Resource scheduling method for distributed energy system

Abstract

The invention relates to the technical field of data mining and intelligent scheduling, in particular to a distributed energy system-oriented resource scheduling method, which comprises the following steps of: acquiring historical data, and constructing a typical scene set according to the historical data; scene matching is carried out according to the current real-time data, each scheduling scheme of the current real-time data at least comprises a distributed power supply type, a grid-connected node position, capacity configuration and an energy storage charging and discharging time sequence, and each scheduling scheme is regarded as a whale individual; multi-subgroup collaborative search is introduced when the optimal resource scheduling scheme is searched based on the whale optimization algorithm; and selecting an individual which meets the power grid power balance constraint, the node voltage stability constraint, the energy storage charge state constraint and the renewable energy output power constraint and has the highest fitness as an optimal individual in the population. According to the method, multi-subgroup collaborative search, adaptive weight adjustment and dynamic variation mechanisms are introduced into the whale optimization algorithm, so that the global optimization ability and convergence stability of a nonlinear and multi-target scheduling problem are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of data mining and intelligent scheduling technology, specifically to a resource scheduling method for distributed energy systems. Background Technology

[0002] With the rapid development of new power systems and smart interconnected systems, the scale of multi-source distributed resources (including energy, computing, and manufacturing equipment) is continuously expanding, becoming a crucial support for achieving coordinated optimization of "source, grid, load, and storage" and flexible system operation. In distributed energy systems, common resource types include adjustable energy sources such as gas-fired combined cooling, heating, and power (CCHP) and micro gas turbines; fluctuating renewable energy sources such as photovoltaics and wind power; and electrochemical systems with energy storage capabilities. The heterogeneity, diversity, and dynamic coupling of these resources significantly increase the complexity of system operation and scheduling.

[0003] However, existing technologies still have significant shortcomings when dealing with the scheduling problem of multi-source heterogeneous resources:

[0004] (1) Insufficient data quality and consistency: There are duplicates, missing values ​​and outliers in multi-source time series data, and the sampling resolution is inconsistent, which seriously affects the training and prediction accuracy of the model.

[0005] (2) Poor representativeness of typical dispatch scenarios: Traditional methods generate scenarios based only on experience or single features, without fully considering the coupled effects of seasonal changes, electricity price fluctuations and extreme weather.

[0006] (3) Insufficient intelligence of scheduling algorithms: Most optimization algorithms cannot simultaneously take into account nonlinearity, multi-objective and uncertainty constraints, resulting in low solution efficiency and poor stability.

[0007] (4) Lack of cross-domain universality: Existing solutions are mostly designed for specific energy scenarios and lack universal methodologies that can be adapted to computing, manufacturing and other fields that require time-series scheduling. Summary of the Invention

[0008] To address the problems of unstable multi-source data quality, insufficient construction of typical scheduling scenarios, and low adaptability of optimization algorithms in existing systems, this invention proposes a resource scheduling method for distributed energy systems. This method treats each scheduling scheme as an individual whale and searches for the optimal scheduling scheme based on the whale algorithm. Specifically, it includes the following steps:

[0009] Acquire historical data and construct a set of typical scenarios based on the historical data, including at least high electricity price-high load scenarios, extreme weather fluctuation scenarios, and renewable energy high output scenarios;

[0010] Scenario matching is performed based on the current real-time data. Each scheduling scheme for the current real-time data includes at least the distributed power source type, grid connection node location, capacity configuration, and energy storage charging and discharging sequence. Each scheduling scheme is regarded as an individual whale.

[0011] When searching for the optimal resource scheduling scheme based on the whale optimization algorithm, a multi-subgroup collaborative search is introduced. In the high electricity price-high load scenario, the search is conducted with the minimum energy loss; in the extreme weather fluctuation scenario, the search is conducted with the minimum node voltage deviation index; and in the renewable high output scenario, the search is conducted with the maximum renewable energy penetration rate.

[0012] During the search, the individual with the highest fitness that satisfies the grid power balance constraint, node voltage stability constraint, energy storage state of charge constraint, and renewable energy output constraint is selected as the optimal individual in the population.

[0013] This invention introduces multi-subgroup collaborative search, adaptive weight adjustment and dynamic mutation mechanism into the whale optimization algorithm, which effectively improves the global optimization ability and convergence stability of nonlinear and multi-objective scheduling problems. Attached Figure Description

[0014] Figure 1 This is a detailed flowchart of a resource scheduling method for distributed energy systems according to the present invention;

[0015] Figure 2 This is a schematic diagram illustrating the preprocessing effect of typical renewable energy (solar power generation) output time series data according to the present invention.

[0016] Figure 3 This is a schematic diagram illustrating the preprocessing effect of typical renewable energy (wind power) output time series data according to the present invention.

[0017] Figure 4 This is a diagram showing the results of the optimized scheduling of renewable distributed energy resources driven by power data according to the present invention. Detailed Implementation

[0018] 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.

[0019] This invention proposes a resource scheduling method for distributed energy systems, treating each scheduling scheme as an individual whale and searching for the optimal scheduling scheme based on the whale algorithm, such as... Figure 1 Specifically, it includes the following steps:

[0020] Acquire historical data and construct a set of typical scenarios based on the historical data, including at least high electricity price-high load scenarios, extreme weather fluctuation scenarios, and renewable energy high output scenarios;

[0021] Scenario matching is performed based on the current real-time data. Each scheduling scheme for the current real-time data includes at least the distributed power source type, grid connection node location, capacity configuration, and energy storage charging and discharging sequence. Each scheduling scheme is regarded as an individual whale.

[0022] When searching for the optimal resource scheduling scheme based on the whale optimization algorithm, a multi-subgroup collaborative search is introduced. In the high electricity price-high load scenario, the search is conducted with the minimum energy loss; in the extreme weather fluctuation scenario, the search is conducted with the minimum node voltage deviation index; and in the renewable high output scenario, the search is conducted with the maximum renewable energy penetration rate.

[0023] Each search begins by filtering individuals with the highest fitness, selecting those that satisfy grid power balance constraints, node voltage stability constraints, energy storage state of charge constraints, and renewable energy output constraints as the optimal individuals in the population.

[0024] The data in this embodiment is based on platforms such as smart meters, Supervisory Control and Data Acquisition (SCADA) systems, and Distributed Energy Management Systems (DEMS), collecting multi-source time-series power data covering the entire year, including but not limited to load demand, photovoltaic and wind power output, energy storage status of charge, electricity price data, grid operating parameters, and meteorological factors.

[0025] The raw power data collected is processed to improve its quality, including but not limited to: removing duplicate records and redundant data; using an adaptive method based on the sliding standard deviation to detect anomalies in the power data; filling in missing data using wind-solar hybrid interpolation or reconstruction of historical similar samples; and using time series aggregation and downsampling methods to transform high-resolution data into hourly feature vectors to improve time series consistency and model usability.

[0026] This embodiment constructs feature vectors based on time dimensions (such as season, month, weekday / holiday labels), economic dimensions (such as time-of-use electricity price volatility, average electricity price), meteorological dimensions (such as temperature, irradiance, wind speed, humidity), and operational dimensions (such as daily load curve shape, renewable energy output volatility). Using K-means or DBSCAN clustering algorithms, it performs cluster analysis on multi-source operational data throughout the year, dividing the data into several classes. Each class represents a typical operating mode of the system under specific disturbance conditions. The clustering results form a set of typical scenarios, including but not limited to weekday scenarios, holiday scenarios, and extreme weather scenarios.

[0027] Preferably, the present invention includes at least high electricity price-high load scenarios, extreme weather fluctuation scenarios, and renewable high output scenarios.

[0028] The purpose of clustering is to identify the energy supply and demand balance characteristics of a system under different external disturbances and output statistical characteristic parameters for various typical scenarios (such as average load, average renewable energy output, electricity price level, and volatility). These parameters are used to calculate the fitness function in the subsequent scheduling phase, quantitatively reflecting the economic efficiency, safety, and environmental friendliness of the system operation under each scenario, thereby establishing a mapping relationship between typical scenarios and scheduling optimization objectives.

[0029] During the real-time operation phase of the system, real-time data streams (including current weather, electricity price, and load information) are acquired. The Euclidean distance between the real-time feature vector and the center vector of each typical scenario is calculated, and the most similar scenario in the typical scenario set is determined. Based on this, the system invokes the corresponding scenario's optimized scheduling strategy, realizing the automatic identification and real-time matching of typical scheduling scenarios, thereby improving the system's scheduling response capability under different operating conditions.

[0030] To address the constraints and parameters in the fitness function, this invention employs multi-timescale load and renewable energy output prediction, wherein:

[0031] Based on historical load, power output and meteorological data, LSTM or ST-GCN networks are used to achieve rolling forecasts for the next 1 to 24 hours, improving the foresight of short-term dispatch.

[0032] A CNN-LSTM hybrid model is used to extract time-periodic and spatially correlated features to achieve daily, weekly, and monthly trend predictions, reflecting seasonal changes and load structure evolution patterns.

[0033] Multi-step rolling prediction is achieved by using a sliding time window strategy, which enhances the model's ability to capture long-term time-series fluctuations and improves prediction stability and generalization performance.

[0034] The load curve and renewable energy output curve output by the forecasting module serve as key input parameters for subsequent dispatch optimization, constraining the energy balance equations and operational boundary conditions in the optimization model. Short-term forecast results directly impact real-time and intraday dispatching of unit output allocation, energy storage charging and discharging strategies, and demand response execution plans; medium- and long-term forecast results are used to optimize power generation plans, maintenance schedules, and energy storage capacity utilization plans. Through the linkage between the forecasting and typical scenario identification modules, a closed-loop coupling mechanism of "scenario-driven—forecast-guided—dispatch optimization" is achieved, enabling the system to achieve dynamic, forward-looking, and economical energy management at different time scales.

[0035] To achieve coordinated and optimized scheduling of multi-source distributed energy systems under uncertain conditions, this invention, based on the typical operating scenarios and multi-timescale prediction results obtained in the preceding steps, first determines the set of input characteristic parameters for each scenario, including time features (such as season, time period, weekday / holiday labels); meteorological features (such as temperature, irradiance, wind speed, humidity); economic features (such as time-of-use pricing, market clearing pricing); and system features (such as load forecasts, renewable energy output forecasts). Based on these multi-dimensional parameters, this invention uses the Monte Carlo simulation method to introduce uncertainty perturbations into the prediction results, obtaining sample sets under different perturbation levels. Each sample corresponds to a set of system operating states under renewable energy output, load demand, and electricity price conditions, thus forming a typical scheduling sample set. These samples not only reflect the operational uncertainties of the system under different scenarios but also provide distributed robust training and validation data support for the optimization model.

[0036] During the real-time operation phase, the system dynamically selects the fitness function based on the current scenario category identified by the scenario matching module, thereby achieving scenario-based adaptive adjustment of the scheduling target. For example, in the "high electricity price-high load" scenario, the system prioritizes reducing energy loss (preferably using formula (5) as the fitness function); in the "extreme weather fluctuation" scenario, it prioritizes maintaining voltage stability (preferably using formula (6) as the fitness function); and in the "high output of renewable energy" scenario, it prioritizes enhancing the absorption and energy storage smoothing capabilities of renewable energy (preferably using formula (7) as the fitness function).

[0037] After obtaining a typical scheduling sample set, a multi-source distributed energy grid connection and optimal scheduling model is established from three aspects: decision variables, system constraints, and objective function. Decision variables include, but are not limited to: distributed power source type, grid connection node location, capacity configuration, and energy storage charging and discharging timing. System constraints include, but are not limited to: grid power balance, node voltage stability, energy storage state of charge, and renewable energy output power.

[0038] (1)

[0039] (2)

[0040] (3)

[0041] (4)

[0042] Among them, G i For node device i, there is a collection of conventional and renewable distributed power sources connected to the grid. E represents the output power of the j-th renewable power generation unit in the equipment at time t. i For node device i, it is the set of energy storage devices connected to the grid. Let be the discharge power of device i at time t. The load power requirement of device i at node t; Let be the amount of power exchanged between node device i and its neighboring node device m at time t. Let i be the set of adjacent node devices; Let be the minimum voltage amplitude of node device i at time t; Let be the maximum voltage amplitude of node device i at time t; Let be the voltage amplitude of node device i at time t; This indicates the state of charge of the energy storage device at time t+1; To improve energy storage charging efficiency; The charging power of the energy stored in node device i at time t; The time step; For energy storage discharge efficiency; The discharge power of energy stored in node device i at time t; The minimum state of charge of the energy storage device at time t; The maximum value of the state of charge of the energy storage device at time t; The maximum charging power for energy storage of node device i; The maximum discharge power for storing energy in node device i; Let be the output power of the i-th renewable power generation unit in the equipment at time t. Let be the maximum available power of the i-th renewable power generation unit in the node device at time t.

[0043] The objective function (i.e., the fitness function of the scheduling algorithm) includes, but is not limited to, the system energy loss F. loss Minimize the node voltage deviation index F voltage Minimize the renewable energy penetration rate F renew Maximize, where:

[0044] (5)

[0045] (6)

[0046] (7)

[0047] in, This refers to system energy loss. This is a set of lines in the power grid topology, where each line consists of two node devices; Let be the resistance of the line between node device i and node device j; Let be the active power on the branch between node device i and node device j in the t-th iteration; Let be the reactive power on the branches of node device i and node device j in the t-th iteration; This refers to the node voltage deviation index. It is the set of device nodes in the system; This is the voltage reference value for the node device, typically 1 pu; For renewable energy penetration rate; This refers to the set of all renewable energy devices in the system.

[0048] This invention employs an intelligent optimization algorithm capable of solving nonlinear, multi-objective, and nonconvex problems, namely the Whale Algorithm (WOA), as the basis for the scheduling algorithm. Within the WOA framework, it introduces multi-subgroup cooperative search, an electricity-constrained adaptive weight adjustment strategy, and dynamic mutation and elite retention mechanisms to improve the algorithm's convergence stability and global search capability for nonlinear and multi-objective optimization problems. Specifically, it includes:

[0049] 101. In the whale optimization algorithm that incorporates multi-subgroup cooperative search, the position iteration process of the i-th individual whale in a certain subgroup includes:

[0050]

[0051] in, X represents the position of the i-th individual whale (in the distributed resource scheduling scheme) in the (t+1)-th iteration; best,k (t) represents the optimal individual in the k-th subgroup during the t-th iteration; A k and C k λ represents the control coefficient of the k-th subgroup in the Whale Optimization Algorithm, determined by the random number and shrinkage factor in the WOA algorithm; Xbest g(t) is the globally optimal individual (the scheduling scheme that satisfies the system constraints and has the best fitness value) among all subgroups at the t-th iteration; λ is the global information interaction weight between subgroups, used to adjust the balance between local search and global exploration; k=1,2,……,K are the subgroup numbers, and the initialization parameters are different for each subgroup. Those skilled in the art can set the initialization parameters based on experience or simulation data. This represents the position of the i-th individual whale in the t-th iteration; This indicates that the absolute value is being calculated.

[0052] Multi-subgroup cooperative search achieves global optimization and improved search diversity under multiple operating conditions through parallel subgroup evolution and global optimum sharing.

[0053] 102. The present invention provides a power constraint-guided adaptive weight adjustment strategy, which utilizes the power imbalance and energy storage state-of-charge deviation of the system at that moment to adjust the control coefficient A of the k-th subgroup in the whale optimization algorithm. kThe updated control parameters are then used to update the individual whale positions, with control coefficient A. k The update process is represented as follows:

[0054] (9)

[0055] in, The control coefficients are updated using the current power imbalance and energy storage state of charge deviation; η1 is the power balance feedback weight coefficient, which determines the response strength of the scheduling algorithm to the deviation when there is a deviation between the power generation and the load power; ΔP(t) is the system power imbalance, which describes the degree of energy balance deviation of the system at time t and is a dimensionless index that measures the difference between the power generation and the load demand; η2 is the energy storage constraint feedback weight coefficient, which determines the magnitude of the scheduling algorithm's contraction or adjustment of the search direction when the SOC deviates from the safe range; ΔSOC(t) is the energy storage state of charge deviation, which characterizes the degree of deviation between the current energy storage battery state of charge and the safe range.

[0056] The power constraint-oriented adaptive weight adjustment strategy adaptively adjusts the weight coefficients of the optimization scheduling algorithm by sensing the power constraint status in real time, so as to ensure that the system achieves efficient coordinated scheduling under power security constraints.

[0057] 103. This invention employs a dynamic mutation and elite preservation mechanism, that is, for the whale optimization algorithm, a mutation operation is performed on the i-th individual whale, and the position of the i-th individual after the mutation operation is represented as:

[0058] (10)

[0059] in, Let be the dynamic mutation intensity of the t-th iteration, which decreases as the number of iterations increases; It is a standard normally distributed random disturbance; p mut The probability of dynamic mutation; rand is a random number between [0,1]. Let represent the fitness function of the i-th whale individual at the t-th iteration; Let represent the fitness function of the globally optimal individual among all subgroups at the t-th iteration.

[0060] This invention employs an improved whale optimization algorithm to solve the distributed resource scheduling problem. Based on the standard whale optimization algorithm, the improved whale optimization algorithm innovatively integrates multi-subgroup cooperative search, power constraint-oriented adaptive weight adjustment strategy, dynamic mutation and elite retention mechanism for the multi-source heterogeneous distributed resource scheduling problem. This algorithm is used to achieve efficient optimization scheduling in dynamic and complex energy scheduling environments.

[0061] Figures 2-3 This paper demonstrates the effectiveness of time-series data preprocessing based on one year's worth of historical photovoltaic (PV) and wind power output data (collected every 15 minutes, totaling 35,040 data points). Using the preprocessing method proposed in this invention (including deduplication, outlier identification based on moving standard deviation, and outlier repair based on historical similar samples), the paper compares the differences in PV and wind power output curves before and after repair at 96 time points (every 15 minutes throughout the day) using three randomly selected outlier dates as examples. The repaired data curves reduce abrupt spikes caused by collection errors or missing data, improving data continuity and reliability, and providing stable data support for subsequent scheduling simulations.

[0062] After clarifying the scheduling scenario (load demand data and renewable energy output data) based on the preprocessing of time-series power data, the IEEE 33-node distribution network was selected as the test system, and a photovoltaic generator and energy storage device were connected to this distribution network. In this embodiment, the purpose of connecting renewable distributed energy to the distribution network is to optimize the daily active power loss and node voltage deviation of the power system (the two optimization objectives have no subjective decision-making preference). Figure 4 The results demonstrate the coordinated scheduling of sources (renewable distributed power sources), grid (main grid power supply), load (user load demand), and storage (energy storage) using an improved whale optimization algorithm. The scheduling results show that, under the premise of safe grid connection, renewable distributed power sources can operate in coordination with energy storage devices, thereby effectively improving the overall operational performance of the distribution network.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A resource scheduling method for a distributed energy system, characterized in that, Each scheduling scheme is regarded as a whale individual, and the optimal scheduling scheme is searched based on a whale algorithm, specifically including the following steps: Historical data is acquired, and a typical scenario set is constructed according to the historical data, including at least a high-price-high-load scenario, an extreme weather fluctuation scenario and a renewable high-output scenario; Each scheduling scheme is regarded as a whale individual, and the optimal scheduling scheme is searched based on a whale algorithm, specifically including the following steps: In the whale optimization algorithm with the multi-swarm collaborative search, the position iteration process of the i-th whale individual includes: In the high-price-high-load scenario, the minimum energy loss is searched, the fitness function is the inverse of the system energy loss, and the system energy loss is represented as: 2.The resource scheduling method for distributed energy system according to claim 1, wherein, In the extreme weather fluctuation scenario, the minimum node voltage deviation index is searched, the fitness function is the inverse of the node voltage deviation index, and the node voltage deviation index is represented as: in, X represents the position of the i-th individual whale in the (t+1)-th iteration; best,k (t) represents the optimal individual in the k-th subgroup during the t-th iteration; A k and C k denoted as the control coefficient of the k-th subgroup in the whale optimization algorithm; Xbest g(t) is the globally optimal individual among all subgroups at the t-th iteration; λ is the weight of global information interaction between subgroups. This represents the position of the i-th individual whale in the t-th iteration; This indicates that the absolute value is being calculated. 3.The resource scheduling method for distributed energy system according to claim 2, wherein, At each time, the control coefficient A of the kth subgroup in the whale optimization algorithm is updated by using the power imbalance degree and the state of charge deviation of the system at this time k The updating process is as follows: the control parameters are updated, the position of the whale individual is updated by using the updated control parameters, and the control coefficient A k ​ ； wherein, is the updated control coefficient using the current time power imbalance and the energy storage state of charge deviation; η1 is the power balance feedback weight coefficient; ΔP(t) is the system power imbalance; η2 is the energy storage constraint feedback weight coefficient; and ΔSOC(t) is the energy storage state of charge deviation.

4. The resource scheduling method for distributed energy systems according to claim 1 or 2, characterized in that, In the renewable high-output scenario, the maximum renewable energy penetration rate is searched, the fitness function is the renewable energy penetration rate, and the renewable energy penetration rate is represented as: wherein, is the system energy loss; is a set of lines in the grid topology, each line consisting of two node devices; is the resistance of the line between node device i and node device j; is the real power on the branch between node device i and node device j at iteration t; is the reactive power on the branch between node device i and node device j at iteration t; is the voltage magnitude of node device i at time t.

5. The method of claim 1 or 2, wherein, The power balance constraint of the power grid is represented as: wherein, is a node voltage deviation index; is a set of device nodes in the system; is a voltage magnitude of the node device i at time t; is a voltage reference value of the node device.

6. The resource scheduling method for distributed energy system according to claim 1 or 2, characterized in that, The node voltage stability constraint is represented as: wherein, is the renewable energy penetration rate; is the set of all renewable energy devices in the system; is the power output of the i-th node device at time t; is the set of device nodes in the system; is the power output of the j-th node device at time t.

7. The method of claim 1, wherein, The energy storage state of charge constraint is represented as: The renewable energy output power constraint is represented as: For the whale optimization algorithm, the mutation operation is performed on the i-th whale individual, and the position of the i-th whale individual after the mutation operation is updated as: The scenario matching according to the current real-time data includes: wherein, G i is the set of conventional and renewable distributed power sources connected to node device i at time t; is the output power of the jth renewable power generation unit in node device i at time t; E i is the set of energy storage devices connected to node device i at time t; is the discharge power of node device i at time t; is the load power demand of node device i at time t; is the power exchange of node device i to its adjacent node device m at time t, is the set of adjacent node devices of node device i; is the minimum voltage amplitude of node device i at time t; is the maximum voltage amplitude of node device i at time t; is the voltage amplitude of node device i at time t; denotes the state of charge of the energy storage device at time t+1; is the energy storage charging efficiency; is the charging power of the energy storage of node device i at time t; is the time step; is the energy storage discharging efficiency; is the discharging power of the energy storage of node device i at time t; is the minimum value of the state of charge of the energy storage device at time t; is the maximum value of the state of charge of the energy storage device at time t; is the maximum charging power of the energy storage of node device i; is the maximum discharging power of the energy storage of node device i; is the output power of the ith renewable power generation unit in node device i at time t; is the available maximum power of the ith renewable power generation unit in node device i at time t.

8. The resource scheduling method for distributed energy systems according to claim 1 or 2, characterized in that, The data of the system for a whole year is clustered, and is at least clustered into a high-price-high-load scenario, an extreme weather fluctuation scenario and a renewable high-output scenario, and the characteristics in the clustering include time characteristics, economic characteristics, weather characteristics and operation characteristics; where, denotes the position of the i-th whale individual at the t-th iteration; is the dynamic variation strength for the t-th iteration, which decreases with the increase of iteration number; is a standard normal distribution random disturbance; p mut is a dynamic variation probability; rand is a random number between [0, 1]; is the global optimal individual among all subgroups at the t-th iteration; denotes the fitness function of the i-th whale individual at the t-th iteration; denotes the fitness function of the global optimal individual among all subgroups at the t-th iteration. 9.The method of claim 1, wherein, When the real-time data is matched, the sum of the Euclidean distances of the current time characteristics, economic characteristics, weather characteristics and operation characteristics and each characteristic in each sample in each scenario is calculated, and the scenario with the shortest distance is matched for the current real-time data. ​ ​

Images