for each time period in the park is defined as the sum of the total output of all power generation units in the park, the discharge power of the energy storage system, and the power supply of the external power grid at any scheduling time, which is equal to the sum of the total load demand of the park, the charging power of the energy storage system, and the network loss. The recursive constraint of the state of charge of the energy storage system is characterized by the fact that the state of charge at the current moment is equal to the sum of the state of charge at the previous moment and the increment of the charging and discharging power under the influence of time interval and charging and discharging efficiency. It is used to describe the energy evolution process of the energy storage system in each scheduling period. The upper and lower limits of the state of charge of the energy storage system and the consistency constraints of the state of charge at the beginning and end of the time are used to prevent overcharging and over-discharging of the energy storage and to ensure the periodic continuity of the energy storage operation within the scheduling cycle. The output boundary constraints of photovoltaic power generation systems and wind turbine generators limit the actual dispatch output of photovoltaic and wind power to between zero and the predicted output, so as to realize the operation mechanism that allows wind and solar curtailment but prohibits over-generation; at the same time, it includes equipment operation status logical constraints, using 0-1 integer variables to represent the unit start-up and shutdown status, energy storage charging and discharging mutual exclusion status, and controllable load switching status, and avoids conflicting operation status of the same equipment at the same time through logical constraint relationships; The bidirectional power constraint on the transmission power of the tie line and the capacity limit constraint on the main transformer are used to limit the power exchange range between the park and the external power grid, and to ensure that the tie line and transformer equipment operate within the safe operating capacity range.

[0024] The aforementioned operational constraints in this application establish power balance constraints for each time period within the park to ensure a dynamic matching relationship between power generation, energy storage, external power supply, and load demand. By introducing recursive state-of-charge (SOC) constraints and upper and lower limit constraints for the energy storage system, the continuity and safety control of the energy storage evolution process are achieved. Based on this, combined with the output boundary constraints of photovoltaic and wind power and the logical constraints of equipment operating status, over-limit operation of equipment and state conflicts are avoided, ensuring the coordinated operation of multiple resources. Furthermore, through bidirectional power constraints on tie lines and capacity limitations of main transformers, the power exchange range between the park and the external power grid is standardized, thereby constructing a complete and rigorous operational constraint system to improve dispatch feasibility and safety.

[0025] In some embodiments, the method for obtaining load forecasting data is as follows: Historical power curve data of various loads in the park are obtained, and the historical power curve data are preprocessed, including the removal of abnormal data and the imputation of missing data, to obtain load time series data. Feature extraction is performed on the load time series data to obtain time features, including hourly features, weekday features, monthly features, and holiday identifiers; at the same time, meteorological data related to the load are acquired, including temperature, humidity, and irradiance. The time features and meteorological data are aligned and fused to construct a multi-source feature dataset. The multi-source feature dataset is then used to construct samples based on the sliding window method to obtain a training sample sequence dataset. Based on the training sample sequence dataset, a long short-term memory neural network load prediction model was constructed and trained. At the same time, the model hyperparameters were tuned using the validation set. The trained load forecasting model is used to predict the load for each time period within the future scheduling cycle, and the load forecast values ​​for each time period are obtained.

[0026] The load forecasting data acquisition method described in this application improves the quality of raw data by removing anomalies and imputing missing data in historical load power curves; it achieves synergistic characterization of multi-source influencing factors by extracting time features and fusing meteorological data such as temperature, humidity, and irradiance; based on this, it constructs training samples using a sliding window and performs model training and parameter optimization based on a long short-term memory neural network to achieve effective learning of load time-series characteristics; thereby obtaining high-precision load forecasting results for future scheduling periods and improving the foresight and accuracy of scheduling decisions.

[0027] In some embodiments, the method for generating renewable energy output forecast data is as follows: Acquire historical irradiance data and component temperature data of photovoltaic power generation systems, as well as historical wind speed data of wind turbine generators; Historical irradiance data, component temperature data, and historical wind speed data are preprocessed to obtain standardized input data; Based on standardized input data, photovoltaic power output prediction models and wind power output prediction models are constructed respectively. The photovoltaic power output prediction model is a power calculation model that considers a temperature correction coefficient, and the wind power output prediction model is a wind speed-power conversion model based on the wind turbine power curve. Acquire meteorological forecast data for the future scheduling cycle, including irradiance forecast data and wind speed forecast data; Meteorological forecast data are input into photovoltaic power output forecast models and wind power output forecast models respectively to calculate renewable energy output forecast data for each period and generate renewable energy output forecast curves.

[0028] For example, in Low-Carbon Industrial Park A, a photovoltaic power generation system collected 15-minute historical data from the past year, including solar irradiance data (unit: W / m²). 2The system collects irradiance and component temperature data (in °C), and simultaneously collects historical wind speed data (in m / s) for the corresponding time period from the wind turbine generator. First, the raw data is aligned to a unified time scale, and then the Min-Max normalization method is used to standardize the irradiance, temperature, and wind speed data to eliminate the influence of different units on model training. In the photovoltaic power output modeling process, standardized irradiance is used as the basic input, and a component temperature correction coefficient is introduced. For example, when the component temperature rises above 25 °C, the theoretical output is linearly reduced, thus constructing a photovoltaic power output prediction model of "irradiance—temperature correction—power output". In the wind power output modeling process, based on the power curve provided by the generator manufacturer, the wind speed range is divided into a cut-in zone, a linear growth zone, and a rated plateau zone. Based on this, a mapping relationship model from wind speed to power is established to achieve a nonlinear conversion from wind speed input to wind power output. During the forecasting phase, meteorological forecast data for the future scheduling cycle is acquired and input into the trained or calibrated photovoltaic (PV) power output forecasting model and wind power output forecasting model, respectively, to calculate the predicted PV power output and wind power output values ​​for the corresponding time periods. Finally, the PV power output forecasting results and wind power output forecasting results for each time period are spliced ​​and summarized to form a unified renewable energy power output forecasting curve. For example, under a typical sunny day scenario, PV power output shows a single-peak curve with a clear midday peak, while wind power output exhibits a random fluctuation curve that varies with wind speed, thus providing an input basis for subsequent integrated energy scheduling and flexible resource optimization in the park.

[0029] The renewable energy output forecasting data generation method described in this application improves the consistency and reliability of input data by preprocessing and standardizing historical meteorological data such as irradiance, module temperature, and wind speed. It achieves refined characterization of the output characteristics of different types of renewable energy by constructing a photovoltaic output forecasting model considering temperature correction and a wind power output forecasting model based on wind turbine power curves. Based on this, it performs model-driven calculations using meteorological forecast data for future scheduling cycles to generate output forecasting results and continuous forecast curves for each time period, thereby improving the accuracy and stability of renewable energy output forecasting and providing reliable support for optimized scheduling.

[0030] In some embodiments, the electricity price information includes time-of-use pricing schemes, demand charge rules, interruptible load compensation pricing policies, and distributed generation feed-in tariffs published by the power grid company. The time-of-use pricing scheme divides 24 hours into peak, mid-peak, flat, and off-peak periods, each corresponding to a different electricity price. The demand charge rules calculate fees based on the maximum demand value within the billing cycle at a unit price. The interruptible load compensation pricing policy provides corresponding price compensation when the park responds to grid demand-side management instructions and proactively reduces load. The distributed generation feed-in tariff specifies the purchase price for surplus electricity generated by renewable energy sources in the park after self-consumption and then fed back to the grid. Based on the electricity price information, an economic objective function is constructed, including electricity purchase cost, demand electricity expenditure, interruptible load compensation revenue, and electricity sales revenue. This enables the optimized scheduling process to: prioritize energy storage charging and transferable load electricity consumption during off-peak electricity price periods, and prioritize the release of energy storage power and transfer of controllable loads during peak electricity price periods; and simultaneously reduce demand electricity expenditure by optimizing the interaction power with the grid in each time period to reduce the maximum demand value.

[0031] In some embodiments, the multi-objective optimization model includes: An economic objective function is used to minimize the total operating cost of the park during the scheduling cycle. The total operating cost includes the cost of purchased electricity, gas, equipment depreciation, and operation and maintenance, and deducts the revenue from interruptible load compensation and surplus electricity fed into the grid. An environmental objective function is used to minimize the total carbon emissions of the park, which includes indirect carbon emissions from the grid side corresponding to purchased electricity and direct carbon emissions from gas consumption within the park. The renewable energy consumption rate objective function is used to maximize the utilization ratio of renewable energy and improve the local consumption level of clean energy by minimizing the amount of wind and solar power curtailment. The load fluctuation mitigation objective function is used to minimize the inter-period fluctuation amplitude of the power interaction between the park and the power grid, so as to reduce the impact on the external power grid. The weighted coefficient method is used to linearly combine multiple objective functions into a single objective optimization problem. The weight coefficients are set by the park's operators based on actual preferences and policy guidance. Alternatively, the constraint method can be used to transform some objectives into constraints.

[0032] For example, in the intraday optimization scheduling scenario of Low-Carbon Industrial Park A, the scheduling cycle is 24 hours and the time resolution is 15 minutes. The park needs to coordinate the output of purchased electricity, gas turbine units, energy storage systems, and renewable energy sources simultaneously. In the economic objective function, assuming that the park's purchased electricity is 12,000 kWh in a certain scheduling cycle and the average electricity price is 0.8 yuan / kWh, the cost of purchased electricity is 9,600 yuan; the gas consumption cost of the gas turbine units is 3,500 yuan; the total equipment depreciation and maintenance cost is 1,200 yuan; and the compensation revenue of 800 yuan is obtained from implementing interruptible load regulation, and the revenue from selling surplus electricity to the grid is 600 yuan. The total operating cost is calculated using the method of "total cost minus revenue" to obtain the final economic evaluation value. Regarding the environmental objective function, assuming the grid-side indirect carbon emission factor is 0.6 kg CO2 / kWh, the carbon emissions corresponding to purchased electricity are 7200 kg CO2; the direct carbon emissions from gas consumption are 1800 kg CO2, resulting in a total carbon emission of 9000 kg CO2 for the park. This indicator is used to measure the low-carbon level of the dispatch scheme. In the renewable energy absorption rate objective function, assuming the total available power generation from photovoltaic and wind power in the park is 8000 kWh, with actual absorption of 7200 kWh, the amount of wind and solar power curtailed is 800 kWh. Minimizing this curtailment maximizes the renewable energy absorption rate, thereby improving the efficiency of clean energy utilization. In the load fluctuation smoothing objective function, the exchange power between the park and the grid during adjacent dispatch periods is statistically analyzed. For example, if the exchange power is 5MW and 3MW in two adjacent periods, the fluctuation range is 2MW. Minimizing this fluctuation value during optimization reduces the impact on the external grid and improves the smoothness of system operation. In the multi-objective fusion stage, the weighted coefficient method is used to linearly combine the above-mentioned economic, environmental, absorption rate, and volatility objective functions. For example, the economic weight is set to 0.4, the environmental weight to 0.3, the absorption rate weight to 0.2, and the volatility weight to 0.1, thereby constructing a unified single-objective optimization function. Alternatively, in scenarios with strong policy constraints, the total carbon emissions or curtailment rate is set as the upper limit of the constraint, and the remaining objectives are used for minimization optimization, thereby achieving flexible scheduling decisions under different operating strategies.

[0033] The multi-objective optimization model described in this application minimizes the overall operating cost of the industrial park by establishing an economic objective function centered on the costs of purchased electricity, gas, and equipment operation and maintenance. It also constructs an environmental objective function by introducing indirect carbon emissions from the power grid and direct carbon emissions from gas consumption within the park, effectively constraining carbon emission levels. Furthermore, it constructs a renewable energy absorption rate objective function by minimizing wind and solar power curtailment and a load smoothing objective function by constraining interactive power fluctuations, achieving synergistic optimization of multi-dimensional operational performance. Finally, it employs a weighted coefficient method or constraint transformation method to uniformly process the multiple objectives, thereby improving the overall optimality and flexibility of the scheduling results while meeting multiple operational needs.

[0034] In some embodiments, the method for generating initial candidate solutions using the chaotic mapping method is as follows: Control parameters and initial conditions are set based on a chaotic mapping model to generate an initial chaotic sequence; Iterative calculations are performed on the chaotic mapping model to obtain a sequence of chaotic variables whose values ​​are located within a preset interval; The chaotic variable sequence is mapped to the range of values ​​of each decision variable in the optimization problem to obtain the corresponding initial values ​​of the decision variables; For different types of decision variables, initial values ​​of the decision variables are constrained to obtain the variable values. The values ​​of each decision variable are combined to obtain the initial candidate solutions.

[0035] For example, when solving the flexible resource scheduling problem in low-carbon industrial parks based on an improved optimization algorithm, a Logistic chaotic mapping model is selected as the initial solution generation tool, with control parameters set to μ=3.99 and initial seed value x0=0.37. Under these parameter conditions, a chaotic sequence of length N is generated through iterative calculation, resulting in a distribution characteristic of quasi-random but highly ergodic values ​​within the interval (0,1). Subsequently, the obtained chaotic variable sequence is mapped to the domain of each decision variable in the optimization problem. For example, the range of the energy storage charging and discharging power variable is [-2.5MW, 2.5MW], so the chaotic variable is converted into the corresponding initial power value through a linear mapping relationship; the range of the controllable load regulation rate variable is [0,1], so the initial regulation coefficient is obtained directly by proportional mapping; the range of the tie-line power exchange variable is [-5MW, 5MW], and the initialization is also completed through normalization mapping. After mapping, constraint corrections are applied to different types of variables. For example, charging and discharging mutual exclusion constraints are applied to the energy storage power variable, the controllable load variable is piecewise corrected to meet user comfort boundary conditions, and upper and lower limits are truncated for tie-line power to ensure that all variables meet physical feasibility constraints. Finally, the legal values ​​of each decision variable are combined and encoded according to the structure of "energy storage status - load regulation - tie-line power - distributed power output" to form a set of initial candidate solutions that meet the constraints and have good distribution, thus providing a high-quality initial population foundation for improving the global search of the parrot optimization algorithm.

[0036] The initial candidate solution generation method described in this application generates a chaotic sequence with good ergodicity by constructing a chaotic mapping model and setting control parameters and initial conditions. By performing interval mapping on the chaotic sequence, it is converted into the initial values ​​of each decision variable, achieving uniform coverage of the solution space. On this basis, the initial values ​​are modified in combination with the value constraints of different types of decision variables to ensure the feasibility of candidate solutions. Thus, it generates initial candidate solutions with uniform distribution and high diversity, improving the global search capability and convergence performance of the subsequent optimization process.

[0037] In some embodiments, the method for constructing a corresponding reverse candidate solution for each initial candidate solution is as follows: For each decision variable in the initial candidate solution, based on its value range and variable type, construct the inverse value symmetric about the center point or boundary within the domain of the corresponding variable; Based on the combined values ​​of the inverse variables corresponding to each decision variable, an inverse candidate solution individual corresponding to each initial candidate solution individual is constructed.

[0038] The reverse candidate solution construction method described in this application generates reverse values ​​symmetrical about the center point or boundary within the corresponding domain for each decision variable in the initial candidate solution, based on its value range and variable type, thereby achieving a symmetrical expansion of the original solution space. On this basis, the reverse variable values ​​are combined to construct reverse candidate solution individuals that correspond one-to-one with the initial candidate solutions, thus effectively improving population diversity and solution space coverage.

[0039] In some embodiments, the initial population is constructed as follows: Statistical analysis of historical optimal solutions yields the statistical characteristics of each decision variable; Based on statistical characteristics, the candidate solutions in the candidate solution individual set are guided and corrected. Constraint violation degree is calculated for individual candidate solutions after guided correction, and fitness is evaluated based on a multi-objective optimization model; Candidate solutions are ranked according to the fitness assessment results, and candidates whose fitness meets the preset screening conditions are selected according to the preset population size to construct the initial population.

[0040] For example, when solving a multi-objective optimization scheduling problem, several sets of historical optimal solutions obtained during historical scheduling are first statistically analyzed. The distribution characteristics of each decision variable are extracted. For instance, the charging and discharging power of the energy storage system is mostly concentrated in the range of [-1.8MW, 1.5MW], the controllable load regulation rate is mostly concentrated between 0.3 and 0.7, and the tie-line power exchange is mostly distributed in the range of [-3MW, 3MW]. This forms the statistical mean and variance of each variable. In the guided correction stage, the candidate solutions based on chaotic mapping or random generation are compared with the above statistical characteristics. For variable values ​​that deviate significantly from the historical optimal distribution range, such as the energy storage charging power exceeding the upper limit of the historical high-frequency range, a certain proportion is used to shrink and correct towards the statistical mean to improve the feasibility and convergence quality of the initial solution. During the constraint violation calculation phase, each corrected candidate solution is checked to see if it satisfies the power balance constraint, energy storage state boundary constraint, and equipment operation logic constraint. For example, if the energy storage simultaneously satisfies that both charging and discharging variables are non-zero at a certain moment, it is judged as a logical conflict and the corresponding violation penalty term is calculated. At the same time, the fitness of each candidate solution is evaluated based on a multi-objective optimization model that considers economic efficiency, environmental performance, and renewable energy absorption rate, and a comprehensive evaluation value is obtained. Subsequently, all candidate solutions are ranked according to the fitness evaluation results. For example, the top 30% of individuals with better fitness and lower constraint violation are selected as preferred candidate solutions. Combined with the preset population size requirement (e.g., population size of 100), truncation or supplementation is performed to finally construct an initial population with both diversity and high quality characteristics, thereby providing a better search starting point for the subsequent optimization iteration process and accelerating the convergence process.

[0041] The initial population construction method described in this application extracts the distribution characteristics of each decision variable through statistical analysis of historical optimal solutions, thereby providing prior guidance for candidate solutions. It then guides and corrects individual candidate solutions based on statistical characteristics, improving their initial quality and feasibility. Furthermore, it comprehensively evaluates and ranks candidate solutions by combining constraint violation calculation and multi-objective fitness assessment. This process selects high-quality individuals that meet preset conditions to construct the initial population, improving the overall quality and optimization efficiency of the population.

[0042] In some embodiments, the method for iteratively optimizing the initial population based on the improved parrot optimization algorithm is as follows: Select the individual with the best fitness from the initial population as the global best individual; The initial population is divided into neighborhoods according to a preset ring topology, and the optimal individual in the neighborhood is determined within the corresponding neighborhood of each individual. For each individual to be updated, a random number is generated and compared with a preset following probability threshold. If the random number is less than the following probability threshold, the individual to be updated follows the global optimal individual; otherwise, it follows the neighborhood optimal individual. Based on the current position of the individual to be updated, the position of its following objects, and the preset step size factor, the position of the individual to be updated is updated, and a random perturbation term is introduced during the update process; Boundary constraints are applied to the updated individual positions to ensure that the values ​​of each decision variable are within the preset domain. Repeat the above steps until the preset termination condition is met, and output the optimization result. For example, when solving the multi-objective optimization scheduling problem of flexible resources in low-carbon parks, the comprehensive fitness value of each candidate solution individual is first calculated from the initial population. The individual with the best fitness usually corresponds to the scheduling scheme with the lowest operating cost and lowest carbon emissions, and is used as the global optimal individual to guide the search direction of the entire population. Subsequently, the population is divided into neighborhoods according to a preset ring topology structure. For example, 100 individuals are divided into 10 neighborhood rings in sequence, with each neighborhood containing 10 adjacent individuals. The individual with the best fitness within each neighborhood is selected as the neighborhood optimal individual, thereby establishing a two-layer information transmission mechanism between global guidance and local exploration. In the individual update stage, for any individual to be updated, a random number r in the interval [0,1] is generated and compared with a preset following probability threshold (e.g., 0.7): when r is less than the threshold, the individual prioritizes moving towards the global optimal individual; otherwise, it learns from the optimal individual in its neighborhood, thereby realizing the adaptive switching between "global exploration and local development". During the position update process, the current individual position X(t), the target position Xbest, and the step size factor are used. Based on this, an update formula is constructed, and a random perturbation term following a normal or uniform distribution is introduced. This is used to enhance population diversity and avoid premature convergence. For example, the energy storage power variable may be adjusted from -1.2MW to -0.8MW after the update, and a small random shift may occur under the influence of perturbation, thereby enhancing the search's ability to escape local optima. In the boundary processing stage, the updated individuals are subjected to dimension-by-dimensional constraint corrections. For example, energy storage power exceeding [-2.5MW, 2.5MW] is truncated to the boundary value, and the load regulation rate is limited to the [0,1] interval, thereby ensuring that all candidate solutions meet the physical and operational feasibility constraints. Finally, by continuously repeating the cyclic process of "fitness evaluation - neighborhood update - location iteration - constraint correction" until the preset termination condition (such as the maximum number of iterations or fitness convergence threshold) is met, the final optimization result is output, thereby obtaining the optimal scheduling scheme that meets the goals of economy, low carbon emissions, and new energy consumption.

[0043] The iterative optimization method described in this application selects the individual with the best fitness as the global guide and constructs a neighborhood optimum mechanism based on a ring topology to achieve the coordinated use of global and local information. By setting a following probability threshold, the individual adaptively selects the following object between the global optimum and the neighborhood optimum, and combines the step size factor and random perturbation for position updates, enhancing the flexibility and diversity of the search process. On this basis, boundary constraints are used to ensure the feasibility of the solution, and the population distribution is continuously optimized during the iteration process, thereby improving the convergence speed and optimization accuracy of the algorithm and achieving stable and efficient optimization solutions.

[0044] In some embodiments, compared with the existing parrot optimization algorithm, the improved parrot optimization algorithm introduces a dual population construction method combining a chaotic mapping generation mechanism and a reverse learning strategy in the algorithm initialization stage, and uses historical optimal solution statistics to guide and correct the initial candidate solutions, thereby effectively improving the uniformity of the initial population distribution and the coverage of high-quality solutions in the solution space. In the iterative optimization stage, by constructing a dual following mechanism based on the globally optimal individual and the neighborhood optimal individual, individuals can adaptively select the search guidance direction according to random probability, and the search trajectory is dynamically adjusted by combining the step size factor and random perturbation term, thereby enhancing the algorithm's ability to escape local optima in complex non-convex solution spaces. At the same time, by introducing a feedback mechanism based on population dispersion index and optimization progress index, the exploration intensity parameter and randomness parameter are dynamically and adaptively adjusted, enabling the algorithm to automatically balance according to the convergence state in the search stage. The algorithm enhances convergence stability and optimization efficiency by combining global exploration and local development capabilities. Furthermore, a population hierarchical mechanism based on individual performance indicators divides the population into high-performance, medium-performance, and low-performance subgroups. Differentiated update strategies—such as refined development, standard follow-up updates, and enhanced exploration or re-initialization—are employed for each subgroup to achieve collaborative evolution among individuals of varying quality, thereby improving the overall population's evolutionary efficiency. Additionally, by constructing an elite individual set and introducing an elite-guided mutation mechanism, high-performance individuals are not only used to retain high-quality solutions but also to guide and correct the search direction of non-elite individuals, effectively preventing premature convergence and enhancing population diversity. Finally, a multi-condition joint termination mechanism enables the algorithm to adaptively terminate when it meets constraints on the number of iterations, convergence stagnation, or computation time, thereby improving its applicability and real-time performance in engineering scheduling scenarios.

[0045] In some embodiments, the population dispersion index is calculated as follows: The position of the population centroid is obtained by averaging the position vectors of each individual in the current population along their dimensions. Based on the location of the population centroid, calculate the Euclidean distance between the position vector of each individual and the location of the population centroid. The population dispersion index is obtained by averaging the Euclidean distances of each individual.

[0046] For example, at a certain iteration number t, the population size is N=5, and each candidate solution individual consists of three decision variables, representing energy storage power, controllable load regulation rate, and tie-line switching power, respectively. Assume that the position vectors of the current 5 individuals are: X1= (1.0, 0.2, -2.0), X2= (0.5, 0.4, -1.5), X3= (1.5, 0.3, -2.5), X4=(0.8, 0.6, -1.0), X5= (1.2, 0.5, -2.2).

[0047] First, the mean of each dimension is calculated to obtain the position Y of the population centroid: Y=((1.0+0.5+1.5+0.8+1.2) / 5, (0.2+0.4+0.3+0.6+0.5) / 5, (-2.0-1.5-2.5-1.0-2.2) / 5); That is, Y = (1.0, 0.4, -1.84).

[0048] Then, the Euclidean distance between each individual and the population centroid is calculated, for example, the distance between an individual and the population centroid. The distance d1 between the center of gravity and the centroid is: .

[0049] Similarly, calculations can be performed separately. , , , .

[0050] After obtaining the distances of all individuals to the center of gravity, for , , , , The average value is then used to obtain an index of the current population dispersion.

[0051] The method for calculating the population dispersion index described in this application determines the location of the population centroid by calculating the mean of the position vectors of each individual in the population, thus achieving a centralized representation of the overall distribution state; it characterizes the relative distribution differences of individuals by calculating the Euclidean distance between each individual and the population centroid; and on this basis, it obtains the dispersion index by averaging each distance, thereby quantitatively reflecting the dispersion level of the population and providing a basis for subsequent dynamic adjustment of search strategies.

[0052] In some embodiments, the optimization process metric is calculated as follows: During the iterative optimization process, the global optimal fitness value for each generation is obtained and stored; At the current iteration, extract the historical global optimal fitness values ​​within the current generation and the preset number of generations prior from the stored global optimal fitness values; Using the current generation's global optimal fitness value as a benchmark, calculate the fitness difference between it and the historical global optimal fitness value; The absolute values ​​of each fitness difference are processed, and the absolute value results are averaged to obtain the optimization process index.

[0053] For example, in the iterative solution of the multi-objective optimization scheduling problem in low-carbon industrial park A, assume that the global optimal fitness values ​​for the most recent 5 generations (t-4 to t) have been recorded as: 120.5, 118.2, 117.0, 116.8, and 116.7. A smaller fitness value indicates lower overall operating costs and better optimization performance. First, using the current generation t's global optimal fitness value of 116.7 as a benchmark, calculate the differences between it and the historical optimal fitness values: -3.8, -1.5, -0.3, -0.1, and 0. Then, take the absolute value of these differences and average the absolute values ​​to obtain the optimization progress index P: P = (3.8 + 1.5 + 0.3 + 0.1 + 0) / 5 = 1.14. In practical applications, a large optimization progress index P indicates a significant improvement in the recent optimal solutions, suggesting the algorithm is still in a rapid convergence or exploration phase. When the index gradually decreases and approaches 0, it indicates that the global optimal fitness value has stabilized during continuous iterations, and the algorithm has entered the convergence phase. Based on this index, the exploration intensity and random perturbation parameters in the improved parrot optimization algorithm can be dynamically adjusted to enhance overall optimization efficiency and convergence stability.

[0054] The optimization process index calculation method described in this application continuously tracks historical optimization information by recording the global optimal fitness value of each generation during the iteration process; it extracts the historical optimal fitness of the current generation and several generations prior, and calculates the difference based on the current optimal value to characterize the improvement range during the optimization process; based on this, it performs absolute value and average processing on the difference to obtain the optimization process index, thereby quantitatively reflecting the convergence trend and optimization progress of the algorithm, and providing a basis for dynamic parameter adjustment.

[0055] In some embodiments, the method for dynamically adjusting the exploration intensity parameter and the randomness parameter is as follows: Set the maximum and minimum values ​​of the exploration intensity parameter and the randomness parameter, as well as the maximum number of iterations, and establish a decreasing relationship between the exploration intensity parameter, the randomness parameter, and the number of iterations; In the current iteration, calculate the basic parameter values ​​of the exploration intensity parameter and the randomness parameter based on the decreasing relationship; Obtain the dispersion index and optimization progress index of the current population; Determine whether the dispersion index of the current population is lower than the dispersion judgment threshold. If not, keep the randomness parameter unchanged or update it according to the preset decreasing relationship; if so, increase the randomness parameter. Determine whether the current population's optimization progress indicators meet the stagnation condition. If not, update the exploration intensity parameter and randomness parameter according to the preset decreasing relationship; if they meet, increase the exploration intensity parameter and randomness parameter.

[0056] For example, in the t-th iteration of the multi-objective optimization scheduling of a low-carbon industrial park A, the range of the exploration intensity parameter μ is set to [0.1, 1.0], the range of the randomness parameter β is set to [0.05, 0.5], the maximum number of iterations is 200, and a linear decay relationship decreasing with the number of iterations is adopted as the basic update strategy. In the early stage of iteration (e.g., t=10), μ=0.85 and β=0.42 are calculated according to the decay relationship. At this time, the algorithm maintains strong exploration ability and high random perturbation to expand the search range. In the middle stage (e.g., t=100), the basic parameters decay to μ=0.45 and β=0.22. At this time, the algorithm gradually transitions from global exploration to local development. Furthermore, in the current iteration stage, the population dispersion index D=0.35 is obtained, and the dispersion judgment threshold is set to 0.3. Since D>0.3, it indicates that the population still has high dispersion. Therefore, the randomness parameter β maintains its original decreasing value or is updated according to the predetermined decay rule without additional amplification adjustment. Simultaneously, the optimization progress index P=0.05 is obtained, and the stagnation threshold is set to 0.02. Since P>0.02, it indicates that the optimization is still in the continuous improvement stage. Therefore, the exploration intensity parameter μ and the randomness parameter β are updated normally according to the preset decreasing relationship without triggering enhancement adjustment. When the population dispersion significantly decreases (e.g., D=0.15) and the optimization progress index enters a stagnant state (e.g., P≤0.02) in the later stage of iteration (e.g., t=180), it is determined that the algorithm may be trapped in a local convergence region. At this time, the randomness parameter β is increased (e.g., from 0.08 to 0.15), and the exploration intensity parameter α is adjusted to increase in stages (e.g., from 0.12 to 0.25) to enhance the search perturbation ability and escape the local optimum region.

[0057] The method for dynamically adjusting the parameters described in this application achieves basic adaptive updates of the parameters by setting the value ranges of the exploration intensity parameter and the randomness parameter and constructing a relationship of change with decreasing iteration count. It also introduces population dispersion index and optimization progress index to evaluate the current search state in real time. Furthermore, when the population dispersion decreases or the optimization process stagnates, the randomness parameter and exploration intensity parameter are increased to enhance search perturbation and the ability to escape local optima. This achieves dynamic adaptive control of the parameters, improving the coordination and optimization efficiency between global search and local development.

[0058] In some embodiments, the method for dividing the initial population into high-performance, medium-performance, and low-performance subpopulations based on individual performance indicators is as follows: The individuals in the current population are sorted based on their fitness values ​​to obtain the corresponding fitness ranking. Based on fitness value, fitness ranking, and fitness change within a preset iteration interval, the performance indicators of each individual are determined. Based on performance indicators, the population is divided into different subgroups: individuals that meet the first preset condition are classified as high-performance subgroups, individuals that meet the second preset condition are classified as medium-performance subgroups, and individuals that meet the third preset condition are classified as low-performance subgroups.

[0059] For example, in a certain iteration t, the population size is N=10. First, all individuals are ranked according to their comprehensive fitness value (e.g., a weighted result considering operating costs, carbon emissions, and renewable energy absorption rates), resulting in a fitness ranking sequence from best to worst, where a higher ranking indicates a better solution quality. Based on this, a comprehensive evaluation is further performed by combining the fitness value itself with its change within the most recent preset iteration interval (e.g., the last 5 generations). For example, if an individual currently has a high fitness ranking but its fitness has only increased slightly in the last 5 generations, it indicates limited improvement potential; conversely, if some individuals with medium rankings show a continuous and rapid increase in fitness in recent iterations, it indicates a strong evolutionary trend and search potential. Based on the above fitness value, ranking information, and change rate, a unified performance index is constructed, such as a weighted fusion of the ranking normalized score, fitness normalized value, and change rate to obtain the performance index value for each individual. In the classification phase, the first preset condition is set as "performance indicators are in the top 30% and fitness change rate is positive and relatively high". Individuals that meet this condition are classified as high-performance subgroups, for example, for fine development and elite guidance. The second preset condition is set as "performance indicators are in the middle range (e.g., 30% to 70%)", which are classified as medium-performance subgroups for regular follow-up updates and stable evolution. The third preset condition is set as "performance indicators are in the bottom 30% or fitness is continuously declining", which are classified as low-performance subgroups for enhanced perturbation search or re-initialization to improve population diversity.

[0060] The population classification method described in this application obtains the relative superiority or inferiority of individuals by ranking their fitness values. By integrating fitness values, fitness rankings, and their changes within a preset iteration interval, a comprehensive performance index is constructed to achieve dynamic evaluation of individual development potential. Based on this, the population is divided into high-performance, medium-performance, and low-performance subgroups according to the performance index, thereby achieving hierarchical management of the population structure and providing a basis for subsequent differentiated optimization strategies.

[0061] In some embodiments, the method of using a differentiated strategy for position updating for different subgroups is as follows: For individuals in the high-performance subgroup, a position update strategy combining elite preservation and small step-size perturbation is adopted, with a small step-size factor and a low random perturbation intensity set. For individuals in the medium-performance subgroup, a standard position update strategy is adopted, and a preset step size factor and random perturbation intensity are set so that each individual updates its position based on the globally optimal individual or the neighborhood optimal individual. For individuals in the low-performance subgroup, an enhanced exploration position update strategy is adopted, with a larger step size factor or a higher random perturbation intensity set, while re-initialization is performed on individuals that meet the preset conditions. Based on the corresponding step size factor and random perturbation intensity, position updates are performed on individuals in the high-performance subgroup, medium-performance subgroup, and low-performance subgroup respectively to obtain the updated individual positions.

[0062] For example, in iteration t, assume the current population has been divided into three categories: high-performance subgroup, medium-performance subgroup, and low-performance subgroup. Individuals in the high-performance subgroup typically correspond to schemes close to the optimal scheduling solution, such as those with small energy storage power fluctuations and relatively stable tie-line power. For individuals in the high-performance subgroup, taking a certain energy storage scheduling solution (0.8MW, 0.4, -1.2MW) as an example, its update process employs a strategy combining elite preservation and small-step perturbation. For instance, centered on the global optimal solution, only a small perturbation (such as a step size factor of 0.05 and a low random perturbation strength ε) is applied, causing its update X′ to be fine-tuned near the original solution, thus avoiding damage to existing high-quality structures. For individuals in the medium-performance subgroup, a standard follow-up update strategy is adopted. For example, if an individual's current solution is (0.3MW, 0.6, -0.8MW), it chooses to follow the global optimal solution or the neighborhood optimal solution based on random probability, and updates using a medium step size factor (such as 0.2) and moderate random perturbation, thereby maintaining stability while continuing to approach the high-quality region. For individuals in low-performance subgroups, such as solutions with poor fitness and numerous constraint violations, an enhanced exploration strategy is adopted. In this case, the step size factor is significantly increased (e.g., 0.6–0.9), the intensity of random perturbations is increased, and in some cases, individuals are reinitialized, for example, by remapping variables such as energy storage power and load regulation rate to random locations within the feasible region. This enhances population diversity and avoids stagnation in inferior solution regions. Ultimately, by performing position update operations of varying intensities on the three types of subgroups, the high-performance subgroup focuses on "refined development and solution stability maintenance," the medium-performance subgroup focuses on "balanced search and convergence advancement," and the low-performance subgroup focuses on "global exploration and solution space expansion." This achieves collaborative evolution within the population and significantly improves overall optimization efficiency and global optimization capabilities.

[0063] The differentiated position update method described in this application achieves stable preservation of high-quality solutions and refined local search by employing a strategy combining elite preservation and small-step perturbation for high-performance subgroups; it balances search efficiency and convergence performance by adopting a standard update strategy based on global or neighborhood optima for medium-performance subgroups; and it further enhances the exploration strategy of large-step and high-random perturbation for low-performance subgroups by combining it with a re-initialization mechanism, thereby improving their ability to escape local optima. This achieves collaborative optimization of different subgroups and improves overall search efficiency and global optimization capability.

[0064] In some embodiments, the method for dynamically adjusting the grading results based on the optimization effect of each update strategy is as follows: After each iteration cycle, the fitness value of each individual is calculated based on the updated individual position, and the performance index of each individual is updated accordingly. Based on the updated performance metrics, the individuals in the population are reordered to obtain the updated fitness ranking. Based on the updated fitness ranking, the population is reclassified and divided into updated high-performance subpopulation, medium-performance subpopulation and low-performance subpopulation. Based on the results of the reclassification, the subgroups to which each individual belongs are adjusted to achieve the redistribution of individuals among different subgroups; Within a preset iteration interval, the fitness changes of each subgroup are statistically analyzed, and the optimization effect of the corresponding differentiated position update strategy for each subgroup is determined accordingly. When the optimization effect is lower than the policy adjustment threshold, the policy parameters of the corresponding subgroup are adjusted to update the policy configuration in subsequent iterations. Output the updated population classification results and corresponding policy parameters for use in the population optimization process of the next iteration.

[0065] For example, after the t-th iteration of the scheduling optimization in Low-Carbon Industrial Park A, the fitness value is first recalculated based on the updated individual positions. For instance, the fitness of an individual whose comprehensive operating cost decreases from 12,000 yuan to 11,500 yuan is improved, and its performance index is updated accordingly. Subsequently, all individuals are reordered according to the new performance index, causing some individuals that were originally in the medium-performance range to rise to the high-performance range due to rapid convergence, while some individuals whose performance has declined are downgraded to the low-performance range. In the reclassification stage, for example, some individuals in the original high-performance subgroup are reclassified to the medium-performance subgroup due to a significant decrease in fitness after the update; at the same time, some individuals in the medium-performance subgroup that show a continuous optimization trend are promoted to the high-performance subgroup, thereby realizing the dynamic reconstruction of the subgroup structure. In the strategy effect statistics stage, the changes in fitness of each subgroup are statistically analyzed using a preset iteration interval (such as the last 5 generations) as a window. For example, the average fitness of the high-performance subgroup only improves by 0.5%, while that of the medium-performance subgroup improves by 3.2% and that of the low-performance subgroup by 1.8%. This indicates that the small-step fine-update strategy used by the current high-performance subgroup has limited effectiveness, while the standard strategy for the medium-performance subgroup has better overall optimization efficiency. When the optimization effect of a certain subgroup is detected to be lower than the policy adjustment threshold (e.g., the improvement is less than 1%), the policy parameters corresponding to that subgroup are dynamically adjusted. For example, the perturbation intensity is appropriately increased for the high-performance subgroup to enhance its local escape capability, or the re-initialization ratio is increased for the low-performance subgroup to enhance its global exploration capability. Finally, after each iteration cycle, the updated population classification results and corresponding policy parameter configurations are output, enabling the next iteration cycle to continue optimization based on the latest population structure and policy parameters. This forms a "performance feedback-driven dynamic hierarchical optimization closed-loop mechanism," effectively improving the algorithm's adaptability and convergence performance in complex multi-objective scheduling problems.

[0066] The hierarchical dynamic adjustment method described in this application achieves dynamic evaluation of individual states by real-time correction of performance indicators based on the updated individual fitness during the iteration process; it enables adaptive flow of individuals among high-performance, medium-performance, and low-performance subgroups by reordering and hierarchically dividing the population; based on this, it evaluates the effectiveness of the differentiated strategy by statistically analyzing the optimization effect of each subgroup within a preset iteration interval, and adjusts the strategy parameters when the effect is insufficient; thereby achieving coordinated adaptive optimization of population hierarchization and update strategy, improving overall optimization efficiency and algorithm stability.

[0067] In some embodiments, the preset termination condition includes at least one or a combination of the following: Iteration termination condition: The number of iterations reaches the preset maximum number of iterations, wherein the maximum number of iterations is set according to the size of the optimization problem and the computation time constraints; Convergence stagnation termination condition: The global optimal fitness value has not improved within a preset number of consecutive generations, and the fitness change in the most recent generations is less than the preset convergence accuracy threshold, indicating that the algorithm has entered a stable convergence state. Calculation time termination condition: The optimized calculation time reaches the preset time limit to meet the time window constraints of the actual scheduling calculation; Combined termination conditions: simultaneously satisfying the minimum number of iterations and the fact that the global optimal fitness value has not significantly improved within several consecutive generations, in order to avoid premature termination or misjudgment of convergence due to accidental fluctuations; Trend determination termination condition: Based on the trend of the change of the global optimal fitness value in the most recent several generations of the sliding window statistics, when the average change is less than the preset convergence accuracy threshold, the convergence termination condition is determined to be met. When any of the preset termination conditions are met, the iteration process stops, and the current global best individual is output as the final optimized solution.

[0068] The termination condition setting method described in this application constrains the optimization process by setting a maximum number of iterations and an upper limit on computation time, ensuring that the algorithm runs within a controllable range of computational resources. By introducing a convergence stagnation judgment and trend judgment mechanism, the change of the global optimal fitness is dynamically monitored to accurately identify the convergence state of the algorithm. On this basis, the minimum number of iterations and multiple condition combination judgments are combined to avoid premature termination or misjudgment of convergence. Thus, while ensuring optimization accuracy, computational efficiency is improved, and the current global optimal solution is output as the final result when any termination condition is met.

[0069] In some embodiments, the method for optimal decoding into standardized scheduling instructions for each flexible resource in each scheduling period is as follows: The optimal solution is analyzed according to the preset decision variable coding rules to obtain the power variables and operating status variables of each flexible resource in each scheduling period. Based on power variables and operating status variables, scheduling information for each flexible resource is generated. Among them, the energy storage system corresponds to the charging and discharging power of each time period, the controllable load equipment corresponds to the adjustable power or transfer power of each time period, the photovoltaic power generation system and wind turbine generator set correspond to the scheduling output value of each time period, and the start-stop status of each equipment corresponds to the 0-1 operating status command. The scheduling information of each flexible resource is organized according to the scheduling time period sequence to generate a power sequence and a state sequence with the scheduling time period as the unit; The power sequence and state sequence are subjected to a feasibility check to determine whether they meet the physical constraints of the equipment. The physical constraints include at least the power ramp-up constraint, the minimum start-up time constraint, and the minimum downtime constraint. When a constraint violation is detected, the power sequence is smoothed or the state sequence is time-adjusted to meet the constraint conditions. After passing the feasibility verification, standardized scheduling instructions are generated. The standardized scheduling instruction file includes at least equipment identification information, scheduling time period information, power setting value, and operating status instructions.

[0070] For example, after optimizing 24 scheduling periods per day in Low-Carbon Industrial Park A, a set of optimal solution vectors is obtained. Each dimension corresponds to the energy storage system, controllable load, photovoltaic output, wind power output, and the operating status variables of each device. After parsing according to the preset coding rules, the following structured information can be obtained: In the 10th scheduling period, the power variable corresponding to the energy storage system is -1.2MW (indicating charging), the controllable load adjustment power is 0.5MW, the photovoltaic output is 3.8MW, the wind power output is 2.1MW, and the energy storage operating status is 1 (operating), while the status of some controllable load devices is 0 (interruptible or shut down). Subsequently, the above resource information is reorganized according to the scheduling period to form a power sequence and a status sequence in time series. For example, the energy storage system forms a power sequence consisting of charging and discharging power over 24 scheduling periods, where the charging and discharging power corresponding to the first 3 scheduling periods can be represented as "[-0.8MW, -1.2MW, 0.5MW]"; the controllable load forms a regulation sequence consisting of the regulation power of each scheduling period, where the regulation power corresponding to the first 3 scheduling periods can be represented as "[0.3MW, 0.5MW, 0.4MW]"; photovoltaic and wind power each form their corresponding time-period output sequences, and the equipment start-up and shutdown states form a 0-1 state matrix to describe the operation of each device in each time period. During the feasibility verification phase, for example, if it is found that the power change of the energy storage system exceeds the maximum ramp constraint between the 12th and 13th time periods (e.g., exceeding ±1.5MW / 15min), the power sequence is smoothed to bring the change back to the allowable range; if it is found that a gas turbine unit violates the minimum start-up time constraint (e.g., it is shut down before continuous operation for 3 time periods), the state sequence is time-adjusted to meet the continuous operation requirements. After completing the above corrections, a standardized dispatch instruction file is generated. For example, in time period 10, the following instruction structure is formed: the equipment is identified as "Energy Storage System No. 1", the dispatch power is set to -1.2MW, and the operating status is "Charging Operation"; the controllable load equipment is identified as "Air Conditioning System A", the adjustment power is 0.5MW, and the status is "Adjustable Operation"; the photovoltaic and wind power equipment correspond to the actual available output set values. Finally, the standardized dispatch instructions for all time periods are summarized in chronological order to form a standardized dispatch instruction set, which can be directly used in the park's energy management system (EMS) or dispatch execution layer, realizing a seamless conversion from optimization results to engineering control instructions, thereby ensuring the actual executability of the optimization model and the ability to be implemented in engineering projects.

[0071] The optimal solution decoding method described in this application obtains the power variables and operating status variables of each flexible resource in each scheduling period by encoding and parsing the optimal solution, thereby realizing a structured expression of the scheduling results. By constructing scheduling information covering energy storage charging and discharging power, controllable load adjustment power, and new energy output, and combining it with equipment start-up and shutdown status, a time-series power sequence and a status sequence are generated. On this basis, the feasibility of the power and status sequences is verified and constraints are corrected to ensure that the scheduling results meet the physical operating constraints of the equipment. Thus, standardized scheduling instructions are generated, realizing the effective transformation of optimization results into actual control execution.

[0072] To verify the effectiveness of the flexible resource optimization scheduling method for low-carbon industrial parks based on the improved parrot optimization algorithm disclosed in this application under actual working conditions, the following uses a comprehensive energy demonstration park as a case study.

[0073] A comprehensive energy demonstration park, comprising photovoltaic power generation systems, wind turbine generators, energy storage systems, controllable loads, and electric vehicle charging facilities, experienced high peak-hour electricity purchase costs, insufficient renewable energy absorption, and significant load fluctuations during the high-temperature, high-load operation period of the summer of 2025. Traditional scheduling methods based on empirical rules or single-objective optimization struggled to balance economic efficiency with low-carbon requirements. To improve the overall operational level of the park, a flexible resource optimization scheduling platform was constructed based on the technical solution of this application. This platform performs unified modeling and intelligent scheduling decisions on multi-source data, achieving coordinated optimization control of the park's energy system.

[0074] First, a unified model was created for the park's flexible resource allocation information, obtaining data showing an energy storage system capacity of 20MWh, a charge / discharge efficiency of 92%, a controllable load adjustment range of 15% of the total load, an average daily electric vehicle connection of approximately 300 vehicles, a photovoltaic installed capacity of 15MW, a wind power installed capacity of 10MW, and a minimum technical output of 30% for the combined heat and power (CHP) units. Based on this, historical load data, meteorological data, and electricity price data for the park were collected, and a multi-source dataset was constructed to provide foundational support for subsequent optimization.

[0075] In the forecasting phase, a long short-term memory neural network load forecasting model was constructed based on historical operating data, and photovoltaic and wind power output forecasting models were established in conjunction with meteorological forecast data. The forecast results show that the peak photovoltaic output at noon the following day can reach 12.6MW, while the evening peak load will reach 28MW, indicating a significant peak-to-valley difference and time mismatch problem.

[0076] In the optimization modeling process, a multi-objective optimization model incorporating economic costs, carbon emissions, and renewable energy absorption rates is constructed, and a unified solution is obtained by combining power balance constraints, energy storage state of charge constraints, and equipment operation logic constraints. The chaotic mapping and back-learning strategy described in this application is used to generate an initial population, and iterative optimization is performed based on the improved parrot optimization algorithm. Efficient optimization is achieved through population hierarchical management and dynamic parameter adjustment mechanisms.

[0077] Optimization results show that, within a typical dispatch cycle, the energy storage system charges during peak photovoltaic output periods and discharges during evening peak periods, with controllable load shifting to off-peak periods, and electric vehicle charging load achieving time-sharing dispatch. Compared to traditional methods, the park's externally purchased electricity decreased by approximately 18%, the curtailment rate decreased from 12.4% to 3.1%, total carbon emissions decreased by approximately 14.7%, and the fluctuation range of power interaction between the park and the grid decreased by approximately 21%.

[0078] Based on this, the optimization results are decoded and standardized scheduling instructions are generated, which are then issued to various types of equipment for execution. Actual operation results show that all flexible resources respond promptly, scheduling instructions are executed accurately, and no violations of equipment constraints occur, verifying the feasibility of the scheduling scheme.

[0079] During 30 consecutive days of operation tracking, the park's overall operating costs decreased by an average of approximately 11.3%, energy storage utilization increased to over 85%, and the local consumption level of renewable energy significantly improved. Through feedback analysis of the optimization process data, the algorithm parameters were further adaptively adjusted and the model updated, making subsequent scheduling results more stable and reliable.

[0080] In summary, the technical solution presented in this application achieves efficient collaborative scheduling of various types of flexible resources in low-carbon industrial parks by integrating multi-source prediction data with a multi-objective optimization model and combining it with an adaptive search mechanism of the improved parrot optimization algorithm. Compared with existing technologies, it demonstrates significant advantages in reducing operating costs, increasing renewable energy absorption rates, and suppressing load fluctuations, and has good engineering application value and prospects for widespread application.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A flexible resource optimization scheduling method for low-carbon industrial parks based on an improved parrot optimization algorithm, characterized in that: Includes the following steps: Based on the flexible resource allocation information and operational constraints of low-carbon industrial parks, and combined with the load forecast data, renewable energy output forecast data and electricity price information of the current scheduling cycle, a multi-objective optimization model for flexible resource collaborative scheduling is constructed. A fitness evaluation function is defined based on a multi-objective optimization model, and an initial candidate solution is generated using a chaotic mapping method. Based on the reverse learning strategy, a corresponding reverse candidate solution individual is constructed for each initial candidate solution individual. At the same time, the initial candidate solution individual and the reverse candidate solution individual constitute a candidate solution individual set, and an initial population is constructed based on the candidate solution individual set. The initial population was iteratively optimized based on the improved parrot optimization algorithm; During the iterative optimization process, each candidate solution is updated in position by following the current global best solution or a good neighboring solution, and the search space is expanded under the action of random perturbation. During the iterative optimization process, the exploration intensity parameter and the randomness parameter are dynamically adjusted based on the population dispersion index and the optimization progress index to achieve dynamic coordination between global search and local development capabilities. During the iterative optimization process, the initial population is divided into high-performance subgroups, medium-performance subgroups, and low-performance subgroups based on individual performance indicators. Differentiated strategies are used to update the positions of different subgroups, and the hierarchical results are dynamically adjusted according to the optimization effect of each update strategy. When the preset triggering conditions are met, candidate individuals that meet the preset elite determination conditions are selected from the high-performance subgroup based on the fitness evaluation results, and they are constructed into an elite individual set. Based on the elite individual set, guided mutation operations are performed on non-elite individuals to enhance local search capabilities. If the iterative optimization does not reach the preset termination condition, the iterative optimization will continue until the termination condition is reached. If the iterative optimization has reached the preset termination condition, the optimal solution is output, and the optimal solution is decoded into standardized scheduling instructions for each flexible resource in each scheduling period. The standardized scheduling instructions are then sent to the corresponding flexible resource equipment to achieve collaborative optimization scheduling of the low-carbon park. The method for generating initial candidate solutions is as follows: Control parameters and initial conditions are set based on a chaotic mapping model to generate an initial chaotic sequence; Iterative calculations are performed on the chaotic mapping model to obtain a sequence of chaotic variables whose values ​​are located within a preset interval; The chaotic variable sequence is mapped to the range of values ​​of each decision variable in the optimization problem to obtain the corresponding initial values ​​of the decision variables; For different types of decision variables, initial values ​​of the decision variables are constrained to obtain the variable values. The values ​​of each decision variable are combined to obtain the initial candidate solutions. The method for iteratively optimizing the initial population based on the improved parrot optimization algorithm is as follows: Select the individual with the best fitness from the initial population as the global best individual; The initial population is divided into neighborhoods according to a preset ring topology, and the optimal individual in the neighborhood is determined within the corresponding neighborhood of each individual. For each individual to be updated, a random number is generated and compared with a preset following probability threshold. If the random number is less than the following probability threshold, the individual to be updated follows the global optimal individual; otherwise, it follows the neighborhood optimal individual. Based on the current position of the individual to be updated, the position of its following objects, and the preset step size factor, the position of the individual to be updated is updated, and a random perturbation term is introduced during the update process. Boundary constraints are applied to the updated individual positions to ensure that the values ​​of each decision variable are within the preset domain. Repeat the above steps until the preset termination condition is met, and output the optimized result.

2. The method for flexible resource optimization scheduling in low-carbon industrial parks according to claim 1, characterized in that, The multi-objective optimization model includes: An economic objective function is used to minimize the total operating cost of the park during the scheduling cycle. The total operating cost includes the cost of purchased electricity, gas, equipment depreciation, and operation and maintenance, and deducts the revenue from interruptible load compensation and surplus electricity fed into the grid. An environmental objective function is used to minimize the total carbon emissions of the park, which includes indirect carbon emissions from the grid side corresponding to purchased electricity and direct carbon emissions from gas consumption within the park. The renewable energy consumption rate objective function is used to maximize the utilization ratio of renewable energy and improve the local consumption level of clean energy by minimizing the amount of wind and solar power curtailment. The load fluctuation mitigation objective function is used to minimize the inter-period fluctuation amplitude of the power interaction between the park and the power grid, so as to reduce the impact on the external power grid. The weighted coefficient method is used to linearly combine multiple objective functions into a single objective optimization problem. The weight coefficients are set by the park's operators based on actual preferences and policy guidance. Alternatively, the constraint method can be used to transform some objectives into constraints.

3. The method for flexible resource optimization scheduling in low-carbon industrial parks according to claim 1, characterized in that, The method for constructing a corresponding reverse candidate solution for each initial candidate solution is as follows: For each decision variable in the initial candidate solution, based on its value range and variable type, construct the inverse value symmetric about the center point or boundary within the domain of the corresponding variable; Based on the combined values ​​of the inverse variables corresponding to each decision variable, an inverse candidate solution individual corresponding to each initial candidate solution individual is constructed.

4. The flexible resource optimization scheduling method for low-carbon industrial parks according to claim 3, characterized in that, The initial population is constructed as follows: Statistical analysis of historical optimal solutions yields the statistical characteristics of each decision variable; Based on statistical characteristics, the candidate solutions in the candidate solution individual set are guided and corrected. Constraint violation degree is calculated for individual candidate solutions after guided correction, and fitness is evaluated based on a multi-objective optimization model; The candidate solutions are ranked according to the fitness assessment results, and the candidate solutions whose fitness meets the preset screening conditions are selected according to the preset population size to construct the initial population.

5. The method for flexible resource optimization scheduling in low-carbon industrial parks according to claim 1, characterized in that, The method for converting the optimal decoding into standardized scheduling instructions for each flexible resource in each scheduling period is as follows: The optimal solution is analyzed according to the preset decision variable coding rules to obtain the power variables and operating status variables of each flexible resource in each scheduling period. Based on power variables and operating status variables, scheduling information for each flexible resource is generated. Among them, the energy storage system corresponds to the charging and discharging power of each time period, the controllable load equipment corresponds to the adjustable power or transfer power of each time period, the photovoltaic power generation system and wind turbine generator set correspond to the scheduling output value of each time period, and the start-stop status of each equipment corresponds to the 0-1 operating status command. The scheduling information of each flexible resource is organized according to the scheduling time period sequence to generate a power sequence and a state sequence with the scheduling time period as the unit; The power sequence and state sequence are subjected to a feasibility check to determine whether they meet the physical constraints of the equipment. The physical constraints include at least the power ramp-up constraint, the minimum start-up time constraint, and the minimum downtime constraint. When a constraint violation is detected, the power sequence is smoothed or the state sequence is time-adjusted to meet the constraint conditions. After passing the feasibility verification, standardized scheduling instructions are generated. The standardized scheduling instruction file includes at least equipment identification information, scheduling time period information, power setting value, and operating status instructions.

Images