Energy management method, device and system of distributed energy system, and storage medium

Abstract

The embodiment of the invention provides an energy management method, device and system for a distributed energy system, and a storage medium, and relates to the technical field of distributed energy systems and energy management. The method comprises the following steps: based on historical operation data, extracting a typical load characteristic mode through clustering analysis; based on the typical load characteristic mode and external prediction information, performing short-term load prediction through a neural network model; establishing a multi-objective optimization model comprising an economical objective and an environmental objective, wherein decision variables of the optimization model at least comprise output plans of controllable energy units in the distributed energy system; and based on a short-term load prediction result, an optimization algorithm is adopted to solve the multi-objective optimization model, and a scheduling scheme of the distributed energy system is generated. The method is beneficial for improving the utilization rate of renewable energy sources, promoting effective balance between operation cost and environmental emission, enhancing the capability of an algorithm to jump out of local optimum, and improving the overall operation performance of a distributed energy system.

Description

Technical Field

[0001] This application relates to the field of distributed energy systems and energy management technology, specifically to an energy management method and apparatus for a distributed energy system, a distributed energy system, and a storage medium. Background Technology

[0002] Currently, with the increasing penetration of renewable energy in distributed energy systems, the operation of these systems exhibits significant randomness, intermittency, and multi-source coupling characteristics. Traditional centralized energy management models struggle to adapt to the volatility and dispersion of distributed energy, leading to low system efficiency, insufficient power supply reliability, and difficulty in fully utilizing local renewable energy sources. Especially in scenarios with high proportions of photovoltaic (PV) integration, rapid changes in sunlight conditions often cause power supply-demand imbalances, posing severe challenges to grid security and economic operation.

[0003] Existing technologies employ local control strategies based on rules or empirical thresholds, or introduce optimized scheduling models with a single economic objective. For example, basic power balance and operating cost control are achieved by setting priority orders for photovoltaic power output, fixed thresholds for energy storage charging and discharging, or start-stop conditions for diesel generators. Some methods also attempt to use optimization algorithms, such as linear programming or dynamic programming, to optimize a certain operating objective of the system in order to improve economic efficiency.

[0004] In the process of implementing the embodiments of this application, at least the following problems were found in the related technology:

[0005] The adoption of related technologies has enhanced the system's autonomous operation capability to some extent and reduced its dependence on the external power grid. However, in practical applications, these technologies often focus on optimizing a single objective or rely on simplified models and empirical rules, failing to fully consider the coupling relationship between load forecasting accuracy, multi-objective coordination, and global optimization capabilities. This results in scheduling schemes often struggling to achieve an effective balance between economy, environmental protection, and renewable energy utilization, and are prone to getting trapped in local optima, thus limiting further improvement in the overall performance of distributed energy systems. Summary of the Invention

[0006] This application provides an energy management method and apparatus for a distributed energy system, a distributed energy system, and a storage medium.

[0007] A first aspect of this application provides an energy management method for a distributed energy system, comprising:

[0008] Based on historical operational data, typical load characteristic patterns are extracted through cluster analysis;

[0009] Short-term load forecasting is performed using a neural network model based on typical load characteristic patterns and external forecasting information.

[0010] Establish a multi-objective optimization model that includes both economic and environmental objectives. The decision variables of the optimization model should include at least the output plans of each controllable energy unit in the distributed energy system.

[0011] Based on short-term load forecasting results, an optimization algorithm is used to solve the multi-objective optimization model and generate a scheduling scheme for the distributed energy system.

[0012] In one optional embodiment of this application, the distributed energy system includes at least one of a photovoltaic power generation unit, an energy storage unit, a diesel generator, a public grid interface, and a load unit; and / or, the controllable energy unit includes at least one of an energy storage unit, a diesel generator, and a public grid interface.

[0013] In an optional embodiment of this application, a multi-objective optimization model including economic and environmental objectives is established, comprising:

[0014] Construct an operating cost objective function, which includes at least two of the following: purchased electricity cost, diesel generator fuel cost, energy storage unit loss cost, and renewable energy curtailment penalty cost;

[0015] An emission target function is constructed to quantify the pollutant emissions generated by the operation of diesel generators and the purchase of electricity from the public grid.

[0016] In one optional embodiment of this application, the renewable energy curtailment penalty cost is related to the curtailment power of the photovoltaic power generation unit; and / or, the pollutants quantified by the emission objective function include at least one of carbon dioxide, sulfur dioxide, and nitrogen oxides.

[0017] In an optional embodiment of this application, establishing a multi-objective optimization model that includes economic and environmental objectives further includes:

[0018] Establish system power balance constraints to ensure that at least one of the outputs from photovoltaic power generation units, energy storage units, diesel generators, and public grid interfaces is balanced with load demand and power curtailment; and / or,

[0019] Establish upper and lower limits for the output of each energy unit; and / or,

[0020] Establish operating constraints for energy storage units, which include at least one of the following: state of charge constraints, charge / discharge power constraints, and state of charge update model.

[0021] In an optional embodiment of this application, typical load characteristic patterns are extracted based on historical operating data through cluster analysis, including:

[0022] Clustering algorithms are used to cluster historical daily data sequences, which must include at least electricity price sequences, outdoor temperature sequences, and load sequences.

[0023] Each cluster center sequence obtained after clustering is output as a typical load characteristic pattern.

[0024] In one optional embodiment of this application, the clustering algorithm is the K-means algorithm; and / or, Euclidean distance is used as a similarity measure; and / or, the change in cluster centers is less than a preset threshold is used as a convergence condition.

[0025] In an optional embodiment of this application, before clustering the historical daily data sequence using a clustering algorithm, the method further includes:

[0026] The data in each dimension of the historical daily data series are normalized; the historical daily data series also includes features used to identify date types or seasons.

[0027] In an optional embodiment of this application, short-term load forecasting is performed using a neural network model based on typical load characteristic patterns and external forecasting information, including:

[0028] Based on the predicted electricity price, predicted outdoor temperature, typical load characteristic patterns of the corresponding period, and historical load data at the same time, the input feature vector of the neural network is constructed.

[0029] The input feature vector is fed into the feedforward neural network model;

[0030] Receive the predicted load value output by the feedforward neural network model.

[0031] In an optional embodiment of this application, the load data at the same historical time includes load data at the same time on the day before the prediction date, and load data at the same time of the same type on the week before the prediction date; and / or, the activation function of the hidden layer of the feedforward neural network adopts the Sigmoid function or the ReLU function.

[0032] In an optional embodiment of this application, based on short-term load forecasting results, an optimization algorithm is used to solve a multi-objective optimization model to generate a dispatch scheme for the distributed energy system, including:

[0033] Use short-term load forecasting results as input parameters for load demand in the multi-objective optimization model;

[0034] The initial solution population is formed by using the power output plans of each controllable energy unit in each future time period as decision variables;

[0035] Iterative solutions are obtained based on a one-to-one optimization algorithm. In each iteration, the population is updated by a one-to-one comparison and relocation mechanism between solution vectors, based on the objective function value and the constraint satisfaction.

[0036] When the iteration termination condition is met, the optimal solution vector in the current population is output as the scheduling scheme.

[0037] In an optional embodiment of this application, the one-to-one optimization algorithm evaluates the solution vector based on the objective function value and the degree of violation of system operation constraints during population updates.

[0038] In an optional embodiment of this application, after performing short-term load forecasting using a neural network model based on typical load characteristic patterns and external forecasting information, the method further includes:

[0039] The prediction results are evaluated using prediction accuracy evaluation indicators to quantify prediction performance; among them, prediction accuracy evaluation indicators include at least two of the following: mean absolute error, root mean square error, mean absolute percentage error, and coefficient of determination.

[0040] A second aspect of this application provides an energy management device for a distributed energy system, including a processor and a memory storing program instructions. The processor is configured to execute the energy management method for a distributed energy system as described in the first aspect of this application when running the program instructions.

[0041] A third aspect of this application provides a distributed energy system, including:

[0042] The distributed energy system itself; and,

[0043] The energy management device of the distributed energy system, as described in the second aspect of the embodiments of this application, is installed on the distributed energy system body.

[0044] A fourth aspect of this application provides a computer-readable storage medium storing program instructions that, when executed, cause a computer to perform an energy management method for a distributed energy system as described in the first aspect of this application.

[0045] The energy management method and apparatus, distributed energy system, and storage medium of the distributed energy system provided in the embodiments of this application have the following beneficial effects:

[0046] This application's embodiments extract typical load characteristic patterns based on historical operating data and using cluster analysis, providing a more representative input basis for subsequent load forecasting and thus helping to improve the accuracy of load forecasting. Furthermore, combining these typical load characteristic patterns with external forecasting information, short-term load forecasting using a neural network model can better capture the complex nonlinear relationship between load and various influencing factors, further improving forecast accuracy and providing a reliable basis for optimized scheduling. Based on this, a multi-objective optimization model that simultaneously includes economic and environmental objectives is established, incorporating operating costs and pollutant emissions into the optimization framework, enabling scheduling decisions to balance economic efficiency and environmental protection. Finally, based on the short-term load forecast results and using an optimization algorithm to solve this multi-objective optimization model, it can collaboratively optimize the output plans of each controllable energy unit while satisfying system constraints, generating a better scheduling scheme in a global or near-global sense. This helps to improve the utilization rate of renewable energy while promoting an effective balance between operating costs and environmental emissions, enhancing the algorithm's ability to escape local optima, and improving the overall operating performance of the distributed energy system. Attached Figure Description

[0047] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0048] Figure 1 This is a schematic diagram of an energy management method for a distributed energy system provided in an embodiment of this application;

[0049] Figure 2 This is a schematic diagram of another energy management method for a distributed energy system provided in an embodiment of this application;

[0050] Figure 3 This is a schematic diagram of an energy management device for a distributed energy system provided in an embodiment of this application.

[0051] Figure label:

[0052] 800: Energy management device for distributed energy systems; 801: Processor; 802: Memory; 803: Communication interface; 804: Bus. Detailed Implementation

[0053] To make the technical solutions and advantages of the embodiments of this application clearer, the exemplary embodiments of this application will be described in further detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not an exhaustive list of all embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0054] Figure 1 and Figure 2 This is a schematic diagram of an energy management method for a distributed energy system provided in an embodiment of this application. Any of the following methods can be executed in the distributed energy system or in a server or terminal device that is connected to the distributed energy system.

[0055] Combination Figure 1 As shown in the figure, this application provides an energy management method for a distributed energy system, including:

[0056] S01, based on historical operating data, extracts typical load characteristic patterns through cluster analysis.

[0057] S02, based on typical load characteristic patterns and external forecast information, performs short-term load forecasting through a neural network model.

[0058] S03. Establish a multi-objective optimization model that includes economic and environmental objectives. The decision variables of the optimization model shall include at least the output plans of each controllable energy unit in the distributed energy system.

[0059] S04. Based on the short-term load forecast results, an optimization algorithm is used to solve the multi-objective optimization model to generate a scheduling scheme for the distributed energy system.

[0060] The energy management method for distributed energy systems provided in this application extracts typical load characteristic patterns based on historical operating data and using cluster analysis, providing a more representative input basis for subsequent load forecasting and thus improving the accuracy of load forecasting. Furthermore, combining these typical load characteristic patterns with external forecasting information, a neural network model is used for short-term load forecasting, which can better capture the complex nonlinear relationship between load and various influencing factors, further improving forecast accuracy and providing a reliable basis for optimized scheduling. Based on this, a multi-objective optimization model is established that simultaneously includes economic and environmental objectives, incorporating operating costs and pollutant emissions into the optimization framework, enabling scheduling decisions to balance economic efficiency and environmental protection. Finally, based on the short-term load forecast results and using an optimization algorithm to solve this multi-objective optimization model, it can collaboratively optimize the output plans of each controllable energy unit while satisfying system constraints, generating a better scheduling scheme in a global or near-global sense. This helps to improve the utilization rate of renewable energy while promoting an effective balance between operating costs and environmental emissions, enhancing the algorithm's ability to escape local optima, and improving the overall operating performance of the distributed energy system.

[0061] In one optional embodiment of this application, the distributed energy system includes at least one of a photovoltaic power generation unit, an energy storage unit, a diesel generator, a public grid interface, and a load unit; and / or, the controllable energy unit includes at least one of an energy storage unit, a diesel generator, and a public grid interface.

[0062] In this way, by configuring the distributed energy system to include at least one of photovoltaic power generation units, energy storage units, diesel generators, a public grid interface, and load units, and limiting the controllable energy units to at least one of energy storage units, diesel generators, and public grid interfaces, the system can integrate renewable energy, dispatchable backup power, and grid interaction capabilities. This provides a diversified physical basis for subsequent short-term forecasting and multi-objective optimization based on typical load characteristic patterns. Photovoltaic power generation units directly utilize solar energy, reducing dependence on traditional energy sources. Energy storage units smooth supply and demand fluctuations by storing and releasing energy. Diesel generators serve as reliable backups to enhance system stability. The public grid interface supports electricity trading to utilize time-of-use pricing. Load units, as demand-side components, are managed. The coordinated configuration of these units allows the system to flexibly adjust the output plans of each controllable energy unit during operation. This creates conditions for extracting typical load characteristic patterns through cluster analysis, neural network prediction, and multi-objective optimization to solve scheduling schemes. Furthermore, it helps to comprehensively consider economic and environmental objectives in the optimization model, achieving an effective balance between operating costs and environmental emissions.

[0063] In an optional embodiment of this application, a multi-objective optimization model including economic and environmental objectives is established, including: constructing an operating cost objective function, which includes at least two of the following: purchased electricity cost, diesel generator fuel cost, energy storage unit loss cost, and renewable energy curtailment penalty cost; and constructing an emission objective function, which quantifies the pollutant emissions generated by the operation of the diesel generator and the purchase of electricity from the public grid.

[0064] By constructing the operating cost objective function to include at least two of the following: purchased electricity cost, diesel generator fuel cost, energy storage unit loss cost, and renewable energy curtailment penalty cost, this model can comprehensively reflect the economic impact of each controllable energy unit (such as the public grid interface, diesel generators, and energy storage units) in a distributed energy system. Purchased electricity cost is related to grid interaction expenditures; diesel generator fuel cost is related to backup power fuel consumption; energy storage unit loss cost is related to the depletion of energy storage equipment's cycle life; and renewable energy curtailment penalty cost incentivizes the priority consumption of photovoltaic power generation units through an economic penalty mechanism. This comprehensively reduces the overall system operating cost and improves renewable energy utilization during the optimization process. Simultaneously, by constructing an emission objective function specifically quantifying the pollutant emissions generated from diesel generator operation and electricity purchased from the public grid, environmental impact factors are directly incorporated into the optimization framework, enabling dispatch decisions to consider pollutant reduction targets. The combination of these two objective functions allows the multi-objective optimization model to collaboratively consider economic and environmental factors during subsequent solutions. Under the premise of satisfying constraints such as system power balance, it automatically seeks a balance between operating costs and emission levels, thereby contributing to the generation of more economical and environmentally friendly dispatch schemes.

[0065] In one optional embodiment of this application, the renewable energy curtailment penalty cost is related to the curtailment power of the photovoltaic power generation unit; and / or, the pollutants quantified by the emission objective function include at least one of carbon dioxide, sulfur dioxide, and nitrogen oxides.

[0066] By explicitly linking the cost of renewable energy curtailment penalties to the curtailed power of photovoltaic (PV) power generation units, a direct economic adjustment mechanism is introduced into the operating cost objective function. This allows the optimization model to impose clear cost penalties on behaviors that reduce or waste PV power during the solution process. Consequently, the optimization algorithm tends to generate scheduling schemes that prioritize the use of PV power generation and reduce unnecessary curtailment. This helps improve the actual utilization rate of PV power generation units and reduce dependence on the public power grid or diesel generators. By specifying that the pollutants quantified in the emission objective function include at least one of carbon dioxide, sulfur dioxide, and nitrogen oxides, the assessment of environmental impact becomes more specific and comprehensive, covering major greenhouse gases and air pollutants. This ensures that the environmental protection goals pursued in the optimization process are not limited to a single emission indicator but comprehensively consider the different environmental impacts of multiple pollutants. This makes the final scheduling scheme more comprehensive and has more practical environmental benefits in reducing pollutant emissions. The combination of these two features refines the multi-objective optimization model from both economic incentive and environmental constraint dimensions, enabling it to more accurately guide the system operation in the direction of improving renewable energy consumption and reducing overall environmental hazards.

[0067] In an optional embodiment of this application, the multi-objective optimization model, which includes economic and environmental objectives, is further comprising: establishing system power balance constraints to ensure that at least one of the outputs of the photovoltaic power generation unit, energy storage unit, diesel generator, and public grid interface is balanced with load demand and curtailed power; and / or establishing upper and lower limit constraints on the output of each energy unit; and / or establishing energy storage unit operation constraints, wherein the energy storage unit operation constraints include at least one of state of charge constraints, charge and discharge power constraints, and state of charge update model.

[0068] In this way, by establishing system power balance constraints, the output of photovoltaic power generation units, energy storage units, diesel generators, and public grid interfaces is ensured to remain balanced with load demand and curtailed power at all times. This provides a basic energy conservation foundation for optimized scheduling and prevents system instability caused by power shortages or surpluses. By establishing upper and lower limit constraints on the output of each energy unit, the output of photovoltaic power generation units, energy storage units, diesel generators, and public grid interfaces is limited to their respective safe operating ranges, avoiding equipment overload or underload operation and ensuring equipment lifespan and operational safety. By establishing operational constraints for energy storage units, including at least one of state-of-charge constraints, charge / discharge power constraints, and state-of-charge update models, it is ensured that energy storage units operate within a safe state-of-charge range, charge / discharge power is controlled, and their energy state changes can be accurately simulated, thereby preventing overcharging and over-discharging and maintaining the availability and durability of energy storage units. The establishment of the above constraints enables the multi-objective optimization model to strictly follow the physical limitations and safety requirements of the actual system during subsequent solutions, thus generating a scheduling scheme that satisfies both operational feasibility and economic and environmental objectives.

[0069] In an optional embodiment of this application, typical load characteristic patterns are extracted based on historical operating data through cluster analysis, including: clustering historical daily data sequences using a clustering algorithm, wherein the historical daily data sequences include at least an electricity price sequence, an outdoor temperature sequence, and a load sequence; and outputting each cluster center sequence as a typical load characteristic pattern.

[0070] Thus, a clustering algorithm is used to cluster historical daily data sequences that contain at least electricity price sequences, outdoor temperature sequences, and load sequences. Electricity price and outdoor temperature are key external factors affecting load changes, and the combination of these three factors can comprehensively reflect the actual fluctuation patterns of load under different operating conditions. The clustering algorithm can group historical daily data with similar change characteristics into one category, removing redundant information and random fluctuations from the data. Furthermore, outputting the central sequence of each cluster as a typical load characteristic pattern allows for the extraction of common patterns in load changes under different operating conditions, forming a representative load characteristic template. This avoids analytical biases caused by using scattered and disorganized raw historical data, providing a more accurate and targeted input foundation for short-term load forecasting based on typical load characteristic patterns, while reducing the complexity of subsequent data processing.

[0071] In one optional embodiment of this application, the clustering algorithm is the K-means algorithm; and / or, Euclidean distance is used as a similarity measure; and / or, the change in cluster centers is less than a preset threshold is used as a convergence condition.

[0072] Thus, by employing the K-means algorithm to cluster historical daily data sequences, leveraging its simplicity and high computational efficiency, massive historical daily data sequences can be quickly divided into several categories. Using Euclidean distance as a similarity metric, the geometric distance between multidimensional data points composed of electricity price sequences, outdoor temperature sequences, and load sequences can be intuitively calculated, effectively distinguishing the similarity of energy consumption patterns on different dates. By using a convergence condition where the change in cluster centers is less than a preset threshold, the clustering iteration process can be ensured to terminate stably, avoiding unlimited oscillations and thus obtaining a stable and representative sequence of cluster centers. The combination of these algorithmic features makes the clustering process more efficient and the results more stable, thus helping to extract typical load characteristic patterns that better reflect the common laws under different combinations of electricity prices, temperatures, and loads. This provides a higher-quality, less noisy input basis for subsequent short-term load forecasting based on this pattern.

[0073] In an optional embodiment of this application, before clustering the historical daily data sequence using a clustering algorithm, the method further includes: normalizing the data of each dimension in the historical daily data sequence; wherein the historical daily data sequence also includes features for identifying date type or season.

[0074] By normalizing the data across all dimensions in the historical daily data series, the differences in units and ranges between electricity price, outdoor temperature, and load series are eliminated. This allows subsequent clustering algorithms to fairly consider the influence of each dimension, preventing any single dimension from dominating the clustering results due to its large absolute value. This helps extract more accurate typical load characteristic patterns that reflect the inherent similarity between data points. Furthermore, adding features to the historical daily data series to identify date type or season provides crucial contextual classification information for the clustering process. This enables the algorithm to more effectively group date data with the same weekday / weekend attributes or belonging to the same season. Consequently, the extracted typical load characteristic patterns are not only based on numerical similarity but also incorporate time-cycle patterns, better representing real load behavior under different scenarios. This provides a more accurate and generalizable input foundation for subsequent short-term load forecasting based on this pattern.

[0075] In an optional embodiment of this application, short-term load forecasting is performed using a neural network model based on typical load characteristic patterns and external forecasting information. This includes: constructing an input feature vector for the neural network based on the predicted electricity price for the forecast period, the predicted outdoor temperature, the typical load characteristic patterns for the corresponding period, and load data from the same historical time; inputting the input feature vector into the feedforward neural network model; and receiving the predicted load value output by the feedforward neural network model.

[0076] In this way, by constructing the input feature vector of the neural network based on the predicted electricity price, predicted outdoor temperature, typical load characteristic patterns for the corresponding period, and load data from the same historical time, multiple key factors influencing load changes can be integrated. The predicted electricity price reflects the economic guidance of electricity market price signals on future electricity consumption behavior; the predicted outdoor temperature represents the direct impact of meteorological conditions on load; the typical load characteristic patterns extracted from historical data clustering provide a baseline load profile for the period under similar scenarios; and the load data from the same historical time captures the short-term cycles and recent changes in load itself. Using these multi-source heterogeneous features as input to the feedforward neural network model allows the model to simultaneously learn and capture the complex nonlinear coupling relationships between load and electricity price, temperature, historical patterns, and recent behavior. Furthermore, by inputting the constructed input feature vector into the feedforward neural network model and receiving its output predicted load value, the powerful nonlinear fitting capability of the neural network can be utilized to comprehensively process the above-mentioned multi-dimensional features, thereby generating more accurate short-term load forecast results than traditional methods that rely on single or a few factors.

[0077] In an optional embodiment of this application, the load data at the same historical time includes load data at the same time on the day before the prediction date, and load data at the same time of the same type on the week before the prediction date; and / or, the activation function of the hidden layer of the feedforward neural network adopts the Sigmoid function or the ReLU function.

[0078] By specifically defining historical load data at the same time as the prediction date, including load data at the same time the day before the prediction date, and load data at the same time of the same type of day in the week before the prediction date, the neural network can simultaneously capture two key temporal patterns of load changes. Data at the same time the day before the prediction date reflects the latest, short-term trends and inertia of load changes, while data of the same type of day in the previous week (e.g., all weekdays or weekends) reflects the long-term habitual patterns of load on a weekly basis. Combining these two provides a more comprehensive historical reference for prediction, helping to improve the predictive model's ability to characterize the temporal features of load. By specifying that the hidden layer activation function of the feedforward neural network should use either the Sigmoid function or the ReLU function, a key nonlinear transformation capability is introduced to the model. The Sigmoid function can smoothly map the neuron output to a specific interval, helping the model learn complex nonlinear patterns, while the ReLU function can effectively alleviate the gradient vanishing problem in deep network training and accelerate the model's training convergence speed. The above methods enhance both the temporal structure of the input data and the processing capabilities within the model, enabling the feedforward neural network model to more effectively learn and fuse the complex nonlinear relationships between multi-source input features, thereby helping to generate more accurate short-term load forecasts.

[0079] In an optional embodiment of this application, based on short-term load forecasting results, an optimization algorithm is used to solve a multi-objective optimization model to generate a scheduling scheme for a distributed energy system. This includes: using the short-term load forecasting results as input parameters for load demand in the multi-objective optimization model; initializing an initial solution population using the output plans of each controllable energy unit in future time periods as decision variables; iteratively solving the solution using a one-to-one optimization algorithm, updating the population in each iteration based on the objective function value and constraint satisfaction through a one-to-one comparison and relocation mechanism between solution vectors; and outputting the optimal solution vector in the current population as the scheduling scheme when the iteration termination condition is met.

[0080] By using short-term load forecasts obtained from neural networks as input parameters for load demand in the multi-objective optimization model, the optimization process is based on predictions that more closely reflect future actual energy consumption, thus improving the foresight and applicability of the scheduling scheme. Initializing the initial solution population by using the output plans of each controllable energy unit (such as energy storage units, diesel generators, and public grid interfaces) for future time periods as decision variables provides a clear search space and feasible starting point for subsequent intelligent optimization algorithms. Iterative solving based on a one-to-one optimization algorithm, utilizing its mechanism of updating the population characteristics in each iteration based on the objective function value (including economic and environmental objectives) and constraint satisfaction, through one-to-one comparison and relocation between solution vectors, enables the algorithm to perform efficient global exploration and refined local search within the vast solution space. This helps prevent the population from prematurely converging to suboptimal solutions, thereby enhancing the algorithm's ability to escape local optima. Finally, when the iteration termination condition is met, the optimal solution vector in the current population is output as the scheduling scheme.

[0081] In an optional embodiment of this application, the one-to-one optimization algorithm evaluates the solution vector based on the objective function value and the degree of violation of system operation constraints during population updates.

[0082] Combination Figure 2 As shown in the embodiment of this application, another energy management method for a distributed energy system is provided, including:

[0083] S01, based on historical operating data, extracts typical load characteristic patterns through cluster analysis.

[0084] S02, based on typical load characteristic patterns and external forecast information, performs short-term load forecasting through a neural network model.

[0085] S21. The prediction results are evaluated using prediction accuracy evaluation indicators to quantify prediction performance; wherein the prediction accuracy evaluation indicators include at least two of the following: mean absolute error, root mean square error, mean absolute percentage error, and coefficient of determination.

[0086] S03. Establish a multi-objective optimization model that includes economic and environmental objectives. The decision variables of the optimization model shall include at least the output plans of each controllable energy unit in the distributed energy system.

[0087] S04. Based on the short-term load forecast results, an optimization algorithm is used to solve the multi-objective optimization model to generate a scheduling scheme for the distributed energy system.

[0088] The energy management method for distributed energy systems provided in this application employs at least two prediction accuracy evaluation indicators, including mean absolute error, root mean square error, mean absolute percentage error, and coefficient of determination, to evaluate short-term load forecasting results obtained based on typical load characteristic patterns and neural network models. This allows for a comprehensive and quantitative measurement of prediction performance from multiple dimensions. Mean absolute error reflects the average absolute level of prediction error; root mean square error is more sensitive to larger prediction deviations; mean absolute percentage error reflects the proportion of prediction error relative to the actual value; and the coefficient of determination can be used to evaluate the overall goodness of fit between the prediction curve and the actual curve. This multi-indicator comprehensive evaluation mechanism ensures that the performance of the prediction model no longer depends on a single evaluation standard, more comprehensively exposing the advantages and disadvantages of the prediction model in different aspects. This provides clear directions and basis for model selection, parameter adjustment, or subsequent optimization, helping to ensure higher reliability and accuracy of load forecasting data input into the multi-objective optimization model.

[0089] In practical applications, before establishing a multi-objective optimization model that includes economic and environmental objectives, it is necessary to first perform mathematical modeling of the distributed energy system to provide an accurate physical description basis for the optimization problem.

[0090] The specific modeling process includes establishing the input-output relationship of each energy unit and the overall power balance relationship of the system.

[0091] For renewable energy power generation units, such as photovoltaic (PV) power generation units, an output model is established. In one implementation, the output power of the PV power generation unit is represented by an empirical model related to irradiance and temperature, as shown in the formula:

[0092]

[0093] Among them, P PV P represents the actual output power of the photovoltaic array under current operating conditions. ref The rated output power of the photovoltaic array under standard test conditions; K t T is the temperature coefficient of a photovoltaic module, usually a negative value, representing the proportion of power output reduction caused by an increase in module temperature; amb G represents ambient temperature; T represents current solar irradiance; G represents ambient temperature; T represents current solar irradiance. ref For reference temperature, 25℃ is generally taken; G ref For reference irradiance, 1000 W / m is generally used. 2 .

[0094] For controllable standby power generation units, such as diesel generators, a fuel consumption model is established. In one implementation, the fuel consumption of the diesel generator is described using a linear approximation model, with the following formula:

[0095] Q(t)=αP(t)+βP r

[0096] Where Q(t) is the fuel consumption of the diesel generator at time t (e.g., L / h); α is the fuel consumption coefficient proportional to the active power output; β is the fixed fuel consumption coefficient related to the rated capacity; P(t) is the actual active power output of the diesel generator at time t; P r This refers to the rated power of the diesel generator.

[0097] For energy storage units, such as battery energy storage systems, energy state update models are established for both charging and discharging states. In the charging state, the battery power / energy update formula is:

[0098]

[0099] Under discharge conditions, the battery power / energy update formula is:

[0100]

[0101] in, and P represents the state of energy (or equivalent power) of the battery at time t under charging or discharging conditions; BAT (t-1) represents the battery's state of energy or equivalent power at the previous moment; σ represents the battery's self-discharge rate; N PV P represents the number of photovoltaic modules connected in parallel; PV (t) represents the photovoltaic output power at time t; η CON For DC / DC or DC / AC converter efficiency; P Load (t) represents the total system load power; η INV Inverter efficiency; η BAT Battery charging efficiency (represented by discharge efficiency in the discharge formula).

[0102] For each load unit, its structural model is established. The total load power is obtained by adding the power of the flexible load and the power of the critical load, as shown in the formula:

[0103] P t,Load =P t,flexible +P t,essential

[0104] Among them, P t,Load P represents the total load power at time t. t,flexible For movable and dispatchable flexible load power; P t,essential This refers to the critical load power that must be met and cannot be interrupted.

[0105] Furthermore, the power balance relationship of the system on the AC side is established. At the converter, the power balance formula is:

[0106] P Grid (t)=P PV (t)+P bat (t)+P DG (t)-P Load (t)

[0107] Among them, P Grid (t) represents the net power exchange with the public power grid at time t; a value greater than 0 indicates that electricity was purchased from the grid. bat (t) represents the net power of the battery on the AC side at time t, with positive for discharging and negative for charging; P DG (t) represents the output power of the diesel generator at time t.

[0108] By accurately modeling renewable energy generation units, backup generation units, energy storage units, and load units, and establishing the system power balance relationship, the physical objects and their operating characteristics in the distributed energy system are transformed into variables and constraints that can be used for subsequent optimization solutions, providing an engineering-feasible mathematical description basis for establishing a multi-objective optimization model.

[0109] In practical applications, a multi-objective optimization model that includes economic and environmental objectives is established. Specifically, this includes constructing an operating cost objective function, constructing an emission objective function, and establishing various constraints that the system must satisfy to operate.

[0110] An operating cost objective function is constructed to quantify the economic operation level of the distributed energy system. This objective function is a comprehensive goal that includes at least two of the following: purchased electricity cost, diesel generator fuel cost, energy storage unit loss cost, and renewable energy curtailment penalty cost. In one implementation, the objective is to minimize the total daily operating cost, expressed as:

[0111]

[0112] Where T represents the total number of scheduling periods.

[0113] Purchased electricity cost C grid (t) represents the cost of purchasing electricity from the public grid, for example Where P grid (t) represents the power (P) purchased from the grid at time t. grid (t)>0 indicates electricity purchase), This refers to the time-of-use electricity price for the power grid.

[0114] Diesel generator fuel cost C diesel (t) represents the fuel consumption cost of the diesel generator. It can be calculated using a linear model, for example... Where P diesel(t) represents the output power of the diesel generator at time t. This represents the fuel cost per unit output power. A more accurate quadratic function model can also be used, such as... Where C f Let P be the fuel unit price, a, b, c be the fitting coefficients of the diesel engine cost function, and P be the fuel unit price. DG (j0 represents the diesel engine output during the j-th time period.)

[0115] Energy storage unit loss cost C battery (t) characterizes the cost of battery life loss due to cycle charge and discharge, such as C. battery =α·E bat (t), where α is the battery life cost coefficient, E bat (t) represents the charge / discharge energy of the battery at time t.

[0116] Renewable energy curtailment penalty cost C RE-Curt (t) is used to incentivize systems to prioritize the use of renewable energy and reduce waste, for example... Where β is the power curtailment penalty coefficient. Let t be the photovoltaic power that is cut off or abandoned. The cost of renewable energy curtailment can be related to the curtailed power of the photovoltaic power generation unit, or it can be related to the curtailed power of other renewable energy sources (such as wind power).

[0117] An emission target function is constructed to quantify the environmental impact of the system operation, specifically the pollutant emissions generated from the operation of the diesel generator and the purchase of electricity from the public grid. In one embodiment, the emission target function is expressed as:

[0118]

[0119] In the emission objective function, the emissions from diesel generators DG The formula for calculating (T) is as follows:

[0120]

[0121] Where, N DG This refers to the number of diesel generators. These are the emission factors of carbon dioxide, sulfur dioxide, and nitrogen oxides per unit of electricity output of the j-th diesel generator; Let be the output power of the j-th diesel generator at time T.

[0122] In the emission objective function, the emission from electricity purchased by the grid is... Grid The formula for calculating (T) is as follows:

[0123]

[0124] in, These are the emission factors for carbon dioxide, sulfur dioxide, and nitrogen oxides generated per unit of electricity purchased from the public power grid; P Grid (T) represents the power purchased from the grid at time T. The pollutants quantified by the emission objective function include at least one of carbon dioxide, sulfur dioxide, and nitrogen oxides.

[0125] To ensure the feasibility and security of the scheduling scheme, the following constraints need to be established for system operation:

[0126] 1. System power balance constraints are used to ensure that the output of photovoltaic power generation units, energy storage units, diesel generators, and public grid interfaces are balanced with load demand and abandoned power. A specific expression is:

[0127] P grid (t)+P PV (t)+P bat,disch (t)+P DG (t)=P load (t)+P bat,chg (t)+P curt (t)

[0128] Among them, P bat,disch (t) represents the battery discharge power, P bat,chg (t) represents the battery charging power, P curt (t represents the amount of renewable energy power that is reduced at time t.)

[0129] 2. Output limits for each energy unit, including:

[0130] Power constraint at public grid interface: P grid,min <P grid (t) <P grid,max During the off-network operation period [t] s ,t e Within [the specified area], P also needs to be satisfied. grid (t) = 0. Output constraint of the photovoltaic power generation unit: P PV,min (t)≤P PV (t)≤P PV,max (t). Diesel generator output constraints: Where 'i' represents the diesel engine number. Furthermore, the diesel generator must also meet the minimum stable output constraint: in This is the minimum stable output of the diesel engine.

[0131] 3. Energy storage unit operating constraints, including:

[0132] Definition and mutual exclusion constraints of charge and discharge power: Net battery power Pbat (t)=P bat,ch (t)-P bat,disch (t), and to avoid simultaneous charging and discharging, it must satisfy

[0133] Charge and discharge power boundary constraints:

[0134]

[0135] State of Charge (SOC) Boundary Constraints: SOC bat,min ≤SOC bat (t)≤SOC bat,max .

[0136] State of charge update models are used to describe the dynamic changes in SOC, for example:

[0137]

[0138] Where SOC(t) is the battery state of charge at time t, η ch and η disch These represent the battery's charging efficiency and discharging efficiency, E. bat Δt represents the battery's rated capacity and the time step.

[0139] AC power constraints considering converter efficiency:

[0140]

[0141] Where, η conv For converter efficiency. Constraints on diesel generator charging the battery: P bat,ch (t)≤η conv ·P DG (t).

[0142] Through the above steps, a multi-objective optimization model that simultaneously considers economic objectives (operating costs) and environmental objectives (pollutant emissions) and includes complete system operation constraints was constructed, laying the foundation for subsequent optimization algorithms to solve for the optimal scheduling scheme.

[0143] In practical applications, typical load characteristic patterns are extracted based on historical operating data through cluster analysis. Specifically, a clustering algorithm is used to cluster historical daily data sequences, which must include at least electricity price sequences, outdoor temperature sequences, and load sequences. Then, the center sequence of each cluster is output as a typical load characteristic pattern.

[0144] The specific implementation of the clustering algorithm includes the following steps: Clustering the historical daily data sequence using a clustering algorithm. First, the data of each dimension in the historical daily data sequence is normalized. In one implementation, a normalized feature vector is constructed as the input to the clustering algorithm, for example:

[0145]

[0146] Among them, EP i (t) represents the electricity price data at time t on day i. Let L be the outdoor temperature data for the corresponding time on day i. i (t) represents the load data at time t on day i; max(EP) i (t)), max(L i (t) represents the maximum value of the corresponding data within the sample period, used to normalize the data of each dimension to a similar numerical range to eliminate the influence of units. Historical daily data series may also include features to identify date types or seasons, such as feature labels to distinguish between weekdays and weekends.

[0147] Clustering is performed using a similarity metric. During clustering, the similarity (or distance) between each sample vector and the cluster center vectors needs to be calculated. One implementation uses Euclidean distance as the similarity metric, and its calculation formula is as follows:

[0148]

[0149] Where, x p Let c represent the sample vector to be clustered at the p-th node (i.e., a normalized sequence of historical daily data). q Let q be the cluster center vector. and These represent the values ​​of the sample vector and the cluster center vector in the i-th feature dimension, respectively, where n is the total number of feature dimensions. Other similarity measures, such as Manhattan distance and cosine similarity, can also be used.

[0150] Set the iterative convergence condition. The clustering algorithm updates the cluster centers iteratively until the preset convergence condition is met. A typical convergence condition is that the algorithm is considered to have converged and stops iterating when the change in cluster centers between two consecutive iterations is less than a certain threshold. The mathematical expression for this convergence condition is, for example:

[0151]

[0152] Where k is the preset number of clusters, Let represent the i-th cluster center vector obtained after the t-th iteration, and ε be a preset convergence threshold. The clustering algorithm can be K-means, hierarchical clustering, or fuzzy C-means clustering, etc.

[0153] Through the above clustering process, historical days with similar electricity prices, temperature and load change patterns are classified into the same category, and the center sequence of each category (i.e. the feature vector sequence that best represents the commonality of the category) is extracted as the typical load feature pattern in the corresponding scenario, providing a clear and representative input basis for subsequent short-term load forecasting.

[0154] In practical applications, short-term load forecasting is performed using a neural network model based on typical load characteristic patterns and external forecasting information. Specifically, the following steps are included:

[0155] First, an input feature vector for the neural network is constructed based on the predicted electricity price, predicted outdoor temperature, typical load characteristic patterns for the corresponding period, and historical load data at the same time. External prediction information includes, but is not limited to, predicted electricity price and predicted outdoor temperature; typical load characteristic patterns for the corresponding period are extracted from historical operating data through cluster analysis; historical load data at the same time includes load data at the same time the day before the prediction date, and load data at the same time for the same day type (e.g., both weekdays or both weekends) within the week before the prediction date. These features are combined to form a multi-dimensional input feature vector X(t), used to predict the load at time t.

[0156] Next, the constructed input feature vector is fed into a trained neural network model. The neural network model learns the complex nonlinear mapping relationship between the input features and the load, and outputs a predicted load value. In one implementation, this mapping relationship can be represented as... in Let t be the predicted load value at time t, NN(·) represent the function implemented by the neural network model, and θ be the set of parameters of the neural network model, including the connection weights and biases between neurons in each layer.

[0157] The specific structure of a neural network model can be implemented in various ways. For example, a feedforward neural network can be used. This feedforward neural network contains an input layer, at least one hidden layer, and an output layer. The number of neurons in the input layer is the same as the dimension of the input feature vector X(t). The hidden layer contains several neurons, and its activation function can be a nonlinear function such as the sigmoid function or the Rectified Unit Function (ReLU) to introduce nonlinear transformation capability. The output layer typically contains one neuron and uses a linear activation function to directly regress the output predicted load value. Neural network models can also be recurrent neural networks or convolutional neural networks, or other models capable of nonlinear fitting.

[0158] Finally, the predicted load value output by the neural network model is received as an input parameter for the load demand in future periods in the multi-objective optimization model, providing a basis for scheduling optimization.

[0159] In practical applications, based on short-term load forecasting results, optimization algorithms are used to solve multi-objective optimization models to generate scheduling schemes for distributed energy systems. Specifically, the following steps are included:

[0160] First, the short-term load forecast results are used as input parameters for load demand in the multi-objective optimization model.

[0161] Secondly, using the output plans of each controllable energy unit in future time periods as decision variables, an initial solution population is formed. Specifically, all decision variables within the scheduling period are organized into a solution vector. For a candidate solution... Its form is Where m is the total number of decision variables, and each decision variable x i,d This represents the planned output of a controllable energy unit at a specific future time period, such as the switching power of a public grid interface, the charging power of an energy storage unit, the discharging power of an energy storage unit, the predicted power of a photovoltaic power generation unit, and the output power of a diesel generator; N is the population size, i.e., the number of candidate solutions. Each candidate solution is randomly initialized within the feasible range of the decision variables. For the d-th decision variable of the i-th candidate solution, its initial value is determined by formula x. i,d =lb d +rand(0,1)·(ub d -lb d Generate, where d = 1,...,m. Here, lb d and ub d These are the lower and upper bounds of the d-th decision variable, respectively, determined by the technical parameters and operational constraints of the corresponding controllable energy unit; rand(0,1) represents a random number uniformly distributed in the interval [0,1]. All candidate solutions constitute the population matrix. Its form is:

[0162]

[0163] Simultaneously, the comprehensive objective function value corresponding to each candidate solution is calculated, forming an objective function value vector. in It is the i-th candidate solution The evaluation value is calculated based on the economic objective function and the environmental objective function in the multi-objective optimization model.

[0164] Then, an iterative solution is performed based on a one-to-one optimization algorithm. The goal of the optimization algorithm is to solve a multi-objective optimization problem of the form minJ(cost target, emission target), while satisfying all system operation constraints, including power balance constraints, upper and lower limits of output for each energy unit, and operation constraints for energy storage units. In each iteration, based on the objective function values ​​of candidate solutions and their degree of violation of system operation constraints, the population is updated through a one-to-one comparison and relocation mechanism between solution vectors, driving the population to evolve towards a direction that satisfies the constraints and has a better objective function.

[0165] When the preset iteration termination condition is met (e.g., reaching the maximum number of iterations or achieving quality convergence of the solution), the iteration stops, and the solution vector with the best overall evaluation in the current population is output as the final distributed energy system scheduling scheme. This scheduling scheme clarifies the specific output plan of each controllable energy unit (such as public grid interface, energy storage unit, diesel generator, etc.) in each future time period.

[0166] In practical applications, after short-term load forecasting is performed using a neural network model based on typical load characteristic patterns and external forecasting information, the forecasting results are evaluated using forecasting accuracy evaluation indicators to quantify forecasting performance.

[0167] Prediction accuracy evaluation metrics are used to objectively measure the degree of deviation between the predicted values ​​of a neural network model and the actual observed values, thereby assessing the accuracy of the prediction model. In one implementation, the prediction accuracy evaluation metrics include at least two of the following: absolute error metric, relative error metric, goodness-of-fit metric, and standardized error metric. For example, the absolute error metric may be the mean absolute error, the relative error metric may be the mean absolute percentage error, the goodness-of-fit metric may be the coefficient of determination or the correlation coefficient, and the standardized error metric may be the root mean square error or the root mean square error in the form of the coefficient of variation.

[0168] Specifically, one or more of the following indicators can be used for evaluation:

[0169] Mean Absolute Error: Used to measure the average level of the absolute value of prediction error; its calculation formula is:

[0170]

[0171] Where N is the total number of sample points participating in the evaluation; e i Let f be the prediction error for the i-th sample point, and let f be the predicted value for the i-th sample point. i With the i-th measured value m i The difference, i.e., e i =f i -m i .

[0172] Root mean square error (RMSE): This measures the dispersion of prediction error and is more sensitive to larger errors. Its calculation formula is as follows:

[0173]

[0174] Mean Absolute Percentage Error: This measures the magnitude of the prediction error relative to the actual value. The formula for calculating it is:

[0175]

[0176] Coefficient of determination: Used to measure the degree to which the prediction result explains the fluctuations in the actual value, reflecting the goodness of fit of the prediction curve. Its calculation formula is as follows:

[0177] R 2 =r 2

[0178] Where r is the correlation coefficient between the predicted value and the measured value, and its calculation formula is:

[0179]

[0180] Here, f is the average of all predicted values. This is the average of all measured values.

[0181] Root mean square error in the form of coefficient of variation: used to eliminate the influence of dimensions and facilitate comparison between data of different orders of magnitude; its calculation formula is as follows:

[0182]

[0183] in, This represents the average of the measured values.

[0184] By calculating one or more of the above indicators, the performance of short-term load forecasting models can be comprehensively and quantitatively evaluated from multiple perspectives, thus providing a reliable basis for model selection, optimization, or validation.

[0185] Combination Figure 3 As shown, this application embodiment provides an energy management device 800 for a distributed energy system, including a processor 801 and a memory 802. Optionally, the device may further include a communication interface 803 and a bus 804. The processor 801, communication interface 803, and memory 802 can communicate with each other via the bus 804. The communication interface 803 can be used for information transmission. The processor 801 can call logical instructions in the memory 802 to execute the energy management method of the distributed energy system described in the above embodiment.

[0186] Furthermore, the logic instructions in the aforementioned memory 802 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0187] The memory 802, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this application. The processor 801 executes functional applications and data processing by running the program instructions / modules stored in the memory 802, thereby realizing the energy management method of the distributed energy system in the above embodiments.

[0188] The memory 802 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 802 may include high-speed random access memory and may also include non-volatile memory.

[0189] This application provides a distributed energy system, including: a distributed energy system body and the aforementioned energy management device 800. The energy management device 800 is installed on the distributed energy system body. The installation relationship described herein is not limited to placement within the distributed energy system, but also includes installation connections with other components of the distributed energy system, including but not limited to physical connections, electrical connections, or signal transmission connections. Those skilled in the art will understand that the energy management device 800 can be adapted to any feasible distributed energy system body, thereby realizing other feasible embodiments.

[0190] This application provides a computer-readable storage medium storing computer-executable instructions configured to execute the energy management method of the distributed energy system described above.

[0191] The technical solutions of this application embodiment can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this application embodiment. The aforementioned storage medium can be a non-transitory storage medium, including: USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, and other media capable of storing program code.

[0192] The technical solutions of this application embodiment can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this application embodiment. The aforementioned storage medium can be a non-transitory storage medium, including: USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, and other media capable of storing program code.

[0193] The foregoing description and accompanying drawings fully illustrate embodiments of this application to enable those skilled in the art to practice them. Other embodiments may include structural, logical, electrical, procedural, and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the terminology used in this application is for describing embodiments only and is not intended to limit the claims. As used in the description of embodiments and claims, the singular forms “a,” “an,” and “the” are intended to equally include the plural forms unless the context clearly indicates otherwise. Similarly, the term “and / or,” as used herein, means including one or more of the associated listed items and all possible combinations thereof. Additionally, when used in this application, the term "comprise" and its variations "comprises" and / or "comprising" refer to the presence of stated features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof. Without further limitations, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element. In this document, each embodiment may focus on the differences from other embodiments, and similar or identical parts between embodiments can be referred to mutually. For methods, products, etc., of the embodiments claimed, if they correspond to the method section of the embodiments claimed, then the relevant parts can be referred to the description of the method section.

[0194] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments claimed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0195] The methods and products (including but not limited to devices and equipment) disclosed in the embodiments herein can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units may be merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to implement this embodiment according to actual needs. In addition, the functional units in the embodiments of this application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0196] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description; sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

Claims

1. An energy management method for a distributed energy system, characterized in that, include: Based on historical operational data, typical load characteristic patterns are extracted through cluster analysis; Short-term load forecasting is performed using a neural network model based on typical load characteristic patterns and external forecasting information. Establish a multi-objective optimization model that includes both economic and environmental objectives. The decision variables of the optimization model should include at least the output plans of each controllable energy unit in the distributed energy system. Based on short-term load forecasting results, an optimization algorithm is used to solve the multi-objective optimization model and generate a scheduling scheme for the distributed energy system.

2. The method according to claim 1, characterized in that, A distributed energy system includes at least one of the following: a photovoltaic power generation unit, an energy storage unit, a diesel generator, a public grid interface, and a load unit; and / or, The controllable energy unit includes at least one of the following: energy storage unit, diesel generator, and public grid interface.

3. The method according to claim 1, characterized in that, Establish a multi-objective optimization model that includes both economic and environmental objectives, including: Construct an operating cost objective function, which includes at least two of the following: purchased electricity cost, diesel generator fuel cost, energy storage unit loss cost, and renewable energy curtailment penalty cost; An emission target function is constructed to quantify the pollutant emissions generated by the operation of diesel generators and the purchase of electricity from the public grid.

4. The method according to claim 3, characterized in that, The cost of renewable energy curtailment penalties is related to the curtailed power output of photovoltaic (PV) power generation units; and / or, The pollutants quantified by the emission target function include at least one of carbon dioxide, sulfur dioxide, and nitrogen oxides.

5. The method according to claim 3, characterized in that, Establishing a multi-objective optimization model that includes both economic and environmental objectives also includes: Establish system power balance constraints to ensure that at least one of the outputs from photovoltaic power generation units, energy storage units, diesel generators, and public grid interfaces is balanced with load demand and power curtailment; and / or, Establish upper and lower limits for the output of each energy unit; and / or, Establish operating constraints for energy storage units, which include at least one of the following: state of charge constraints, charge / discharge power constraints, and state of charge update model.

6. The method according to any one of claims 1 to 5, characterized in that, Based on historical operational data, typical load characteristic patterns are extracted through cluster analysis, including: Clustering algorithms are used to cluster historical daily data sequences, which must include at least electricity price sequences, outdoor temperature sequences, and load sequences. Each cluster center sequence obtained after clustering is output as a typical load characteristic pattern.

7. The method according to claim 6, characterized in that, The clustering algorithm is the K-means algorithm; and / or, Use Euclidean distance as a similarity measure; and / or, The convergence condition is that the change in cluster centers is less than a preset threshold.

8. The method according to claim 6, characterized in that, Before clustering historical daily data sequences using clustering algorithms, the following steps are also included: The data in each dimension of the historical daily data series are normalized; the historical daily data series also includes features used to identify date types or seasons.

9. The method according to any one of claims 1 to 5, characterized in that, Based on typical load characteristic patterns and external forecasting information, short-term load forecasting is performed using a neural network model, including: Based on the predicted electricity price, predicted outdoor temperature, typical load characteristic patterns of the corresponding period, and historical load data at the same time, the input feature vector of the neural network is constructed. The input feature vector is fed into the feedforward neural network model; Receive the predicted load value output by the feedforward neural network model.

10. The method according to claim 9, characterized in that, Historical load data for the same time period includes load data for the same time period on the day preceding the forecast date, and load data for the same time period on the same day of the week preceding the forecast date; and / or, The activation function of the hidden layer in a feedforward neural network is either the Sigmoid function or the ReLU function.

11. The method according to any one of claims 1 to 5, characterized in that, Based on short-term load forecasting results, an optimization algorithm is used to solve the multi-objective optimization model, generating a dispatch scheme for the distributed energy system, including: Use short-term load forecasting results as input parameters for load demand in the multi-objective optimization model; The initial solution population is formed by using the power output plans of each controllable energy unit in each future time period as decision variables; Iterative solutions are obtained based on a one-to-one optimization algorithm. In each iteration, the population is updated by a one-to-one comparison and relocation mechanism between solution vectors, based on the objective function value and the constraint satisfaction. When the iteration termination condition is met, the optimal solution vector in the current population is output as the scheduling scheme.

12. The method according to claim 11, characterized in that, When updating the population, the one-to-one optimization algorithm evaluates the solution vector based on the objective function value and the degree of violation of system operation constraints.

13. The method according to any one of claims 1 to 5, characterized in that, Based on typical load characteristic patterns and external forecasting information, short-term load forecasting using a neural network model also includes: The prediction results are evaluated using prediction accuracy evaluation indicators to quantify prediction performance; among them, prediction accuracy evaluation indicators include at least two of the following: mean absolute error, root mean square error, mean absolute percentage error, and coefficient of determination.

14. An energy management device for a distributed energy system, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to execute, when running the program instructions, the energy management method of the distributed energy system as described in any one of claims 1 to 13.

15. A distributed energy system, characterized in that, include: The distributed energy system itself; and, The energy management device for the distributed energy system as described in claim 14 is installed on the distributed energy system body.

16. A computer-readable storage medium storing program instructions, characterized in that, When the program instructions are executed, they cause the computer to perform the energy management method of the distributed energy system as described in any one of claims 1 to 13.

Images