Distributed power resource collaborative optimization system and method

Abstract

The invention relates to the technical field of intelligent power grids, in particular to a collaborative optimization system and method for distributed power resources, and the system comprises a data collection module, a time-space correlation analysis module, a user response module, a data privacy protection module, a time-of-use electricity price optimization module and a scheduling control module. The space-time correlation analysis module performs interval prediction of distributed photovoltaic power by using a Bayesian-long short-term memory network model, the user response module designs a hierarchical response strategy according to user categories, and the data privacy protection module protects power utilization data through a differential privacy mechanism. The time-of-use electricity price optimization module dynamically adjusts the electricity price according to the peak-valley load difference, and the scheduling control module achieves the cooperative regulation and control of distributed resource integration, local-level power grid coordination and provincial-level power grid optimization. The technical problems of an existing power system in the aspects of efficiently integrating distributed resources, improving scheduling flexibility and optimizing an electricity price strategy are solved.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and in particular to a system and method for collaborative optimization of distributed power resources. Background Art

[0002] In modern power systems, distributed power resources (such as distributed generation, energy storage equipment, demand response, etc.) are widely used. However, how to efficiently integrate these distributed resources and achieve optimized regulation has become a core challenge in the development of the current power system. The existing power grid system usually relies on a centralized dispatching method, which is mainly aimed at centralized optimization of large-scale power generation, but does not take into account the diversity and regional characteristics of distributed resources. This is specifically reflected in the following aspects:

[0003] Limitations of centralized dispatch strategies: The current dispatch model is centered on large-scale power generation, ignoring the uniqueness of distributed resources in different regions and making it difficult to fully utilize their potential.

[0004] Static demand response model: Most existing demand response strategies are based on fixed rules and fail to adapt to dynamic changes in user behavior or price fluctuations in the market in real time, reducing the flexibility of the system.

[0005] Single-layer optimization model: The existing dispatching model usually only optimizes a certain level of the provincial or prefectural power grid, lacks cross-level integrated collaborative optimization capabilities, and leads to inefficient resource integration and regulation.

[0006] In addition, these prior arts also have the following specific defects:

[0007] The temporal and spatial characteristics of distributed resources are not fully utilized: The temporal (periodicity, volatility) and spatial (geographical distribution) characteristics of distributed resources have not been fully studied and utilized, resulting in a lack of targeted scheduling strategies and difficulty in achieving optimal resource allocation.

[0008] The user-side response mechanism is single: the existing user response strategy fails to be designed in a hierarchical manner according to the different demand characteristics of industrial, commercial and residential users. The response mode is too single, resulting in low user participation.

[0009] Insufficient provincial and local coordinated optimization capabilities: The optimization scheduling between provincial power grids and prefecture-level power grids is usually carried out independently, lacking a cross-level resource integration mechanism, making it difficult to achieve global optimization.

[0010] Insufficient flexibility in market-based incentives: Current market price signals lack the ability to adjust dynamically, and the incentive mechanism design is not comprehensive enough, making it difficult to stimulate users' potential to actively participate in resource regulation. Summary of the invention

[0011] In view of the above problems existing in the prior art, the present invention is proposed.

[0012] Therefore, the technical problem to be solved by the present invention is: in view of the shortcomings of the prior art, the present invention collects historical operating data of distributed resources, extracts their time characteristics (such as daily load curve, volatility) and spatial characteristics (geographical location and resource distribution in the region); studies the complementary characteristics between resources (such as the complementarity between the fluctuation of distributed power generation and the smoothing function of energy storage equipment), and solves the problem of insufficient research on time and space characteristics; the present invention establishes a resource scheduling model based on time and space characteristics, clarifies the allocation of each distributed resource in time and space, and solves the problems of insufficient collaborative optimization capabilities and a single response mechanism.

[0013] To solve the above technical problems, the present invention provides the following technical solutions: a collaborative optimization system for distributed power resources, comprising: a data acquisition module, a spatiotemporal correlation analysis module, a user response module, a data privacy protection module, a time-of-use electricity price optimization module and a dispatching control module.

[0014] As a preferred solution of the distributed power resource collaborative optimization system described in the present invention, the data acquisition module includes a module for collecting real-time data of distributed photovoltaic, energy storage equipment and loads.

[0015] As a preferred solution of the distributed power resource collaborative optimization system described in the present invention, the spatiotemporal correlation analysis module includes using the Bayesian-long short-term memory network model to perform interval prediction on distributed photovoltaic power; combining Pearson correlation analysis to perform feature decomposition on the historical data of the distributed photovoltaic cluster to enhance the accuracy of spatial correlation modeling;

[0016] The Bayesian-long short-term memory network model achieves high-precision prediction of distributed photovoltaic output range under complex and highly fluctuating weather conditions by dynamically adjusting training parameters.

[0017] As a preferred solution of a collaborative optimization system of distributed power resources described in the present invention, the user response module includes designing a hierarchical response strategy based on user categories and response willingness, combining demand willingness with marginal cost optimization scheduling; a hierarchical response strategy based on the marginal utility of interruption cost is designed, wherein user categories include industrial users, commercial users and residential users; load reduction targets are flexibly adjusted according to actual user needs, and the economic benefits of user participation are maximized by optimizing incentive strategies.

[0018] As a preferred solution of the distributed power resource collaborative optimization system described in the present invention, the data privacy protection module includes protecting the power consumption data through a differential privacy mechanism; injecting gamma distribution noise into the user data collection stage through a local differential privacy mechanism to achieve privacy protection, while ensuring the efficiency of power load statistical data;

[0019] The differential privacy mechanism includes optimizing the allocation of electricity prices during peak, sharp, valley and concave periods in combination with user time-sharing electricity consumption data.

[0020] As a preferred solution of the coordinated optimization system of distributed power resources described in the present invention, the time-of-use electricity price optimization module includes dynamically adjusting the electricity price according to the peak-valley load difference.

[0021] As a preferred solution of the distributed power resource collaborative optimization system described in the present invention, the dispatch control module includes a multi-layer optimization framework, including a distributed resource integration layer, a prefecture-level power grid coordination layer, and a provincial power grid optimization layer, to achieve collaborative regulation within the region; adaptive regulation between the distributed resource integration layer, the prefecture-level power grid coordination layer, and the provincial power grid optimization layer is achieved through a distributed optimization algorithm;

[0022] The multi-layer optimization includes a coordinated dispatch strategy to dynamically adjust the charging and discharging plans of distributed energy storage to smooth supply and demand fluctuations.

[0023] Another object of the present invention is to provide a collaborative optimization method for distributed power resources, which can achieve accurate scheduling and optimization of distributed photovoltaic, energy storage equipment and loads through a multi-layer optimization framework and a dynamic adjustment strategy for time-of-use electricity prices, thereby improving the flexibility and economic benefits of the power system while protecting user data privacy.

[0024] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for coordinated optimization of distributed power resources, comprising: the dynamic adjustment of peak and valley loads includes dynamically balancing the electricity price benefits and system loads, setting peak and valley time periods to adjust the electricity price elasticity coefficient in real time;

[0025] The scheduling strategy includes decoupling the relationship between distributed photovoltaic scale and prediction error through affine function to achieve adaptive optimization of distributed resources; and reducing the scheduling risk of virtual power plants under the conditions of photovoltaic output and market electricity price fluctuations through conditional risk value method.

[0026] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of a distributed power resource collaborative optimization system as described above when executing the computer program.

[0027] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of a distributed power resource collaborative optimization system as described above are implemented.

[0028] The beneficial effects of the present invention are as follows: the B-LSTM model significantly improves the prediction accuracy of photovoltaic power and reduces the uncertainty in scheduling; the hierarchical response strategy is combined with a flexible incentive scheme to improve user participation enthusiasm; the differential privacy protection mechanism achieves a balance between data privacy protection and statistical utility; the multi-layer scheduling control framework reduces the peak-valley load difference and improves the reliability and flexibility of power grid operation.

[0029] Under severe volatile weather conditions, the coverage of the forecast interval increased by 9.4% and the average width was reduced by more than 20%, significantly optimizing dispatch decisions; improving the reliability of the forecast of photovoltaic power output intervals and meeting the grid dispatch's demand for accurate forecasts.

[0030] The graded response strategy significantly improved the participation rate of demand response, achieving a win-win situation for system demand and user benefits; flexible adjustment of reduction targets reduced user inconvenience and optimized the flexibility of grid dispatch; and improved the accuracy of load reduction and system economic benefits.

[0031] In the data collection of smart meters, gamma distribution noise is injected to achieve local differential privacy protection, ensuring that user personal data cannot be directly inferred; the peak and valley load curves are statistically analyzed using noisy data, and the peak and valley electricity prices are adjusted based on the elasticity coefficient to encourage users to use electricity during off-peak hours; by optimizing the privacy protection budget parameters, the data distortion rate is reduced to ensure a balance between electricity price optimization and user data privacy protection.

[0032] The risk value was reduced by 23% and the return was increased by 13%, which verified the advantages of large-scale photovoltaic aggregation. The affine decoupling model significantly improved the scheduling efficiency and risk control capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0034] Figure 1 A module diagram of a distributed power resource collaborative optimization system provided for one embodiment of the present invention.

[0035] Figure 2 A flow chart of a distributed power resource collaborative optimization system provided by an embodiment of the present invention.

[0036] Figure 3 A VPP scheduling flow chart of a distributed power resource collaborative optimization system provided by an embodiment of the present invention.

[0037] Figure 4 A flow chart of spatiotemporal correlation feature extraction and interval prediction for a distributed power resource collaborative optimization system provided by an embodiment of the present invention. Figure 5 A flow chart of modeling spatiotemporal correlation characteristics of a distributed power resource collaborative optimization system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0039] Example 1, reference Figure 1-Figure 4 , is an embodiment of the present invention, which provides a coordinated optimization system for distributed power resources, including:

[0040] Data acquisition module, spatiotemporal correlation analysis module, user response module, data privacy protection module, time-of-use electricity price optimization module and dispatch control module.

[0041] Data acquisition modules, such as Figure 1 S1 is used to collect real-time data of distributed photovoltaic, energy storage equipment and loads; spatiotemporal correlation analysis module, such as Figure 1 In S2, the Bayesian-Long Short-Term Memory Network (B-LSTM) model is used to perform interval prediction of distributed photovoltaic power; the user response module, such as Figure 1 In S3, a hierarchical response strategy is designed according to user categories and response willingness, and scheduling is optimized by combining demand willingness and marginal cost; data privacy protection module, such as Figure 1 In S4, the electricity consumption data is protected through the differential privacy mechanism; the time-of-use electricity price optimization module, such as Figure 1 In S5, the electricity price is adjusted dynamically according to the peak-valley load difference; the dispatch control module, such as Figure 1 In S6, a multi-layer optimization framework is adopted, including a distributed resource integration layer, a prefecture-level power grid coordination layer, and a provincial power grid optimization layer to achieve coordinated regulation within the region.

[0042] Combine advanced smart meters and edge computing technologies to collect distributed photovoltaic, energy storage and load data in real time; use a distributed architecture to transmit data to the central control system; decompose photovoltaic power sequences based on the CEEMDAN algorithm, extract multi-scale modal features, and use the B-LSTM model combined with the Bayesian method to dynamically adjust training parameters to achieve high-precision interval prediction under complex weather conditions; design a hierarchical response strategy based on marginal cost utility, divide users into three categories: industrial, commercial and residential, optimize the incentive strategy through demand response, and dynamically adjust the load reduction target; adopt a local differential privacy mechanism to achieve privacy protection through gamma distribution noise, while maintaining the availability of statistical data to ensure the security of user data during collection and analysis; focus on peak and valley periods, use an elasticity coefficient model to dynamically adjust electricity prices, encourage users to use electricity off-peak and reduce peak-valley differences; achieve multi-level collaborative scheduling through distributed optimization algorithms, and use flexible strategies between layers to achieve the optimal charging and discharging plan for distributed energy storage.

[0043] In addition, low-power wide area network technology based on LoRa or NB-IoT communication protocol can be used for data collection; the B-LSTM model can be replaced by a deep learning model based on Transformer to further improve the ability to extract spatiotemporal features; the differential privacy mechanism can be replaced by homomorphic encryption or federated learning framework to enhance data security; the time-of-use electricity price can be dynamically adjusted by introducing genetic algorithm or particle swarm optimization algorithm to improve optimization efficiency and response speed;

[0044] The spatiotemporal correlation analysis module further combines Pearson correlation analysis to perform feature decomposition on the historical data of distributed photovoltaic clusters to enhance the accuracy of spatial correlation modeling; the Bayesian-LSTM model dynamically adjusts the training parameters to achieve high-precision prediction of the output range of distributed photovoltaics under complex and highly volatile weather conditions;

[0045] Temporal and spatial correlation feature extraction: The historical output data of multiple photovoltaic sites are calculated by Pearson correlation coefficient, and the sites with strong correlation are screened out. The correlation threshold is set to 0.95. The photovoltaic power series is decomposed into several modal components using CEEMDAN (Complete Integrated Empirical Mode Decomposition) method to extract multi-scale features.

[0046] The interval prediction model is constructed based on the LSTM network. The Dropout layer is added to construct the B-LSTM model. The neural network weights and bias parameters are regarded as probability distributions. The Bayesian update rule is used to dynamically optimize the model parameters, and the interval prediction is achieved through the following formula:

[0047]

[0048] Among them, P n is the predicted value, is the standard deviation of the uncertainty quantification, Z a / 2 is the quantile of the confidence interval;

[0049] Optimization under complex weather conditions: select typical data sets of sunny and strong volatile weather for training and verification, and optimize network hyperparameters such as Dropout probability (0.2-0.5 range);

[0050] Optimize complex weather conditions, combine diversified scenario data sets (sunny and strongly fluctuating weather) to optimize the training process, adjust hyperparameters (such as Dropout rate and learning rate), and improve the adaptability of the model in uncertain environments.

[0051] Combining the decision-dependent uncertainty (DDU) and spatial correlation of distributed photovoltaics, an optimized risk scheduling model is established to improve the accuracy and reliability of distributed photovoltaic output prediction, and solve the impact of photovoltaic output uncertainty on the virtual power plant scheduling strategy; the characteristics are:

[0052] The active power output of distributed photovoltaics mainly depends on the light intensity, the conversion rate of photovoltaic modules and the total area of ​​photovoltaic module panels; the predicted probability density function of light intensity is:

[0053]

[0054] Among them, α and β are distribution parameters, which are obtained by fitting the actual data;

[0055] The predicted power output of distributed photovoltaic is

[0056]

[0057] Where r is solar radiation; r max is the maximum solar radiation; S PV is the photovoltaic panel area; η is the photoelectric conversion efficiency;

[0058] Due to the spatial smoothing effect, the prediction error of the distributed photovoltaic cluster is reduced, and equations (1) and (2) are no longer applicable. Considering the probability density function of the light intensity of the DDU and the distributed photovoltaic output, it can be modified as follows:

[0059]

[0060] α'=ωα ⑸

[0061] β'=σβ ⑹

[0062] Among them, ω and σ are coefficients that characterize the relationship between the scale of distributed photovoltaics and the distribution parameters of the predicted probability density function. This paper obtains the corresponding coefficients by comparing the measured light intensity information of a single observation point with the light intensity information of multiple observation points.

[0063] Considering the unit distributed photovoltaic power prediction scenario that is irrelevant to decision-making, it is generally possible to generate multiple sets of different unit distributed photovoltaic power prediction results using the Monte Carlo method based on the light intensity prediction probability density function. Under this condition, the power prediction scenario K of the distributed photovoltaic cluster considering spatial correlation is * The predicted power of a single distributed photovoltaic system can be obtained by transforming it according to equations (3) to (6).

[0064] However, considering the distributed photovoltaic cluster of DDU, when the scale of aggregated distributed photovoltaic is different, the probability density function of light intensity will also change, that is, K * Transformed to the scale of distributed photovoltaics with respect to the selection of aggregation x k,t Function K * (x k,t ), x k,t This paper decouples the relationship between distributed photovoltaics and light intensity probability density functions under different aggregation scales, and presets a set of power prediction scenarios for possible aggregation scales in advance. And construct affine functions of decisions and corresponding scenarios.

[0065] In view of the volatility of distributed photovoltaic power, the spatiotemporal correlation analysis module adopts a photovoltaic output interval prediction method based on CEEMDAN decomposition and improved B-LSTM model. By extracting spatiotemporal correlation features, it achieves high-precision prediction of distributed photovoltaic short-term power under complex weather conditions. It features:

[0066] Using LSTM and fully connected neural network as the basis, the Dropout layer is added to it to form an improved B-LSTM neural network for short-term photovoltaic output range prediction.

[0067] Compared with traditional neural networks, the weight parameters and bias parameters of Bayesian neural networks (BNNs) are regarded as uncertain random variables, and all weight and bias parameters are represented by probability distributions. The BNN structure can obtain the posterior probability distribution of weight and bias parameters based on photovoltaic output sample data and obtain the optimal parameter results. However, there are still deficiencies in practical applications. This paper constructs a Bayesian neural network model by adding a Dropout layer to the LSTM neural network. Specifically, each neuron in the LSTM obeys a Bernoulli binomial distribution with a probability of p. Therefore, after adding the Dropout layer, the probability of each neuron being discarded is 1-p, so that it does not play a role in the forward propagation process of the corresponding round of neural network training.

[0068] The distributed photovoltaic output interval prediction process based on spatiotemporal correlation features and B-LSTM is as follows. 1) Collect the power generation of each photovoltaic station and perform data preprocessing. 2) Use the Pearson correlation coefficient to analyze the spatiotemporal correlation of each station and screen out N stations with strong correlation. 3) Use CEEMDAN to decompose the power generation of N stations into multiple modal components, and use each component to characterize the spatiotemporal correlation characteristics between each station, thereby achieving effective and refined characterization. 4) Establish a B-LSTM photovoltaic output interval prediction model, using the powerful memory capacity and feature extraction function of the LSTM neural network. 5) Train the modal components of different stations, extract spatiotemporal correlation features, add a Dropout layer to Bayesianize the recurrent neural hidden layer to obtain the optimal weight bias parameters, and then output the cluster power interval prediction results.

[0069] The interval evaluation index is used to evaluate the reliability and accuracy of the interval prediction results. icp The indicator represents the probability that the actual value of photovoltaic output is within the upper and lower limits of the prediction interval, P icp The larger the value, the more reliable the interval prediction model is.

[0070]

[0071] Where N is the total number of photovoltaic output points to be predicted; S n is the Boolean function of the nth photovoltaic output point to be predicted.

[0072] The prediction interval average width (PIAW) characterizes the clarity of the interval prediction model. The smaller the interval average width value at the same confidence level, the narrower the interval prediction and the better the model interval prediction.

[0073]

[0074] Among them, P s , P x are the upper and lower bound power values ​​of interval prediction respectively.

[0075] The user response module designs a hierarchical response strategy based on marginal interruption costs, which not only distinguishes the demand characteristics of industrial users, commercial users and residential users, but also optimizes the incentive strategy, further improving the economic benefits of user participation; its characteristics are:

[0076] In order to comprehensively consider the energy consumption characteristics of users, the user response willingness is introduced in the design of the IBDR incentive function, thereby parameterizing the user type; the user classification method is based on the user interruption characteristics, and the users are classified according to their response willingness; users with the same response willingness are divided into a user level and their power outage cost function is constructed. This classification method facilitates the simultaneous dispatch of multiple types of demand response loads. Finally, the demand response model is considered from the perspective of the balance of benefits between users and VPP operators.

[0077] The user's interruption cost refers to the power outage loss caused by load reduction when the user participates in demand response; user load reduction will change their original electricity consumption habits, resulting in a decrease in electricity consumption satisfaction. The more load is reduced, the higher the user's interruption cost; the stronger the user's willingness to respond, the lower the marginal interruption cost and the higher the marginal benefit.

[0078] Therefore, the construction of the user outage cost function has the following assumptions: (1) As the load reduction increases, the user's power outage cost gradually increases, and the user's marginal benefit decreases accordingly; (2) The user outage cost function is a convex function, that is, the rate of change of the marginal cost increases with the increase of the user's load reduction.

[0079] The quadratic function containing the demand response load reduction x and the demand response user type θ represents the outage cost. Among them, the user type is parameterized as θ according to the demand response willingness, and the value range is [0,1]. When θ=0, it belongs to the direct load control demand response type, that is, under the premise of notifying the user in advance, the direct load control actuator can remotely control the start and stop of the power user's equipment to avoid the peak of electricity consumption. In this case, the user's response willingness is the lowest and the marginal cost is the highest; when θ>0, it belongs to interruptible load control, that is, the user's consent is required before execution to control the power equipment, and the larger the θ value, the stronger the response willingness. When θ=1, it means that the user's response willingness is the highest and the marginal cost is the lowest. The value of θ in the actual problem can be obtained from the public utility database.

[0080] When users are ranked according to marginal cost, if the user outage cost is c(θ,x), the cost incurred by the θ-type user to reduce the x load itself is c. If users are ranked from the lowest to the highest response willingness, equation (9) is satisfied.

[0081]

[0082] Similarly, if users are ranked from highest to lowest response willingness, equation (10) is satisfied.

[0083]

[0084] When the user's reduction amount is 0, the reduction cost should be 0. As long as the power outage cost function is not a decreasing function, any of the above sorting methods can be selected. This paper sorts the users from low to high according to their willingness to respond, that is, in the case of formula (9), the user's power outage cost function should satisfy formulas (11) to (13).

[0085]

[0086] Among them, x ≥ 0; 0 ≤ θ ≤ 1. Therefore, the user power outage cost function can be expressed as:

[0087] c(θ,x)=a1x 2 +a2x-a2xθ ⒁

[0088] Among them, a1 and a2 are DR cost coefficients. -a2xθ ensures that different θ values ​​correspond to different Value (i.e., the user's marginal cost), as the value of θ increases, the marginal cost decreases.

[0089] User benefit refers to the benefits that customers gain from participating in demand response. For IBDR, the benefit is the subsidy b obtained, and the cost is the load interruption cost c(θ,x). Therefore, the benefit can be expressed as:

[0090] B1(θ,x,b)=bc(θ,x) ⒂

[0091] Among them, b is the subsidy obtained by the user's participation in the response, and b≥c(θ,x).

[0092] VPP operators mainly consider the overall benefits of VPP. In a system containing IBDR, VPP should increase its own profits while incentivizing users to participate in the response, thereby improving social welfare.

[0093] VPP utility refers to the satisfaction generated by the implementation of the DR project, which is quantified as the marginal benefit generated by reducing power generation minus the user subsidy cost. It can be expressed as:

[0094] B2(θ,λ)=λx-b ⒃

[0095] Among them, λ is the regional marginal price, which means that the benefit of not supplying electricity to users is λ US dollars / MWh. The model should also meet the consumer profit protection constraint:

[0096]

[0097] Formula (17) is used to ensure that customer benefits are always positive to encourage users to participate in the response. Formula (18) is an incentive-compatible constraint to ensure that users receive appropriate compensation for load reduction.

[0098] The VPP operator dispatches all generators (traditional DG units and renewable DG units) and DR units under its jurisdiction based on the principle of maximizing the use of renewable energy generation. At the same time, VPP as a whole can participate in the electricity market to trade electricity, purchase electricity from the power grid when the electricity price is low, reduce internal power generation to reduce costs, sell electricity to the power grid when the electricity price is high, and increase internal power generation to increase the VPP operator's revenue. Since wind power generation and photovoltaic power generation have strong randomness, this paper describes the wind speed model with Weibull distribution function and photovoltaic power generation with Beta distribution function. Finally, the model optimizes the dispatch with the lowest VPP power generation cost and the maximum demand response utility as the objective function.

[0099] The data privacy protection module adds gamma distribution noise to the electricity consumption data through the differential privacy mechanism, which not only protects user privacy, but also improves the peak-shaving and valley-filling effect of the power grid load in the time-of-use electricity price optimization strategy, and realizes load balancing and electricity cost optimization; its characteristics are:

[0100] In the smart grid, it is assumed that the power consumption data reported by the smart meter SM at each fixed time interval ranges from [0, θ], where θ is the upper limit of the power consumption data. The control center CC selects an appropriate privacy protection budget ε based on data security and data availability requirements, and calculates the Laplace noise scale λ according to formula (1). Then CC generates a message m = {n, λ} and forwards it to the SM of each RU through ES. Where n is the total number of smart meters in the area.

[0101]

[0102] Assume that at time t, SM's meter reading is recorded as d. According to the received message m, SM samples two independent and identically distributed random numbers ξ1(n,λ) and ξ2(n,λ) from the gamma distribution with probability density function (20) as random noise. The random noise is added to the electricity consumption data according to (21).

[0103]

[0104] After completing the noise addition of the electricity consumption data, SM sends the electricity consumption data to ES and CC. After data aggregation by CC, it can be used for subsequent electricity load peak and valley division, time-of-use electricity price decision optimization and other data analysis links.

[0105] After completing the collection of user electricity consumption data, CC draws a daily electricity load curve based on the changes in electricity load in a single operation cycle T (usually 24 hours). The maximum and minimum values ​​on the curve are selected as feature points, and the operation cycle T is subdivided into four time periods: peak load period, peak load period, valley load period and concave valley load period based on the original peak and valley period division by using fuzzy mathematics theory and membership function (such as: small semi-trapezoidal membership function); peak electricity price, peak electricity price, valley electricity price and concave valley electricity price are respectively applied to the above time periods.

[0106] Assuming that the duration of each electricity price period satisfies μ ≥ 1h, use and They represent the start and end time of the peak electricity price, and the duration of the peak electricity price is Δt h ; and They represent the start and end time of the valley electricity price respectively, and the duration of the valley electricity price is Δt1; and They represent the start and end time of the ith peak electricity price (i∈[1,M]), and the duration of the ith peak electricity price is Δt H,i ; and They represent the start and end time of the jth (j∈[1,N]) valley electricity price, and the duration of the jth valley electricity price is Δt L,j , the constraints of each time period are shown in formula (22):

[0107]

[0108] According to the above principles, assuming that there are K feasible peak-valley time division methods f k , then the total peak-valley time division set is F = {f k ,k=1,2,…,K}. In the selected partitioning method f k Under this circumstance, a corresponding multi-objective optimization model for peak-valley time-of-use electricity prices is established. The peak-valley difference of the load curve, the economic benefits of the power market and the changes in users' electricity costs before and after optimization are compared. The optimal partitioning mode is selected from the set F, and then the optimal peak-valley time-of-use electricity price model is obtained.

[0109] CC aims to achieve peak load reduction and valley filling of the overall power load in the smart grid, and constructs the objective function Z1 with the goal of minimizing the peak load in the daily power load curve; and constructs the objective function Z2 with the goal of minimizing the peak-to-valley difference in the daily power load curve. The expressions of the above objective functions are shown in Equations (23) and (24).

[0110]

[0111] in, and are the maximum load value and the minimum load value of the daily load curve, Q is the function for solving the power load, P L,j is the jth valley electricity price, P H,i is the i-th peak electricity price, P1 is the valley electricity price, and P h is the peak electricity price. At the same time, considering factors such as electricity market benefits, user benefits and power generation costs, the model should meet the following constraints.

[0112] In order to ensure the overall benefits of the power market and the enthusiasm of power supply companies, the power supply companies should increase their revenue R after implementing the peak and valley time-of-use electricity prices. a Should not be less than the profit R0 before implementing the peak-valley time-of-use electricity price and the saved investment cost The difference between the two, the constraint expression is shown in equations (25) to (27), where: To calculate the electricity consumption during peak and off-peak hours before implementing time-of-use electricity prices, q H,i ,q L.j The power consumption during the i-th peak period and the j-th valley period is:

[0113]

[0114] Residential users benefit. Residential users are sensitive to changes in electricity prices, and only by benefiting them can they be motivated. The average electricity cost of residential users after implementing the peak and valley time-of-use electricity price plan is C a Should not be higher than the average electricity cost C before the implementation of the plan b , the constraint expression is

[0115] C a ≤C b (28)

[0116] Power generation cost. After the implementation of peak-valley time-of-use electricity prices, the peak electricity price should not be higher than the power generation cost of small generators of power supply enterprises C c , the valley electricity price should not be lower than the marginal cost C during the valley period e , the constraint expression is:

[0117] p H,i ≤C c ;p Lj ≤C e (29)

[0118] In summary, the peak-valley time-of-use electricity price optimization decision model is shown in formula (30):

[0119]

[0120] CC is based on the variable load in the system that can be transferred during period D. The proportion of electricity consumption in period i is φ i,D The impact of the electricity price change in period i on the electricity consumption in period i and period j φ ij ,φ ij According to formula (31) and formula (32), the autoelastic coefficient ω is calculated respectively: ii and the cross elastic coefficient ω ij , and the electricity consumption and price elasticity matrix E shown in formula (33) is obtained.

[0121]

[0122] CC uses the differential evolution algorithm (DE) to solve the peak-valley time-of-use electricity price model to obtain the optimal peak electricity price, optimal peak electricity price, optimal valley electricity price and optimal concave valley electricity price. According to formula (34), the user electricity load in different time periods under the above optimal time-of-use electricity price is calculated, the electricity load curve is redrawn, and the peak-valley changes of the electricity load curve before and after the electricity price optimization are analyzed.

[0123]

[0124] Combined with the risk scheduling model of virtual power plants, the relationship between distributed photovoltaic scale and prediction error is decoupled through affine function, the aggregation scale of distributed resources is dynamically optimized, and the risk of profit volatility in virtual power plant scheduling is reduced based on the conditional value at risk (CVaR) method.

[0125] Based on the above technical solutions, the collaborative optimization system of distributed power resources collects real-time data through the data acquisition module, combines the spatiotemporal correlation analysis model to predict the distributed photovoltaic power range, designs the user classification response strategy, introduces the differential privacy protection mechanism and optimizes the time-of-use electricity price, which solves the technical problems of efficient integration and scheduling of distributed resources, realizes multi-layer collaborative optimization within the region, and improves the resource utilization efficiency and user response rate; by combining the spatiotemporal characteristics of distributed photovoltaics, user response willingness, data privacy protection and multi-layer optimization framework, it solves the technical problems of efficient integration and scheduling of distributed resources, realizes multi-layer collaborative optimization within the region, and improves the economic benefits of the system, user response rate and scheduling reliability.

[0126] The above is a schematic scheme of a distributed power resource collaborative optimization system of this embodiment. It should be noted that the technical scheme of this distributed power resource collaborative optimization system and the technical scheme of the above-mentioned distributed power resource collaborative optimization system belong to the same concept, and the details of the technical scheme of a distributed power resource collaborative optimization method in this embodiment that are not described in detail can all be referred to the description of the technical scheme of the above-mentioned distributed power resource collaborative optimization system.

[0127] Embodiment 2 is an embodiment of the present invention, which provides a method for collaborative optimization of distributed power resources, including:

[0128] The peak-valley load dynamic adjustment includes dynamically balancing the electricity price benefit and system load, setting the peak and valley periods to adjust the electricity price elasticity coefficient in real time;

[0129] The scheduling strategy includes decoupling the relationship between distributed photovoltaic scale and prediction error through affine function to achieve adaptive optimization of distributed resources; and reducing the scheduling risk of virtual power plants under the conditions of photovoltaic output and market electricity price fluctuations through conditional risk value method.

[0130] This embodiment further provides a computing device, which is applicable to a distributed power resource collaborative optimization system, including:

[0131] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement a distributed power resource collaborative optimization system as proposed in the above embodiment.

[0132] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, a coordinated optimization system for distributed power resources as proposed in the above embodiment is implemented.

[0133] The storage medium proposed in this embodiment and the collaborative optimization system of distributed power resources proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0134] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0135] Logic and / or steps otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch instructions from and execute instructions on, an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" may be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0136] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiment, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0137] Example 3, reference Figure 5 An embodiment of the present invention provides a coordinated optimization system for distributed power resources. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0138] Taking a photovoltaic station in a certain area as an example, historical output data of 40 stations at 15-minute intervals were collected for analysis: Pearson correlation coefficient was used to screen out 34 highly correlated stations. The power series was decomposed by CEEMDAN to obtain 5 main modal components: Mode 1 (low-frequency signal) corresponds to trend changes; Modes 2-5 represent high-frequency fluctuation characteristics.

[0139] The B-LSTM model is used to predict photovoltaic power under sunny and strong fluctuating weather conditions. The comparison of PICP and PIAW indicators is shown in Table 1:

[0140] Table 1 Index comparison

[0141]

[0142] like Figure 5As shown in the figure, spatiotemporal correlation analysis can be combined with dynamic network graph modeling technology to conduct more refined mining of spatial features; the model framework can be replaced with an improved deep learning model based on GRU or Transformer to further optimize the long time series prediction effect; the data source can be expanded to satellite observations and drone inspection data to enhance the comprehensiveness of spatiotemporal correlation analysis.

[0143] The user response module designs a hierarchical response strategy based on the marginal utility of interruption costs. User categories include industrial users, commercial users and residential users. The response module flexibly adjusts the load reduction target according to the actual needs of users, and maximizes the economic benefits of user participation by optimizing the incentive strategy.

[0144] According to the user's electricity consumption behavior and interruption cost characteristics, users are divided into three categories: industrial, commercial and residential. Industrial users focus on peak shaving response, commercial users focus on incentive benefits, and residential users tend to respond at low cost; a hierarchical incentive strategy is adopted to convexly optimize the interruption cost function to ensure that the incentive benefits of different users are maximized. For different user types, the marginal utility model is optimized to increase user participation in load reduction; real-time monitoring of user load and electricity price changes, combined with user response willingness to dynamically adjust the reduction target, to ensure a balance between user interests and system benefits.

[0145] In addition, incentive strategies can be combined with blockchain technology to achieve transparent incentive distribution through smart contracts; interruption cost models can use deep reinforcement learning to optimize user behavior predictions and achieve more efficient response strategy design; demand response solutions can be expanded to comprehensive load management systems including electric vehicles and distributed energy storage.

[0146] Combined with the risk scheduling model of virtual power plants, the relationship between distributed photovoltaic scale and prediction error is decoupled through affine function to achieve adaptive optimization of distributed resources; the conditional value at risk (CVaR) method is used to reduce the scheduling risk of virtual power plants under the conditions of photovoltaic output and market electricity price fluctuations.

[0147] Affine function decoupling modeling is used to model the relationship between distributed photovoltaic scale and prediction error as the following affine function:

[0148]

[0149] Among them, α'=ωα, β'=αβ are adjustment parameters, reflecting the relationship between aggregation scale and error.

[0150] CVaR risk optimization, risk objective function is: max f = (1-β)E[β] + βCVαR r .

[0151] Among them, CVαR rIt represents the conditional value at risk under the confidence level γ, and β is the risk aversion coefficient.

[0152] Risk-optimized scheduling of virtual power plants:

[0153] Taking a virtual power plant in a certain region as an example, assuming that the prediction error of the photovoltaic cluster is 30% (small scale) and 20% (large scale), the impact of different aggregation scales on benefits and risks is analyzed. The operation results under multiple scenarios are shown in Table 2:

[0154] Table 2 Operation results

[0155]

[0156] Result analysis: Under scenario 1, the return increased by 13% and the risk decreased by 23%, proving the effectiveness of large-scale aggregation and optimization strategies.

[0157] In addition, the risk assessment method can be combined with Monte Carlo simulation or stochastic programming to further improve the meticulousness of modeling; the scheduling algorithm can adopt deep reinforcement learning to optimize dynamic scheduling through real-time data training; the affine function model can be replaced by a distributed generalized linear model (GLM) to enhance its applicability in complex scenarios.

[0158] The data privacy protection module injects gamma distribution noise in the user data collection stage through the local differential privacy mechanism to achieve privacy protection, while ensuring the efficiency of power load statistical data; the differential privacy protection algorithm optimizes the allocation of electricity prices during peak, spike, valley and concave periods in combination with user time-sharing electricity consumption data.

[0159] In the data collection of smart meters, gamma distribution noise is injected to achieve local differential privacy protection, ensuring that user personal data cannot be directly inferred; the peak and valley load curves are statistically analyzed using noisy data, and the peak and valley electricity prices are adjusted based on the elasticity coefficient to encourage users to use electricity during off-peak hours; by optimizing the privacy protection budget parameters, the data distortion rate is reduced to ensure a balance between electricity price optimization and user data privacy protection.

[0160] In addition, the data protection mechanism can be changed to federated learning to further enhance the reliability of data privacy protection; the optimization algorithm can be combined with fuzzy mathematics methods to enhance the robustness of the electricity price optimization plan; the data acquisition equipment can be upgraded to a smart terminal that supports multiple protocols to adapt to a wider range of application scenarios.

[0161] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A distributed power resource collaborative optimization system, characterized in that: include: Data acquisition module, spatiotemporal correlation analysis module, user response module, data privacy protection module, time-of-use electricity price optimization module and dispatch control module.

2. A distributed power resource collaborative optimization system as claimed in claim 1, characterized in that: The data acquisition module includes a module for collecting real-time data of distributed photovoltaic, energy storage equipment and loads.

3. A distributed power resource collaborative optimization system as claimed in claim 2, characterized in that: The spatiotemporal correlation analysis module includes using the Bayesian-long short-term memory network model to perform interval prediction on distributed photovoltaic power; combining Pearson correlation analysis to perform feature decomposition on the historical data of the distributed photovoltaic cluster to enhance the accuracy of spatial correlation modeling; The Bayesian-long short-term memory network model achieves high-precision prediction of distributed photovoltaic output range under complex and highly fluctuating weather conditions by dynamically adjusting training parameters.

4. A distributed power resource collaborative optimization system as claimed in claim 3, characterized in that: The user response module includes designing a hierarchical response strategy based on user categories and response willingness, combining demand willingness with marginal cost optimization scheduling; designing a hierarchical response strategy based on the marginal utility of interruption cost, in which user categories include industrial users, commercial users and residential users; flexibly adjusting the load reduction target according to the actual needs of users, and maximizing the economic benefits of user participation by optimizing the incentive strategy.

5. A distributed power resource collaborative optimization system as claimed in claim 4, characterized in that: The data privacy protection module includes protecting electricity consumption data through a differential privacy mechanism; injecting gamma distribution noise into the user data collection stage through a local differential privacy mechanism to achieve privacy protection, while ensuring the efficiency of electricity load statistical data; The differential privacy mechanism includes optimizing the allocation of electricity prices during peak, sharp, valley and concave periods in combination with user time-sharing electricity consumption data.

6. A distributed power resource collaborative optimization system as claimed in claim 5, characterized in that: The time-of-use electricity price optimization module includes dynamically adjusting the electricity price according to the peak-valley load difference.

7. A distributed power resource collaborative optimization system as claimed in claim 6, characterized in that: The dispatch control module includes a multi-layer optimization framework, including a distributed resource integration layer, a prefecture-level power grid coordination layer, and a provincial power grid optimization layer, to achieve coordinated regulation within the region; Through distributed optimization algorithms, adaptive control between distributed resource integration layer, prefecture-level power grid coordination layer and provincial power grid optimization layer is achieved; The multi-layer optimization includes a coordinated dispatch strategy to dynamically adjust the charging and discharging plans of distributed energy storage to smooth supply and demand fluctuations.

8. A method for collaborative optimization of distributed power resources based on claims 6 and 7, characterized in that: The peak-valley load dynamic adjustment includes dynamically balancing the electricity price benefit and system load, setting the peak and valley periods to adjust the electricity price elasticity coefficient in real time; The scheduling strategy includes decoupling the relationship between distributed photovoltaic scale and prediction error through affine function to achieve adaptive optimization of distributed resources; and reducing the scheduling risk of virtual power plants under the conditions of photovoltaic output and market electricity price fluctuations through conditional risk value method.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a distributed power resource collaborative optimization system as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a distributed power resource collaborative optimization system as described in any one of claims 1 to 7 are implemented.

Images