Intelligent Linkage Method and System for Distributed Equipment Groups in Industrial Parks Based on Load Characteristics

Abstract

This invention provides an intelligent linkage method and system for distributed energy equipment groups in a park based on load characteristics, relating to the field of energy dispatching technology. The method includes acquiring real-time load data from multiple energy-consuming units within the park and inputting it into a deep neural network model to generate load characteristic curves. Then, based on the load characteristic curves, a multi-objective optimization model is constructed, and an improved particle swarm optimization algorithm is used to solve for the optimal operating scheme. Finally, each distributed energy device is controlled according to the optimal operating scheme, and a dynamic optimization correction mechanism is triggered in case of anomalies or sudden load changes. This invention, through load characteristic analysis and intelligent optimization control, can minimize system operating costs, maximize energy utilization efficiency, and achieve coordinated operation and dynamic load balance of the distributed energy equipment group in the park.

Description

Technical Field

[0001] This invention relates to energy dispatching technology, and more particularly to a method and system for intelligent linkage of distributed equipment groups in a park based on load characteristics. Background Technology

[0002] Industrial park energy management is a crucial means to improve energy efficiency and reduce operating costs. With the rapid development of distributed energy technologies, more and more industrial parks are adopting distributed energy devices such as photovoltaics, energy storage, and electric cooling to meet their energy needs. How to effectively coordinate and control these devices to ensure their coordinated operation and dynamically adjust operating strategies based on load changes has become a key challenge for industrial park energy management.

[0003] Traditional park energy management methods typically employ rule-based control strategies or simple optimization algorithms, which struggle to fully consider the complex influences of load characteristics, equipment characteristics, and environmental factors, resulting in low energy utilization efficiency and high operating costs. For example, some methods control equipment start-up and shutdown solely based on pre-set time schedules, failing to adapt to dynamic load changes; others employ single-objective optimization strategies, considering only economic or environmental benefits, making it difficult to achieve multi-objective collaborative optimization.

[0004] The existing technology has the following defects and shortcomings:

[0005] I. Lack of in-depth analysis of load characteristics. Traditional energy management methods are usually based on simple load forecasting models, which make it difficult to accurately capture the complex characteristics of load peak and valley distribution, changing trends, and correlations between different energy-consuming units, resulting in control strategies that are difficult to adapt to dynamic load changes.

[0006] Second, the collaborative control capability of distributed energy devices is insufficient. Traditional control methods typically control each device independently, lacking consideration for the interaction between devices, making it difficult to achieve coordinated and optimized operation of distributed energy devices, thus restricting the improvement of overall energy utilization efficiency.

[0007] Third, there is a lack of adaptability to dynamic environmental factors. In actual operation, factors such as ambient temperature and light intensity can significantly affect the operation of loads and distributed energy equipment. Traditional energy management methods typically lack the ability to effectively perceive and adapt to these dynamic environmental factors, making it difficult to achieve optimal control results. Summary of the Invention

[0008] This invention provides a method and system for intelligent linkage of distributed equipment groups in a park based on load characteristics, which can solve the problems in the prior art.

[0009] A first aspect of the present invention provides a method for intelligent linkage of a distributed device group in a campus based on load characteristics, comprising:

[0010] The system acquires real-time load data from multiple energy-consuming units within the park, including electricity load data, heating load data, and cooling load data. The real-time load data is then input into a pre-trained deep neural network model, which is trained based on historical load data and environmental parameter data to generate load characteristic curves for the multiple energy-consuming units. These load characteristic curves include peak-valley distribution characteristics, load change trend characteristics, and load correlation characteristics.

[0011] Based on the load characteristic curve, a multi-objective optimization model for the distributed energy equipment group in the park is constructed. The optimization objectives of the multi-objective optimization model include minimizing system operating costs and maximizing energy utilization efficiency. The constraints of the multi-objective optimization model include equipment operating parameter constraints, energy balance constraints, and environmental constraints. The multi-objective optimization model is solved using an improved particle swarm optimization algorithm to obtain the optimal operating scheme for the distributed energy equipment group in the park. The optimal operating scheme includes the power generation of photovoltaic power generation equipment, the charging and discharging power of energy storage equipment, and the cooling power of electric cooling equipment. The improved particle swarm optimization algorithm adopts an adaptive local search strategy and an elite solution dynamic maintenance mechanism based on the particle swarm optimization algorithm.

[0012] According to the optimal operating plan, control commands are sent to each distributed energy device through the park's energy management system. The control commands include the device's start / stop status, operating power, and operating parameters. The operating status of each distributed energy device is monitored in real time. When an abnormal device operation or a sudden load change is detected, a dynamic optimization and correction mechanism is triggered. The dynamic optimization and correction mechanism dynamically updates the optimal operating plan based on the rolling time domain to ensure the coordinated operation of the park's distributed energy device group and dynamic load balance.

[0013] The real-time load data is input into a pre-trained deep neural network model, which is trained based on historical load data and environmental parameter data to generate load characteristic curves for the multiple energy-consuming units, including:

[0014] The peak and valley difference index of the maximum load, minimum load and average load within 24 hours corresponding to the real-time load data is calculated based on the deep neural network model. The rate of change of the peak and valley difference in the time dimension is obtained to obtain the intraday peak and valley change rate. The peak duration index is obtained by weighted calculation of the duration of each peak interval. The weight coefficient of each duration interval in the peak duration index is determined based on the load characteristics.

[0015] The real-time load data is subjected to multi-scale wavelet decomposition to obtain multi-level detail components and approximate components; the trend change rate and acceleration characteristics are calculated based on the approximate components; energy density analysis is performed on the detail components to calculate the energy value of each scale detail component and its proportion of the total energy, and the energy proportion is obtained. The total energy includes the sum of the energy of the detail components and the energy of the approximate components.

[0016] The dynamic correlation coefficient between the loads of different energy-consuming units is calculated using the sliding window method. The dynamic correlation coefficient is calculated based on the covariance and standard deviation of the load data within the window. The correlation fluctuation index is calculated based on the dynamic correlation coefficient, and the correlation fluctuation index characterizes the stability of the correlation coefficient within the observation period.

[0017] The load characteristic vector is constructed by combining the peak and valley difference index, intraday peak and valley change rate, peak duration index, trend change rate, acceleration characteristics, energy proportion, dynamic correlation coefficient, and correlation fluctuation index. The characteristic importance is calculated based on the variance of each load characteristic vector. The characteristic importance is determined by the ratio of the variance of the load characteristic vector to the sum of the variances of all load characteristic vectors, thereby generating the load characteristic curves of the multiple energy-consuming units.

[0018] Based on the load characteristic curve, a multi-objective optimization model for the distributed energy equipment group in the park is constructed. The optimization objectives of the multi-objective optimization model include minimizing system operating costs and maximizing energy utilization efficiency, including:

[0019] The fuel cost of each device in the distributed energy equipment group in the park is calculated based on the output power, equipment efficiency and unit fuel price of each device at each time. The fuel cost of the device, the pre-acquired equipment operation and maintenance cost and the pre-acquired equipment start-up and shutdown cost are added together to obtain the operating cost of a single device. The operating costs of all devices are accumulated in the time dimension and the device dimension to obtain the total operating cost of the system.

[0020] The system obtains the output energy and input energy of each device, calculates the ratio of the output energy to the input energy to obtain the overall system efficiency; obtains the available energy and total energy of various energy forms, calculates the weighted sum of the ratios of available energy and total energy of various energy forms based on preset weighting coefficients to obtain the energy quality coefficient; and multiplies the overall system efficiency by the energy quality coefficient to obtain the system energy utilization efficiency.

[0021] A multi-objective optimization model is constructed, with the total operating cost of the system as the first optimization objective and the negative of the system's energy utilization efficiency as the second optimization objective; the constraints of the multi-objective optimization model include equipment operating parameter constraints, energy balance constraints, and environmental constraints.

[0022] The improved particle swarm optimization algorithm is used to solve the multi-objective optimization model, resulting in the optimal operating scheme for the distributed energy equipment group in the park, including:

[0023] Construct particle position vectors and velocity vectors representing the operating state of the device at different time periods, limit the velocity vectors to a preset velocity range, and initialize the particle swarm using time-based adaptive inertia weights. The adaptive inertia weights decrease non-linearly with the increase of the number of iterations to obtain the initialized particle swarm.

[0024] For each particle in the initial particle swarm, a dual-objective fitness function is constructed based on system operating cost and energy utilization efficiency for evaluation. Non-dominated sorting is used to determine the dominance level of each particle, and the crowding distance between particles is calculated as a diversity index. Based on the dominance level and the crowding distance, a dynamic learning factor is used to update the particle position vector and velocity vector.

[0025] For the updated particle position vector, a mixture of Gaussian perturbations with different scale standard deviations is constructed for local search. The weights of perturbations at different scales are adjusted by adaptive mixing coefficients. A solution space density function is constructed based on the crowding degree of the current solution. The solution space density function is used to dynamically adjust the local search direction to obtain a local optimized solution.

[0026] An elite solution evaluation system is established for the local optimization solution, including dominance level, distance diversity, and target spatial distribution. The comprehensive score of the solution is calculated based on the elite solution evaluation system. A dynamic capacity strategy that adaptively adjusts according to the current particle swarm size is adopted to maintain the elite solution set.

[0027] The convergence index is calculated based on the continuous change in particle position of the elite solution set; density analysis and uniformity evaluation are performed on the converged non-dominated solution set, and finally the optimal Pareto solution set with uniform distribution is output. The optimal Pareto solution set corresponds to the optimal operation scheme of the distributed energy equipment group.

[0028] The method further includes:

[0029] A dynamic search step size is constructed based on the number of iterations, which decreases exponentially with the number of iterations. A Gaussian mixture distribution with different scale standard deviations is constructed, and the mixture coefficients with different scale standard deviations are combined using an adaptively adjusted mixing coefficient according to the iteration process to obtain a Gaussian mixture perturbation vector. The distance distribution between solutions in the current solution set is calculated, and a density function reflecting the crowding degree of the solution space is established. The dynamic search step size, the Gaussian mixture perturbation vector, and the density function are multiplied to construct an enhanced local search operator.

[0030] For the solution set optimized by the enhanced local search operator, the dominance level is obtained by calculating the number of other solutions that each solution is dominated by. The distance diversity index is calculated based on the minimum Euclidean distance between solutions. The target space distribution index is obtained by calculating the deviation of each solution from the average value in the target space. The dominance level, distance diversity index and target space distribution index are combined by weighting coefficients to construct an elite solution comprehensive scoring function.

[0031] Based on the current population size of the particle swarm, the dynamic capacity limit of the elite solution set is linearly determined within the preset minimum and maximum capacity range; the solution set with the best dominance level and whose number does not exceed the dynamic capacity limit is selected to form the updated elite solution set.

[0032] For the updated elite solution set, the position change of the solution between adjacent iterations is calculated, and a population convergence index is constructed based on the statistical characteristics of the position change. Non-dominated solutions that satisfy the population convergence index and are evenly distributed are selected from the updated elite solution set, and finally an optimal Pareto solution set with even distribution is formed.

[0033] According to the optimal operating scheme, the control commands sent to each distributed energy device through the park energy management system include:

[0034] Construct a set of device start / stop instructions, a set of power control instructions, and a set of parameter adjustment instructions. The set of device start / stop instructions includes start / stop status identifiers for each device, the set of power control instructions includes target power values ​​for each device, and the set of parameter adjustment instructions includes operating parameter vectors for each device.

[0035] Based on the energy efficiency coefficient, response speed coefficient, and flexibility coefficient of each device, the energy efficiency coefficient, response speed coefficient, and flexibility coefficient are weighted and combined using a weighting coefficient to obtain the importance index of each device. The execution priority of the control commands of each device is determined according to the importance index.

[0036] A device startup timing matrix is ​​established, wherein the matrix elements in the device startup timing matrix represent the startup response delay between two corresponding devices. Based on the device startup timing matrix, instruction execution time windows are divided, and the key execution time nodes within the time windows are determined sequentially according to the constraint relationship of the device startup timing matrix.

[0037] Based on the importance index and startup timing matrix of each device, the start-stop status identifiers in the device start-stop instruction set are decomposed according to priority order to obtain the device start-stop instruction sequence. Similarly, the target power values ​​in the power control instruction set are decomposed according to priority order to obtain the power control instruction sequence.

[0038] The device start / stop instruction sequence and power control instruction sequence are combined according to the divided time windows and key execution time nodes to form a phased control instruction containing the execution timing. The phased control instruction includes the start / stop status identifier, target power value and operating parameter vector of each device at each key time node.

[0039] A second aspect of the present invention provides an intelligent linkage system for a distributed equipment group in a park based on load characteristics, comprising:

[0040] The first unit is used to acquire real-time load data of multiple energy-consuming units in the park. The real-time load data includes electricity load data, heating load data, and cooling load data. The real-time load data is input into a pre-trained deep neural network model. The deep neural network model is trained based on historical load data and environmental parameter data to generate load characteristic curves of the multiple energy-consuming units. The load characteristic curves include peak-valley distribution characteristics, load change trend characteristics, and load correlation characteristics.

[0041] The second unit is used to construct a multi-objective optimization model for the distributed energy equipment group in the park based on the load characteristic curve. The optimization objectives of the multi-objective optimization model include minimizing system operating costs and maximizing energy utilization efficiency. The constraints of the multi-objective optimization model include equipment operating parameter constraints, energy balance constraints, and environmental constraints. The improved particle swarm optimization algorithm is used to solve the multi-objective optimization model to obtain the optimal operating scheme of the distributed energy equipment group in the park. The optimal operating scheme includes the power generation of photovoltaic power generation equipment, the charging and discharging power of energy storage equipment, and the cooling power of electric cooling equipment. The improved particle swarm optimization algorithm adopts an adaptive local search strategy and an elite solution dynamic maintenance mechanism based on the particle swarm optimization algorithm.

[0042] The third unit is used to send control commands to each distributed energy device through the park energy management system according to the optimal operation plan. The control commands include the device start / stop status, operating power and operating parameters. It monitors the operating status of each distributed energy device in real time. When an abnormal operation or load change is detected, it triggers a dynamic optimization and correction mechanism. The dynamic optimization and correction mechanism dynamically updates the optimal operation plan based on the rolling time domain to ensure the coordinated operation of the park's distributed energy device group and dynamic load balance.

[0043] Third aspect of the present invention

[0044] An electronic device is provided, comprising:

[0045] processor;

[0046] Memory used to store processor-executable instructions;

[0047] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0048] Fourth aspect of the present invention,

[0049] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0050] The beneficial effects of this application are as follows:

[0051] 1. Reduce system operating costs and improve energy efficiency: By using a multi-objective optimization model that comprehensively considers minimizing system operating costs and maximizing energy efficiency, and by using an improved particle swarm optimization algorithm to solve for the optimal operating scheme, the operating costs of the park's energy system can be effectively reduced and the energy efficiency can be improved.

[0052] 2. Improve load forecasting accuracy and equipment operation stability: Based on the deep neural network model, historical load data and environmental parameter data are trained to accurately predict the load characteristic curves of energy-consuming units, including peak and valley distribution characteristics, load change trend characteristics and load correlation characteristics, thereby improving load forecasting accuracy and ensuring equipment operation stability.

[0053] 3. Achieve dynamic load balancing and enhance system adaptability: The dynamic optimization and correction mechanism can monitor equipment operating status and load changes in real time, and dynamically update the optimal operating scheme based on rolling time domain when equipment operation is abnormal or load changes suddenly occur, ensuring the coordinated operation and dynamic load balancing of the distributed energy equipment group in the park, and enhancing the system's adaptability to emergencies. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the intelligent linkage method for a distributed equipment group in a park based on load characteristics, according to an embodiment of the present invention.

[0055] Figure 2 This is a schematic diagram of the structure of a campus distributed equipment group intelligent linkage system based on load characteristics, according to an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0057] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0058] Figure 1 This is a flowchart illustrating the intelligent linkage method for distributed equipment groups in a park based on load characteristics, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0059] S101. Obtain real-time load data of multiple energy-consuming units within the park, the real-time load data including electricity load data, heating load data and cooling load data; input the real-time load data into a pre-trained deep neural network model, the deep neural network model being trained based on historical load data and environmental parameter data, to generate load characteristic curves of the multiple energy-consuming units, wherein the load characteristic curves include peak-valley distribution characteristics, load change trend characteristics and load correlation characteristics;

[0060] S102. Based on the load characteristic curve, a multi-objective optimization model for the distributed energy equipment group in the park is constructed. The optimization objectives of the multi-objective optimization model include minimizing system operating costs and maximizing energy utilization efficiency. The constraints of the multi-objective optimization model include equipment operating parameter constraints, energy balance constraints, and environmental constraints. The multi-objective optimization model is solved using an improved particle swarm optimization algorithm to obtain the optimal operating scheme for the distributed energy equipment group in the park. The optimal operating scheme includes the power generation of photovoltaic power generation equipment, the charging and discharging power of energy storage equipment, and the cooling power of electric cooling equipment. The improved particle swarm optimization algorithm adopts an adaptive local search strategy and an elite solution dynamic maintenance mechanism based on the particle swarm optimization algorithm.

[0061] S103. According to the optimal operation plan, control commands are sent to each distributed energy device through the park energy management system. The control commands include the device start / stop status, operating power, and operating parameters. The operating status of each distributed energy device is monitored in real time. When an abnormal operation or load change is detected, a dynamic optimization correction mechanism is triggered. The dynamic optimization correction mechanism dynamically updates the optimal operation plan based on the rolling time domain to ensure the coordinated operation of the park's distributed energy device group and dynamic load balance.

[0062] In one optional implementation, the real-time load data is input into a pre-trained deep neural network model, which is trained based on historical load data and environmental parameter data to generate load characteristic curves for the plurality of energy-consuming units, including:

[0063] The peak and valley difference index of the maximum load, minimum load and average load within 24 hours corresponding to the real-time load data is calculated based on the deep neural network model. The rate of change of the peak and valley difference in the time dimension is obtained to obtain the intraday peak and valley change rate. The peak duration index is obtained by weighted calculation of the duration of each peak interval. The weight coefficient of each duration interval in the peak duration index is determined based on the load characteristics.

[0064] The real-time load data is subjected to multi-scale wavelet decomposition to obtain multi-level detail components and approximate components; the trend change rate and acceleration characteristics are calculated based on the approximate components; energy density analysis is performed on the detail components to calculate the energy value of each scale detail component and its proportion of the total energy, and the energy proportion is obtained. The total energy includes the sum of the energy of the detail components and the energy of the approximate components.

[0065] The dynamic correlation coefficient between the loads of different energy-consuming units is calculated using the sliding window method. The dynamic correlation coefficient is calculated based on the covariance and standard deviation of the load data within the window. The correlation fluctuation index is calculated based on the dynamic correlation coefficient, and the correlation fluctuation index characterizes the stability of the correlation coefficient within the observation period.

[0066] The load characteristic vector is constructed by combining the peak and valley difference index, intraday peak and valley change rate, peak duration index, trend change rate, acceleration characteristics, energy proportion, dynamic correlation coefficient, and correlation fluctuation index. The characteristic importance is calculated based on the variance of each load characteristic vector. The characteristic importance is determined by the ratio of the variance of the load characteristic vector to the sum of the variances of all load characteristic vectors, thereby generating the load characteristic curves of the multiple energy-consuming units.

[0067] A method for generating load characteristic curves of energy-consuming units driven by real-time load data includes the following steps:

[0068] First, acquire historical load data and environmental parameter data. Historical load data includes the electricity consumption data of each energy-consuming unit over a past period; for example, hourly electricity consumption data for the past year can be collected. Environmental parameter data includes temperature, humidity, weather conditions, etc., which can be obtained through weather stations or sensors. For example, the average daily temperature, maximum temperature, minimum temperature, and humidity data for the past year can be collected.

[0069] Then, the acquired historical load data and environmental parameter data are used to train the deep neural network model. The deep neural network model can employ structures such as convolutional neural networks, recurrent neural networks, or their variants. During training, historical load data is used as the output, and environmental parameter data and corresponding time information are used as the input. For example, historical load data and environmental parameter data from the past year can be arranged chronologically and input into the deep neural network model for training, enabling the model to learn the relationship between load data and environmental parameters.

[0070] Next, the real-time load data is input into a trained deep neural network model to generate load characteristic curves for multiple energy-consuming units. Real-time load data refers to the current electricity consumption data, which can be collected in real time through devices such as smart meters. For example, by inputting the current electricity consumption data and corresponding environmental parameter data into the trained deep neural network model, the model will output predicted load data for each energy-consuming unit for the next 24 hours, thereby generating load characteristic curves.

[0071] Based on the output of the deep neural network model, the maximum, minimum, and average load values ​​within 24 hours corresponding to the real-time load data are calculated. For example, the maximum and minimum values ​​can be found from the predicted 24-hour load data, and the average value of the 24-hour load data can be calculated. Then, the peak-to-valley difference, i.e., the difference between the maximum and minimum values, is calculated. To obtain the rate of change of the peak-to-valley difference over time, the change in the peak-to-valley difference between adjacent time points can be calculated, thus obtaining the intraday peak-to-valley change rate.

[0072] To calculate the peak duration index, the peak intervals must first be defined. For example, intervals exceeding the average load by a certain percentage can be defined as peak intervals. Then, the duration of each peak interval is calculated separately. Finally, the durations of each peak interval are weighted and summed according to pre-set weighting coefficients to obtain the peak duration index. The weighting coefficients can be adjusted based on load characteristics; for example, longer-duration peak intervals can be assigned a larger weight.

[0073] Multi-scale wavelet decomposition is performed on real-time load data. Wavelet decomposition can break down a signal into components of different frequencies, thereby extracting signal features. Multi-scale wavelet decomposition can decompose a signal into multiple levels of detail components and an approximate component. The detail components represent the high-frequency part of the signal, and the approximate component represents the low-frequency part of the signal.

[0074] Based on the approximate components obtained from wavelet decomposition, the rate of change of trend and acceleration characteristics are calculated. The rate of change of trend can be obtained by calculating the difference between the approximate components at adjacent time points. The acceleration characteristics can be obtained by calculating the difference between the rates of change of trend at adjacent time points.

[0075] Energy density analysis is performed on the detail components obtained from wavelet decomposition. The energy values ​​of the detail components at each scale and the proportion of each scale's energy to the total energy are calculated to obtain the energy percentage. The total energy includes the sum of the energies of all detail components and approximate components.

[0076] The sliding window method is used to calculate the dynamic correlation coefficient between the loads of different energy-consuming units. The sliding window method involves sliding a fixed-size window across the time series and calculating the correlation coefficient of the data within the window. The dynamic correlation coefficient is calculated based on the covariance and standard deviation of the load data within the window.

[0077] The correlation volatility index is calculated based on the dynamic correlation coefficient. The correlation volatility index characterizes the stability of the correlation coefficient within the observation period; for example, its volatility can be measured by calculating the standard deviation of the dynamic correlation coefficient.

[0078] Finally, a load characteristic vector is constructed by combining the peak-valley difference index, intraday peak-valley change rate, peak duration index, trend change rate, acceleration characteristics, energy share, dynamic correlation coefficient, and correlation fluctuation index. The characteristic importance is calculated based on the variance of each load characteristic vector. The characteristic importance is determined by the ratio of the variance of the load characteristic vector to the sum of the variances of all load characteristic vectors. Ultimately, load characteristic curves for multiple energy-consuming units are generated.

[0079] The beneficial effects of this method can be summarized in the following three aspects:

[0080] 1. High accuracy: By utilizing deep neural network models and multi-scale wavelet decomposition techniques, features of load data can be extracted more accurately, leading to more accurate predictions of future loads and the generation of more precise load characteristic curves. For example, learning the complex relationships between historical data and environmental parameters through deep learning models can improve the accuracy of load forecasting.

[0081] 2. Comprehensiveness: This method comprehensively considers multiple dimensions of load data, including peak-to-valley differences, rate of change, duration, trend changes, energy distribution, and correlations between different energy-consuming units, thus providing a more comprehensive description of the load characteristics of energy-consuming units. For example, wavelet decomposition can be used to analyze load change characteristics at different time scales, and correlation analysis can reveal the mutual influences between different energy-consuming units.

[0082] 3. High practicality: The load characteristic curves generated by this method can provide important reference for power system planning, operation, and management. For example, future load peaks can be predicted based on the load characteristic curves, thereby optimizing power dispatch strategies and improving energy utilization efficiency. Simultaneously, this method can also be used to identify abnormal electricity consumption behavior, improving power system security.

[0083] In one optional implementation, based on the load characteristic curve, a multi-objective optimization model for the distributed energy equipment group in the park is constructed. The optimization objectives of the multi-objective optimization model include minimizing system operating costs and maximizing energy utilization efficiency, including:

[0084] The fuel cost of each device in the distributed energy equipment group in the park is calculated based on the output power, equipment efficiency and unit fuel price of each device at each time. The fuel cost of the device, the pre-acquired equipment operation and maintenance cost and the pre-acquired equipment start-up and shutdown cost are added together to obtain the operating cost of a single device. The operating costs of all devices are accumulated in the time dimension and the device dimension to obtain the total operating cost of the system.

[0085] The system obtains the output energy and input energy of each device, calculates the ratio of the output energy to the input energy to obtain the overall system efficiency; obtains the available energy and total energy of various energy forms, calculates the weighted sum of the ratios of available energy and total energy of various energy forms based on preset weighting coefficients to obtain the energy quality coefficient; and multiplies the overall system efficiency by the energy quality coefficient to obtain the system energy utilization efficiency.

[0086] A multi-objective optimization model is constructed, with the total operating cost of the system as the first optimization objective and the negative of the system's energy utilization efficiency as the second optimization objective; the constraints of the multi-objective optimization model include equipment operating parameter constraints, energy balance constraints, and environmental constraints.

[0087] A multi-objective optimization method for constructing a distributed energy equipment cluster in a park based on load characteristic curves aims to minimize system operating costs and maximize energy utilization efficiency. The specific steps of this method are as follows:

[0088] First, obtain the park's load characteristic curve, which describes the park's energy demand, such as electricity and heating, at different times. For example, hourly electricity demand data for the coming week, as well as heating and cooling demand data, can be obtained based on historical data, forecasting models, or actual measurements.

[0089] Next, the composition of the distributed energy equipment cluster in the park will be determined, including, for example, gas-fired internal combustion engines, gas-fired boilers, photovoltaic power generation systems, and energy storage systems. Parameters for each device will be obtained, including rated power, efficiency characteristic curve, fuel type, unit fuel price, operation and maintenance costs, and start-up and shutdown costs. For example, the efficiency characteristic curve of a gas-fired internal combustion engine describes the efficiency value at different output power levels; the unit fuel price of a gas-fired boiler is the price per cubic meter of natural gas; and the operation and maintenance costs of a photovoltaic power generation system include daily maintenance and cleaning expenses.

[0090] Then, the distributed energy equipment group in the park is scheduled and optimized to meet the park's load demand. Scheduling optimization needs to consider equipment operating parameter constraints, such as the output power range and start / stop limits; energy balance constraints, i.e., at any given time, the sum of the output energy of all equipment equals the park's load demand; and environmental constraints, such as emission limits. The result of scheduling optimization is the output power of each device at each time. For example, at 9:00 AM, the photovoltaic power generation system outputs 100kW, and the gas-fired internal combustion engine outputs 50kW, together meeting the park's 150kW power demand.

[0091] Based on the scheduling optimization results, the total system operating cost is calculated. First, the fuel cost of each device is calculated based on its output power, efficiency, and unit fuel price at each time point. Then, the fuel cost, pre-acquired equipment maintenance cost, and pre-acquired equipment start-up and shutdown cost are added together to obtain the operating cost of a single device. Finally, the operating costs of all devices are summed over both the time and device dimensions to obtain the total system operating cost. For example, if a gas-fired internal combustion engine has an output power of 50kW, an efficiency of 40%, and a unit fuel price of 3 yuan / cubic meter at a certain time, then the fuel cost at that time is 50kW / 40% * 3 yuan / cubic meter * operating time.

[0092] Simultaneously, the system's energy utilization efficiency is calculated. First, the output energy and input energy of each device are obtained, and the ratio of output energy to input energy is calculated to obtain the overall system efficiency. Then, the available energy and total energy of various energy forms are obtained, such as the ratio of available energy of natural gas to its total energy, the ratio of available energy of electricity to its total energy, etc. Based on preset weighting coefficients, a weighted sum of the ratios of available energy of various energy forms to total energy is calculated to obtain the energy quality coefficient. Finally, the overall system efficiency is multiplied by the energy quality coefficient to obtain the system energy utilization efficiency. For example, if the overall system efficiency is 80% and the energy quality coefficient is 90%, then the system energy utilization efficiency is 72%.

[0093] Finally, a multi-objective optimization model is constructed, with the total system operating cost as the first optimization objective and the inverse of the system's energy utilization efficiency as the second. This model is solved using a multi-objective optimization algorithm, such as a genetic algorithm or particle swarm optimization. Ultimately, a set of Pareto optimal solutions is obtained, representing different trade-offs between system operating cost and energy utilization efficiency. Decision-makers can then select the appropriate solution based on actual needs.

[0094] The beneficial effects of this method can be summarized in the following three aspects:

[0095] 1. Reduce park energy costs: By optimizing the operation strategy of distributed energy equipment clusters, the park's energy consumption can be effectively reduced, thereby lowering energy costs.

[0096] 2. Improve energy utilization efficiency: This method takes into account both the overall efficiency and quality of energy, and can effectively improve the overall utilization efficiency of energy.

[0097] 3. Reduce environmental pollution: By optimizing the operation of the energy system, pollutant emissions can be reduced, thereby protecting the environment.

[0098] In one optional implementation, the multi-objective optimization model is solved using an improved particle swarm optimization algorithm to obtain the optimal operating scheme for the distributed energy equipment group in the park, including:

[0099] Construct particle position vectors and velocity vectors representing the operating state of the device at different time periods, limit the velocity vectors to a preset velocity range, and initialize the particle swarm using time-based adaptive inertia weights. The adaptive inertia weights decrease non-linearly with the increase of the number of iterations to obtain the initialized particle swarm.

[0100] For each particle in the initial particle swarm, a dual-objective fitness function is constructed based on system operating cost and energy utilization efficiency for evaluation. Non-dominated sorting is used to determine the dominance level of each particle, and the crowding distance between particles is calculated as a diversity index. Based on the dominance level and the crowding distance, a dynamic learning factor is used to update the particle position vector and velocity vector.

[0101] For the updated particle position vector, a mixture of Gaussian perturbations with different scale standard deviations is constructed for local search. The weights of perturbations at different scales are adjusted by adaptive mixing coefficients. A solution space density function is constructed based on the crowding degree of the current solution. The solution space density function is used to dynamically adjust the local search direction to obtain a local optimized solution.

[0102] An elite solution evaluation system is established for the local optimization solution, including dominance level, distance diversity, and target spatial distribution. The comprehensive score of the solution is calculated based on the elite solution evaluation system. A dynamic capacity strategy that adaptively adjusts according to the current particle swarm size is adopted to maintain the elite solution set.

[0103] The convergence index is calculated based on the continuous change in particle position of the elite solution set; density analysis and uniformity evaluation are performed on the converged non-dominated solution set, and finally the optimal Pareto solution set with uniform distribution is output. The optimal Pareto solution set corresponds to the optimal operation scheme of the distributed energy equipment group.

[0104] An optimized operation method for distributed energy device clusters aims to reduce system operating costs and improve energy utilization efficiency. This method employs an improved particle swarm optimization algorithm to search for the optimal operating scheme and utilizes a series of strategies to enhance the algorithm's optimization performance and solution set quality.

[0105] First, the particle swarm is initialized. To characterize the operating state of the devices at different time periods, particle position vectors and velocity vectors are constructed. The velocity vectors are limited to a preset range; for example, the rate of change of the device's output power is limited to a certain range. A time-based adaptive inertia weight is used to initialize the particle swarm. The inertia weight decreases non-linearly with the number of iterations; for example, an exponential decreasing function is used, giving the algorithm strong global search capabilities in the early stages and focusing on local optimization in later stages. Assuming there are 10 devices, each time period is 1 hour, and the operating cycle is 24 hours, then the position vector of each particle contains 240 elements, each element representing the output power of the corresponding device in the corresponding time period.

[0106] Then, the fitness of each particle in the initial particle swarm is evaluated. A dual-objective fitness function is constructed based on system operating cost and energy efficiency. For example, the operating cost objective can consider fuel cost, maintenance cost, etc., while the energy efficiency objective can consider energy conversion efficiency, energy loss, etc. Non-dominated ranking is used to determine the dominance level of each particle, and the crowding distance between particles is calculated as a diversity index. The crowding distance reflects the distribution density of solutions around the particle. Assuming there are two objective functions, each particle is non-dominated ranked according to its objective function value to obtain its dominance level.

[0107] Next, the particle position and velocity vectors are updated. A dynamic learning factor is used to update the particle's position and velocity vectors based on its dominance level and crowding distance. The learning factor can be adjusted according to the particle's dominance level; for example, particles with higher dominance levels have larger learning factors, allowing them to better maintain their advantage. Assuming a particle has a dominance level of 1 and a crowding distance of 0.5, the learning factor for that particle is calculated according to a preset dynamic learning factor formula.

[0108] A local search is performed on the updated particle position vector. A mixture of Gaussian perturbations with different standard deviations is constructed for the local search; for example, three different standard deviations are set to represent small, medium, and large-scale perturbations, respectively. The weights of perturbations at different scales are adjusted using adaptive mixing coefficients. These mixing coefficients can be adjusted based on the crowding level of the current solution; for example, in crowded regions, the weight of larger-scale perturbations is increased to escape local optima. A solution space density function is constructed based on the crowding level of the current solution, and this function is used to dynamically adjust the local search direction to obtain locally optimized solutions. For example, in sparse regions, the search is preferentially directed towards regions with higher density.

[0109] An elite solution evaluation system is established for locally optimized solutions. This system includes dominance level, distance diversity, and target space distribution. A comprehensive score for the solution is calculated based on this evaluation system. For example, the comprehensive score can be calculated according to the weights of dominance level, distance diversity, and target space distribution. A dynamic capacity strategy is adopted to maintain the elite solution set, with the dynamic capacity adaptively adjusted according to the current particle swarm size. For example, the larger the population size, the larger the capacity of the elite solution set.

[0110] Convergence metrics are calculated based on the continuous changes in particle positions within the elite solution set. For example, the variance of particle positions in the elite solution set is calculated; if the variance is less than a preset threshold, the algorithm is considered converged. Density analysis and uniformity evaluation are performed on the converged non-dominated solution set, ultimately outputting a uniformly distributed optimal Pareto solution set. For example, the distances between solutions in the non-dominated solution set are calculated; if the distances are uniformly distributed, the solution set is considered uniformly distributed. The optimal Pareto solution set corresponds to the optimal operating scheme for a distributed energy device cluster. For example, based on the particle position vectors in the optimal Pareto solution set, the output power of each device in each time period can be determined.

[0111] The beneficial effects of this method can be summarized in the following three aspects:

[0112] 1. Improved optimization performance: By employing strategies such as adaptive inertia weights, dynamic learning factors, and Gaussian mixture perturbations, the algorithm's global search capability and local optimization capability are enhanced, enabling it to find better solutions more quickly.

[0113] 2. Improve the quality of the solution set: By using methods such as non-dominated sorting, crowding distance and elite solution evaluation system, the diversity and uniformity of the Pareto solution set are guaranteed, providing more diverse selection options.

[0114] 3. Enhanced practicality: This method can flexibly adjust parameters according to actual conditions, such as the number of devices, operating cycle, objective function, etc., and is suitable for optimizing the operation of different types of distributed energy systems.

[0115] In one optional implementation, the method further includes:

[0116] A dynamic search step size is constructed based on the number of iterations, which decreases exponentially with the number of iterations. A Gaussian mixture distribution with different scale standard deviations is constructed, and the mixture coefficients with different scale standard deviations are combined using an adaptively adjusted mixing coefficient according to the iteration process to obtain a Gaussian mixture perturbation vector. The distance distribution between solutions in the current solution set is calculated, and a density function reflecting the crowding degree of the solution space is established. The dynamic search step size, the Gaussian mixture perturbation vector, and the density function are multiplied to construct an enhanced local search operator.

[0117] For the solution set optimized by the enhanced local search operator, the dominance level is obtained by calculating the number of other solutions that each solution is dominated by. The distance diversity index is calculated based on the minimum Euclidean distance between solutions. The target space distribution index is obtained by calculating the deviation of each solution from the average value in the target space. The dominance level, distance diversity index and target space distribution index are combined by weighting coefficients to construct an elite solution comprehensive scoring function.

[0118] Based on the current population size of the particle swarm, the dynamic capacity limit of the elite solution set is linearly determined within the preset minimum and maximum capacity range; the solution set with the best dominance level and whose number does not exceed the dynamic capacity limit is selected to form the updated elite solution set.

[0119] For the updated elite solution set, the position change of the solution between adjacent iterations is calculated, and a population convergence index is constructed based on the statistical characteristics of the position change. Non-dominated solutions that satisfy the population convergence index and are evenly distributed are selected from the updated elite solution set, and finally an optimal Pareto solution set with even distribution is formed.

[0120] A multi-objective optimization method based on an elite solution set strategy is proposed to find multiple optimal solutions in a complex search space. This method constructs an enhanced local search operator through dynamic search step size, Gaussian mixture perturbation, and density function, and combines dominance level, distance diversity, and objective space distribution indices to screen elite solutions, ultimately obtaining a uniformly distributed Pareto optimal solution set.

[0121] First, an initial population containing multiple solutions is initialized, each representing a potential solution to a multi-objective optimization problem. For example, in the case of designing an aircraft, each solution can represent different design options for the aircraft, including parameters such as wing shape, fuselage length, and engine thrust.

[0122] Next, an iterative search process is performed. In each iteration, the dynamic search step size is first calculated based on the current iteration number. The dynamic search step size decreases exponentially with the number of iterations; for example, the step size can be set to 0.9 raised to the power of the initial value. This ensures a larger exploration range in the early stages of the search, followed by a more refined local search in later stages.

[0123] Next, a Gaussian mixture distribution is constructed, which is a combination of multiple Gaussian distributions with different standard deviations. For example, three Gaussian distributions with standard deviations of 0.1, 0.5, and 1 can be used. Mixing coefficients are used to control the proportion of Gaussian distributions with different standard deviations in the mixture distribution, and these mixing coefficients are adaptively adjusted during the iteration process. For example, in the early stages of the search, Gaussian distributions with larger standard deviations can be given higher weights for broader exploration; while in the later stages of the search, Gaussian distributions with smaller standard deviations can be given higher weights for finer local searches. The dynamic search step size is multiplied by the Gaussian mixture perturbation vector to obtain a perturbation vector.

[0124] Simultaneously, the distance distribution between solutions in the current solution set is calculated, and a density function is established to reflect the crowding level of the solution space. For example, a Gaussian kernel density estimation method can be used to construct the density function. Multiplying the perturbation vector by the density function constructs an enhanced local search operator. This operator can guide the search process to avoid crowded regions and explore sparser regions.

[0125] The enhanced local search operator is applied to each solution in the current population to generate a new solution. The newly generated solutions and the original solutions are evaluated, and the objective function value for each solution is calculated. For example, in the aircraft design case, the objective function could be the aircraft's flight speed and fuel efficiency.

[0126] Then, the number of times each solution is dominated by other solutions is calculated to obtain the dominance level. Solutions with lower dominance levels have better performance. Simultaneously, a distance diversity index is calculated based on the minimum Euclidean distance between solutions, reflecting the diversity of the solution set. Furthermore, a target space distribution index is obtained by calculating the deviation of each solution from the mean in the target space, reflecting the distribution of the solution set in the target space. The dominance level, distance diversity index, and target space distribution index are weighted and combined using weighting coefficients to construct an elite solution comprehensive scoring function. For example, the weight of the dominance level can be set to 0.5, the weight of the distance diversity index to 0.3, and the weight of the target space distribution index to 0.2.

[0127] Based on the current population size, the dynamic capacity upper limit of the elite solution set is linearly determined within a preset minimum and maximum capacity range. For example, assuming a minimum capacity of 10, a maximum capacity of 100, and a current population size of 50, the dynamic capacity upper limit is 60. The solution set with the best dominance level and a number not exceeding the dynamic capacity upper limit is selected to form the updated elite solution set.

[0128] For the updated elite solution set, the positional changes of solutions between adjacent iterations are calculated, and a population convergence index is constructed based on the statistical characteristics of these positional changes. For example, the mean or variance of the positional changes can be calculated as the population convergence index. Non-dominated solutions that satisfy the population convergence index and are uniformly distributed are selected from the updated elite solution set, ultimately forming a uniformly distributed optimal Pareto solution set.

[0129] The beneficial effects of this method can be summarized in the following three aspects:

[0130] 1. Improved search efficiency: By using dynamic search step size and mixed Gaussian perturbation, the search space can be explored more effectively and the optimal solution can be found faster.

[0131] 2. Enhanced population diversity: By using density functions and distance diversity indices, population diversity can be effectively maintained, avoiding the search process from getting trapped in local optima.

[0132] 3. A uniformly distributed Pareto optimal solution set was obtained: By using the elite solution set strategy and the target space distribution index, a uniformly distributed Pareto optimal solution set can be obtained, providing decision-makers with more choices.

[0133] In one optional implementation, according to the optimal operating scheme, sending control commands to each distributed energy device through the park energy management system includes:

[0134] Construct a set of device start / stop instructions, a set of power control instructions, and a set of parameter adjustment instructions. The set of device start / stop instructions includes start / stop status identifiers for each device, the set of power control instructions includes target power values ​​for each device, and the set of parameter adjustment instructions includes operating parameter vectors for each device.

[0135] Based on the energy efficiency coefficient, response speed coefficient, and flexibility coefficient of each device, the energy efficiency coefficient, response speed coefficient, and flexibility coefficient are weighted and combined using a weighting coefficient to obtain the importance index of each device. The execution priority of the control commands of each device is determined according to the importance index.

[0136] A device startup timing matrix is ​​established, wherein the matrix elements in the device startup timing matrix represent the startup response delay between two corresponding devices. Based on the device startup timing matrix, instruction execution time windows are divided, and the key execution time nodes within the time windows are determined sequentially according to the constraint relationship of the device startup timing matrix.

[0137] Based on the importance index and startup timing matrix of each device, the start-stop status identifiers in the device start-stop instruction set are decomposed according to priority order to obtain the device start-stop instruction sequence. Similarly, the target power values ​​in the power control instruction set are decomposed according to priority order to obtain the power control instruction sequence.

[0138] The device start / stop instruction sequence and power control instruction sequence are combined according to the divided time windows and key execution time nodes to form a phased control instruction containing the execution timing. The phased control instruction includes the start / stop status identifier, target power value and operating parameter vector of each device at each key time node.

[0139] The park's energy management system achieves optimal energy operation by sending control commands to distributed energy devices. The specific implementation method of this control approach is described in detail below:

[0140] First, the system collects real-time operating data from all distributed energy devices within the park, including device type, current power, coefficient of performance (COP), response speed, and adjustable parameter range. For example, it collects data showing that gas turbine A has a current power of 50kW, a COP of 0.4, a response speed of 10kW / min, and adjustable parameters including gas flow rate and rotational speed; and it collects data showing that photovoltaic array B has a current power of 20kW, a COP of 0.9, a response speed of 5kW / min, and adjustable parameters including inverter output voltage and current.

[0141] Next, the equipment control instruction set is constructed. The instruction set consists of three parts: the equipment start-stop instruction set, which records the target start-stop status of each device, such as "start" or "stop"; the power control instruction set, which records the target power value of each device, such as the target power of gas turbine A being 60kW and the target power of photovoltaic array B being 30kW; and the parameter adjustment instruction set, which records the target operating parameter vector of each device, such as adjusting the gas flow rate of gas turbine A to 80% and the speed to 6000rpm.

[0142] Next, the importance of each device is assessed. An importance index is calculated by weighting the device's energy efficiency coefficient, response speed coefficient, and flexibility coefficient. For example, if the weight of the energy efficiency coefficient is set to 0.5, the weight of the response speed coefficient to 0.3, and the weight of the flexibility coefficient to 0.2, the importance index of gas turbine A is 0.5*0.4 + 0.3*10 + 0.2*1 = 3.2; the importance index of photovoltaic array B is 0.5*0.9 + 0.3*5 + 0.2*1 = 2.15. Based on the importance index, the priority of control command execution for each device is determined. Devices with higher importance index values ​​have higher priority; for example, gas turbine A has a higher priority than photovoltaic array B.

[0143] Establish a device startup timing matrix. This matrix describes the startup dependencies and time delays between devices. For example, after gas turbine A starts, it takes 5 minutes before waste heat boiler C can start; therefore, the element value from A to C in the matrix is ​​5. Based on this matrix, divide the instruction execution time window and determine the key execution time nodes within the time window sequentially according to the constraints of the device startup timing matrix. For example, if the time window is set to 10 minutes, then 0 minutes, 5 minutes, and 10 minutes can be set as key execution time nodes.

[0144] Based on the equipment importance indicators and startup timing matrix, the equipment start-up and shutdown command sets and power control command sets are decomposed into command sequences. According to priority, the equipment start-up and shutdown status indicators and target power values ​​are assigned to various key execution time nodes. For example, at 0 minutes, gas turbine A starts with a power setting of 55kW; at 5 minutes, waste heat boiler C starts; at 10 minutes, the power of gas turbine A is adjusted to 60kW, and the power of photovoltaic array B is adjusted to 30kW.

[0145] Finally, the equipment start-up and shutdown command sequence, power control command sequence, and parameter adjustment command set are combined according to the divided time windows and key execution time nodes to form a phased control command containing the execution sequence. For example, at time 0 minutes, the control command for gas turbine A is "start, power 55kW, gas flow 75%, speed 5500rpm"; at time 5 minutes, the control command for waste heat boiler C is "start"; at time 10 minutes, the control command for gas turbine A is "run, power 60kW, gas flow 80%, speed 6000rpm", and the control command for photovoltaic array B is "run, power 30kW, inverter output voltage 500V, current 60A".

[0146] The beneficial effects of this method are reflected in three aspects:

[0147] First, improve energy efficiency. By considering the energy efficiency coefficient and operating status of equipment, optimize energy allocation and utilization to minimize energy waste and improve overall energy efficiency.

[0148] Second, enhance system response speed. By considering the response speed coefficient of equipment, prioritize scheduling equipment with fast response times to quickly respond to changes in energy demand and ensure the stability and reliability of energy supply.

[0149] Third, improve system operational flexibility. By considering the equipment's flexibility coefficient and adjustable parameter range, the operating status of the equipment can be flexibly adjusted to adapt to different energy demand scenarios, thereby improving the overall operational flexibility of the system.

[0150] Figure 2This is a schematic diagram of the intelligent linkage system for a distributed equipment group in a park based on load characteristics, as described in an embodiment of the present invention. Figure 2 As shown, the system includes:

[0151] The first unit is used to acquire real-time load data of multiple energy-consuming units in the park. The real-time load data includes electricity load data, heating load data, and cooling load data. The real-time load data is input into a pre-trained deep neural network model. The deep neural network model is trained based on historical load data and environmental parameter data to generate load characteristic curves of the multiple energy-consuming units. The load characteristic curves include peak-valley distribution characteristics, load change trend characteristics, and load correlation characteristics.

[0152] The second unit is used to construct a multi-objective optimization model for the distributed energy equipment group in the park based on the load characteristic curve. The optimization objectives of the multi-objective optimization model include minimizing system operating costs and maximizing energy utilization efficiency. The constraints of the multi-objective optimization model include equipment operating parameter constraints, energy balance constraints, and environmental constraints. The improved particle swarm optimization algorithm is used to solve the multi-objective optimization model to obtain the optimal operating scheme of the distributed energy equipment group in the park. The optimal operating scheme includes the power generation of photovoltaic power generation equipment, the charging and discharging power of energy storage equipment, and the cooling power of electric cooling equipment. The improved particle swarm optimization algorithm adopts an adaptive local search strategy and an elite solution dynamic maintenance mechanism based on the particle swarm optimization algorithm.

[0153] The third unit is used to send control commands to each distributed energy device through the park energy management system according to the optimal operation plan. The control commands include the device start / stop status, operating power and operating parameters. It monitors the operating status of each distributed energy device in real time. When an abnormal operation or load change is detected, it triggers a dynamic optimization and correction mechanism. The dynamic optimization and correction mechanism dynamically updates the optimal operation plan based on the rolling time domain to ensure the coordinated operation of the park's distributed energy device group and dynamic load balance.

[0154] A third aspect of the embodiments of the present invention,

[0155] An electronic device is provided, comprising:

[0156] processor;

[0157] Memory used to store processor-executable instructions;

[0158] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0159] Fourth aspect of the present invention,

[0160] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0161] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent linkage of distributed equipment groups in a park based on load characteristics, characterized in that, include: The system acquires real-time load data from multiple energy-consuming units within the park, including electricity load data, heating load data, and cooling load data. The real-time load data is input into a pre-trained deep neural network model, which is trained based on historical load data and environmental parameter data to generate load characteristic curves of the multiple energy-consuming units. The load characteristic curves include peak-valley distribution characteristics, load change trend characteristics, and load correlation characteristics. Based on the load characteristic curve, a multi-objective optimization model for the park's distributed energy equipment group is constructed. The fuel cost of the equipment is calculated based on the output power, equipment efficiency, and unit fuel price of each equipment in the park's distributed energy equipment group at each time. The fuel cost of the equipment, the pre-acquired equipment operation and maintenance cost, and the pre-acquired equipment start-up and shutdown cost are added together to obtain the operating cost of a single equipment. The operating costs of all equipment are accumulated in the time dimension and the equipment dimension to obtain the total operating cost of the system. The system's overall efficiency is obtained by acquiring the output energy and input energy of each device and calculating the ratio of the output energy to the input energy. The available energy and total energy of various energy forms are obtained, and the energy quality coefficient is obtained by calculating the weighted sum of the ratios of available energy and total energy of various energy forms based on preset weighting coefficients; the system energy utilization efficiency is obtained by multiplying the overall system efficiency by the energy quality coefficient. A multi-objective optimization model is constructed, with the total operating cost of the system as the first optimization objective and the negative of the system's energy utilization efficiency as the second optimization objective; the constraints of the multi-objective optimization model include equipment operating parameter constraints, energy balance constraints, and environmental constraints. The improved particle swarm optimization algorithm is used to solve the multi-objective optimization model to obtain the optimal operation scheme of the distributed energy equipment group in the park. The optimal operation scheme includes the power generation of photovoltaic power generation equipment, the charging and discharging power of energy storage equipment and the cooling power of electric cooling equipment. The improved particle swarm optimization algorithm adopts an adaptive local search strategy and an elite solution dynamic maintenance mechanism on the basis of the particle swarm optimization algorithm. According to the optimal operating plan, control commands are sent to each distributed energy device through the park energy management system. The control commands include the device start / stop status and operating parameters. The system monitors the operating status of each distributed energy device in real time. When an abnormal operation or sudden load change is detected, a dynamic optimization and correction mechanism is triggered. The dynamic optimization and correction mechanism updates the optimal operating scheme based on the rolling time domain to ensure the coordinated operation of the distributed energy device group in the park and the dynamic balance of the load.

2. The method according to claim 1, characterized in that, The real-time load data is input into a pre-trained deep neural network model, which is trained based on historical load data and environmental parameter data to generate load characteristic curves for the multiple energy-consuming units, including: The peak-valley difference index of the maximum load, minimum load and average load within 24 hours corresponding to the real-time load data is calculated based on the deep neural network model. The rate of change of the peak-valley difference in the time dimension is obtained to obtain the intraday peak-valley change rate. The peak duration index is obtained by weighted calculation of the duration of each peak interval. The weight coefficient of each duration interval in the peak duration index is determined based on the load characteristics. The real-time load data is subjected to multi-scale wavelet decomposition to obtain multi-level detail components and approximate components; the trend change rate and acceleration characteristics are calculated based on the approximate components; energy density analysis is performed on the detail components to calculate the energy value of each scale detail component and its proportion of the total energy, and the energy proportion is obtained. The total energy includes the sum of the energy of the detail components and the energy of the approximate components. The dynamic correlation coefficient between the loads of different energy-consuming units is calculated using the sliding window method. The dynamic correlation coefficient is calculated based on the covariance and standard deviation of the load data within the window. The correlation fluctuation index is calculated based on the dynamic correlation coefficient, and the correlation fluctuation index characterizes the stability of the correlation coefficient within the observation period. The load characteristic vector is constructed by combining the peak-valley difference index, intraday peak-valley change rate, peak duration index, trend change rate, acceleration characteristics, energy proportion, dynamic correlation coefficient, and correlation fluctuation index. The characteristic importance is calculated based on the variance of each load characteristic vector. The characteristic importance is determined by the ratio of the variance of the load characteristic vector to the sum of the variances of all load characteristic vectors, thereby generating the load characteristic curves of the multiple energy-consuming units.

3. The method according to claim 1, characterized in that, The improved particle swarm optimization algorithm is used to solve the multi-objective optimization model, resulting in the optimal operating scheme for the distributed energy equipment group in the park, including: Construct particle position vectors and velocity vectors representing the operating state of the device at different time periods, limit the velocity vectors to a preset velocity range, and initialize the particle swarm using time-based adaptive inertia weights. The adaptive inertia weights decrease non-linearly with the increase of the number of iterations to obtain the initialized particle swarm. For each particle in the initial particle swarm, a dual-objective fitness function is constructed based on system operating cost and energy utilization efficiency for evaluation. Non-dominated sorting is used to determine the dominance level of each particle, and the crowding distance between particles is calculated as a diversity index. Based on the dominance level and the crowding distance, a dynamic learning factor is used to update the particle position vector and velocity vector. For the updated particle position vector, a mixture of Gaussian perturbations with different scale standard deviations is constructed for local search. The weights of perturbations at different scales are adjusted by adaptive mixing coefficients. A solution space density function is constructed based on the crowding degree of the current solution. The solution space density function is used to dynamically adjust the local search direction to obtain a local optimized solution. An elite solution evaluation system is established for the local optimization solution, including dominance level, distance diversity, and target spatial distribution. The comprehensive score of the solution is calculated based on the elite solution evaluation system. A dynamic capacity strategy that adaptively adjusts according to the current particle swarm size is adopted to maintain the elite solution set. The convergence index is calculated based on the continuous change in particle position of the elite solution set; density analysis and uniformity evaluation are performed on the converged non-dominated solution set, and finally the optimal Pareto solution set with uniform distribution is output. The optimal Pareto solution set corresponds to the optimal operation scheme of the distributed energy equipment group.

4. The method according to claim 3, characterized in that, The method further includes: A dynamic search step size is constructed based on the number of iterations, which decreases exponentially with the number of iterations. A Gaussian mixture distribution with different scale standard deviations is constructed, and the mixture coefficients with different scale standard deviations are combined using an adaptively adjusted mixing coefficient according to the iteration process to obtain a Gaussian mixture perturbation vector. The distance distribution between solutions in the current solution set is calculated, and a density function reflecting the crowding degree of the solution space is established. The dynamic search step size, the Gaussian mixture perturbation vector, and the density function are multiplied to construct an enhanced local search operator. For the solution set optimized by the enhanced local search operator, the dominance level is obtained by calculating the number of other solutions that each solution is dominated by. The distance diversity index is calculated based on the minimum Euclidean distance between solutions. The target space distribution index is obtained by calculating the deviation of each solution from the average value in the target space. The dominance level, distance diversity index and target space distribution index are combined by weighting coefficients to construct an elite solution comprehensive scoring function. Based on the current population size of the particle swarm, the dynamic capacity limit of the elite solution set is linearly determined within the preset minimum and maximum capacity range; the solution set with the best dominance level and whose number does not exceed the dynamic capacity limit is selected to form the updated elite solution set. For the updated elite solution set, the position change of the solution between adjacent iterations is calculated, and a population convergence index is constructed based on the statistical characteristics of the position change. Non-dominated solutions that satisfy the population convergence index and are evenly distributed are selected from the updated elite solution set, and finally an optimal Pareto solution set with even distribution is formed.

5. The method according to claim 1, characterized in that, According to the optimal operating scheme, the control commands sent to each distributed energy device through the park energy management system include: Construct a set of device start / stop instructions, a set of power control instructions, and a set of parameter adjustment instructions. The set of device start / stop instructions includes start / stop status identifiers for each device, the set of power control instructions includes target power values ​​for each device, and the set of parameter adjustment instructions includes operating parameter vectors for each device. Based on the energy efficiency coefficient, response speed coefficient, and flexibility coefficient of each device, the energy efficiency coefficient, response speed coefficient, and flexibility coefficient are weighted and combined using a weighting coefficient to obtain the importance index of each device. The execution priority of the control commands of each device is determined according to the importance index. A device startup timing matrix is ​​established, wherein the matrix elements in the device startup timing matrix represent the startup response delay between two corresponding devices. Based on the device startup timing matrix, instruction execution time windows are divided, and the key execution time nodes within the time windows are determined sequentially according to the constraint relationship of the device startup timing matrix. Based on the importance index and startup timing matrix of each device, the start-stop status identifiers in the device start-stop instruction set are decomposed according to priority order to obtain the device start-stop instruction sequence. Similarly, the target power values ​​in the power control instruction set are decomposed according to priority order to obtain the power control instruction sequence. The device start / stop instruction sequence and power control instruction sequence are combined according to the divided time windows and key execution time nodes to form a phased control instruction containing the execution timing. The phased control instruction includes the start / stop status identifier, target power value and operating parameter vector of each device at each key time node.

6. A smart linkage system for distributed equipment groups in a park based on load characteristics, used to implement the method as described in any one of claims 1-5, characterized in that, include: The first unit is used to acquire real-time load data of multiple energy-consuming units within the park. The real-time load data includes electricity load data, heating load data, and cooling load data. The real-time load data is input into a pre-trained deep neural network model, which is trained based on historical load data and environmental parameter data to generate load characteristic curves of the multiple energy-consuming units. The load characteristic curves include peak-valley distribution characteristics, load change trend characteristics, and load correlation characteristics. The second unit is used to construct a multi-objective optimization model for the distributed energy equipment group in the park based on the load characteristic curve. The optimization objectives of the multi-objective optimization model include minimizing system operating costs and maximizing energy utilization efficiency. The constraints of the multi-objective optimization model include equipment operating parameter constraints, energy balance constraints, and environmental constraints. The improved particle swarm optimization algorithm is used to solve the multi-objective optimization model to obtain the optimal operating scheme of the distributed energy equipment group in the park. The optimal operating scheme includes the power generation of photovoltaic power generation equipment, the charging and discharging power of energy storage equipment, and the cooling power of electric cooling equipment. The improved particle swarm optimization algorithm adopts an adaptive local search strategy and an elite solution dynamic maintenance mechanism based on the particle swarm optimization algorithm. The third unit is used to send control commands to each distributed energy device through the park energy management system according to the optimal operation plan. The control commands include the device start / stop status and operating parameters. It monitors the operating status of each distributed energy device in real time. When an abnormal operation or load change is detected, it triggers a dynamic optimization and correction mechanism. The dynamic optimization and correction mechanism dynamically updates the optimal operation plan based on the rolling time domain to ensure the coordinated operation of the park's distributed energy device group and dynamic load balance.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images