Personalized prediction method and system for virtual power plant behavior characteristics

Abstract

The invention discloses a personalized prediction method and system for behavior characteristics of a virtual power plant, and relates to the technical field of electricity and automation thereof, and the method comprises the steps: constructing all distributed resource models in the virtual power plant; each distributed resource model in the virtual power plant is summarized into a polyhedral multi-cycle operation model with linear inequality constraints; based on the uncertainty of the distributed resources, through an internal approximation method of flexible aggregation of a feasible region, performing aggregation modeling on each distributed resource in the virtual power plant, and generating a flexible boundary of the overall available resources of the virtual power plant; according to the method, historical operation data and meteorological data are utilized to analyze a coupling relationship between environmental factors and distributed resource output, a distributed resource power prediction model based on an Informer-network network is constructed, and overall resources of a virtual power plant are predicted. According to the method, the problem of complexity in virtual power plant resource scheduling is effectively solved, the precision of power prediction and the flexibility of resource scheduling are improved, and the operation efficiency of a power grid is optimized.

Description

Technical Field

[0001] The present invention relates to the field of electrical and automation technology, and in particular to a method and system for personalized prediction of behavior characteristics of a virtual power plant. Background Art

[0002] Virtual Power Plant (VPP) relies on the Internet and modern information and communication technology to aggregate multiple types of distributed resources into a whole, showing the characteristics of a power plant to the outside world. It can provide multiple services such as power energy balance, backup, frequency regulation, and peak regulation for the power grid, which is of great significance for improving the safe and stable operation level of the system and promoting the consumption of new energy. However, since virtual power plants involve multiple distributed resources (such as photovoltaics, wind power, energy storage, and adjustable loads, etc.), their operating characteristics are complex and mutually coupled, and how to accurately model and predict the power output of resources has become a major problem. Most of the existing prediction methods rely on traditional statistical models or simplified assumptions, which cannot effectively capture the complex spatiotemporal coupling relationship and uncertainty, resulting in low prediction accuracy and inability to meet real-time scheduling requirements. Summary of the invention

[0003] The present invention provides a method and system for personalized prediction of behavior characteristics of a virtual power plant, which can effectively solve the problems pointed out in the background technology.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] A personalized prediction method for virtual power plant behavior characteristics, comprising:

[0006] Considering the output characteristics and spatiotemporal characteristics of different types of distributed renewable energy, the distributed resource models within the virtual power plant are constructed;

[0007] Combining the coupling of active power, energy and regulation services of the distributed resources, and based on the distributed resource models within the virtual power plant, a multi-period operation model defined by linear inequality constraints is established;

[0008] Based on the uncertainty of distributed resources, the distributed resources in the virtual power plant are modeled through the inner approximation method of flexible aggregation of feasible domain, the flexibility boundary of the overall available resources of the virtual power plant is obtained, and the overall available resource range of the virtual power plant is determined;

[0009] Based on the flexibility boundary of the virtual power plant and the overall available resource range, historical operation data and meteorological data are used to analyze the coupling relationship between environmental factors and distributed resource output, and a distributed resource power prediction model based on the Informer-network is constructed to predict the overall resources of the virtual power plant.

[0010] Furthermore, each of the distributed resource models can be characterized as:

[0011] Energy storage equipment,

[0012]

[0013] In the formula, and are the charging power, discharging power and remaining energy of ES i respectively; is the energy dissipation rate; and They are the maximum charging power, maximum discharging power and climbing limit of ES i respectively; and are the maximum and minimum remaining energies allowed subject to the state-of-charge boundary constraints of ES i; are the charging efficiency and discharging efficiency of ES i, respectively, and ΔT is the change in ambient temperature;

[0014] Thermally controllable loads,

[0015]

[0016] In the formula, and are the power consumption and indoor temperature of TCR i respectively; and are the maximum and minimum power consumption and climbing limit of TCR i respectively; and are the lower and upper bounds of the ideal indoor temperature respectively; w TCR (t) is the ambient temperature; is the dissipation rate of TCR i; TCR is the conversion coefficient of active power to temperature; is the influencing factor of ambient temperature;

[0017] Delayed load,

[0018]

[0019] In the formula, and are the input energy and residual energy of DL i, respectively; and are the dissipation rate and conversion coefficient of DL i, respectively; They are the maximum and minimum charging power and rated charging capacity respectively; represents the energy required for DL ​​i; and are the start and end time of DL i respectively.

[0020] Furthermore, in combination with the coupling of active power, energy and regulation services of the distributed resources, a multi-period operation model defined by linear inequality constraints is established based on each distributed resource model within the virtual power plant, further comprising:

[0021] Considering the coupling of active power, energy and regulation services, a general multi-period operation model of a single distributed resource is obtained;

[0022] Expressing the constraints of active power, energy and regulation service of each distributed resource as linear inequalities, and constructing the feasible domain of each distributed resource within multiple cycles;

[0023] All linear inequality constraints are integrated into a polyhedral model to represent the feasible operation space of all resources.

[0024] Furthermore, the general multi-cycle operation model of the single distributed resource is expressed as:

[0025]

[0026] In the formula, and e i (t) are the input power, output power and remaining energy of a single distributed resource i; The upper limit of regulation for regulation service; θ i is the energy dissipation rate, and are the maximum input power, maximum output power and ramp limit of distributed resource i, respectively, and e i (t) and are the maximum and minimum remaining energies allowed subject to the state of charge boundary constraints of distributed resource i, are the energy input efficiency and energy output efficiency of distributed resource i, respectively, and w i (t) is the ambient temperature; ω i is the influence factor of ambient temperature, η i is the conversion coefficient of active power of distributed resource i to temperature.

[0027] Furthermore, based on the uncertainty of distributed resources, the distributed resources in the virtual power plant are modeled through the inner approximation method of flexible aggregation of the feasible domain, the flexibility boundary of the overall available resources of the virtual power plant is obtained, and the overall available resource range of the virtual power plant is determined, further including:

[0028] Establish an independent feasible domain model for each distributed resource in the virtual power plant, representing the operating range of the resource in different time periods;

[0029] Use Minkowski and to aggregate the feasible domains of multiple distributed resources in the virtual power plant to obtain the comprehensive feasible domain of all resources and form the available resource range of the virtual power plant as a whole;

[0030] For the comprehensive feasible domain of each distributed resource, an inner approximation method based on a homogeneous polyhedron is adopted to optimize the resource aggregation model by adjusting the scaling factor and the conversion factor to form the flexibility boundary of the overall available resources of the virtual power plant.

[0031] Furthermore, the comprehensive feasible domain is:

[0032]

[0033] In the formula φ i ≥0 is the scaling factor of distributed resource i, is the conversion factor of distributed resource i, is the basic homogeneous polyhedron of the aggregate k; and is the coefficient matrix that determines the feasible domain polyhedron, X0 is the controllable variable vector in the basic homogeneous polyhedron; X i represents the i-th controllable variable vector, and X is the controllable variable vector of the entire virtual power plant after aggregation.

[0034] Furthermore, the flexibility boundary representation of the overall available resources of the virtual power plant is formed, that is, the aggregation feasible domain of the virtual power plant aggregation model is:

[0035]

[0036] In the formula, is the feasible domain of aggregation, is the controllable variable vector, And R k =[r k (t) |t＝1,…,T ] T , are the input power and output power of the kth distributed resource respectively; r k (t) provides scope for the regulation service of the kth distributed resource, and denote the scaling factor and conversion factor respectively, and is the coefficient matrix that determines the feasible domain polyhedron.

[0037] Furthermore, for the historical operation data and meteorological data, a feature selection method based on the maximum mutual information coefficient is used to reduce the model input feature dimension, and the calculation formula is:

[0038]

[0039] Where I(a,b) is the mutual information value between variables a and b; A and B are the number of segments divided along the a and b directions, that is, the grid distribution, the values ​​of A and B increase from 1, and A*B≤Q is satisfied; the size of Q is 0.6 of the total amount of data; k(A i ,B i ) is the coordinate point (a, b) that falls into A at the same time i , B i The probability of k(A i )、k(B i ) are coordinate points (a, b) that fall into A i , B i probability; log2min(A,B) is a normalization factor used to control the MIC value range between [0,1]. The larger the MIC value between the two variables, the stronger the correlation. Conversely, the smaller the MIC value, the smaller the correlation between the two variables.

[0040] Furthermore, the distributed resource power prediction model based on the Informer-network includes:

[0041] Input variable embedding module: Input variables are embedded through data encoding, position encoding and timestamp encoding to generate low-dimensional embedding vectors for subsequent encoders and decoders to use;

[0042] Self-attention module, used to filter out more important features from input information;

[0043] The encoder module is used to extract long-term dependencies in the input data;

[0044] A decoder module is used to generate power prediction outputs for distributed resources.

[0045] A personalized prediction system for virtual power plant behavior characteristics, the system comprising:

[0046] The distributed model building module considers the output characteristics and spatiotemporal characteristics of different types of distributed renewable energy sources and builds the distributed resource models within the virtual power plant.

[0047] A multi-period model building module, combining the coupling of active power, energy and regulation services of the distributed resources, and building a multi-period operation model defined by linear inequality constraints based on each distributed resource model within the virtual power plant;

[0048] The flexibility boundary generation module, based on the uncertainty of distributed resources, models the distributed resources in the virtual power plant through the inner approximation method of flexible aggregation of the feasible domain, obtains the flexibility boundary of the overall available resources of the virtual power plant, and determines the overall available resource range of the virtual power plant;

[0049] The distributed model construction module, based on the flexibility boundary of the virtual power plant and the overall available resource range, uses historical operation data and meteorological data to analyze the coupling relationship between environmental factors and distributed resource output, constructs a distributed resource power prediction model based on the Informer-network network, and predicts the overall resources of the virtual power plant.

[0050] The technical solution of the present invention can achieve the following technical effects:

[0051] It effectively solves the complexity problem in virtual power plant resource scheduling, improves the accuracy of power forecasting and the flexibility of resource scheduling, and optimizes the efficiency of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 A flowchart of a personalized prediction method for virtual power plant behavior characteristics;

[0054] Figure 2 Schematic diagram of distributed resource aggregation for virtual power plants;

[0055] Figure 3 It is a diagram of the flexible aggregation method based on homogeneous polyhedron;

[0056] Figure 4 This is the structure diagram of the Informer-network model. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0059] Embodiment 1

[0060] like Figure 1 As shown, the present invention provides a method for personalized prediction of virtual power plant behavior characteristics, the method comprising:

[0061] S1: Considering the output characteristics and spatiotemporal characteristics of different types of distributed renewable energy, construct the distributed resource models within the virtual power plant;

[0062] Specifically, for each type of distributed resource (such as photovoltaic, wind power, energy storage equipment, adjustable load, etc.), their power output characteristics need to be analyzed separately. The power output of photovoltaic and wind power resources is affected by the fluctuation of sunshine intensity and wind speed, and has significant temporal and spatial variability; energy storage equipment is mainly affected by charging and discharging characteristics, energy storage limitations and equipment efficiency; and adjustable loads depend on the demand changes of the power grid for dynamic adjustment. On this basis, for each resource, a corresponding mathematical model is established, and linear inequalities and dynamic equations are used to describe its operating behavior in different time periods. These models not only consider the power output of the resources, but also their energy storage, charging and discharging capabilities, and regulation service capabilities. Factors such as service, so as to accurately reflect the operating characteristics and coupling relationships of various resources within the virtual power plant, provide important support for subsequent resource aggregation modeling and power prediction.

[0063] S2: Combining the coupling of active power, energy and regulation services of distributed resources, a multi-period operation model defined by linear inequality constraints is established based on the distributed resource models within the virtual power plant;

[0064] Specifically, by considering the coupling relationship between the active power, energy and regulation services of each distributed resource in the virtual power plant, the behavior of these resources is summarized into a multi-period operation model in the form of a polyhedron. This modeling method can accurately describe the output characteristics of various resources in the virtual power plant in different time periods and their interactions, reflecting the complex spatiotemporal coupling relationship between resources. By adopting the polyhedron form of linear inequality constraints, the operation constraints of resources can be converted into a geometric model, so that the operational range of resources can be clearly expressed in the time dimension. This polyhedron model can provide a basis for virtual power plant resource scheduling, help optimize resource allocation, and improve the overall efficiency and stability of the power system. At the same time, this method can also effectively deal with external uncertainties (such as weather changes, demand fluctuations, etc.), provide a more robust solution for the scheduling of the power system, and provide a computationally feasible and efficient framework for subsequent scheduling optimization and decision-making.

[0065] S3: Based on the uncertainty of distributed resources, the distributed resources in the virtual power plant are modeled through the inner approximation method of flexible aggregation of the feasible domain, the flexibility boundary of the overall available resources of the virtual power plant is obtained, and the overall available resource range of the virtual power plant is determined;

[0066] Specifically, we first need to consider the output characteristics and uncertainties of each distributed resource in the virtual power plant, such as wind power, photovoltaic power, energy storage equipment and adjustable load. The output of these resources is not only affected by the natural environment (such as weather changes, seasonal fluctuations, etc.), but also by external factors such as grid demand and market prices. Therefore, how to reasonably aggregate each distributed resource under the condition of considering uncertainty has become the key to improving the dispatching capability of the virtual power plant. Through the inner approximation method of flexible aggregation of feasible domain, the feasible domain of each distributed resource is adjusted and aggregated. First, the feasible domain of each resource is modeled as a homogeneous polyhedron, and then the flexibility aggregation of each resource is achieved by adjusting the scaling factor and conversion factor. Through this method, the flexibility boundary of the overall resources in the virtual power plant can be accurately described under uncertain environmental conditions. This flexibility boundary not only considers the dynamic characteristics of the resources themselves, but also reflects the interaction between resources and the uncertainty of the external environment; finally, through the aggregated flexibility boundary, more accurate decision support can be provided for the dispatching and resource management of the virtual power plant, ensuring that the virtual power plant can provide stable, reliable and efficient power services in the face of uncertainty.

[0067] S4: Based on the flexibility boundary of the virtual power plant and the overall available resource range, using historical operation data and meteorological data, analyze the coupling relationship between environmental factors and distributed resource output, build a distributed resource power prediction model based on the informer-network network, and predict the overall resources of the virtual power plant.

[0068] Specifically, it is necessary to first collect the historical operating data of each distributed resource in the virtual power plant (such as power output, load changes, etc.) and meteorological data related to the environment (such as temperature, humidity, wind speed, light intensity, etc.), and pre-process and clean these data. Then, by analyzing the coupling relationship between environmental factors and the output of distributed resources, a mathematical model is established to reveal the impact of different environmental factors on the output of each resource; next, a power prediction model based on Informer-network is constructed. Informer-network is a time series prediction model based on deep learning that can effectively capture long-term dependencies. In this model, historical data and meteorological data are converted into low-dimensional vectors through the input embedding layer, and position encoding and timestamp encoding are added to ensure that the model can handle the time relationship in the time series data. Using the self-attention mechanism and multi-head self-attention layer, the model can focus on the information of important time points and predict future power output based on historical data and external meteorological factors. The output of the model is the power forecast of the virtual power plant in the future period, which can be used for power dispatch, load balancing and energy optimization. In addition, the informer-network can also provide a certain degree of uncertainty estimation to help power system operators better cope with unforeseen changes. Ultimately, based on this prediction model, the virtual power plant can achieve more accurate resource dispatch and improve the overall system operation efficiency and stability.

[0069] In view of the different output characteristics of distributed resources in virtual power plants and the complex mapping relationship between internal resource characteristics and external economic characteristics, this invention proposes a personalized prediction method for virtual power plant behavior characteristics that takes into account multiple types of power commodity transactions. Compared with the commonly used methods, this method significantly improves the accuracy and completeness of the virtual power plant model and the prediction accuracy of virtual power plant output. It is beneficial to the power dispatching and peak-shaving of the power system, and improves the efficiency, economy and stability of the power system.

[0070] As a preference for the above embodiment, each distributed resource model can be characterized as follows:

[0071] Energy storage equipment,

[0072]

[0073]

[0074] In the formula, and are the charging power, discharging power and remaining energy of ES i respectively; is the energy dissipation rate; and They are the maximum charging power, maximum discharging power and climbing limit of ES i respectively; and are the maximum and minimum remaining energies allowed subject to the state-of-charge boundary constraints of ES i; are the charging efficiency and discharging efficiency of ES i, respectively;

[0075] Thermally controllable loads,

[0076]

[0077] In the formula, and are the power consumption and indoor temperature of TCR i respectively; and are the maximum and minimum power consumption and climbing limit of TCR i respectively; and are the lower and upper bounds of the ideal indoor temperature respectively; w kCR (t) is the ambient temperature; is the dissipation rate of TCR i; TCR is the conversion coefficient of active power to temperature; is the influencing factor of ambient temperature;

[0078] Delayed load,

[0079]

[0080] In the formula, and are the input energy and residual energy of DL i, respectively; and are the dissipation rate and conversion coefficient of DL i, respectively; They are the maximum and minimum charging power and rated charging capacity respectively; represents the energy required for DL ​​i; and are the start and end time of DL i respectively.

[0081] Specifically, energy storage devices: batteries and other energy storage devices can store / release energy and realize the conversion of energy between different time periods. Considering the daily charging and discharging strategies and charge state boundaries of energy storage devices, the energy storage devices are modeled; thermally controllable loads: the dynamic characteristics of TCR can be modeled using derivatives and then converted into discrete forms; deferrable loads: electric vehicle charging loads are a typical DL, and this type of distributed resource needs to meet energy requirements within a predetermined time.

[0082] As a preferred embodiment of the above, in combination with the coupling of active power, energy and regulation services of distributed resources, a multi-period operation model defined by linear inequality constraints is established based on each distributed resource model within the virtual power plant, further comprising:

[0083] S21: Considering the coupling of active power, energy and regulation services, a general multi-period operation model of a single distributed resource is obtained;

[0084] S22: Expressing the constraints of active power, energy and regulation service of each distributed resource as a linear inequality, and constructing a feasible domain of each distributed resource within multiple cycles;

[0085] S23: Integrate all linear inequality constraints into a polyhedral model to represent the feasible operation space of all resources.

[0086] Specifically, first, in step S21, the mutual coupling relationship between the active power, energy storage and regulation services of different types of distributed resources (such as energy storage equipment, wind power, photovoltaics, adjustable loads, etc.) in the virtual power plant is considered. Each resource has its own specific output characteristics. For example, the charging and discharging power of the energy storage equipment is affected by the remaining energy and regulation requirements, while the output of photovoltaic and wind resources is affected by the fluctuations of environmental factors (such as light and wind speed). Therefore, for each distributed resource, it is necessary to establish a corresponding multi-period operation model, that is, to consider the power output, energy change and ability to provide regulation services of the resource in different time steps. In this way, the model can reflect the dynamic behavior of the resource and predict its changes and uncertainties in future time periods; then, in step S22, it is necessary to convert the constraints of active power, energy and regulation services of each distributed resource into the form of linear inequalities. These constraints include: the power output of each resource is subject to maximum and minimum limits, the charging and discharging power of the energy storage device is subject to the constraints of the device capacity and the remaining energy, and the ability of the regulation service is also subject to physical limitations such as the power change rate (climbing rate). All these constraints are described in the form of linear inequalities, thereby providing a feasible operating space (feasible domain) for each distributed resource in multiple time periods. Clear boundaries are set. These linear inequality constraints ensure that each resource will not violate its physical limitations during the actual scheduling process and provide a basis for optimized scheduling. Finally, in step S23, the linear inequality constraints of all distributed resources are integrated to obtain a polyhedron model. The polyhedron model is a geometric representation method that represents the constraints of each resource in different time periods as the boundaries of a polyhedron, thereby forming a feasible operating space for the resources. This geometric representation method can clearly describe the possible operating states of all resources in the virtual power plant in multiple time periods, and can also reflect the mutual influence and constraints between different resources. Through the geometric representation of the polyhedron, the overall resource scheduling of the virtual power plant can more accurately calculate the operating range and scheduling restrictions of each resource, ensuring efficient resource scheduling and optimization while meeting all resource constraints.

[0087] As a preference for the above embodiment, a general multi-cycle operation model of a single distributed resource is expressed as:

[0088]

[0089] In the formula, and e i (t) are the input power, output power and remaining energy of a single distributed resource i; The upper limit of regulation for regulation service; θ i is the energy dissipation rate, and are the maximum input power, maximum output power and ramp limit of distributed resource i, respectively, and e i (t) and are the maximum and minimum remaining energies allowed subject to the state of charge boundary constraints of distributed resource i, are the energy input efficiency and energy output efficiency of distributed resource i, respectively, and w i (t) is the ambient temperature; ω i is the influence factor of ambient temperature, η i is the conversion coefficient of active power of distributed resource i to temperature.

[0090] Specifically, an equivalent aggregation model of a virtual power plant considering uncertainty is constructed. ,In addition to the coupling relationship between power and energy, many distributed resources can ,further provide regulation services to the power system.,Therefore, considering the coupling of active power, energy, and regulation services, ,a general multi-period operation model of a single distributed resource can be obtained.

[0091] The scope of regulation service provision should be determined by the maximum acceptable regulation range set by the virtual power plant, the active power variation law and the residual energy state. The specific formula is as follows:

[0092]

[0093]

[0094] in, is the maximum acceptable range, h R is the energy reserve ratio of the specified service. Note that the mentioned regulation service can represent different types of ancillary services in actual implementation, such as frequency regulation capacity or hot standby service. Since the distributed resources are heterogeneous, the core parameters and feasible domains are different. Without losing generality, considering active power import, active power export and regulation capacity provision as controllable variables, they can be rearranged into the following compact form, which is essentially a polyhedron constrained by linear inequalities.

[0095]

[0096] in represents the controllable variable vector, And R i =[r i (t) |t＝1,…,T ] T ;M i and N i is the coefficient matrix.

[0097] As a preferred embodiment of the above, based on the uncertainty of distributed resources, the distributed resources in the virtual power plant are modeled by the inner approximation method of flexible aggregation of the feasible domain, the flexibility boundary of the overall available resources of the virtual power plant is obtained, and the overall available resource range of the virtual power plant is determined, which further includes:

[0098] S31: Establish an independent feasible domain model for each distributed resource in the virtual power plant, representing the operating range of the resource in different time periods;

[0099] S32: Aggregate the feasible domains of multiple distributed resources in the virtual power plant using Minkowski and obtain the comprehensive feasible domain of all resources to form the available resource range of the virtual power plant as a whole;

[0100] S33: For the comprehensive feasible domain of each distributed resource, an inner approximation method based on a homogeneous polyhedron is adopted to optimize the resource aggregation model by adjusting the scaling factor and the conversion factor to form a flexibility boundary of the overall available resources of the virtual power plant.

[0101] Specifically, first, in step S31, an independent feasible domain model is established for each distributed resource in the virtual power plant (such as photovoltaic, wind power, energy storage equipment and adjustable load, etc.). The feasible domain of each resource represents its operating range in different time periods, taking into account factors such as its active power, energy state and regulation capability. For example, the feasible domain of the energy storage equipment is limited by the charging and discharging power and the remaining energy, while the power output of wind power and photovoltaic is affected by meteorological conditions. By modeling these resources, their operating range in different time periods can be accurately described; in step S32, the Minkowski sum is used to aggregate the feasible domains of multiple distributed resources in the virtual power plant. The Minkowski sum operation merges the feasible domains of each resource into a comprehensive feasible domain to form the overall available resource range of the virtual power plant. This comprehensive feasible domain reflects the dispatchable state of all resources in the virtual power plant in multiple time periods and ensures the synergy between the resources. Through this aggregation method, the constraints of all resources can be uniformly considered, providing a global operation framework for the overall dispatch of the virtual power plant; finally, in step S33, the comprehensive feasible domain of each distributed resource is optimized using the inner approximation method based on homogeneous polyhedron. By adjusting the scaling factor and conversion factor of each resource, the aggregation model of the resource is optimized. This method enhances the flexibility boundary of the virtual power plant by accurately adjusting the shape of the resource feasible domain, ensuring that it can adapt to external uncertainties and changes in grid demand. Finally, these optimized flexibility boundaries provide a more accurate and efficient basis for resource scheduling and optimization of virtual power plants.

[0102] As a preferred embodiment of the above, the comprehensive feasible domain is:

[0103]

[0104] In the formula φ i ≥0 is the scaling factor of distributed resource i, is the conversion factor of distributed resource i, is the basic homogeneous polyhedron of aggregate k; and is the coefficient matrix that determines the feasible domain polyhedron, X0 is the controllable variable vector in the basic homogeneous polyhedron; X i represents the i-th controllable variable vector, and X is the controllable variable vector of the entire virtual power plant after aggregation.

[0105] As a preferred embodiment of the above, the flexibility boundary representation of the overall available resources of the virtual power plant is formed, that is, the aggregation feasible domain of the virtual power plant aggregation model is:

[0106]

[0107] In the formula, is the feasible domain of aggregation, is the controllable variable vector, And R k =[r k (t) |t＝1,…,T ] T , are the input power and output power of the kth distributed resource respectively; r k (t) provides scope for the regulation service of the kth distributed resource, and denote the scaling factor and conversion factor respectively, and is the coefficient matrix that determines the feasible domain polyhedron.

[0108] Specifically, based on the flexible aggregation method of homogeneous polyhedron feasible domain, a virtual power plant aggregation model considering the uncertainty of distributed resources is established, such as Figure 2 shown.

[0109] The deterministic feasible domain of a set of distributed resources (Ω k AGG ) can be characterized by the Minkowski sum of each distributed resource, which is defined as:

[0110]

[0111] in, Calculate for the Minkowski sum.

[0112] The present invention adopts a basic homogeneous polyhedron to approximate a distributed resource polyhedron. In order to enhance the similarity, a basic homogeneous polyhedron having the same structure as a general distributed resource polyhedron is adopted. The method aims to Compute an inner approximate polyhedral feasible region for each distributed resource For the feasible domain of distributed resources from the same homogeneous polyhedron, the Minkowski sum calculation can be simplified to:

[0113]

[0114] The aggregation method based on homogeneous polyhedrons only needs to perform arithmetic addition when calculating the Minkowski sum, which has high computational efficiency. Figure 3 As shown in Figure 2, based on the basic homogeneous polyhedron and considering the uncertainty of distributed resources, the maximum inner approximate feasible domain of each distributed resource can be expressed as:

[0115]

[0116] Where ε∈(0,1) is the allowed out-of-bounds probability.

[0117] By scaling (φ i ) and conversion Basic homogeneous polyhedron The model is modeled as a chance constraint, and the inner approximation result of the original feasible region of the distributed resource s is obtained under uncertain conditions, aiming to expand the approximate feasible region of each distributed resource and maximize the aggregation accuracy. In order to convert it into a solvable form, an auxiliary variable is introduced Convert the maximum inner approximation problem into a linear programming problem:

[0118]

[0119] stα i ,U i ≥0;

[0120]

[0121]

[0122] Where L dim =16T-4 is the dimension of the N matrix in (2); U i (s,:) and M i (s,:) respectively represent U i and M i The sth row of ; is the uncertainty matrix N i The sth element of and σ i (s) is the predicted value and is given by N i The standard deviation of the prediction error of (s). For ES devices with non-negligible charge and discharge losses, β i (t)|t∈{1,…,2T} should be set to 0 to ensure that the charging and discharging power is simultaneously 0.

[0123] Through calculation, the scaling factor and conversion factor of a single distributed resource are obtained. According to formula (8), the scaling factor of the virtual power plant aggregation model is and conversion factor It can be calculated as follows:

[0124]

[0125] For the virtual power plant aggregation model, the aggregation feasible domain can be expressed as:

[0126]

[0127] As a preferred embodiment of the above, for the historical operation data and meteorological data, a feature selection method based on the maximum mutual information coefficient is used to reduce the model input feature dimension, and the calculation formula is:

[0128]

[0129] Where I(a,b) is the mutual information value between variables a and b; A and B are the number of segments divided along the a and b directions, that is, the grid distribution, the values ​​of A and B increase from 1, and A*B≤Q is satisfied; the size of Q is 0.6 of the total amount of data; k(A i ,B i ) is the coordinate point (a, b) that falls into A at the same time i , B i The probability of k(A i )、k(B i ) are coordinate points (a, b) that fall into Ai and B respectively i probability; log2min(A,B) is a normalization factor used to control the MIC value range between [0,1]. The larger the MIC value between the two variables, the stronger the correlation. Conversely, the smaller the MIC value, the smaller the correlation between the two variables.

[0130] Specifically, the maximum mutual information coefficient (MIC) is selected as a feature selection method mainly because it can capture the linear and nonlinear relationships between features. In the power prediction of virtual power plants, the input data usually contains meteorological data and historical operation data, and there are often complex nonlinear relationships between these data. Traditional feature selection methods mainly target linear relationships and cannot handle this situation. MIC can measure both linear and nonlinear associations at the same time, which is very suitable for such data. MIC can also effectively process high-dimensional data. By gridding the data, it can identify the features with the strongest correlation with the target variable (such as power output), thereby reducing redundant features and improving the computational efficiency of the model. In addition, MIC can also automatically remove irrelevant features, avoid the interference of redundant information on the model, and improve the prediction accuracy. In addition, MIC can process different types of data, including continuous data and discrete data, which makes it more flexible in processing various data in virtual power plants. Therefore, the use of MIC can help accurately select features related to power prediction, thereby optimizing the resource scheduling of virtual power plants and the operation efficiency of power systems.

[0131] As a preferred embodiment of the above, Figure 4 As shown in FIG. 1 , the distributed resource power prediction model based on the Informer-network includes:

[0132] Input variable embedding module: Input variables are embedded through data encoding, position encoding and timestamp encoding to generate low-dimensional embedding vectors for subsequent encoders and decoders to use;

[0133] Self-attention module, used to filter out more important features from input information;

[0134] The encoder module is used to extract long-term dependencies in the input data;

[0135] A decoder module is used to generate power prediction outputs for distributed resources.

[0136] Specifically, first, the input variable embedding module converts the original high-dimensional input data into a low-dimensional embedding vector through data encoding, position encoding and timestamp encoding. Data encoding extracts local features of the input data through one-dimensional convolution operations, position encoding adds time series information to each data point through sine and cosine functions, and timestamp encoding converts time information into a vector of fixed dimension. These encoded embedding vectors are summed and provided as input to subsequent encoder and decoder modules; next, the self-attention module uses the self-attention mechanism to screen out the most important features for power prediction. The self-attention mechanism dynamically allocates attention points by calculating the correlation between the various parts of the input data, allowing the model to automatically focus on the data that has the strongest relationship with the prediction target. region, this mechanism can capture the complex temporal dependencies of the input data and effectively handle the long-term and short-term dependencies in the data; then, the encoder module extracts the long-term dependencies in the input data through multiple self-attention layers to capture the global information of the entire sequence. The multi-head self-attention mechanism in the encoder enables the model to process multiple feature subspaces in parallel, thereby more comprehensively understanding the intrinsic structure of the data. In addition, the encoder also combines convolution operations and pooling operations to further improve the efficiency of feature extraction and reduce computational complexity; finally, the decoder module generates power prediction values ​​for distributed resources based on the output of the encoder. The decoder uses two multi-head self-attention layers to process the features obtained from the encoder and generates predictions for future power output based on these features. At the same time, in order to increase processing speed, the decoder directly receives a compressed input vector, thereby accelerating the prediction process.

[0137] The encoder is mainly used to extract long-term dependencies between long sequence inputs. Its main component is the multi-head probabilistic sparse self-attention mechanism. Since there are redundant combinations in the results of the probabilistic sparse self-attention module, it is necessary to extract the dominant key features through the distillation mechanism. The formula for the distillation operation from layer j to layer j+1 is:

[0138]

[0139] in, It is an important operation output in the multi-head probabilistic sparse self-attention block. Conv1d is a one-dimensional convolution operation, whose activation function is the ELU function, and MaxPool is a pooling operation. Through this distillation operation, the length of the input sequence can be shortened by half, thereby reducing the memory usage and computing time of the encoder.

[0140] Decoder module (Deco distributed resource)

[0141] The model uses a standard decoder, which consists of two multi-head self-attention layers. At the same time, in order to alleviate the processing speed drop of generating prediction results, an input vector is directly provided to the decoder:

[0142]

[0143] At the same time, a masked multi-head attention mechanism is used for the input vector, which can prevent attention from being paid to information at future moments when processing data at a certain moment. Then, the final output is obtained through a fully connected layer. It is worth noting that the decoder structure proposed by this method is a generative structure, that is, the entire prediction sequence can be generated at one time, which greatly reduces the prediction time of decoding. In the encoder and decoder, the connection between each layer uses residual and normalization modules, which can make the network deeper and effectively avoid the problems of gradient disappearance, gradient explosion, and the degradation of deep networks to shallow networks.

[0144] Embodiment 2

[0145] Based on the same inventive concept as the personalized prediction method for virtual power plant behavior characteristics in the aforementioned embodiment, the present invention further provides a personalized prediction system for virtual power plant behavior characteristics, the system comprising:

[0146] The distributed model building module considers the output characteristics and spatiotemporal characteristics of different types of distributed renewable energy sources and builds the distributed resource models within the virtual power plant.

[0147] The multi-period model building module combines the active power, energy and regulation service coupling of distributed resources, and builds a multi-period operation model defined by linear inequality constraints based on the distributed resource models within the virtual power plant;

[0148] The flexibility boundary generation module, based on the uncertainty of distributed resources, models the distributed resources in the virtual power plant through the inner approximation method of flexible aggregation of the feasible domain, obtains the flexibility boundary of the overall available resources of the virtual power plant, and determines the overall available resource range of the virtual power plant;

[0149] The distributed model construction module, based on the flexibility boundary of the virtual power plant and the overall available resource range, uses historical operation data and meteorological data to analyze the coupling relationship between environmental factors and distributed resource output, constructs a distributed resource power prediction model based on the Informer-network network, and predicts the overall resources of the virtual power plant.

[0150] The above-mentioned prediction system in the present invention can effectively implement a personalized prediction method for the behavior characteristics of a virtual power plant, and the technical effects that can be achieved are as described in the above-mentioned embodiments, which will not be repeated here.

[0151] Although the present application has been described in conjunction with specific features and embodiments thereof, it is obvious that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and the accompanying drawings are merely exemplary illustrations of the present application as defined therein, and are deemed to have covered any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A personalized prediction method for virtual power plant behavior characteristics, characterized in that: The method comprises: Considering the output characteristics and spatiotemporal characteristics of different types of distributed renewable energy, the distributed resource models within the virtual power plant are constructed; Combining the coupling of active power, energy and regulation services of the distributed resources, and based on the distributed resource models within the virtual power plant, a multi-period operation model defined by linear inequality constraints is established; Based on the uncertainty of distributed resources, the distributed resources in the virtual power plant are modeled through the inner approximation method of flexible aggregation of feasible domain, the flexibility boundary of the overall available resources of the virtual power plant is obtained, and the overall available resource range of the virtual power plant is determined; Based on the flexibility boundary of the virtual power plant and the overall available resource range, historical operation data and meteorological data are used to analyze the coupling relationship between environmental factors and distributed resource output, and a distributed resource power prediction model based on the Informer-network is constructed to predict the overall resources of the virtual power plant.

2. The personalized prediction method for virtual power plant behavior characteristics according to claim 1 is characterized in that: Each of the distributed resource models can be characterized as: Energy storage equipment, In the formula, and are the charging power, discharging power and remaining energy of ESi respectively; is the energy dissipation rate; and They are the maximum charging power, maximum discharging power and climbing limit of ESi respectively; and are the maximum and minimum remaining energies allowed subject to the state-of-charge boundaries of ESi; are the charging efficiency and discharging efficiency of ES i, respectively, and ΔT is the change in ambient temperature; Thermally controllable loads, In the formula, and are the power consumption and indoor temperature of TCR i respectively; and They are the maximum and minimum power consumption and climbing limit of TCRi respectively; and are the lower and upper bounds of the ideal indoor temperature respectively; w TCR (t) is the ambient temperature; is the dissipation rate of TCR i; TCRi is the conversion coefficient of active power to temperature; is the influencing factor of ambient temperature; Delayed load, In the formula, and are the input energy and residual energy of DLi respectively; and are the dissipation rate and conversion coefficient of DLi respectively; They are the maximum and minimum charging power and rated charging capacity respectively; Indicates the energy required for DLi; and are the start and end time of DLi respectively.

3. The personalized prediction method for virtual power plant behavior characteristics according to claim 1 is characterized in that: In combination with the coupling of active power, energy and regulation services of the distributed resources, a multi-period operation model defined by linear inequality constraints is established based on the distributed resource models within the virtual power plant, further comprising: Considering the coupling of active power, energy and regulation services, a general multi-period operation model of a single distributed resource is obtained; Expressing the constraints of active power, energy and regulation service of each distributed resource as linear inequalities, and constructing the feasible domain of each distributed resource within multiple cycles; All linear inequality constraints are integrated into a polyhedral model to represent the feasible operation space of all resources.

4. The personalized prediction method for virtual power plant behavior characteristics according to claim 3 is characterized in that: The general multi-cycle operation model of the single distributed resource is expressed as: In the formula, and e i (t) are the input power, output power and remaining energy of a single distributed resource i; Regulation ceiling for regulation services; θ i is the energy dissipation rate, and are the maximum input power, maximum output power and ramp limit of distributed resource i, respectively, and e i (t) and are the maximum and minimum remaining energies allowed subject to the state of charge boundary constraints of distributed resource i, are the energy input efficiency and energy output efficiency of distributed resource i, respectively, and w i (t) is the ambient temperature; ω i is the influence factor of ambient temperature, η i is the conversion coefficient of active power of distributed resource i to temperature.

5. The personalized prediction method for virtual power plant behavior characteristics according to claim 1 is characterized in that: Based on the uncertainty of distributed resources, the distributed resources in the virtual power plant are modeled through the inner approximation method of flexible aggregation of the feasible domain, the flexibility boundary of the overall available resources of the virtual power plant is obtained, and the overall available resource range of the virtual power plant is determined, further including: Establish an independent feasible domain model for each distributed resource in the virtual power plant, representing the operating range of the resource in different time periods; Use Minkowski and to aggregate the feasible domains of multiple distributed resources in the virtual power plant to obtain the comprehensive feasible domain of all resources and form the available resource range of the virtual power plant as a whole; For the comprehensive feasible domain of each distributed resource, an inner approximation method based on a homogeneous polyhedron is adopted to optimize the resource aggregation model by adjusting the scaling factor and the conversion factor to form the flexibility boundary of the overall available resources of the virtual power plant.

6. The personalized prediction method for virtual power plant behavior characteristics according to claim 5 is characterized in that: The comprehensive feasible domain is: In the formula φ i ≥0 is the scaling factor of distributed resource i, is the conversion factor of distributed resource i, is the basic homogeneous polyhedron of the aggregate k; and is the coefficient matrix that determines the feasible domain polyhedron, X0 is the controllable variable vector in the basic homogeneous polyhedron; X i represents the i-th controllable variable vector, and X is the controllable variable vector of the entire virtual power plant after aggregation.

7. The personalized prediction method for virtual power plant behavior characteristics according to claim 5 is characterized in that: The flexibility boundary representation of the overall available resources of the virtual power plant is formed, that is, the aggregation feasible domain of the virtual power plant aggregation model is: In the formula, is the feasible domain of aggregation, is the controllable variable vector, And R k =[r k (t) |t＝1,…,T ] T , are the input power and output power of the kth distributed resource respectively; r k (t) provides scope for the regulation service of the kth distributed resource, and denote the scaling factor and conversion factor respectively, and is the coefficient matrix that determines the feasible domain polyhedron.

8. The personalized prediction method for virtual power plant behavior characteristics according to claim 1 is characterized in that: For the historical operation data and meteorological data, a feature selection method based on the maximum mutual information coefficient is used to reduce the model input feature dimension. The calculation formula is: Where I(a,b) is the mutual information value between variables a and b; A and B are the number of segments divided along the a and b directions, that is, the grid distribution, the values ​​of A and B increase from 1, and A*B≤Q is satisfied; the size of Q is 0.6 of the total amount of data; k(A i ,B i ) is the probability that the coordinate point (a, b) falls into Ai and Bi at the same time, k(A i )、k(B i ) are coordinate points (a, b) that fall into A i , B i probability; log2min(A,B) is a normalization factor used to control the MIC value range between [0,1]. The larger the MIC value between the two variables, the stronger the correlation. Conversely, the smaller the MIC value, the smaller the correlation between the two variables.

9. The personalized prediction method for virtual power plant behavior characteristics according to claim 1 is characterized in that: The distributed resource power prediction model based on the Informer-network includes: Input variable embedding module: Input variables are embedded through data encoding, position encoding and timestamp encoding to generate low-dimensional embedding vectors for subsequent encoders and decoders to use; Self-attention module, used to filter out more important features from input information; The encoder module is used to extract long-term dependencies in the input data; A decoder module is used to generate power prediction outputs for distributed resources.

10. A personalized prediction system for virtual power plant behavior characteristics, characterized in that: The system comprises: The distributed model building module considers the output characteristics and spatiotemporal characteristics of different types of distributed renewable energy sources and builds the distributed resource models within the virtual power plant. A multi-period model building module, combining the coupling of active power, energy and regulation services of the distributed resources, and building a multi-period operation model defined by linear inequality constraints based on each distributed resource model within the virtual power plant; The flexibility boundary generation module, based on the uncertainty of distributed resources, models the distributed resources in the virtual power plant through the inner approximation method of flexible aggregation of the feasible domain, obtains the flexibility boundary of the overall available resources of the virtual power plant, and determines the overall available resource range of the virtual power plant; The distributed model construction module, based on the flexibility boundary of the virtual power plant and the overall available resource range, uses historical operation data and meteorological data to analyze the coupling relationship between environmental factors and distributed resource output, constructs a distributed resource power prediction model based on the Informer-network network, and predicts the overall resources of the virtual power plant.

Images