Method, system, device, medium and product for quantitatively evaluating adjustable capacity of demand side flexible resource of power distribution network

Abstract

This invention discloses a method, system, equipment, medium, and product for quantitatively assessing the demand-side adjustability of distribution networks, relating to the field of power system analysis and optimization technology. First, a dual-objective optimization model is constructed with the optimization objectives of maximizing renewable energy absorption and minimizing system operating costs. Then, multiple sets of training samples are generated based on this dual-objective optimization model and preprocessed to construct a labeled dataset. Subsequently, the labeled dataset is input into a pre-set deep neural network for training, resulting in a rapid prediction model for flexible resource adjustability. Finally, the distribution network data to be predicted on the demand side of the target distribution network is input into this rapid prediction model, outputting the quantitative assessment result of resource adjustability. The model-driven approach uses the dual-objective optimization model as its core, while the data-driven approach uses a deep neural network as its carrier, enabling rapid response in the assessment process. The combination of these two approaches effectively solves the technical problem of existing flexible resource adjustability assessments struggling to balance accuracy and real-time response efficiency.

Description

Technical Field

[0001] This invention relates to the field of power system analysis and optimization technology, and in particular to a method, system, equipment, medium and product for quantitatively evaluating the adjustability of demand-side flexible resources in distribution networks. Background Technology

[0002] With the continuous expansion of new energy power generation, represented by distributed photovoltaic and wind power, and its penetration rate in the distribution network constantly increasing, the high dependence of its power generation output on weather conditions leads to significant intermittent and fluctuating characteristics. Traditional power grid operation modes are facing severe challenges, and operational pressures such as system voltage stability and frequency regulation are becoming increasingly prominent. Against this backdrop, various new resources with regulation potential, such as distributed energy storage, are gradually penetrating the power system. These resources are not only electricity consumption units but also contain enormous regulation potential. Their accurate assessment and efficient utilization have become key issues in improving the resilience of power grid operation.

[0003] Currently, the assessment methods for the adjustability of such flexible resources are mainly divided into three categories: physical modeling-based methods, although possessing clear physical meaning and theoretical foundation, face problems such as high model complexity, heavy computational burden, and difficulty in obtaining parameters in practical applications; data-driven methods based on machine learning can learn adjustment patterns from historical data, but heavily rely on the quality and quantity of historical data, have insufficient generalization ability in new scenarios, and poor model interpretability; and empirical rule-based methods, although simple and easy to use, have limited assessment accuracy and cannot adapt to complex and ever-changing operating scenarios.

[0004] However, these existing approaches generally have a core limitation: a single assessment approach is difficult to meet the real-time response requirements in dynamic scenarios while ensuring assessment accuracy. Furthermore, they lack comprehensive consideration of multi-dimensional characteristics such as adjustment response time, resource adjustment rate, and duration, resulting in assessment results that cannot effectively support the timeliness requirements of power grid operation decisions and restricting the full release of the potential of flexible resource adjustment. Summary of the Invention

[0005] This invention provides a method, system, equipment, medium, and product for quantitatively assessing the adjustability of flexible resources on the demand side of a distribution network, which solves the technical problem that existing assessments of the adjustability of flexible resources are difficult to balance accuracy and real-time response efficiency.

[0006] The first aspect of this invention provides a method for quantitatively assessing the adjustability of demand-side flexible resources in a distribution network, comprising:

[0007] Construct a dual-objective optimization model with the goals of maximizing new energy consumption and minimizing system operating costs;

[0008] Multiple sets of training samples are generated based on the dual-objective optimization model, and the training samples are preprocessed to construct a label dataset;

[0009] The labeled dataset is used as input to a pre-set deep neural network for training, resulting in a rapid prediction model for flexible resource adjustment capabilities.

[0010] The fast prediction model for flexible resource regulation capacity is input with the demand-side data of the target distribution network and outputs a quantitative assessment result of resource regulation capacity.

[0011] Optionally, the construction of a dual-objective optimization model with the objectives of maximizing renewable energy consumption and minimizing system operating costs includes:

[0012] A distributed energy storage model is constructed based on power constraints, energy dynamic constraints, state of charge constraints, and ramp constraints.

[0013] A temperature-controlled load model is constructed based on thermodynamic model constraints, power constraints, comfort zone limitation constraints, precooling strategy constraints, and equipment heterogeneity constraints.

[0014] An adjustable industrial load model is constructed based on the constraints of adjustable power, energy conservation, time-of-use electricity price response, capacity variation, and simultaneity coefficient.

[0015] An electric vehicle model is constructed based on power boundary constraints, energy dynamic constraints, energy limit constraints, energy satisfaction constraints, charge-discharge mutual exclusion constraints, and SOC distribution constraints.

[0016] A data center model is constructed based on adjustable load power boundary constraints, backup power duration constraints, and energy conservation constraints.

[0017] A 5G base station model is constructed based on adjustable service load boundary constraints, sleep mode energy saving ratio constraints, active base station number constraints, adjustment range constraints, and traffic diversity constraints.

[0018] With the optimization objectives of maximizing the absorption of new energy and minimizing the system operating cost, a dual-objective optimization model is constructed by coupling the distributed energy storage model, the temperature-controlled load model, the adjustable industrial load model, the electric vehicle model, the data center model, and the 5G base station model, and system safety operation constraints are set.

[0019] The system's safe operation constraints include power balance constraints, new energy output constraints, flexible resource cluster adjustment constraints, system node voltage constraints, and line current carrying capacity constraints.

[0020] Optionally, the step of generating multiple sets of training samples based on the bi-objective optimization model and preprocessing the training samples to construct a labeled dataset includes:

[0021] Using the key parameters of the dual-objective optimization model as the sampling objects, the value space of the sampling parameters is determined. The key parameters include the configuration parameters of flexible resources, system constraint parameters, and scenario parameters.

[0022] Based on the value space of the sampling parameters, multiple sets of parameter combinations are generated;

[0023] Each of the parameter combinations is input into the bi-objective optimization model for solution, and the benchmark results corresponding to each parameter combination are obtained.

[0024] Each set of parameter combinations is used as input features, and the corresponding benchmark results are used as output labels to form multiple sets of training samples.

[0025] The training samples are preprocessed, including data cleaning, outlier handling, and standardization.

[0026] Multiple preprocessed training samples are divided according to a preset partitioning ratio to obtain a labeled dataset.

[0027] Optionally, generating multiple sets of parameter combinations based on the sampling parameter value space includes:

[0028] Based on the value space of the sampling parameters, uniform sampling is performed using the Latin hypercube sampling method to generate multiple sets of parameter combinations.

[0029] Optionally, before inputting the target distribution network demand-side data into the rapid prediction model for flexible resource regulation capacity, the method further includes:

[0030] The preprocessing is performed on the demand-side distribution network data to be predicted for the target distribution network.

[0031] Optionally, the quantitative evaluation results of resource adjustability include maximum adjustable capacity, resource adjustability time, adjustment response time, resource adjustment rate, individual resource adjustment gradient, resource adjustment accuracy, resource adjustment load rate, and resource adjustment volatility.

[0032] A second aspect of the present invention provides a quantitative assessment system for the demand-side flexible resource adjustability of a distribution network, comprising:

[0033] The model-driven evaluation engine module is used to build a dual-objective optimization model with the optimization objectives of maximizing the consumption of new energy and minimizing the system operating cost.

[0034] The data processing module is used to generate multiple sets of training samples based on the dual-objective optimization model, and to preprocess the training samples to construct a label dataset;

[0035] The data-driven evaluation engine module is used to train a pre-set deep neural network using the labeled dataset as input, and to obtain a rapid prediction model of flexible resource adjustment capability.

[0036] The standardized evaluation module is used to input the target distribution network demand-side data into the flexible resource adjustment capability rapid prediction model and output the quantitative evaluation result of resource adjustability capability.

[0037] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the quantitative assessment method for the demand-side flexible resource adjustability of the distribution network as described above.

[0038] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed, it implements the quantitative assessment method for the adjustable capacity of demand-side flexible resources in a distribution network as described above.

[0039] The fifth aspect of the present invention provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein, when the program instructions are executed by a computer, the computer performs the quantitative assessment method for the adjustability of flexible resources on the demand side of the distribution network as described above.

[0040] As can be seen from the above technical solutions, the present invention has the following advantages:

[0041] This invention provides a method, system, equipment, medium, and product for quantitatively assessing the demand-side adjustability of flexible resources in distribution networks. It employs an innovative model-data dual-drive approach. First, a dual-objective optimization model is constructed, with the optimization objectives of maximizing renewable energy absorption and minimizing system operating costs. Then, multiple sets of training samples are generated based on this dual-objective optimization model and preprocessed to construct a labeled dataset. Subsequently, the labeled dataset is input into a pre-built deep neural network for training, resulting in a rapid prediction model for flexible resource adjustability. Finally, the distribution network data to be predicted on the demand side of the target distribution network is input into this rapid prediction model, outputting the quantitative assessment result of resource adjustability. The model-driven approach, with the dual-objective optimization model at its core, provides solid theoretical and data support for assessment accuracy. The data-driven approach, using a deep neural network as a carrier, enables rapid response in the assessment process. The organic combination of these two approaches effectively solves the technical problems of existing flexible resource adjustability assessments, which struggle to balance accuracy and real-time response efficiency, and cannot comprehensively assess multi-dimensional characteristics. Furthermore, the construction of the dual-objective optimization model closely aligns with the actual operating rules of flexible resources. The generated training samples possess high accuracy and representativeness, providing a high-quality data foundation for the training of deep neural networks. This ensures the accuracy of the final evaluation results and avoids the shortcomings of single data-driven models, which rely on historical data, have insufficient generalization ability, and are difficult to guarantee accuracy. The fully trained deep neural network can achieve rapid mapping and prediction of the data to be predicted without repeatedly solving complex optimization models, significantly improving the real-time response efficiency of the evaluation. This solves the problems of high computational complexity and slow response of single physical modeling methods. At the same time, the dual-objective optimization model incorporates multi-dimensional operational constraints of flexible resources during its construction process. The generated training samples cover multi-dimensional features related to flexible resource adjustment, enabling the trained dual-drive rapid prediction model to output comprehensive quantitative evaluation results of resource adjustability. This effectively achieves a comprehensive evaluation of the multi-dimensional adjustment characteristics of flexible resources, breaking the limitations of existing evaluation methods that cannot fully cover the adjustment characteristics of flexible resources and have limited practicality of evaluation results. This provides accurate and efficient technical support for the efficient scheduling and rational utilization of flexible resources on the demand side of the distribution network. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 A flowchart illustrating the steps of a method for quantitatively evaluating the adjustability of demand-side flexible resources in a power distribution network, as provided in an embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram of the distribution network topology and flexible resource and new energy access nodes provided in an embodiment of the present invention;

[0045] Figure 3 Heatmap of eight core indicators of the adjustability of six types of flexible resources provided in embodiments of the present invention;

[0046] Figure 4 A heat map for predicting the accuracy of flexible loads provided in an embodiment of the present invention;

[0047] Figure 5 This is a histogram of the relative error distribution of distributed energy storage prediction results provided in an embodiment of the present invention.

[0048] Figure 6 This is a histogram showing the relative error distribution of temperature control load prediction results provided in an embodiment of the present invention.

[0049] Figure 7 The relative error distribution histogram of adjustable industrial load forecasting results provided in this embodiment of the invention;

[0050] Figure 8 This is a histogram of the relative error distribution of electric vehicle prediction results provided in an embodiment of the present invention.

[0051] Figure 9 This is a histogram of the relative error distribution of data center prediction results provided in an embodiment of the present invention.

[0052] Figure 10 This is a histogram of the relative error distribution of 5G base station prediction results provided in an embodiment of the present invention.

[0053] Figure 11 A structural block diagram of a quantitative assessment system for the demand-side flexible resource adjustability of a power distribution network, provided in an embodiment of the present invention;

[0054] Figure 12 This is a schematic diagram of the user interface layout of the distribution network demand-side flexible resource adjustability quantitative assessment system provided in an embodiment of the present invention.

[0055] Figure 13 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0056] This invention provides a method, system, device, medium, and product for quantitatively assessing the adjustability of flexible resources on the demand side of a distribution network, considering both model-driven and data-driven approaches. It is applicable to the accurate assessment and efficient utilization of flexible resources in new distribution systems with a high proportion of renewable energy. The invention addresses technical problems by constructing a dual-engine assessment architecture that deeply integrates model-driven and data-driven approaches.

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0058] It should be noted that, in the optional embodiments of the present invention, the data related to object information, etc., requires the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies. Furthermore, the collection, use, and processing of the relevant data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. In other words, if the embodiments of the present invention involve data related to an object, it needs to be obtained with the object's authorization and consent, the authorization and consent of relevant departments, and in accordance with the relevant laws, regulations, and standards of the country and region. If the embodiments involve personal information, the acquisition of all personal information requires the individual's consent. If sensitive information is involved, the separate consent of the information subject is required. The embodiments also need to be implemented with the object's authorization and consent.

[0059] This invention constructs a unified quantitative index system that balances evaluation accuracy and speed, providing effective technical support for the planning and scheduling of new power distribution systems.

[0060] Please see Figure 1 , Figure 1 A flowchart illustrating the steps of a method for quantitatively evaluating the adjustability of demand-side flexible resources in a power distribution network, as provided in an embodiment of the present invention.

[0061] This invention provides a method for quantitatively assessing the demand-side flexible resource adjustability of a distribution network, comprising:

[0062] Step 101: Construct a dual-objective optimization model with the optimization objectives of maximizing the consumption of new energy and minimizing the system operating cost.

[0063] The dual-objective optimization model, with the core optimization objectives of maximizing renewable energy consumption and minimizing system operating costs, couples six flexible resource models: distributed energy storage, temperature-controlled loads, adjustable industrial loads, electric vehicles, data centers, and 5G base stations. It also sets system safety operation constraints, forming a mixed-integer nonlinear programming model to provide high-precision benchmark results for flexible resource regulation capabilities. Training samples: Using the key parameter combination of the dual-objective optimization model as input features, and the benchmark results obtained from solving the model with this parameter combination as output labels, a set of data pairs is formed for training the deep neural network, covering various operating scenarios of the distribution network. Preprocessing: Standardization processing is performed on the training samples or the distribution network data to be predicted, including data cleaning, outlier handling, and standardization operations (to ensure the accuracy of model training and prediction). Labeled dataset: Multiple preprocessed training samples are divided into training, validation, and test sets according to a preset ratio (8:1:1 in this scheme) for parameter fitting, training monitoring, and generalization capability evaluation of the deep neural network. It is the core data support for the data-driven engine; Pre-built deep neural network: This solution adopts a multilayer perceptron structure, adapting to the dimensions of input features (87 dimensions) and output labels (48 dimensions), including an input layer, a hidden layer (including ReLU activation function, batch normalization, and random deactivation mechanism), and an output layer, used to train a fast prediction model; Fast prediction model for flexible resource adjustment capability: The prediction model obtained by training the pre-built deep neural network through the labeled dataset can quickly map the distribution network data to be predicted with the quantitative evaluation results of the flexible resource adjustment capability, achieving millisecond-level accurate prediction, balancing accuracy and efficiency; Distribution network data to be predicted: Various core data on the demand side of the target distribution network, including system basic parameters, six types of flexible resource configuration and operation parameters, scenario and market parameters, and model training correlation parameters, used as input to the fast prediction model to obtain evaluation results; Quantitative evaluation results of resource adjustment capability: Output by the fast prediction model, used to comprehensively characterize the eight core quantitative indicators of the six types of flexible resource adjustment capabilities, providing multi-dimensional quantitative basis for distribution network scheduling decisions.

[0064] In this embodiment of the invention, a dual-objective optimization model is constructed with the optimization objectives of maximizing the absorption of new energy and minimizing the system operating cost. The dual-objective optimization model couples the basic operating constraints of the flexible resources on the demand side of the distribution network and the safety operating constraints of the distribution network system. The basic operating constraints of the flexible resources include related constraints such as power boundary and energy dynamics, while the system safety operating constraints include core constraints such as power balance and node voltage. Through the collaborative construction of the objective function and the constraint conditions, a dual-objective optimization model that fits the actual operating scenario of the distribution network is obtained. Solving this model can yield the benchmark results of the flexible resource adjustment capability under different parameter scenarios.

[0065] Among them, the distributed energy storage model comprehensively considers the boundary constraints of charging and discharging power, the dynamic equation of state of charge (SOC), and the impact of charging and discharging depth on battery life, forming a synergistic fit with the aforementioned power constraints, energy dynamic constraints, state of charge limitation constraints, and ramping constraints; the temperature-controlled load model establishes the dynamic evolution equation of indoor temperature based on equivalent thermal resistance and equivalent heat capacity parameters, and simultaneously sets user comfort temperature range constraints, which conforms to the core requirements of the aforementioned thermodynamic model constraints and comfort range limitation constraints; the adjustable industrial load model focuses on characterizing its interruptible and transferable operation characteristics, establishing operating constraints such as minimum interruption time, maximum number of interruptions, and adjustable power boundaries, echoing the aforementioned adjustable power constraints and energy conservation constraints; the electric vehicle model is based on user... Based on travel survey statistics, a probability distribution model for electric vehicle grid connection / disconnection times is established, taking into account battery capacity constraints, state of charge (SOC) limitations, and user travel energy demand guarantees, matching the design principles of the aforementioned power and energy satisfaction constraints. The data center model establishes the correlation between the spatiotemporal transfer characteristics of computing load and server power consumption, considering service quality constraints, server energy consumption characteristics, and backup power supply continuous operation constraints, maintaining consistency with the aforementioned adjustable load power boundary constraints, backup power supply duration constraints, and energy conservation constraints. The 5G base station model combines the power consumption characteristics of 5G communication equipment and base station sleep control strategies to establish a dynamic correlation model between base station energy consumption and service load, improving the full coverage of the six types of flexible resource models. On this basis, coupling the above six types of flexible resource models, a system-level optimization model is constructed with the dual objectives of maximizing new energy consumption and minimizing system operating costs. This model comprehensively considers system safety operation constraints such as power balance constraints, node voltage constraints, and line current carrying capacity constraints, forming a complete mixed-integer nonlinear programming (MINLP) problem, which is then solved accurately using optimization solvers such as Gurobi.

[0066] First, construct an IEEE 33-node test system, such as Figure 2 As shown, the system fully recreates the topology and operating characteristics of the distribution network. The deployment nodes and functions of each typical flexible resource and distributed power source are described below:

[0067] Distributed energy storage devices (marked with red stars) are connected to node 3 to smooth out fluctuations in renewable energy output and improve the system's peak-shaving capacity; photovoltaic generator sets (marked with yellow photovoltaic panels) are connected to nodes 24 and 30 to simulate the intermittent output characteristics of distributed photovoltaics; 5G base stations (marked with blue squares) are connected to node 25 to simulate the stable power consumption and flexible adjustment potential of communication base stations; wind turbine generator sets (marked with blue windmills) are connected to nodes 8 and 14 respectively to simulate the regional distribution and output fluctuations of wind power; adjustable loads (marked with purple gears) are connected to node 22 to simulate industrial or commercial loads with active response capabilities; data centers (marked with brown-yellow triangles) are connected to node 33 to simulate the dynamic adjustment characteristics of high-energy-consuming loads; temperature-controlled loads (marked with dark green circles) are connected to node 12 to simulate the adjustment characteristics of loads with thermal inertia, such as air conditioning and heating; electric vehicles (marked with light green diamonds) are connected to node 18 to simulate the charging and discharging adjustment potential of mobile energy storage resources. The deployment nodes and installed capacities of each typical flexible resource and distributed power source are shown in the table below:

[0068]

[0069] Wind turbine generators are connected to nodes 24 and 30, with installed capacities of 1.0MW and 1.5MW respectively; photovoltaic generators are connected to nodes 8 and 14, with installed capacities of 1.0MW and 1.5MW respectively; distributed energy storage is connected to node 3, with an installed capacity of 0.2MW; industrial loads (adjustable loads) are connected to node 22, with an installed capacity of 0.09MW; temperature-controlled loads are connected to node 12, with an installed capacity of 0.05MW; electric vehicles are connected to node 18, with an installed capacity of 0.18MW; data centers are connected to node 33, with an installed capacity of 0.2MW; and 5G base stations are connected to node 25, with an installed capacity of 0.2MW.

[0070] Based on this testing system and rigorous physical mechanisms, refined mathematical models are established for six typical flexible resources: distributed energy storage, temperature-controlled loads, adjustable industrial loads, electric vehicles, data centers, and 5G base stations, as detailed below:

[0071] Distributed Energy Storage Model: A mathematical model built based on the operational characteristics of distributed energy storage to characterize its regulation capabilities. Core constraints include power constraints (limiting the charging and discharging power range of distributed energy storage to avoid exceeding the device's rated capacity), energy dynamic constraints (describing the energy change pattern of distributed energy storage at different time steps, relating charging and discharging power, efficiency, and time), state of charge (SOC) constraints (limiting the SOC range of distributed energy storage to prevent overcharging and over-discharging damage to the battery), and ramp constraints (limiting the rate of change of charging and discharging power of distributed energy storage to ensure operational stability). Temperature-Controlled Load Model: A mathematical model built based on the thermodynamic characteristics of temperature-controlled loads and user requirements. Core constraints include... Constraints include thermodynamic model constraints (based on equivalent thermal resistance and equivalent heat capacity parameters, describing the dynamic changes in indoor temperature), power constraints (limiting the operating power range of temperature-controlled loads to match equipment rated parameters), comfort zone constraints (limiting the reasonable range of indoor temperature to ensure user comfort), pre-cooling strategy constraints (based on time-of-use electricity pricing signals, setting pre-cooling trigger conditions to achieve load transfer and energy saving), and equipment heterogeneity constraints (considering the capacity and operating status differences of multiple temperature-controlled loads to characterize cluster regulation characteristics); Adjustable industrial load model: a mathematical model built based on the interruptible and transferable characteristics of adjustable industrial loads, with core constraints including regulating power constraints (limiting the operating power range of temperature-controlled loads to match equipment rated parameters), and comfort zone constraints (limiting the reasonable range of indoor temperature to ensure user comfort), pre-cooling strategy constraints (based on time-of-use electricity pricing signals, setting pre-cooling trigger conditions to achieve load transfer and energy saving), and equipment heterogeneity constraints (considering the capacity and operating status differences of multiple temperature-controlled loads to characterize cluster regulation characteristics); Adjustable industrial load model: a mathematical model built based on the interruptible and transferable characteristics of adjustable industrial loads, with core constraints including regulating power constraints (limiting the operating power range of temperature-controlled loads), and regulating the operating power range of temperature-controlled loads to match equipment rated parameters, ... The constraints include: adjustment range (to avoid affecting normal production operation), energy conservation constraints (to ensure that the total adjustment of industrial load within the optimization period is within a reasonable range and to avoid long-term energy imbalance), time-of-use electricity price response constraints (to guide industrial load to increase load during off-peak hours and decrease load during peak hours according to time-of-use electricity price signals, to achieve peak shaving and valley filling), capacity variation constraints (to characterize the random fluctuation characteristics of the capacity of a single industrial load device), and simultaneity coefficient constraints (to describe the probability of multiple industrial load devices operating simultaneously, reflecting the cluster adjustment potential); electric vehicle model: a mathematical model built based on the travel characteristics of electric vehicle users and the battery operation characteristics, with core constraints including power boundary constraints (limiting the total charging and discharging power range of the electric vehicle cluster). The constraints include: (1) the number of associated vehicles, (2) the charging and discharging power of a single vehicle and (3) the participation ratio of vehicle-to-grid (V2G); (4) the dynamic constraints of power (describe the total power change of the electric vehicle cluster at different time steps); (5) the power limit constraints (limit the total power range of the electric vehicle cluster to match battery capacity characteristics); (6) the energy satisfaction constraints (ensure that the total charging amount within the optimization period meets the daily travel power needs of all vehicles and guarantees service quality); (7) the charging and discharging mutual exclusion constraints (control electric vehicles through 0-1 state variables, so that they can only be in charging or discharging state at the same time to avoid charging and discharging conflicts); and (8) the SOC distribution constraints (describe the initial state of charge (SOC) distribution of the electric vehicle cluster to fit the actual operating scenario).Data Center Model: A mathematical model built based on the load characteristics and power supply reliability requirements of data centers. Core constraints include adjustable load power boundary constraints (limiting the power range of adjustable loads in the data center to avoid affecting normal server operation), backup power duration constraints (limiting the cumulative operating time of backup power to ensure power supply reliability), and energy conservation constraints (ensuring the total load adjustment of the data center within a reasonable range during the optimization period to avoid excessive load reduction). 5G Base Station Model: A mathematical model built based on the power consumption characteristics and communication service requirements of 5G base stations. Core constraints include adjustable service load boundary constraints (limiting the power range of adjustable service loads of 5G base stations to ensure communication service quality), sleep mode energy saving ratio constraints (limiting the energy saving ratio after the base station enters sleep mode to balance energy saving and service quality), active base station number constraints (limiting the number of base stations remaining active during the optimization period to meet regional communication coverage requirements), adjustment range constraints (limiting the load adjustment range of 5G base stations to avoid affecting equipment lifespan and communication stability), and traffic diversity constraints (considering the differences in communication traffic in different regions and at different times). This describes the regulation characteristics of base station clusters; system safety operation constraints include: power balance constraints (ensuring that the active power supply and demand of the distribution network is balanced at each time step (the sum of new energy output, grid power supply, and flexible resource regulation output equals the sum of power consumption of various loads and line losses), which is the core constraint for system safety operation), new energy output constraints (limiting the active power output range of wind power and photovoltaics to ensure that it does not exceed the maximum power output at the current moment, while relating reactive power output to the power factor to ensure the stability of new energy output), flexible resource cluster regulation constraints (limiting the load regulation range of six types of flexible resource clusters to keep them within the upper and lower limits allowed by the equipment, avoiding excessive regulation that could damage the equipment or affect service quality), system node voltage constraints (limiting the voltage range of each node in the distribution network to keep it within the safe threshold, correcting the node voltage through voltage drop calculation to avoid excessively high or low voltage affecting equipment operation and power supply quality), and line current carrying capacity constraints (limiting the transmission power of each line in the distribution network to keep it within the safe threshold of the line's rated capacity, avoiding line overload that could damage the equipment and ensuring the safety of the grid topology).

[0072] Further, step 101 may include the following sub-steps:

[0073] S11. Based on power constraints, energy dynamic constraints, state of charge constraints, and ramping constraints, a distributed energy storage model is constructed.

[0074] In this embodiment of the invention, the power constraint of the distributed energy storage model is:

[0075]

[0076] In the formula, The output of energy storage at time t (discharge is positive, charging is negative). This represents the maximum charging / discharging power for energy storage.

[0077] The energy dynamic constraint is:

[0078]

[0079] In the formula, For the amount of energy stored at time t+1, For the amount of energy stored at time t, For energy storage charging efficiency, For energy storage and discharge efficiency, For the self-discharge rate per hour of energy storage, The charging power of energy storage at time t (the value is positive). Let t be the discharge power of the stored energy at time t (the value is positive). This represents the time step for a single charge / discharge process.

[0080] The state of charge constraint is as follows:

[0081]

[0082] In the formula, Minimum allowable power consumption, This represents the maximum permissible power level.

[0083] The climbing constraint is:

[0084]

[0085] In the formula, This represents the maximum permissible variation in the charging / discharging power of an energy storage device per unit time.

[0086] Through the collaborative construction of the above multi-dimensional constraints, the charging and discharging operation characteristics of distributed energy storage are fully characterized.

[0087] S12. Based on thermodynamic model constraints, power constraints, comfort zone limitation constraints, precooling strategy constraints, and equipment heterogeneity constraints, a temperature control load model is constructed.

[0088] In this embodiment of the invention, the thermodynamic model constraint of the temperature-controlled load is expressed in continuous form as follows:

[0089]

[0090] To adapt to numerical solutions, it is further transformed into a discretized form:

[0091]

[0092] In the formula, Let be the indoor temperature at time t. Let be the outdoor temperature at time t. The coefficient of performance (COP) of an air conditioner. For heat capacity, For thermal resistance, Let be the air conditioning power at time t. For time step.

[0093] The power constraint is:

[0094]

[0095] In the formula, This refers to the rated power of the air conditioner. This is the minimum operating power of the air conditioner.

[0096] The comfort zone limitation is:

[0097]

[0098] In the formula, Set point for room temperature This is a temperature dead zone.

[0099] The precooling strategy is constrained as follows:

[0100]

[0101] In the formula, The signal represents the electricity price (or system load level) at time t. It is a predicted electricity price (or predicted system load level) signal at time t+4, used to predict future load peaks.

[0102] This is used to lower indoor temperatures in advance when future load peaks are predicted, thereby reducing air conditioning power during peak hours; considering the heterogeneity of temperature-controlled load equipment, the power of a single device follows a normal distribution:

[0103]

[0104] In the formula, This is the serial number for the air conditioner (temperature-controlled load equipment), used to distinguish different individual devices. Let be the power of the i-th air conditioner. The variance of the normal distribution is denoted by , which represents the power fluctuation coefficient.

[0105] Based on this, the polymerization power is calculated:

[0106]

[0107] In the formula, For the number of air conditioners, This refers to the polymerization power.

[0108] Introducing a simultaneous rate coefficient to describe the synchronous operation characteristics of multiple devices:

[0109]

[0110] In the formula, This is the simultaneity rate coefficient.

[0111] The final effective polymerization power is obtained as follows:

[0112]

[0113] In the formula, To achieve effective power aggregation.

[0114] Through the collaborative construction of the above multi-dimensional constraints, the thermodynamic characteristics, operational constraints, and aggregation behavior of the temperature-controlled load are fully characterized.

[0115] S13. Based on the constraints of adjustable power, energy conservation, time-of-use electricity price response, capacity variation, and simultaneity coefficient, an adjustable industrial load model is constructed.

[0116] In this embodiment of the invention, the adjustable power constraint for the adjustable industrial load is:

[0117]

[0118] In the formula, For industrial base load, To achieve the maximum adjustment ratio, Let be the load adjustment amount at time t. This means that the constraint holds true for all time steps.

[0119] The energy conservation constraint is:

[0120]

[0121] In the formula, The total number of time periods or time ranges considered for this optimization problem.

[0122] This is used to ensure that the total adjustment during the entire cycle does not exceed ±5% of the total base load, thus avoiding long-term energy imbalance.

[0123] The time-of-use pricing response constraint is:

[0124]

[0125] In the formula, The time-of-use electricity price at time t is used to guide the load to increase during off-peak hours, decrease during peak hours, and allow for small fluctuations during normal hours.

[0126] The capacity variation constraint is:

[0127]

[0128] In the formula, This is the arithmetic mean of the rated maximum power of each user in the industrial load cluster. This is the serial number of the adjustable industrial load equipment. Let i be the actual total power of the i-th industrial load unit. The deviation between the actual power of the i-th industrial load unit and the average power of the cluster. This represents the normalized variance of the maximum power of each user in the industrial load cluster relative to the reference power. With a mean of 0 and a variance of The normal distribution is used to describe the random characteristics of the variation in the capacity of industrial equipment.

[0129] Meanwhile, the coefficient constraints are:

[0130]

[0131] In the formula, The simultaneous coefficient, The number of industrial load units reflects the probability of multiple industrial load units operating simultaneously. The more industrial load units there are, the closer the simultaneous coefficient is to 0.85.

[0132] Through the collaborative construction of the above multi-dimensional constraints, the adjustment characteristics, response behavior and aggregation law of adjustable industrial load are fully characterized.

[0133] S14. Based on power boundary constraints, energy dynamic constraints, energy limit constraints, energy satisfaction constraints, charge-discharge mutual exclusion constraints, and SOC distribution constraints, construct an electric vehicle model;

[0134] In this embodiment of the invention, the power boundary constraint of the electric vehicle is:

[0135]

[0136] In the formula, The total charging power at time t, Let be the total discharge power at time t. For the number of vehicles, This represents the maximum charging power for a single vehicle. This represents the maximum discharge power of a single vehicle. To determine the proportion of participation in V2G (vehicle-to-grid) services, This means that the constraint holds true for all time steps.

[0137] The dynamic constraint on power consumption is:

[0138]

[0139] In the formula, The total charge at time t+1 Let be the total charge at time t. For charging efficiency, For discharge efficiency, For time step.

[0140] The power limit constraint is:

[0141]

[0142] In the formula, This represents the theoretical upper limit of the total battery capacity of the electric vehicle cluster. For the battery capacity of each vehicle.

[0143] The energy satisfies the constraints (quality of service) as follows:

[0144]

[0145] In the formula, This is to determine the daily electricity demand of each vehicle, ensuring that the total energy available during the cycle meets the travel needs of all vehicles. The time step is the time increment in the dynamic constraint of the power quantity.

[0146] The charge / discharge mutual exclusion constraint is:

[0147]

[0148] In the formula, Select a variable for the charging / discharging mode, specifically a 0-1 state variable. A value of 1 indicates that charging is allowed and discharging is prohibited, while a value of 0 indicates that discharging is allowed and charging is prohibited. This refers to the maximum charging power limit for a single electric vehicle. This is the upper limit of the maximum discharge power of a single electric vehicle. Let be the charging power of the electric vehicle cluster at time t. Let be the discharge power of the electric vehicle cluster at time t.

[0149] Considering the SOC distribution characteristics of the electric vehicle cluster, the initial SOC follows a uniform distribution:

[0150]

[0151] In the formula, Let represent the initial state of charge of the i-th energy storage unit (battery) in the electric vehicle cluster when it is connected to the grid. It is a uniform distribution on the interval [0.2, 0.7]. This is the serial number of the energy storage unit. This represents the total number of energy storage units.

[0152] Based on this, calculate the equivalent initial SOC:

[0153]

[0154] In the formula, This represents the equivalent initial state of charge for the electric vehicle cluster.

[0155] Introducing a SOC diversity coefficient to quantify the degree of power dispersion within the cluster:

[0156]

[0157] In the formula, This is an efficiency correction factor related to the state of charge distribution of energy storage units in an electric vehicle cluster. This represents the standard deviation of the initial state of charge of each energy storage unit in the electric vehicle cluster.

[0158] Through the collaborative construction of the above multi-dimensional constraints, the charging and discharging characteristics, energy demand, and SOC distribution patterns of electric vehicle clusters are fully characterized.

[0159] S15. Construct a data center model based on adjustable load power boundary constraints, backup power duration constraints, and energy conservation constraints.

[0160] In this embodiment of the invention, the adjustable load power boundary constraint of the data center is:

[0161]

[0162] In the formula, Let t be the adjustable load power of the data center. This is the maximum adjustable power for the data center. This means that the constraint holds true for all time steps;

[0163] The duration of backup power supply is constrained as follows:

[0164]

[0165] In the formula, For indicator functions, when The value is 1 if the time condition is met, and 0 otherwise. It is used to calculate the cumulative operating time of the backup power supply. This is the maximum permissible duration of the backup power supply. This is the maximum continuous operating time of the data center's backup power supply, used to limit the duration of backup power supply availability to ensure power supply reliability.

[0166] The energy conservation constraint is:

[0167]

[0168] In the formula, Let be the data center load adjustment amount at time t. The net output power of the data center's backup power supply. The adjustment time, which characterizes the duration of the entire optimization cycle, is used to ensure that the total adjustment amount within the cycle is not less than -10% of the maximum adjustable power, thus avoiding excessive load reduction that could affect the normal operation of the data center.

[0169] Through the collaborative construction of the above multi-dimensional constraints, the operational characteristics of adjustable loads in data centers, the limitations of backup power supply usage, and energy balance requirements are fully characterized.

[0170] S16. Based on adjustable service load boundary constraints, sleep mode energy saving ratio constraints, active base station number constraints, adjustment range constraints, and traffic diversity constraints, a 5G base station model is constructed.

[0171] In this embodiment of the invention, the load of a 5G base station consists of both basic communication load and value-added service load, expressed as follows:

[0172]

[0173] In the formula, Let t be the total power consumption of the 5G base station. The communication load of a single active base station, Let t be the number of 5G base stations that are operational in the given area at time t. Let t be the power consumption of the 5G base station used to carry value-added data services.

[0174] Adjustable service load boundary constraints are:

[0175]

[0176] In the formula, For the adjustable value-added service load at time t, Let t be the power consumption of the 5G base station.

[0177] The energy-saving ratio constraint for hibernation mode is:

[0178]

[0179] This is used to characterize the energy-saving effect achieved by base stations in hibernation mode by reducing the load of value-added services.

[0180] The active base station count constraint is:

[0181]

[0182] In the formula, This represents the total number of base stations. This represents the minimum proportion of active base stations, used to ensure the reliability of basic communication coverage.

[0183] The adjustment range constraint is:

[0184]

[0185] In the formula, Let t be the change in power consumption of the 5G base station. This represents the maximum power consumption of a 5G base station at time t for carrying value-added data services, used to limit the upper and lower boundaries of load adjustment.

[0186] To characterize the randomness of traffic across multiple base stations, a traffic diversity coefficient is introduced:

[0187]

[0188] In the formula, This is the simultaneous rate coefficient for traffic (data services).

[0189] Through the collaborative construction of the above multi-dimensional constraints, the load composition, sleep energy-saving characteristics and adjustment capabilities of 5G base stations are fully characterized.

[0190] S17. With the optimization objectives of maximizing the absorption of new energy and minimizing the system operating cost, a dual-objective optimization model is constructed by coupling distributed energy storage model, temperature-controlled load model, adjustable industrial load model, electric vehicle model, data center model and 5G base station model, and system safety operation constraints are set.

[0191] The constraints for safe operation of the system include power balance constraints, new energy output constraints, flexible resource cluster regulation constraints, system node voltage constraints, and line current carrying capacity constraints.

[0192] In this embodiment of the invention, the dual-objective optimization model is a mixed-integer nonlinear programming (MINLP) problem, whose objective function has two dimensions: one is maximizing the absorption of new energy sources, expressed as:

[0193]

[0194] In the formula, This represents the amount of renewable energy consumed by the j-th renewable energy source at time t. The number of new energy sources connected. To optimize the time interval of the cycle.

[0195] Second, minimize the system operating cost, expressed as:

[0196]

[0197] In the formula, The energy storage operating cost is obtained by accumulating the product of the energy storage charging and discharging power and the corresponding cost coefficient over a time interval. The load adjustment cost is obtained by multiplying and summing the adjustment power of various flexible resources and their corresponding cost coefficients. The power loss cost of the power grid is obtained by multiplying and summing the line loss power and the loss cost coefficient; The cost of curtailed electricity is obtained by multiplying the curtailed power of new energy sources by the curtailment cost coefficient and accumulating the results.

[0198] The specific expressions for each cost are as follows:

[0199]

[0200]

[0201]

[0202]

[0203] In the formula, Let be the power that the system purchases from the grid at time t. Let be the unit price of electricity purchased by the system from the grid at time t. Let be the charging power of the new energy source at time t. Let t be the unit price of electricity purchased from new energy sources by the system at time t. Let be the regulating power of the k-th type of flexible load at time t. The compensation price for providing unit regulating power for the k-th type of flexible load at time t. This represents the total number of categories of flexible resources. Let be the active power loss of the l-th line at time t. Let be the unit active power loss cost coefficient for the l-th line. The total number of distribution network lines. Let be the power curtailment of the i-th renewable energy unit at time t. Let be the penalty cost coefficient applied to the unit power curtailment of the i-th renewable energy unit at time t.

[0204] To ensure the safe and stable operation of the system, the model is equipped with multi-dimensional constraints:

[0205] Power balance constraints are used to ensure real-time supply and demand balance in the system, and their expression is:

[0206]

[0207] In the formula, Let i be the active power output of the i-th wind turbine at time t; The active power output of the i-th photovoltaic unit at time t; Let be the power purchased by the lower i grid nodes from the grid at time t; Let be the discharge power of the electric vehicle at time t; This represents the power stored at time t. When the value is greater than 0, energy storage is charged; When the value is less than 0, the stored energy is discharged; Let be the charging power of the electric vehicle at time t; Let be the amount of renewable energy curtailed at time t; The total system load before any flexible resources are involved in adjustment at time t; This is the sum of the base power for all industrial loads; The total regulating power of all industrial loads at time t; For time t, the base adjustable power of all adjustable industrial loads relative to their reference power; Let t be the total aggregate power of all temperature-controlled loads in the system. This is the sum of the baseline power of all data center loads; The net regulation power provided to the power grid by all data center clusters at time t; This represents the sum of the baseline power of all 5G base stations at time t. Let t be the net change in total load of 5G base stations relative to its baseline value at time t.

[0208] The renewable energy output constraint is used to limit the output range of wind power and solar power. Meanwhile, the reactive power output of wind power and solar power is determined by the active power output and the power factor angle, expressed as:

[0209]

[0210]

[0211]

[0212]

[0213] In the formula, Let be the maximum output power of the i-th wind turbine at time t; Let be the maximum output power of the i-th photovoltaic unit at time t. For the reactive power output of the i-th wind turbine at time t, For the reactive power output of the i-th photovoltaic unit at time t, The power factor angle for wind power. The power factor angle for photovoltaics.

[0214] Flexible resource cluster adjustment constraints are used to limit the adjustment range of various flexible resource clusters, and the expression is:

[0215]

[0216] In the formula, The actual load of the flexible resource cluster at time t. As the baseline load, This represents the maximum power that the flexible resource cluster can increase at time t. This represents the limit value for reducing the power of the flexible resource cluster at time t.

[0217] System node voltage constraints are used to maintain node voltages within a safe range, and their expression is:

[0218]

[0219] A simplified power flow calculation formula is used to calculate voltage drop:

[0220]

[0221] In the formula, and These represent the lower and upper limits of the allowable voltage at all nodes (buses) in the power system, respectively. This represents the voltage amplitude at the nth node at time t; This refers to the voltage of the upstream node; This is the correction factor for voltage drop; Branch resistance; The active power of the branch circuit; For branch circuit reactance; The reactive power of the branch circuit; This is the voltage reference value; This is the apparent power reference value.

[0222] Line current carrying capacity constraint is used to limit the line transmission power to not exceed a safety threshold, and its expression is:

[0223]

[0224] In the formula, Let be the active power flowing through the b-th branch at time t; The rated apparent power capacity of the b-th branch is given.

[0225] By coupling various flexible resource models with system-level constraints, this dual-objective optimization model can accurately characterize the operating characteristics and resource regulation potential of the distribution network.

[0226] Step 102: Generate multiple sets of training samples based on the dual-objective optimization model, and preprocess the training samples to construct a label dataset.

[0227] In this embodiment of the invention, based on the dual-objective optimization model constructed in step 101, the key parameters of the model are selected as sampling objects, a reasonable sampling range is determined, and multiple sets of different parameter combinations are generated. Each set of parameter combinations is input into the dual-objective optimization model for solution, and the benchmark results of the adjustable capability of flexible resources under the corresponding parameter scenario are obtained. The parameter combinations are used as input features and the corresponding benchmark results are used as labels to form initial training samples. In order to ensure the reliability and evaluation accuracy of subsequent deep neural network training, the initial training samples are preprocessed to remove invalid data, correct outliers, and perform standardization to eliminate the influence of data dimension differences. After preprocessing, a labeled dataset that meets the training requirements is obtained.

[0228] Furthermore, step 102 may include the following sub-steps:

[0229] S21. Using the key parameters of the dual-objective optimization model as the sampling objects, determine the value space of the sampling parameters. The key parameters include the configuration parameters of flexible resources, system constraint parameters, and scenario parameters.

[0230] S22. Generate multiple sets of parameter combinations based on the value space of the sampling parameters;

[0231] Furthermore, S22 may include the following sub-steps:

[0232] S221. Based on the sampling parameter value space, uniform sampling is performed using the Latin hypercube sampling method to generate multiple sets of parameter combinations.

[0233] S23. Input each parameter combination into the bi-objective optimization model for solution and obtain the baseline results corresponding to each parameter combination.

[0234] S24. Using each set of parameter combinations as input features and the corresponding benchmark results as output labels, multiple sets of training samples are formed.

[0235] S25. Preprocess multiple sets of training samples, including data cleaning, outlier handling and standardization.

[0236] S26. Divide the preprocessed training samples into multiple groups according to a preset division ratio to obtain a labeled dataset.

[0237] Key parameters: These are the core parameters affecting the solution results of the dual-objective optimization model and the flexible resource regulation capability. They cover the configuration parameters of flexible resources, system constraint parameters, and scenario parameters, and are the core objects for sampling and generating training samples. Sampling parameter value space: This clarifies the reasonable value range of various key parameters, forming a high-dimensional parameter space that provides clear boundaries for subsequent uniform sampling, ensuring that sampling covers various operating scenarios of the distribution network. Flexible resource configuration parameters: These are parameters describing the inherent attributes of six types of flexible resource equipment (such as installed capacity, rated power, battery capacity, and number of devices), forming the basis for characterizing the regulation potential of flexible resources. System constraint parameters: These are boundary parameters ensuring the safe operation of the distribution network (such as node voltage upper and lower limits, line current carrying capacity thresholds, reference voltage, and reference capacity), used to set the system safety constraints of the dual-objective optimization model. Scenario parameters: These are parameters reflecting the external operating environment and market signals of the distribution network (such as fluctuations in new energy output, time-of-use pricing, and load level changes), used to simulate different operating scenarios and improve the training sample quality. The comprehensiveness of the sample; parameter combination: a set of key parameters generated by sampling methods based on the sampled parameter value space, each parameter combination corresponding to a distribution network operation scenario; benchmark results: the results obtained by inputting a set of parameter combinations into a bi-objective optimization model and solving it accurately through an optimization solver (such as Gurobi), including renewable energy consumption, system operating costs, and various flexible resource adjustment quantities, which serve as the output labels of the training samples; input features: the part of the training samples used to input the deep neural network, i.e., a set of key parameter combinations, covering various core parameters of the system, flexible resources, and scenarios; output labels: the benchmark results in the training samples corresponding to the input features, i.e., the relevant evaluation indicators of flexible resource adjustment capabilities, used for parameter fitting of the deep neural network; preset partitioning ratio: the partitioning rule of the preprocessed training samples. This scheme adopts an 8:1:1 ratio, corresponding to the training set (model parameter fitting), validation set (training effect monitoring, parameter adjustment), and test set (model generalization ability evaluation), respectively. Latin hypercube sampling method: A high-dimensional parameter space uniform sampling method used to generate multiple parameter combinations based on the sampling parameter value space. It can achieve uniform coverage of the parameter space, avoid sample redundancy and local bias, and ensure that the generated parameter combinations fully represent various operating scenarios of the distribution network, providing support for the generation of high-quality training samples.

[0238] In this embodiment of the invention, the key parameters of the dual-objective optimization model are used as sampling objects to determine the value space of the sampling parameters. The key parameters include configuration parameters of flexible resources, system constraint parameters, and scenario parameters. The flexible resource configuration parameters cover equipment attributes such as installed capacity and adjustment ratios of six typical flexible resources. The system constraint parameters include safety boundaries such as node voltage upper and lower limits and line current carrying capacity thresholds. The scenario parameters include external operating conditions such as fluctuations in new energy output and changes in load levels. By clarifying the reasonable value range of each parameter, a clear spatial boundary is provided for subsequent sampling. Based on the value space of the sampling parameters, multiple sets of parameter combinations are generated to further... The generation process specifically involves uniform sampling based on the sampling parameter value space using the Latin hypercube sampling method to generate multiple parameter combinations. This sampling method can achieve uniform coverage in the high-dimensional parameter space, avoiding sample redundancy and local bias, and ensuring that the generated parameter combinations can fully represent various operating scenarios of the distribution network. Subsequently, each parameter combination is input into a bi-objective optimization model for solution. The mixed-integer nonlinear programming (MINLP) model is accurately solved using optimization solvers such as Gurobi, yielding the corresponding values ​​for each parameter combination. High-precision benchmark results are generated, including evaluation indicators such as renewable energy consumption, system operating costs, and the adjustment capacity of various flexible resources. Then, each parameter combination is used as input features, and the corresponding benchmark results are used as output labels to form multiple training samples. The input features cover detailed configurations of system parameters, scenario parameters, and six types of flexible resources, totaling 87 dimensions. The output labels are the system optimization objectives and eight standardized evaluation indicators for various resources, totaling 48 dimensions. Afterward, the training samples are preprocessed, including data cleaning, outlier handling, and standardization. Specifically, invalid samples are removed, abnormal data is corrected, and the input features and output labels are standardized. The transformation eliminates the influence of differences in data dimensionality and units, improving the stability and convergence speed of subsequent deep neural network training. Finally, the preprocessed training samples are divided according to a preset ratio to obtain a labeled dataset. In this embodiment, the dataset is divided into a training set, a validation set, and a test set in a ratio of 8:1:1. The training set is used for parameter fitting of the deep neural network, the validation set is used to monitor the training process and adjust the network structure to avoid overfitting, and the test set is used to evaluate the generalization ability of the final model. This provides high-quality, structured data support for subsequent deep neural network training, and fully realizes the sample preparation stage of the data-driven evaluation engine.

[0239] In an optional embodiment, the Interquartile Range (IQR) outlier detection method can be used for outlier identification and truncation during preprocessing. Specifically, the process involves: first calculating the quartiles Q1 (the first quartile, i.e., the 25th percentile) and Q3 (the third quartile, i.e., the 75th percentile) for each input feature and output label data; then calculating the interquartile range IQR = Q3 - Q1; setting the outlier judgment threshold as [Q1 - 1.5 × IQR, Q3 + 1.5 × IQR]; and identifying values ​​exceeding this threshold range as outliers, which are then truncated to correct them. This involves uniformly replacing outliers smaller than Q1-1.5×IQR with Q1-1.5×IQR, and uniformly replacing outliers larger than Q3+1.5×IQR with Q3+1.5×IQR. This avoids reducing the sample size or shifting the data distribution due to outlier removal, ensuring the rationality and completeness of the data. After outlier processing, the cleaned and corrected training samples are standardized using the Z-score standardization rule. This converts the input features and output labels of each dimension into uniform numerical values ​​with a mean of 0 and a standard deviation of 1, eliminating the influence of data dimensional differences and dimensions, and improving the stability and convergence speed of subsequent deep neural network training.

[0240] Step 103: Use the labeled dataset as input to train a pre-set deep neural network to obtain a fast prediction model for flexible resource adjustment capabilities.

[0241] In this embodiment of the invention, the label dataset is divided into a training set and a validation set, and then input into a pre-set deep neural network for model training. The architecture of the pre-set deep neural network is adapted to the input feature dimension and output label dimension of the label dataset. During the training process, the model training effect is monitored using the validation set, and the network parameters are adjusted in a timely manner to avoid overfitting and ensure the reliability of model training. Once the model training reaches the preset accuracy requirement, training is stopped, and a rapid prediction model for flexible resource adjustment capability is obtained. The rapid prediction model for flexible resource adjustment capability can realize the rapid mapping between input data and the evaluation results of flexible resource adjustment capability, providing an efficient model carrier for the rapid evaluation of the data to be predicted in step 104. At the same time, relying on the high-quality label dataset in step 102, the rapid prediction model is guaranteed to have high evaluation accuracy, echoing the core idea of ​​model-data dual-drive, and solving the problem that existing evaluation methods cannot balance accuracy and efficiency.

[0242] In the specific implementation, the pre-built deep neural network adopts a multi-layer perceptron (MLP) structure, consisting of an input layer adapted to 87-dimensional input features, several hidden layers, and an output layer adapted to 48-dimensional output labels. The modified linear unit (ReLU) is used as the activation function in the hidden layers to introduce non-linear mapping capabilities. Batch normalization and dropout mechanisms are also incorporated. Batch normalization stabilizes the training process and accelerates convergence, while dropout suppresses overfitting by randomly hiding some neurons, effectively improving the model's generalization ability. During the model training phase, the adaptive moment estimation (Adam) optimizer is used for parameter updates, with mean squared error (MSE) as the loss function. The network parameters are iteratively optimized using the backpropagation algorithm. An early stopping strategy is introduced during training: when the validation set loss no longer decreases significantly over several consecutive training epochs, training is terminated early to further avoid model overfitting. Based on the high-quality labeled dataset obtained in step 102 by dividing it in an 8:1:1 ratio, the training set is used for network parameter fitting, and the validation set is used to monitor the training effect in real time and dynamically adjust the network parameters. Training is stopped after the model training reaches the preset accuracy requirement, and finally a fast prediction model with flexible resource adjustment capability is obtained. This model can realize the rapid mapping from 87-dimensional input features to 48-dimensional output evaluation indicators, which greatly reduces the evaluation time from the hierarchical level required by the original physical model to the millisecond level. It provides an efficient model carrier for the rapid evaluation of the data to be predicted in step 104, which not only ensures the evaluation accuracy but also significantly improves the response efficiency. It fully implements the core idea of ​​dual-driven model and data, and effectively solves the problem that existing evaluation methods cannot balance accuracy and efficiency.

[0243] Step 104: Input the target distribution network demand-side data into the flexible resource regulation capacity rapid prediction model and output the quantitative assessment results of resource regulation capacity.

[0244] In this embodiment of the invention, based on the trained rapid prediction model for flexible resource adjustment capability, the distribution network data to be predicted on the demand side of the target distribution network is first preprocessed in the same way as in step 102 to ensure that its data format and dimensions match the model input requirements. Then, the preprocessed data to be predicted is input into the rapid prediction model. Relying on the rapid mapping capability formed by the model through training on a high-quality labeled dataset, there is no need to repeatedly solve the dual-objective optimization model in step 101. The accurate quantitative evaluation result of resource adjustability can be quickly output. This realizes the core idea of ​​dual-driven model and data, effectively balances evaluation accuracy and real-time response efficiency, and fully realizes the quantitative evaluation of the flexible resource adjustability of the distribution network demand side. It provides a direct decision reference for the efficient scheduling and rational utilization of flexible resources in the distribution network, and further solves the technical problem that existing evaluation methods cannot balance accuracy and efficiency.

[0245] Furthermore, before inputting the target distribution network demand-side data into the rapid prediction model for flexible resource regulation capabilities, the following steps are also included:

[0246] Preprocess the distribution network data to be predicted on the demand side of the target distribution network.

[0247] It should be noted that the distribution network data to be predicted includes system basic parameters, six types of flexible resource allocation and operation parameters, scenario and market parameters, and model training correlation parameters. The system basic parameters include line resistance, line reactance, line rated capacity, node voltage upper and lower limits, system reference voltage, system reference capacity, optimization cycle, and time step. The six types of flexible resource allocation and operation parameters include the installed capacity of distributed energy storage, maximum charging power per unit, maximum discharging power per unit, current state of charge, charging efficiency, discharging efficiency, and charge / discharge depth threshold; the number of temperature-controlled load devices, rated power per unit, equivalent thermal resistance, equivalent heat capacity, coefficient of performance (COP), current indoor temperature, current outdoor temperature, user comfort temperature range, and pre-cooling trigger price threshold; and the basic load, maximum adjustment ratio, and current load adjustment of adjustable industrial loads. The parameters include: energy saving, time-of-use pricing, and capacity fluctuation coefficients; number of electric vehicles, single battery capacity, maximum charging power per vehicle, maximum discharging power per vehicle, current state of charge distribution, probability of grid connection time, probability of grid disconnection time, vehicle-to-grid participation ratio, and daily travel electricity demand per vehicle; basic load, maximum adjustable power, and maximum continuous operation time of backup power for data centers; basic load, maximum adjustable power, service load rate, sleep strategy trigger threshold, and equipment power consumption coefficient for 5G base stations; scenario and market parameters include time-of-use pricing, baseline load power of various flexible resources, actual active power output of wind power, maximum active power output of wind power, actual active power output of photovoltaic power, and maximum active power output of photovoltaic power; model training correlation parameters include the first quartile, third quartile, and mean and standard deviation used in the data standardization process for the interquartile range anomaly detection method.

[0248] In this embodiment of the invention, the preprocessing here is completely consistent with the preprocessing process of the training samples in step 102, so as to ensure that the distribution characteristics of the data to be predicted are consistent with those of the training data. Specifically, it includes first cleaning the distribution network data to be predicted, removing invalid and missing parameter data, then correcting the values ​​with abnormal fluctuations to ensure the rationality of the data, and then standardizing the cleaned data to be predicted by using the same standardization rules as in step 102 to convert the input features of each dimension into values ​​with uniform dimensions, eliminating the influence of differences in data dimensions, and making the format and dimension of the data to be predicted completely match the input requirements of the deep neural network. This ensures that the subsequent fast prediction model can perform accurate inference based on a consistent data distribution, avoid prediction deviations caused by inconsistent data distribution, and lay a solid data foundation for the rapid output of accurate quantitative evaluation results of resource adjustability in step 104.

[0249] Furthermore, the quantitative assessment results of resource adjustability include maximum adjustable capacity, resource adjustability time, adjustment response time, resource adjustment rate, individual resource adjustment gradient, resource adjustment accuracy, resource adjustment load rate, and resource adjustment volatility.

[0250] Maximum Adjustable Capacity: Characterizes the maximum adjustment potential that a flexible resource group can provide in terms of power, i.e., the maximum difference between the average aggregated power after adjustment and the average baseline load power before adjustment during the adjustment period; Resource Adjustable Time: Reflects the ability of the flexible resource group to continuously provide effective adjustment power, i.e., the longest duration for which the adjusted power of the resource group remains within the effective threshold range; Adjustment Response Time: Measures the response speed of the flexible resource group, i.e., the time delay from receiving the adjustment command to actually starting to execute the adjustment action; Resource Adjustment Rate: Quantifies the power change capability of the flexible resource group per unit time, i.e., the ratio of adjustment capacity to adjustment response time; Individual Resource Adjustment Gradient: Reflects the adjustment gradient of a single flexible resource unit. The source device's adjustment precision, i.e., the ratio of the total adjustment capacity of the flexible resource group to the number of aggregated devices, reflects the minimum or typical adjustment step size of a single device; resource adjustment accuracy: assesses the following stability of the actual output adjustment power of the flexible resources to the target adjustment power, calculated by the coefficient of variation of the adjustment power sequence; the higher the value, the better the following performance; resource adjustment load rate: characterizes the capacity utilization intensity of the flexible resource group during the adjustment process, i.e., the ratio of the average adjustment power during the adjustment process to the baseline power before adjustment; resource adjustment volatility: reflects the degree of fluctuation in the output power of the flexible resource group during the adjustment process, i.e., the ratio of the change in the adjustment power sequence to the average adjustment power; the lower the value, the more stable the adjustment.

[0251] In this embodiment of the invention, the quantitative evaluation result of resource adjustability includes eight core quantitative indicators, which comprehensively characterize the adjustability of flexible resource groups from multiple dimensions such as power potential, duration, response speed, and adjustment fineness. The specific calculation and physical meaning of each indicator are as follows:

[0252] Maximum adjustable capacity characterizes the maximum adjustment potential that a flexible resource cluster can provide in terms of power. It is calculated as the difference between the average aggregated power after adjustment and the average baseline load power before adjustment during the adjustment period, expressed as:

[0253]

[0254] In the formula, Maximum adjustable capacity; The average baseline load power (kW) of the flexible resource group before adjustment during the t-th adjustment period; This represents the actual adjustment power value of the flexible resource cluster during the t-th scheduling period at time t.

[0255] Resource adjustability time reflects the time capability of a flexible resource swarm to continuously provide adjustable power, and is defined as the duration of the resource swarm's adjustable power, expressed as:

[0256]

[0257] In the formula, Duration (h) for adjusting power for flexible resource clusters; The candidate values ​​for duration represent the time from the start time. Initially, the length of time during which the power adjustment threshold condition is continuously met; To adjust the power threshold (ε>0); This refers to the starting point of the adjustment process, that is, the initial time when flexible resources begin to enter an effective adjustment state; In order to be from arrive Throughout the entire time interval, all times t satisfy the following condition. conditions.

[0258] Adjustment response time measures the time delay characteristic of a flexible resource group from receiving an adjustment command to actually starting to execute the adjustment action. The expression is:

[0259]

[0260] In the formula, To adjust the response time (min). For the first valid response time, This is the moment when steady-state adjustment is achieved.

[0261] The resource regulation rate quantifies the ability of a flexible resource swarm to change power per unit time, i.e., the ratio of regulation capacity to required time, expressed as:

[0262]

[0263] In the formula, For resource adjustment rate, Adjustable capacity (kW) for flexible resource clusters.

[0264] The resource unit adjustment gradient reflects the minimum or typical adjustment step size of a single flexible resource device, indicating the fineness of the adjustment. Its expression is:

[0265]

[0266] In the formula, Adjusting gradients for individual resources The number of flexible resource devices being aggregated.

[0267] To assess the tracking stability between the actual output power and the target power of flexible resources in order to accurately adjust resource regulation, we first define the coefficient of variation:

[0268]

[0269] The recalculation accuracy is:

[0270]

[0271] In the formula, The coefficient of variation is 1. To effectively regulate power, To adjust the standard deviation of the power series, To adjust the mean of the power sequence, To improve the precision of resource regulation.

[0272] Resource regulation load rate characterizes the proportion of the regulation power actually used by the flexible resource group to its maximum adjustable capacity during the regulation process, reflecting the intensity of capacity utilization. The expression is:

[0273]

[0274] In the formula, To adjust the load rate for resources, For average regulation power, To adjust the baseline power.

[0275] Resource regulation volatility reflects the degree of fluctuation in the output power of a flexible resource cluster during regulation. It is defined as the ratio of the standard deviation of the change in regulation power to the average regulation power, expressed as:

[0276]

[0277] In the formula, To regulate resource volatility, To adjust the total number of time steps for a period of time, This represents the actual adjusted power value of the flexible resource cluster during the (t-1)th scheduling period.

[0278] By quantifying the above eight indicators, the adjustment capability of flexible resource groups can be comprehensively and accurately assessed, providing multi-dimensional quantitative basis for the dispatching decisions of the distribution network.

[0279] Based on model-driven and data-driven evaluation results, eight standardized evaluation indicators are calculated and output, resulting in radar charts, heat maps, and other visualization reports. This comprehensively quantifies the adjustability of flexible resources, showcasing the achievement of system optimization goals, detailed comparisons of various flexible resource indicators, capacity analysis, dynamic response curves, and grid constraint satisfaction. It provides intuitive and comprehensive data support for the planning, operation, market trading, and dispatching decisions of the distribution system.

[0280] The verification was conducted using a 33-node IEEE test system, and the results show that:

[0281] Model-driven engine performance verification: such as Figure 3As shown, the heatmap of eight standard indicators for flexible loads visually presents the quantitative evaluation results of six typical flexible resources—distributed energy storage, temperature-controlled loads, adjustable industrial loads, electric vehicles, data centers, and 5G base stations—on eight adjustable capacity indicators. Color gradients and numerical labels clearly compare the adjustment characteristics and performance differences of various resources.

[0282] Distributed energy storage performed the best, showing the darkest green block in the "Resource Adjustment Rate" (2042.61kW / min) indicator, reflecting its extremely fast power response and adjustment capabilities; at the same time, it also showed an ultra-long continuous adjustment potential in the "Resource Adjustable Time" (205.9 hours), making it a core resource for smoothing system power fluctuations.

[0283] Adjustable industrial loads have a significant advantage in the "resource adjustable time" (27.85 hours) index, and can provide stable regulating power for a long time, making them suitable for participating in long-term peak shaving and load management of the system.

[0284] In addition, electric vehicles performed well in terms of "maximum adjustable capacity", data centers performed well in terms of "resource adjustment accuracy", 5G base stations were relatively stable in terms of "resource adjustment volatility", and temperature-controlled loads had a clear advantage in terms of "resource adjustment load rate". The differentiated characteristics of various resources were accurately depicted through heat maps.

[0285] At the system level, the model-driven engine also demonstrated excellent performance: it achieved a 100% renewable energy absorption rate, which can fully absorb distributed wind power and photovoltaic power output; at the same time, the system operating cost was reduced by 20.22% compared with the benchmark model, verifying the dual value of the engine in improving renewable energy absorption and optimizing system economy.

[0286] Data-driven engine performance verification: such as Figure 4 As shown in the figure, this is a heatmap of the prediction accuracy of various flexible loads, which intuitively displays the prediction accuracy (R) of the system target and six typical flexible resources (distributed energy storage, temperature-controlled loads, adjustable industrial loads, electric vehicles, data centers, and 5G base stations) on eight core output indicators. 2 (score), color gradient from deep red (R) 2 ≈0) to dark green (R) 2 ≈1) Clearly reflects the differences in the predictive fit of different indicators.

[0287] The verification results show that the data-driven engine performs excellently:

[0288] High prediction accuracy: Average R-value across all 48 output metrics 2Reaching 0.8379, the color blocks corresponding to most indicators are dark green (e.g., distributed energy storage at indicators 2, 3, and 4; adjustable industrial load at indicators 0, 2, and 3; data centers at indicators 0, 2, 3, and 4, etc.), R 2 The scores are generally above 0.9, with some indicators even approaching 1.0, indicating a high degree of agreement between the prediction results and the baseline values ​​of the physical model; only a few indicators (such as indicator 5 for data centers and indicator 1 for electric vehicles) have a lower R-value. 2 The scores are relatively low, but these low-precision indicators account for a very small percentage, and the overall prediction error distribution is concentrated, with the relative error of the vast majority of indicators being less than 10%.

[0289] High prediction efficiency: The prediction time has been significantly reduced from the hierarchical time of the original physical model to 50 milliseconds, achieving rapid evaluation at the millisecond level, which can meet the response requirements of real-time dispatching of the distribution network.

[0290] The above results fully demonstrate that the data-driven engine, while ensuring high prediction accuracy, also possesses strong real-time performance, providing solid technical support for the efficient operation of the power distribution network.

[0291] See Figures 5-10 The study uses six error distribution charts to visually demonstrate the statistical characteristics and accuracy of the data-driven evaluation engine's prediction results for six typical flexible resources: distributed energy storage, temperature-controlled loads, adjustable industrial loads, electric vehicles, data centers, and 5G base stations. Each chart corresponds to a histogram of the prediction error distribution for a specific resource type. From the overall distribution pattern, the prediction errors for all resources exhibit a highly concentrated distribution centered around zero, with the relative prediction error for the vast majority of samples being less than 10%, validating the data-driven model's good fitting ability and generalization performance across various resources.

[0292] The error distribution of distributed energy storage is the most compact, almost perfectly conforming to a normal distribution, reflecting its clear physical model, well-defined operational constraints, and highly predictable scheduling behavior.

[0293] The error distribution in the data center also shows a high degree of concentration, indicating that its modeling based on service quality constraints and load spatiotemporal transfer characteristics is relatively accurate and has high predictive reliability.

[0294] The temperature control load and the electric vehicle error distribution are slightly wider, showing a slight tail or skewed characteristics. This is related to the influence of factors such as the randomness of user behavior, changes in ambient temperature, and the diversity of travel modes, but it is still within a controllable range.

[0295] The error distribution of adjustable industrial loads has a certain multi-peak or asymmetrical trend, which may be due to uncertainties in actual operation such as sudden changes in production plans and differences in electricity price response strategies.

[0296] The error distribution of 5G base stations is relatively uniform, but still concentrated in the low error range, reflecting the prediction challenges introduced by communication traffic fluctuations and base station sleep strategies. The model can still effectively capture its overall regulation pattern.

[0297] Please see Figure 11 , Figure 11 This is a structural block diagram of a quantitative assessment system for the demand-side flexible resource adjustability of a power distribution network, provided in an embodiment of the present invention.

[0298] This invention provides a quantitative assessment system for the demand-side flexible resource adjustability of a distribution network, comprising:

[0299] The model-driven evaluation engine module 1101 is used to construct a dual-objective optimization model with the optimization objectives of maximizing the consumption of new energy and minimizing the system operating cost.

[0300] The data processing module 1102 is used to generate multiple sets of training samples based on the dual-objective optimization model, and to preprocess the training samples to construct a label dataset.

[0301] The data-driven evaluation engine module 1103 is used to train a pre-set deep neural network by inputting a labeled dataset to obtain a rapid prediction model of flexible resource adjustment capability.

[0302] The standardized evaluation module 1104 is used to input the target distribution network demand-side data into the flexible resource adjustment capability rapid prediction model and output the quantitative evaluation result of resource adjustment capability.

[0303] Please see Figure 12 To enable the efficient engineering application of evaluation methods, an integrated software tool was developed, possessing the following characteristics:

[0304] Architecture design: A five-layer architecture is adopted, including the solution engine layer (Gurobi+PyTorch), the computing framework layer (NumPy / Pandas), the data layer, the application layer (business logic encapsulation), and the presentation layer (UI and visualization).

[0305] Functional modules:

[0306] Parameter configuration interface: Supports visual configuration of system parameters, flexible resource parameters, and scene parameters;

[0307] Dual-engine evaluation: Users can use a dual-engine approach of physical models and data to make rapid predictions;

[0308] Results visualization: Outputs multi-dimensional charts such as heat maps, radar charts, load curves, and voltage distribution;

[0309] Report generation: Automatically generates a visual evaluation report (HTML / PDF format) containing eight indicators;

[0310] Advanced features:

[0311] Supports parameter sensitivity analysis and scenario comparison;

[0312] Provide sampling methods such as LHS and Sobol for data generation;

[0313] It supports API calls and batch processing mode, making it easy to integrate into the power grid dispatching platform;

[0314] System Interface: The main interface includes a menu bar, navigation area, work area, and property panel, supporting interactive operation and real-time status monitoring.

[0315] Since the above is a system corresponding to a quantitative assessment method for the adjustable capacity of flexible resources on the demand side of a distribution network, its implementation principle is consistent with that of a quantitative assessment method for the adjustable capacity of flexible resources on the demand side of a distribution network. For the sake of convenience and brevity, those skilled in the art can clearly understand that the specific working process of the system and modules described above can be referred to the corresponding process in the aforementioned method embodiments, and will not be repeated here.

[0316] Please see Figure 13 , Figure 13 This is a structural block diagram of an electronic device provided in an embodiment of the present invention.

[0317] An electronic device according to an embodiment of the present invention includes: a memory 1301 and a processor 1302. The memory 1301 stores a computer program. When the computer program is executed by the processor 1302, the processor 1302 executes the quantitative assessment method for the adjustable capacity of flexible resources on the demand side of the distribution network as described in the above embodiment.

[0318] Memory 1301 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Memory 1301 has storage space 1303 for program code 1313 for performing any of the method steps described above. For example, storage space 1303 for program code may include individual program codes 1313 for implementing the various steps in the methods described above. This program code can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When run by a computing processing device, this code causes the computing processing device to perform the various steps in the methods described above. This program code can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When these codes are run by a computing processing device, the device causes it to perform the steps in the quantitative assessment method for the demand-side flexible resource adjustability of the distribution network described above.

[0319] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the quantitative assessment method for the adjustable capacity of demand-side flexible resources in a distribution network as described in the above embodiments.

[0320] This invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer performs the quantitative assessment method for the adjustable capacity of distribution network demand-side flexible resources as described in the above embodiments.

[0321] 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.

[0322] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only 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. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0323] 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 achieve the purpose of this embodiment according to actual needs.

[0324] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0325] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several 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 methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0326] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for quantitatively assessing the demand-side flexible resource adjustability of a distribution network, characterized in that, include: Construct a dual-objective optimization model with the goals of maximizing new energy consumption and minimizing system operating costs; Multiple sets of training samples are generated based on the dual-objective optimization model, and the training samples are preprocessed to construct a label dataset; The labeled dataset is used as input to a pre-set deep neural network for training, resulting in a rapid prediction model for flexible resource adjustment capabilities. The fast prediction model for flexible resource regulation capacity is input with the demand-side data of the target distribution network and outputs a quantitative assessment result of resource regulation capacity.

2. The method for quantitatively evaluating the demand-side flexible resource adjustability of a distribution network according to claim 1, characterized in that, The construction of the dual-objective optimization model, which aims to maximize the absorption of new energy sources and minimize the system operating cost, includes: A distributed energy storage model is constructed based on power constraints, energy dynamic constraints, state of charge constraints, and ramp constraints. A temperature-controlled load model is constructed based on thermodynamic model constraints, power constraints, comfort zone limitation constraints, precooling strategy constraints, and equipment heterogeneity constraints. An adjustable industrial load model is constructed based on the constraints of adjustable power, energy conservation, time-of-use electricity price response, capacity variation, and simultaneity coefficient. An electric vehicle model is constructed based on power boundary constraints, energy dynamic constraints, energy limit constraints, energy satisfaction constraints, charge-discharge mutual exclusion constraints, and SOC distribution constraints. A data center model is constructed based on adjustable load power boundary constraints, backup power duration constraints, and energy conservation constraints. A 5G base station model is constructed based on adjustable service load boundary constraints, sleep mode energy saving ratio constraints, active base station number constraints, adjustment range constraints, and traffic diversity constraints. With the optimization objectives of maximizing the absorption of new energy and minimizing the system operating cost, a dual-objective optimization model is constructed by coupling the distributed energy storage model, the temperature-controlled load model, the adjustable industrial load model, the electric vehicle model, the data center model, and the 5G base station model, and system safety operation constraints are set. The system's safe operation constraints include power balance constraints, new energy output constraints, flexible resource cluster adjustment constraints, system node voltage constraints, and line current carrying capacity constraints.

3. The method for quantitatively evaluating the demand-side flexible resource adjustability of a distribution network according to claim 1, characterized in that, The process involves generating multiple sets of training samples based on the dual-objective optimization model, preprocessing the training samples, and constructing a labeled dataset, including: Using the key parameters of the dual-objective optimization model as the sampling objects, the value space of the sampling parameters is determined. The key parameters include the configuration parameters of flexible resources, system constraint parameters, and scenario parameters. Based on the value space of the sampling parameters, multiple sets of parameter combinations are generated; Each of the parameter combinations is input into the bi-objective optimization model for solution, and the benchmark results corresponding to each parameter combination are obtained. Each set of parameter combinations is used as input features, and the corresponding benchmark results are used as output labels to form multiple sets of training samples. The training samples are preprocessed, including data cleaning, outlier handling, and standardization. Multiple preprocessed training samples are divided according to a preset partitioning ratio to obtain a labeled dataset.

4. The method for quantitatively evaluating the demand-side flexible resource adjustability of a distribution network according to claim 3, characterized in that, The generation of multiple parameter combinations based on the sampling parameter value space includes: Based on the value space of the sampling parameters, uniform sampling is performed using the Latin hypercube sampling method to generate multiple sets of parameter combinations.

5. The method for quantitatively evaluating the demand-side flexible resource adjustability of a distribution network according to claim 1, characterized in that, Before inputting the target distribution network demand-side data into the rapid prediction model for flexible resource regulation capacity, the method further includes: The preprocessing is performed on the demand-side distribution network data to be predicted for the target distribution network.

6. The method for quantitatively evaluating the demand-side flexible resource adjustability of a distribution network according to claim 1, characterized in that, The quantitative evaluation results of resource adjustability include maximum adjustable capacity, resource adjustability time, adjustment response time, resource adjustment rate, individual resource adjustment gradient, resource adjustment accuracy, resource adjustment load rate, and resource adjustment volatility.

7. A quantitative assessment system for the demand-side flexible resource adjustability of a distribution network, characterized in that, include: The model-driven evaluation engine module is used to build a dual-objective optimization model with the optimization objectives of maximizing the consumption of new energy and minimizing the system operating cost. The data processing module is used to generate multiple sets of training samples based on the dual-objective optimization model, and to preprocess the training samples to construct a label dataset; The data-driven evaluation engine module is used to train a pre-set deep neural network using the labeled dataset as input, and to obtain a rapid prediction model of flexible resource adjustment capability. The standardized evaluation module is used to input the target distribution network demand-side data into the flexible resource adjustment capability rapid prediction model and output the quantitative evaluation result of resource adjustability capability.

8. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the quantitative assessment method for the demand-side flexible resource adjustability of the distribution network as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the quantitative assessment method for the demand-side flexible resource adjustability of the distribution network as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, wherein when the program instructions are executed by a computer, the computer performs the quantitative assessment method for the demand-side flexible resource adjustability of the distribution network as described in any one of claims 1-6.

Images