New energy consumption-oriented demand side resource collaborative scheduling method, device and equipment

Abstract

The invention provides a demand side resource collaborative scheduling method, device and equipment for new energy consumption, and relates to the technical field of new energy power generation. The method comprises the steps of inputting multi-source time sequence data into a pre-trained LSTM neural network model, obtaining a core decision variable of a multi-type load, and inputting the core decision variable as an initial solution into a multi-type demand response collaborative optimization operation model; wherein the core decision variable comprises a translation starting time period and a duration of a translation load, a transfer power and a target time period of the transferable load, and a reduction power and a duration of a reduction load; a target function of the multi-type demand response collaborative optimization operation model takes the minimization of the new energy abandoned electric quantity as a target; performing iterative fine tuning on the initial solution by adopting a particle swarm optimization algorithm, and outputting an optimized adjustment strategy when a preset termination condition is met; wherein the adjusting strategy comprises the response quantity of each type of load per hour and a conventional unit adjusting strategy. The operation flexibility and economy of the power system can be enhanced at the same time.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation technology, and in particular to a method, apparatus and equipment for demand-side resource collaborative scheduling for new energy consumption. Background Technology

[0002] With the rapid development and large-scale integration of new energy sources into the power grid, the flexibility of power system regulation faces significant challenges. Aggregated response of demand-side resources has become a crucial means to enhance the capacity for new energy absorption. However, existing research lacks sufficient understanding of the improvement effects and mechanisms of demand-side response involving a large number of loads in new energy absorption. Furthermore, demand-side resources are diverse in type and have varying regulation characteristics; effectively aggregating and optimizing the scheduling of these resources to achieve new energy absorption remains a pressing issue. Therefore, constructing a demand-side resource aggregation response analysis method for new energy absorption is of great significance for improving grid flexibility and new energy absorption capacity.

[0003] In theoretical research, scholar Ge Xiaolin, in her paper "Resource Aggregation Scheduling Strategy for Uncertain Demand Response Based on Integrated Semi-Ladder Cloud," proposed a stochastic optimization scheduling strategy based on an integrated semi-ladder cloud model to address the uncertainty of demand response. This method analyzes user participation intentions and behavioral characteristics using a Logit-XGBoost model to quantify resource output; however, it has high computational resource requirements and is difficult to implement in real-time in large-scale power grids. The model is complex, has a long optimization time, and its prediction accuracy may be affected by user uncertainty, impacting scheduling effectiveness.

[0004] In his paper "Data-Driven Frequency-Constrained Economic Scheduling Considering Demand-Side Rapid Response Resources," scholar Yao Yu proposed an innovative model that analyzes the potential of demand-side resources such as variable frequency air conditioners, battery energy storage, and electric vehicles in frequency regulation, and constructs a joint economic scheduling framework to optimize both the source and load sides. While data-driven approaches offer significant advantages, they rely on a large amount of historical data, posing challenges to real-time performance and accuracy when the system changes. External factors increase uncertainty, and the model has poor adaptability; changes in disturbances require retraining, increasing computational overhead.

[0005] Therefore, there is a need to provide a demand-side resource forecasting solution that balances computational overhead and renewable energy absorption efficiency. Summary of the Invention

[0006] This invention provides a method, apparatus, and equipment for demand-side resource collaborative scheduling for renewable energy consumption, in order to solve the problem of balancing computational overhead and renewable energy consumption efficiency.

[0007] In a first aspect, embodiments of the present invention provide a demand-side resource collaborative scheduling method for renewable energy consumption, comprising: Multi-source time-series data are input into a pre-trained LSTM neural network model to obtain core decision variables for multiple types of loads; wherein, the multiple types of loads include shiftable loads, transferable loads, and reduceable loads; the core decision variables include the shift start time and duration of shiftable loads, the transfer power and target time of transferable loads, and the reduction power and duration of reduceable loads; The core decision variables are used as initial solutions and input into the multi-type demand response collaborative optimization operation model; wherein, the objective function of the multi-type demand response collaborative optimization operation model is to minimize the amount of renewable energy curtailment; the constraints of the multi-type demand response collaborative optimization operation model include: load shifting constraints, load transfer constraints, load reduction constraints, and power system operation constraints; The initial solution is iteratively fine-tuned using the Particle Swarm Optimization (PSO) algorithm. When a preset termination condition is reached, the optimized adjustment strategy is output. The adjustment strategy includes the hourly response of each type of load and the conventional unit adjustment strategy.

[0008] In one possible implementation, the shiftable load constraints include: a shift start time constraint, a shift duration constraint, and a constraint on the load power after shift within an acceptable time interval. The constraints on transferable loads include: the constraint that the total power consumption remains unchanged before and after the transfer, the constraint that the transfer power is allowed within a certain range, the constraint that the transfer target time period is limited, and the constraint that the minimum continuous operating time is limited.

[0009] The load reduction constraints include: power reduction constraints, minimum continuous reduction time constraints, maximum continuous reduction time constraints, and maximum number of reduction constraints. The power system operation constraints include at least one of the following: real-time power supply and demand balance constraints, upper and lower limits of conventional unit output constraints, system spinning reserve constraints, line power transmission constraints, and node voltage phase angle constraints.

[0010] In one possible implementation, the iterative fine-tuning of the initial solution using the PSO algorithm includes: The core decision variables output by the LSTM are used as elite particles to initialize the particle population. A comprehensive fitness function is constructed with the goal of minimizing the amount of renewable energy curtailment. The comprehensive fitness function is constructed by integrating renewable energy output, load response power, renewable energy curtailment, operating costs, constraints and penalties, and user satisfaction. The particle's quality is evaluated based on the comprehensive fitness function, the individual optimal position and the global optimal position are updated, and the particle's velocity and position are iteratively updated based on the updated individual optimal position and the global optimal position.

[0011] In one possible implementation, the step of outputting an optimized adjustment strategy when a preset termination condition is met includes: When the maximum number of iterations is reached or the fitness function value converges, the iteration is terminated, and the global optimal position at this point is output as the final optimized adjustment strategy.

[0012] In one possible implementation, the PSO algorithm employs a collaborative decomposition strategy, decomposing the decision vector into several sub-vectors according to the load type. Each sub-vector is optimized in parallel by an independent subgroup, and the solutions of each subgroup are merged at a fixed iteration interval to evaluate the overall fitness.

[0013] In one possible implementation, following the optimized adjustment strategy for the output, the following is also included: Based on the adjustment strategy, the curtailment improvement rate of the power system is calculated; wherein, the curtailment improvement rate is the ratio of the reduction in the amount of curtailment before and after optimization to the amount of curtailment before optimization. If the curtailment improvement rate is lower than a preset threshold or the operating cost exceeds a cost threshold, the current adjustment strategy and constraint violation information will be used as feedback samples to fine-tune the Long Short-Term Memory (LSTM) neural network model.

[0014] In one possible implementation, the LSTM neural network model includes an input layer, a first LSTM layer, a second LSTM layer, a third LSTM layer, and a fully connected layer. The input layer receives the multi-source time-series data; the first LSTM layer is used to capture short-term time-series dependencies; the second LSTM layer is used to extract medium-term correlations; the third LSTM layer is used to mine long-term trends; and the fully connected layer is used to output the core decision variables.

[0015] In one possible implementation, the multi-source time-series data includes: wind power sequence, photovoltaic power sequence, historical power curve of load that can be shifted, historical power curve of load that can be transferred, historical power curve of load that can be reduced, and minimum and maximum technical output parameters of conventional units.

[0016] In one possible implementation, prior to inputting the multi-source time-series data into the pre-trained LSTM neural network model, the following is also included: The multi-source time-series data is preprocessed; the preprocessing includes: linear interpolation to fill missing values, outlier removal, and normalization.

[0017] Secondly, embodiments of the present invention provide a demand-side resource collaborative scheduling device for renewable energy consumption, comprising: The core decision variable determination module is used to input multi-source time-series data into a pre-trained LSTM neural network model to obtain core decision variables for multiple types of loads; wherein, the multiple types of loads include movable loads, transferable loads, and reduceable loads; the core decision variables include the movable load's movable start time and duration, the transfer power and target time of the transferable load, and the reduce power and duration of the reduceable load; The adjustment strategy determination module is used to input the core decision variables as initial solutions into the multi-type demand response collaborative optimization operation model, and to use the PSO algorithm to iteratively fine-tune the initial solutions. When the preset termination condition is reached, the optimized adjustment strategy is output. The objective function of the multi-type demand response collaborative optimization operation model is to minimize the amount of renewable energy curtailment. The constraints of the multi-type demand response collaborative optimization operation model include: load shifting constraints, load transfer constraints, load reduction constraints, and power system operation constraints. The adjustment strategies include the hourly response amount of each type of load and the adjustment strategies of conventional generating units.

[0018] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0019] In this embodiment of the invention, a pre-trained LSTM neural network model is first used to process multi-source time-series data, fully adapting to the time-series characteristics of new energy generation and load changes in the power system. This accurately outputs the core decision variables of multiple load types, providing a high-quality initial solution for subsequent optimization and effectively reducing the computational complexity and time consumption of subsequent optimization. Then, this initial solution is input into a multi-type demand response collaborative optimization operation model with the goal of minimizing new energy curtailment. By comprehensively considering the constraints of multiple load types and power system operation constraints, the feasibility and safety of the scheduling scheme are ensured. Finally, the PSO algorithm is used to iteratively fine-tune the initial solution to further approach the optimal target. The final output adjustment strategy can achieve coordinated scheduling of multiple load types and conventional units, significantly improving the new energy absorption capacity, reducing curtailment, and enhancing the flexibility and economy of power system operation. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the implementation of a demand-side resource collaborative scheduling method for new energy consumption according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the technical route of a demand-side resource collaborative scheduling method for new energy consumption provided by an embodiment of the present invention; Figure 3This is a schematic diagram of the optimization solution process provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of system load distribution and power curtailment curves provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the demand-side resource collaborative scheduling device for new energy consumption provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] Scholar Ge Xiaolin proposed a stochastic optimization scheduling strategy based on a comprehensive semi-ladder cloud model. First, the Logit-XGBoost model is used to analyze user participation intentions, and combined with user behavioral characteristics, the output of demand response resources is quantified. Compared to traditional demand response models, this research integrates demand-side resources through a virtual battery (VB) model, accurately characterizing the response characteristics of three types of loads and optimizing the scheduling effect of demand response. Compared to traditional single-resource scheduling methods, this comprehensive model, by quantifying and integrating the response characteristics of resources, can significantly improve the system's scheduling efficiency, especially in peak shaving and valley filling and load curve adjustment. Although the comprehensive semi-ladder cloud model and the Logit-XGBoost model can effectively analyze user participation intentions and behavioral characteristics and accurately quantify the output of demand response resources, these methods have high computational resource requirements, especially in large-scale power grids and complex environments, which may make real-time execution of the scheduling strategy difficult. Furthermore, while the virtual battery model and accurate characterization of the response characteristics of multiple load types can improve scheduling efficiency, they also increase model complexity, making the scheduling process potentially require longer optimization and computation times, which may affect the system's flexibility and response speed in practical operation. In addition, due to the uncertainty of users' willingness and behavioral characteristics, the model's prediction accuracy may also be biased, resulting in a suboptimal scheduling outcome and affecting the effectiveness of demand response.

[0022] In his paper "Data-Driven Frequency-Constrained Economic Dispatch Considering Demand-Side Rapid Response Resources," scholar Yao Yu proposed a model that analyzes the potential of demand-side resources such as variable frequency air conditioners, battery storage, and electric vehicles in power system frequency regulation. This model, by considering the rapid response characteristics of demand-side resources, constructs a joint economic dispatch framework capable of optimizing the dispatch of both the source and load sides of the power system based on frequency security constraints. Compared to traditional power demand response, rapid response resources on the demand side can quickly adjust the load in a short time, thus providing the system with more flexible frequency regulation capabilities, especially as the proportion of renewable energy gradually increases and system inertia decreases. This aggregated response of demand-side resources significantly improves the stability and resilience of the system frequency, especially under high-power disturbances, effectively avoiding frequency overruns and thus improving the safety and reliability of the power system. While this technology has significant advantages in data-driven modeling, it relies on a large amount of historical data for training, and its computational accuracy and real-time performance still face challenges when dealing with frequent system changes and uncertainties. The rapid response characteristics of demand-side resources may be affected by external factors, leading to increased uncertainty during dispatch, which may affect the frequency security of the system in extreme cases. Furthermore, the model has poor adaptability; once the input perturbation conditions change, the model needs to be retrained, which increases the complexity and computational cost of real-time scheduling.

[0023] This invention aims to propose a demand-side flexibility resource aggregation response analysis model for renewable energy consumption, revealing the mechanism of demand-side resource provision flexibility and clarifying the promoting effect of large-scale demand-side load aggregation response on renewable energy consumption. It analyzes the mechanism of demand-side resource provision flexibility and proposes a calculation method for adjusting the demand-side flexibility resource space. Considering multiple types of load constraints such as shiftable loads, transferable loads, and loads that can be reduced, it proposes a multi-type demand response collaborative optimization operation model to promote renewable energy consumption and improve the renewable energy consumption effect.

[0024] In this invention, the power system consists of three parts: conventional generating units, new energy generating units, and demand-side resources.

[0025] Conventional generating units not only need to meet load demand but also need to smooth out fluctuations in renewable energy output. When renewable energy generation exceeds the system's regulation capacity, wind and solar power curtailment occurs. Demand-side resources can provide a certain degree of flexibility to reduce renewable energy curtailment. To maximize renewable energy consumption, during periods of renewable energy curtailment, the actual output of conventional generating units should equal their minimum technical output. In this case, the required demand-side flexibility is to increase demand-side flexibility, corresponding to the space between the curves of "conventional unit minimum technical output + renewable energy generation" and "load + interconnection line power transmission." During periods without renewable energy curtailment, the actual output of conventional generating units should not be less than their minimum technical output. In this case, the required demand-side flexibility is to decrease demand-side flexibility, corresponding to the theoretical maximum adjustable space between the curves of "conventional unit actual output" and "conventional unit minimum technical output."

[0026] This invention proposes a method for calculating the demand-side flexibility resource adjustment space, using the power curtailment of new energy sources as an indicator, and comprehensively considering the power generation of photovoltaic units in the system. Wind turbine power generation The actual grid-connected power output of photovoltaic units The actual grid output of wind turbine units Actual output of conventional units System interconnection line power transmission System load : (1) In the formula, The unit's operating status. This refers to the set of conventional units in the system.

[0027] 1) If , making ,but , For the minimum technical output of conventional generating units, and given that the system experiences renewable energy curtailment at this moment, the required increase in demand-side flexibility resources is: (2) 2) If , making ,but , At this moment, there is no curtailment of renewable energy in the system, and the theoretical maximum reduction in demand-side flexibility resource power is: (3) Reducing demand-side flexibility primarily involves supplying loads that can be shifted, transferred, or reduced. Increasing demand-side flexibility can be achieved by shifting, transferring, or reducing the maximum or minimum technical output of conventional generating units using a certain amount of reduced demand-side flexibility resources, and by reducing some renewable energy curtailment through appropriate demand response strategies.

[0028] This invention designs a multi-type demand-side flexibility resource collaborative optimization mechanism for renewable energy consumption. It allows for the full exploration and utilization of the flexibility potential of loads that can be shifted, transferred, and reduced in the operation of the power system, thereby promoting the efficient consumption of renewable energy while improving the overall operational flexibility and economy of the power system.

[0029] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0030] Figure 1 This is a flowchart illustrating the implementation of a demand-side resource collaborative scheduling method for renewable energy consumption, provided by an embodiment of the present invention. Figure 1 As shown, it includes the following steps: S101, input multi-source time series data into a pre-trained LSTM neural network model to obtain the core decision variables of multiple types of loads; among them, multiple types of loads include transferable loads, transferable loads, and loads that can be reduced; the core decision variables include the transfer start time and duration of transferable loads, the transfer power and target time of transferable loads, and the reduction power and duration of loads that can be reduced.

[0031] The execution subject in various embodiments of the present invention can be a server, processor, microprocessor, or other device with data processing capabilities. In actual implementation, the specific implementation method of the execution subject can be selected according to actual needs. This embodiment does not impose any particular restrictions on this, as long as it is a device with data processing capabilities.

[0032] In one possible implementation, the multi-source time-series data includes: wind power series, photovoltaic power series, historical power curves of loads that can be shifted, historical power curves of loads that can be transferred, historical power curves of loads that can be reduced, and minimum and maximum technical output parameters of conventional units.

[0033] The multi-source time-series data encompasses renewable energy generation data such as wind and solar power, historical power data that can be shifted, transferred, and load reduced, as well as key operating parameters of conventional units. This ensures the comprehensiveness and integrity of the input data, fully reflecting the operating characteristics and correlations on both the source and load sides of the power system. The rich and comprehensive data provides ample learning material for the LSTM neural network model, enabling it to fully explore the intrinsic connections between renewable energy output fluctuations, load change patterns, and conventional unit operating constraints. This results in more accurate core decision variables that better meet actual needs, providing reliable data support for subsequent multi-type demand response collaborative optimization operation models and fine-tuning of the particle swarm optimization algorithm. Ultimately, this ensures that the entire scheduling method can efficiently and accurately achieve the goal of maximizing renewable energy consumption.

[0034] S102, the core decision variables are used as the initial solution and input into the multi-type demand response collaborative optimization operation model; the objective function of the multi-type demand response collaborative optimization operation model is to minimize the amount of renewable energy curtailment; the constraints of the multi-type demand response collaborative optimization operation model include: load shifting constraints, load transfer constraints, load reduction constraints, and power system operation constraints.

[0035] In this embodiment, the amount of abandoned renewable energy includes abandoned wind power and abandoned photovoltaic power.

[0036] In one possible implementation, the load shifting constraints include: a shifting start time constraint, a shifting duration constraint, and a constraint on the load power after shifting within an acceptable time interval. The constraints on transferable loads include: the constraint that the total power consumption remains unchanged before and after the transfer, the constraint that the transfer power is within a certain range, the constraint that the transfer target time period is limited, and the constraint that the minimum continuous operating time is limited.

[0037] The load reduction constraints include: power reduction constraints, minimum continuous reduction time constraints, maximum continuous reduction time constraints, and maximum number of reduction constraints. Power system operation constraints include at least one of the following: real-time power supply and demand balance constraints, upper and lower limits of conventional unit output constraints, system spinning reserve constraints, line power transmission constraints, and node voltage phase angle constraints.

[0038] Among these, the constraints on the start time, duration, and power range of load transferable tasks ensure that the scheduling of these tasks conforms to their characteristics of continuous time and constant power, preventing disorderly transfers from affecting normal load use and system stability. Constraints on power conservation, power range, target time periods, and minimum continuous operating time for transferable loads ensure the rationality of load transfer while avoiding losses caused by frequent equipment start-ups and shutdowns. Constraints on load reduction fully explore peak-shaving potential while considering user satisfaction, preventing excessive reductions from impacting user experience. Power system operation constraints build a robust safety barrier from key dimensions such as supply-demand balance, unit output, and reserve capacity. These constraints work together, layer by layer, to ensure the orderly and compliant scheduling of various demand-side resources, providing reliable guarantees for renewable energy consumption and maintaining the stable operation of the power system.

[0039] S103 uses the PSO algorithm to iteratively fine-tune the initial solution. When the preset termination condition is reached, the optimized adjustment strategy is output. The adjustment strategy includes the hourly response of each type of load and the adjustment strategy of conventional units.

[0040] In this embodiment, a pre-trained LSTM neural network model is first used to process multi-source time-series data, fully adapting to the time-series characteristics of renewable energy generation and load changes in the power system. This accurately outputs the core decision variables of multiple load types, providing a high-quality initial solution for subsequent optimization and effectively reducing the computational complexity and time consumption of subsequent optimization. Then, this initial solution is input into a multi-type demand response collaborative optimization operation model with the goal of minimizing renewable energy curtailment. By comprehensively considering the constraints of multiple load types and power system operation constraints, the feasibility and safety of the scheduling scheme are ensured. Finally, the PSO algorithm is used to iteratively fine-tune the initial solution to further approach the optimal target. The final output adjustment strategy can achieve coordinated scheduling of multiple load types and conventional units, significantly improving the renewable energy absorption capacity, reducing curtailment, and enhancing the flexibility and economy of power system operation.

[0041] Before implementing the demand-side resource collaborative scheduling method for renewable energy consumption provided in this application embodiment, it is necessary to establish a multi-type demand response collaborative optimization operation model and an LSTM neural network model for renewable energy consumption in advance, and pre-train the models. Figure 2 This is a schematic diagram of the technical route of a demand-side resource collaborative scheduling method for renewable energy consumption provided by an embodiment of the present invention. Figure 3 This is a schematic diagram of the optimization solution process provided in an embodiment of the present invention. (Combined with...) Figure 2 and Figure 3 This facilitates understanding of the demand-side resource collaborative scheduling method for new energy consumption provided in the embodiments of this application.

[0042] This application constructs an objective function based on maximizing the absorption of renewable energy (wind power and photovoltaic) by the power system, i.e., minimizing the amount of renewable energy curtailed by the system, and comprehensively considers the power generation capacity of wind turbine units connected to the grid. And the power generation of photovoltaic units connected to the grid. : (4) in, (5) (6) In the formula, t is the time period identifier, t=1,2,…,T; The time interval is T; T is the total number of time segments, which this application takes as... =1h, T=24; A collection of wind turbine units within the system; This refers to the collection of photovoltaic units within the system.

[0043] Among them, the characteristic of a transferable load is that it is time-continuous, and the power magnitude remains unchanged at every moment before and after the transfer. For a given transferable load... Its power distribution vector is: (7) In the formula, The load start time; For duration; The physical meaning is exist Load power at any given time.

[0044] Let the acceptable period of load transfer be the interval. There exists within the interval Use 0-1 variables express At the beginning of time period t, that is to say The translation begins at time t. The initial time period set. for: (8) like This indicates that the load has not been shifted. and This indicates that the load has been shifted from time t onwards. The corresponding translated power distribution vector for: (9) (10) After a load can be moved, there are exactly two possibilities: 1) it is not moved; 2) it is moved to an acceptable time interval, with the constraint: (11) when hour, The corresponding power distribution vector is meaningless, so we set all of them to 0.

[0045] The characteristics of transferable loads are that they can be interrupted in time, the power output can be changed before and after the transfer, and the total power consumption remains unchanged. For a given transferable load... Using 0-1 variables This indicates its transition state at time t, i.e. express Shift to time period t This indicates no transfer. The load power distribution vector that can be transferred is... , It is a vector The t-th element represents the load power transferred out at time t. The load power distribution vector to be transferred in is... , It is a vector The t-th element. The power distribution vector of the transferred load is... .

[0046] (12) The constraints that transferable loads must meet include: the total power consumption remains unchanged before and after the transfer; the power of the transferred load at time t is between the maximum and minimum allowable transfer power; the transfer is within the allowable range; and to avoid transferring to multiple discontinuous time periods, i.e., frequent start-ups and shutdowns of equipment, the minimum continuous operating time of the transferred load must be constrained. That is: (13) (14) In the formula, This is the acceptable transfer range; , These represent the maximum and minimum allowable load transfer power at time t, respectively; This is the minimum continuous running time.

[0047] The characteristic of a load that can be reduced is that power can be reduced. For a given load that can be reduced... Load response reduces users' electricity consumption. (Using 0-1 variables) Indicates that the load can be reduced. In the reduction state of time period t, i.e. express It was reduced during time period t. Indicates no reduction, It is a vector The t-th element represents the load power reduced at time t. (Load after scheduling) for: (15) (16) Considering user satisfaction, constraints also need to be placed on the minimum and maximum consecutive reduction time and the number of reductions, namely: (17) In the formula, This is the acceptable reduction range; Minimum continuous reduction time; This is the maximum continuous reduction time; This represents the maximum number of reductions.

[0048] During operation, the power system must meet constraints including the shortest operating time of conventional generating units, the number of start-ups and shutdowns, real-time power supply and demand balance, system spinning reserve, line power transmission, voltage phase angle, unit output, and system power flow balance. (18) (19) (20) (twenty one) (twenty two) (twenty three) (twenty four) In the formula, This is the shortest operating time for conventional generating units; , These represent the 0-1 variables for unit start-up and shutdown. , These represent the start-up and shutdown operations performed by unit g during time period t; This represents the maximum number of times the unit can be started. Let be the load of node i at time t; A set of nodes within the system; Let be the external transmission power of tie line i at time t; For the collection of connecting lines; The critical load power at time t; , , , These are the sets of loads within the system that can be moved, transferred, reduced, or are considered important. This represents the upper limit of the active power output of thermal power unit g; Reserve capacity for system rotation; This is the system's negative spin-off reserve capacity; Let be the transmission power of line l at time t; The upper limit of the transmission power of line l; This is the lower limit of the transmission power of line l; A collection of transmission lines; Let be the voltage phase angle of node i at time t; This represents the maximum voltage phase angle. This represents the maximum uphill ramp power of unit g. This represents the maximum downhill ramp power of unit g. , These are the upper limits of active power output for wind turbine (w) and photovoltaic (v), respectively. The active power matrix injected into the node at time t; Let g be the power output matrix of the unit at time t; M is the node-unit association matrix. If unit g is connected to node n, the matrix element is 1; otherwise, the matrix element is 0. Let B be the load matrix of the node at time t; and let B be the node admittance matrix. Let be the node voltage phase angle matrix at time t; Let be the vector of active power of each branch at time t; It is a diagonal matrix composed of branch admittances.

[0049] To adapt to the temporal characteristics of new energy power generation and load changes in the power system, and to better achieve the core objective of "coordinated optimization of multi-type demand-side resources to minimize the amount of new energy curtailment", an LSTM long short-term memory network is introduced to construct a complete optimization analysis process.

[0050] First, data input and preprocessing were carried out. Wind power sequences were collected with a sampling step of 1 hour and a period of 24 hours. Photovoltaic power sequence ( Data on new energy power generation, etc.; load shifting. Transferable load It can reduce the load. Historical power curves and other load data; and minimum technical output of conventional units. Maximum technical output Parameters such as these.

[0051] Data cleaning is performed by filling in missing values ​​and removing outliers outside the physical limits using linear interpolation, followed by normalization.

[0052] (25) Map all time series data to The interval is used to eliminate the influence of dimensions.

[0053] In one possible implementation, the LSTM neural network model includes an input layer, a first LSTM layer, a second LSTM layer, a third LSTM layer, and a fully connected layer. The input layer receives multi-source time-series data; the first LSTM layer is used to capture short-term time-series dependencies; the second LSTM layer is used to extract medium-term correlations; the third LSTM layer is used to mine long-term trends; and the fully connected layer is used to output core decision variables.

[0054] To address the time-dependent characteristics of power systems (such as the influence of current load dispatch on renewable energy output in the previous 3 hours), a 3-layer LSTM network is designed, with the following structure: Table 1 LSTM Neural Network Structure

[0055] The core decision variables are output using the LSTM network model described above: Table 2 Core Decision Variable Matrix

[0056] The input layer receives 12-dimensional time-series data containing 6 basic features and 6 derived features. The first two LSTM layers are set with 64 and 32 hidden units respectively and return sequences, capturing short-term time-series dependencies such as intraday load fluctuations and medium-term correlations such as inter-period resource coupling. The third LSTM layer is set with 16 hidden units and does not return sequences, mining long-term trends such as the pattern of power curtailment periods. Finally, the fully connected layer outputs 9-dimensional core decision variables, corresponding to the start time and duration of loads that can be shifted, the transfer power and target time of loads that can be transferred, and the reduction power and duration of loads that can be reduced, accurately matching the scheduling needs of the three types of loads.

[0057] In this embodiment, the multi-layered structure of the LSTM neural network model has a clear division of labor and works synergistically. The input layer comprehensively receives multi-source time-series data, providing rich and complete feature support for the model. This structural design, adapted to the time-series characteristics of the power system, enables the model to fully learn the potential patterns in the data, and the output decision variables are more in line with actual scheduling needs. This lays a solid foundation for subsequent collaborative optimization and particle swarm optimization, improving the accuracy and effectiveness of the overall scheduling scheme.

[0058] As described in the foregoing embodiments, in one possible implementation, before inputting the multi-source time-series data into the pre-trained LSTM neural network model, the method further includes: Preprocessing is performed on multi-source time-series data; preprocessing includes: linear interpolation to fill missing values, removal of outliers, and normalization.

[0059] During the model training phase, the core objective is to "minimize the amount of renewable energy wasted," and a loss function is constructed as follows: (26) In the formula, The actual amount of electricity wasted during time period t. To predict the amount of power wasted by the model; To constrain penalty items; This is the penalty coefficient.

[0060] Among them, the constraint penalty term limits the output by applying hard constraints such as power balance and unit output range using a custom activation function. For example, it can reduce load power to not exceed the maximum allowable value. (27) In the formula, For LSTM hidden states, For the output of the fully connected layer, This is the maximum allowable power reduction.

[0061] For soft constraints such as user satisfaction, the constraint is achieved by increasing the penalty value for violation scenarios: For the training process, historical data from the past year (365 days) was used to divide the training set and validation set into a 7:3 ratio, and the Adam optimizer with a learning rate of 0.001 and a decay rate of 0.9 was used for training. At the same time, an early stopping strategy was adopted, and the model was stopped when the validation set loss did not decrease for 5 consecutive rounds to avoid overfitting, thereby embedding the load and system constraints involved in formulas (8)-(24) in the patent into the model.

[0062] In one possible implementation, the PSO algorithm is used to iteratively fine-tune the initial solution, including: The core decision variables output by the LSTM are used as elite particles to initialize the particle population. A comprehensive fitness function is constructed with the goal of minimizing the amount of renewable energy curtailment. The comprehensive fitness function is constructed by integrating renewable energy output, load response power, renewable energy curtailment, operating costs, constraints and penalties, and user satisfaction. The particle's quality is evaluated based on the comprehensive fitness function, and the individual optimal position and the global optimal position are updated. Based on the updated individual optimal position and the global optimal position, the particle's velocity and position are iteratively updated.

[0063] In actual execution, the core decision variables output by LSTM are first used as the initial solution and substituted into the multi-type demand response collaborative optimization operation model mentioned above. Then, the initial solution is fine-tuned through the PSO algorithm to ensure that the system safety constraints are met while further approaching the goal of "minimizing the amount of abandoned power". Finally, the response amount of various loads per hour and the conventional unit adjustment strategy are output.

[0064] Based on the initial scheduling solution output by the LSTM network, the PSO algorithm is introduced to perform global collaborative fine-tuning of the scheduling scheme for multiple resource types, thereby further improving the feasibility of the solution and minimizing the system's wasted power. This step uses the LSTM output as input to construct an initial population for collaborative optimization. The globally optimal scheduling solution is obtained through adaptive particle swarm optimization iterative search, and the optimization results are fed back to the LSTM model to form a closed-loop learning process.

[0065] The initial scheduling variable vector output by the LSTM network is: ; This includes core parameters such as the start time and duration of loads that can be transferred, the transfer power and target time of loads that can be transferred, and the reduction power and duration of loads that can be reduced. The system constraint parameter set is as follows: ; in, These are the lower and upper limits of the load, respectively. For time constraints, To balance supply and demand; The optimized weight parameters are as follows: ; in, These correspond to operating costs, constraints and penalties, and user satisfaction factors, respectively.

[0066] The initialization phase uses the vector output by the LSTM As elite particles embedded into the initial population of PSO, the initial particle position matrix is ​​constructed as follows: (28) Where S is the population size. The remaining particles are generated using Latin hypercube or uniform random sampling. The initial velocity matrix is: (29) After initialization, the individual optimal values ​​of each particle are obtained. and global optimal .

[0067] With minimizing the system's wasted power as the primary objective, a comprehensive fitness function is constructed: (30) in, Let be the renewable output at time t. The load response power under the particle scheme, This indicates that only positive values ​​are squared to measure the amount of electricity wasted; For the running cost function, To constrain penalties, This is a penalty for user satisfaction. It involves calculating the values ​​of each particle by calling the system's runtime model. And update its individual optimal and global optimal indices.

[0068] If an out-of-bounds or constraint violation occurs after the particle position is updated, a repair operator is used to adjust the variable. For time period and duration variables, the following applies: (31) For power-related variables, the following applies: (32) If the supply and demand balance constraint is still not met Then a penalty term is added to the fitness: (33) If the repaired particle still violates the hard constraints, a large penalty is applied to the fitness function to ensure that feasible solutions are prioritized.

[0069] The velocity and position of the particle swarm are updated according to the following formula: (34) (35) in, As a random factor, Learning factor, inertia weight Using a linear decreasing form: (36) Through iterative updates of the above three formulas, the population optimal solution is continuously improved in multi-generational search until convergence.

[0070] In this embodiment, the core decision variables output by LSTM are used as elite particles to initialize the population, giving the initial population a better starting point, reducing the probability of the algorithm getting stuck in local optima, and improving search efficiency. The constructed comprehensive fitness function comprehensively covers key objectives such as renewable energy curtailment, operating costs, constraint compliance, and user satisfaction, avoiding the neglect of one aspect due to optimization of a single objective, and ensuring the comprehensiveness and practicality of the scheduling scheme. By evaluating the quality of particles and iteratively updating the velocity and position of particles, the algorithm can efficiently search for the global optimum in the solution space, ultimately achieving the core objective of minimizing renewable energy curtailment, while also taking into account the economic efficiency of system operation and user experience, thus improving the overall quality of the scheduling scheme.

[0071] In one possible implementation, the PSO algorithm employs a collaborative decomposition strategy, which decomposes the decision vector into several sub-vectors according to the load type. Each sub-vector is optimized in parallel by an independent subgroup, and the solutions of each subgroup are merged at a fixed iteration interval to evaluate the overall fitness.

[0072] For high-dimensional decision-making problems involving multiple load types, the overall decision vector is decomposed into several sub-vectors according to resource type: (37) Each subvector is optimized by an independent subgroup, and the solutions are combined at a fixed iteration interval k to form a combined solution. The overall fitness is calculated using equation (30). The search reaches the global optimum in the later stages. Perform local refinement on the neighborhood: (38) like Then update the global optimal solution.

[0073] When the maximum number of iterations is reached Or it meets the convergence condition. The calculation terminates when the time is right, and the global optimal solution is output. If the solution satisfies the thresholds for power curtailment improvement rate, operating cost, and user satisfaction, it is output as the final scheduling result; otherwise, the optimal solution and constraint violation information are used to form a feedback sample. Input the LSTM model for fine-tuning, generate a new initial solution, and repeat this step.

[0074] The PSO algorithm employs a collaborative decomposition strategy, which decomposes the high-dimensional decision vector into several sub-vectors according to load type. This allows each independent subgroup to focus on optimizing a single type of load, avoiding mutual interference between decisions of different load types and improving the accuracy of local optimization. At the same time, the parallel optimization method significantly shortens the overall computation time and improves optimization efficiency. By merging the solutions of each subgroup and evaluating the overall fitness at a fixed iteration interval, the algorithm ensures that the scheduling schemes for each type of load are coordinated and adapted to each other, avoiding overall scheduling imbalance caused by local optima. Thus, when dealing with complex multi-type load collaborative scheduling problems, it not only guarantees optimization speed but also improves the coordination and optimality of the global scheduling scheme, better serving the goal of renewable energy consumption.

[0075] Furthermore, setting the maximum number of iterations and the convergence of the fitness function as termination conditions avoids the waste of computational resources and excessive time consumption caused by unlimited iteration of the algorithm, while ensuring timely output of results when the optimization reaches a stable state, thus guaranteeing the efficiency of the optimization process. Using the global optimal position at termination as the final adjustment strategy ensures that the output scheduling scheme is the optimal solution under the current optimization conditions. This fully taps the flexibility potential of demand-side resources, satisfies the operating constraints of the power system, effectively improves the absorption of new energy sources, and ensures the reliability and practicality of the scheduling scheme, facilitating its implementation in real-world applications.

[0076] In one possible implementation, after outputting the optimized adjustment strategy, the following is also included: Based on the adjustment strategy, the curtailment improvement rate of the power system is calculated; whereby the curtailment improvement rate is the ratio of the reduction in the amount of curtailment before and after optimization to the amount of curtailment before optimization. If the curtailment improvement rate is lower than the preset threshold or the operating cost exceeds the cost threshold, the current adjustment strategy and constraint violation information will be used as feedback samples to fine-tune the LSTM neural network model.

[0077] In the specific implementation process, the optimization results are compared with the actual operating data to calculate the power curtailment improvement rate: (39) In the formula, This indicates the rate of improvement in power curtailment. If... If the result is less than 90%, or the cost exceeds the threshold, the result is fed back to the LSTM model for retraining and iterative optimization until the requirements are met.

[0078] In this embodiment, the curtailment improvement rate is calculated after the output adjustment strategy, which can intuitively and quantitatively evaluate the improvement effect of the dispatch scheme on renewable energy consumption and provide a clear basis for judging the effectiveness of the scheme. When the curtailment improvement rate does not reach the preset threshold or the operating cost exceeds the limit, the relevant information is used as feedback samples to fine-tune the LSTM neural network model, so that the model can be continuously optimized in combination with the actual operation, improve the accuracy of the subsequent output core decision variables, and form a closed-loop mechanism of "training-optimization-feedback-retraining". This allows the entire dispatch method to have the ability to continuously iterate and optimize, constantly adapt to the changes in the operation of the power system, and ensure the long-term renewable energy consumption effect and the economic efficiency of system operation.

[0079] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0080] To verify the load absorption effect of the demand-side resource collaborative scheduling method for renewable energy consumption provided in this application, a specific embodiment is used for verification.

[0081] Simulation analysis was performed using data from a typical day. System parameters are shown in Table 1, and the simulation step size was set to 1 hour. The system load distribution and power curtailment curves are shown below. Figure 4As shown, the loads include critical loads, shiftable loads, transferable loads, and loads that can be reduced. The curtailment period is from 01:00 to 14:00, and the renewable energy consumption period is from 15:00 to 24:00. The system's positive and negative spinning reserve capacity is set at 5%. The acceptable shift / transfer interval is [1, 14]. The maximum allowable power for each transferred load is 1.5 times its maximum power during its own duration, and the minimum allowable power is 0.8 times its minimum power. The minimum continuous operating time is 2 hours. The acceptable reduction interval is [1, 24]. The minimum continuous reduction time is 2 hours, the maximum continuous reduction time is 8 hours, the maximum number of reductions is 10, the maximum number of unit starts is 1, and the minimum operating time is 24 hours. It is assumed that all types of loads can achieve full response.

[0082] Table 3 Power Grid System Parameters

[0083] To verify the effectiveness of the proposed model, the following four scenarios were set up for comparative analysis: Scenario 1: Multiple load types responding together; Scenario 2: Shiftable loads respond independently; Scenario 3: Transferable load responds independently; Scenario 4: Load can be reduced to respond independently.

[0084] Table 4 shows a comparison of the operational results of collaborative optimization of multi-type demand response based on demand-side resource flexibility for each scenario.

[0085] Table 4 Response Results for Each Scenario

[0086] In Scenario 1, all shiftable loads participate in the response, shifting from 15:00-24:00 to the 01:00-14:00 period; most transferable loads participate in the response, shifting from 15:00-24:00 to the 01:00-14:00 period, mainly concentrated in the 02:00-06:00 and 13:00-14:00 periods; a small amount of loads that can be reduced participate in the response, distributed in the 01:00-24:00 period; the number of conventional units switching on and off changes, resulting in a reduction of 2475MW in the maximum technical output and 4244MW in the minimum technical output of conventional units, and a total increase of 47740MW·h in renewable energy consumption. The established model can effectively promote renewable energy consumption.

[0087] In Scenario 2, all shiftable loads participate in the response, shifting from the 15:00-24:00 period to the 01:00-14:00 period. When the start-stop status of conventional units can be adjusted, the shifting of a large amount of load changes the original load curve, playing a role in peak shaving and valley filling. This adjusts the status of conventional units, reducing the maximum technical output of the units by 2015MW and the minimum technical output by 1762MW. The increased renewable energy consumption is greater than the response of the shiftable loads. When the start-stop status of units cannot be adjusted, the shifting of a large amount of load changes the original load curve, playing a role in peak shaving and valley filling. However, since the power remains unchanged before and after the shift of each load, accumulating a large amount of load shifts cannot perfectly approximate the required upward adjustment of demand-side flexibility resource power curve shape. Overall, the increased renewable energy consumption is less than the response of the shiftable loads.

[0088] In Scenario 3, all transferable loads participate in the response. When the start-up and shutdown of conventional units can be adjusted, the load shifts from 15:00-24:00 to 01:00-6:00. This significant load shift alters the original load curve, effectively shaving peaks and filling valleys. The on / off status of conventional units is adjusted, resulting in a reduction of 1830MW in maximum technical output and 1450MW in minimum technical output. The increased renewable energy consumption exceeds the response of the transferable loads. When the start-up and shutdown of units cannot be adjusted, the load shifts from 15:00-24:00 to 01:00-8:00. This significant load shift again alters the original load curve, effectively shaving peaks and filling valleys. Compared to load balancing, load transfer is less restricted and more flexible. The response of a large volume of transferred loads can better match the shape of the demand-side flexibility resource power curve, and the increased renewable energy consumption equals the response of the transferable loads.

[0089] In Scenario 4, when the start-up and shutdown of conventional units can be adjusted, a small portion of the load that can be reduced participates in the response, mainly during the period from 17:00 to 19:00, playing a peak-shaving role and changing the original load curve. The on / off status of conventional units is adjusted, and the overall maximum technical output of the units is reduced by 58MW and the minimum technical output is reduced by 52MW. The increased renewable energy consumption is greater than the response amount of the load that can be reduced. When the start-up and shutdown of units cannot be adjusted, the load that can be reduced does not participate in the response. Since the overall maximum and minimum technical output of the units remain unchanged, simply reducing the load response cannot provide the flexibility to adjust the demand side. Therefore, in this case, the load that can be reduced cannot promote the consumption of renewable energy.

[0090] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0091] Figure 5The diagram shows a schematic of a demand-side resource collaborative scheduling device for renewable energy consumption provided in an embodiment of the present invention. For ease of explanation, only the parts relevant to the embodiment of the present invention are shown, and are described in detail below: like Figure 5 As shown, the demand-side resource collaborative scheduling device 5 for new energy consumption includes: The core decision variable determination module 501 is used to input multi-source time series data into a pre-trained LSTM neural network model to obtain core decision variables for multiple types of loads. Among them, the multiple types of loads include movable loads, transferable loads, and reduceable loads. The core decision variables include the movable load's start time and duration, the transfer power and target time of the transferable load, and the reduce power and duration of the reduceable load. The adjustment strategy determination module 502 is used to input the core decision variables as the initial solution into the multi-type demand response collaborative optimization operation model, and to use the PSO algorithm to iteratively fine-tune the initial solution. When the preset termination condition is reached, the optimized adjustment strategy is output. The objective function of the multi-type demand response collaborative optimization operation model is to minimize the amount of renewable energy curtailed. The constraints of the multi-type demand response collaborative optimization operation model include: load shifting constraints, load transfer constraints, load reduction constraints, and power system operation constraints. The adjustment strategies include the hourly response amount of each type of load and the adjustment strategies of conventional units.

[0092] In this embodiment, a pre-trained LSTM neural network model is first used to process multi-source time-series data, fully adapting to the time-series characteristics of renewable energy generation and load changes in the power system. This accurately outputs the core decision variables of multiple load types, providing a high-quality initial solution for subsequent optimization and effectively reducing the computational complexity and time consumption of subsequent optimization. Then, this initial solution is input into a multi-type demand response collaborative optimization operation model with the goal of minimizing renewable energy curtailment. By comprehensively considering the constraints of multiple load types and power system operation constraints, the feasibility and safety of the scheduling scheme are ensured. Finally, the PSO algorithm is used to iteratively fine-tune the initial solution to further approach the optimal target. The final output adjustment strategy can achieve coordinated scheduling of multiple load types and conventional units, significantly improving the renewable energy absorption capacity, reducing curtailment, and enhancing the flexibility and economy of power system operation.

[0093] Figure 6 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 6 As shown, the electronic device 6 of this embodiment includes a processor 60 and a memory 61. The memory 61 stores a computer program 62. When the processor 60 executes the computer program 62, it implements the steps in the various method embodiments described above. Alternatively, when the processor 60 executes the computer program 62, it implements the functions of each module / unit in the various device embodiments described above.

[0094] For example, computer program 62 may be divided into one or more modules / units, which are stored in memory 61 and executed by processor 60 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 62 in electronic device 6.

[0095] Electronic device 6 may include, but is not limited to, processor 60 and memory 61. Those skilled in the art will understand that... Figure 6 This is merely an example of electronic device 6 and does not constitute a limitation on electronic device 6. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 6 may also include input / output devices, network access devices, buses, etc.

[0096] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0097] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0098] The above-described 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, and should all be included within the protection scope of the present invention.

Claims

1. A demand-side resource collaborative scheduling method for renewable energy consumption, characterized in that, include: Multi-source time-series data are input into a pre-trained LSTM neural network model to obtain core decision variables for multiple types of loads; wherein, the multiple types of loads include shiftable loads, transferable loads, and reduceable loads; the core decision variables include the shift start time and duration of shiftable loads, the transfer power and target time of transferable loads, and the reduction power and duration of reduceable loads; The core decision variables are used as initial solutions and input into the multi-type demand response collaborative optimization operation model; wherein, the objective function of the multi-type demand response collaborative optimization operation model is to minimize the amount of renewable energy curtailment; the constraints of the multi-type demand response collaborative optimization operation model include: load shifting constraints, load transfer constraints, load reduction constraints, and power system operation constraints; The initial solution is iteratively fine-tuned using a particle swarm optimization algorithm. When a preset termination condition is reached, the optimized adjustment strategy is output. The adjustment strategy includes the hourly response of each type of load and the adjustment strategy for conventional units.

2. The demand-side resource collaborative scheduling method for new energy consumption as described in claim 1, characterized in that, The load transfer constraints include: a transfer start time constraint, a transfer duration constraint, and a constraint on the load power after transfer within an acceptable time interval. The constraints on the transferable load include: the constraint that the total power consumption remains unchanged before and after the transfer, the constraint that the transfer power is allowed within a certain range, the constraint that the transfer target time period is constrained, and the constraint that the minimum continuous operating time is constrained. The load reduction constraints include: power reduction constraints, minimum continuous reduction time constraints, maximum continuous reduction time constraints, and maximum number of reduction constraints. The power system operation constraints include at least one of the following: real-time power supply and demand balance constraints, upper and lower limits of conventional unit output constraints, system spinning reserve constraints, line power transmission constraints, and node voltage phase angle constraints.

3. The demand-side resource collaborative scheduling method for new energy consumption as described in claim 1, characterized in that, The step of iteratively fine-tuning the initial solution using the particle swarm optimization algorithm includes: The core decision variables output by the LSTM are used as elite particles to initialize the particle population. A comprehensive fitness function is constructed with the goal of minimizing the amount of renewable energy curtailment. The comprehensive fitness function is constructed by integrating renewable energy output, load response power, renewable energy curtailment, operating costs, constraints and penalties, and user satisfaction. The particle's quality is evaluated based on the comprehensive fitness function, the individual optimal position and the global optimal position are updated, and the particle's velocity and position are iteratively updated based on the updated individual optimal position and the global optimal position.

4. The demand-side resource collaborative scheduling method for new energy consumption as described in claim 3, characterized in that, When the preset termination condition is met, the optimized adjustment strategy is output, including: When the maximum number of iterations is reached or the fitness function value converges, the iteration is terminated, and the global optimal position at this point is output as the final optimized adjustment strategy.

5. The demand-side resource collaborative scheduling method for new energy consumption as described in claim 3, characterized in that, The particle swarm optimization algorithm employs a collaborative decomposition strategy, which decomposes the decision vector into several sub-vectors according to the load type. Each sub-vector is optimized in parallel by an independent subgroup, and the solutions of each subgroup are merged at a fixed iteration interval to evaluate the overall fitness.

6. The demand-side resource collaborative scheduling method for renewable energy consumption as described in any one of claims 1 to 5, characterized in that, Following the optimized output adjustment strategy, the following is also included: Based on the adjustment strategy, the curtailment improvement rate of the power system is calculated; wherein, the curtailment improvement rate is the ratio of the reduction in the amount of curtailment before and after optimization to the amount of curtailment before optimization. If the curtailment improvement rate is lower than a preset threshold or the operating cost exceeds a cost threshold, the current adjustment strategy and constraint violation information will be used as feedback samples to fine-tune the LSTM neural network model.

7. The demand-side resource collaborative scheduling method for new energy consumption as described in claim 1, characterized in that, The LSTM neural network model includes an input layer, a first LSTM layer, a second LSTM layer, a third LSTM layer, and a fully connected layer. The input layer receives the multi-source time-series data; the first LSTM layer is used to capture short-term time-series dependencies; the second LSTM layer is used to extract medium-term correlations; the third LSTM layer is used to mine long-term trends; and the fully connected layer is used to output the core decision variables.

8. The demand-side resource collaborative scheduling method for new energy consumption as described in claim 1, characterized in that, The multi-source time-series data includes: wind power series, photovoltaic power series, historical power curves of loads that can be shifted, historical power curves of loads that can be transferred, historical power curves of loads that can be reduced, and minimum and maximum technical output parameters of conventional units.

9. A demand-side resource collaborative scheduling device for new energy consumption, characterized in that, include: The core decision variable determination module is used to input multi-source time-series data into a pre-trained LSTM neural network model to obtain core decision variables for multiple types of loads; wherein, the multiple types of loads include shiftable loads, transferable loads, and reduceable loads; the core decision variables include the shift start time and duration of shiftable loads, the transfer power and target time of transferable loads, and the reduction power and duration of reduceable loads; The adjustment strategy determination module is used to input the core decision variables as initial solutions into the multi-type demand response collaborative optimization operation model, and to use the particle swarm optimization algorithm to iteratively fine-tune the initial solutions. When the preset termination condition is reached, the optimized adjustment strategy is output. The objective function of the multi-type demand response collaborative optimization operation model is to minimize the amount of renewable energy curtailment. The constraints of the multi-type demand response collaborative optimization operation model include: load shifting constraints, load transfer constraints, load reduction constraints, and power system operation constraints. The adjustment strategies include the hourly response amount of each type of load and the adjustment strategies of conventional generating units.

10. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 8.

Images