A multi-objective dynamic switching optimization scheduling method and system based on virtual power plant

Abstract

The present application discloses a multi-objective dynamic switching optimization scheduling method and system based on a virtual power plant, wherein the method includes obtaining real-time status data of a target virtual power plant; analyzing the real-time status data through a multi-model integration method based on a big tree mechanism, and predicting the first prediction data of the target virtual power plant; adaptively correcting the first prediction data according to an improved Kalman filter algorithm to obtain second prediction data; based on multiple scheduling requirements of the target virtual power plant, respectively constructing a scheduling objective function corresponding to each scheduling requirement, and constructing a multi-objective dynamic switching scheduling database according to the real-time status data and each scheduling objective function; screening in the multi-objective dynamic switching scheduling database based on the real-time status data and the second prediction data, and scheduling the target virtual power plant with the target scheduling scheme obtained after screening. The method provided by the present application improves the scheduling and operation efficiency of virtual power plants.

Description

Technical Field

[0001] The present application relates to the field of intelligent scheduling of virtual power plants, and in particular to a multi-objective dynamic switching optimization scheduling method and system based on virtual power plants. Background Art

[0002] With the transformation of the global energy structure and the rapid development of renewable energy, virtual power plants (VPPs) as an emerging form of energy management have gradually attracted widespread attention. Virtual power plants integrate distributed energy (such as wind power, photovoltaics), energy storage systems and adjustable loads to form a unified, flexible and adjustable energy supply system to cope with the volatility and uncertainty in the power system.

[0003] However, existing virtual power plant dispatching methods still face many problems when dealing with complex and changing power grid environments. Most of the current dispatching methods are based on fixed rules and single objectives, and cannot balance the multiple demands of power system operation, especially in terms of real-time response, flexible switching of dispatching objectives, and multi-objective optimization. They are difficult to adapt to the ever-changing market demand and power grid status, resulting in limited comprehensive optimization capabilities of virtual power plants. Summary of the invention

[0004] The present application aims to provide a multi-objective dynamic switching optimization scheduling method and system based on a virtual power plant, so as to improve the scheduling and operation efficiency of the virtual power plant.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a multi-objective dynamic switching optimization scheduling method based on a virtual power plant, including:

[0006] Based on the constructed wind turbine output model, photovoltaic unit output model, energy storage model and adjustable load model of the target virtual power plant, real-time status data of the target virtual power plant is obtained;

[0007] By using a multi-model integration method based on a big tree mechanism, the real-time status data is analyzed to predict and obtain first prediction data of the target virtual power plant;

[0008] Adaptively correcting the first prediction data according to an improved Kalman filter algorithm to obtain second prediction data, wherein the adaptive correction is designed to reduce the impact of uncertainty caused by energy fluctuations through a multi-layer state transition mechanism;

[0009] Based on multiple scheduling requirements of the target virtual power plant, respectively construct a scheduling objective function corresponding to each scheduling requirement, and construct a multi-objective dynamic switching scheduling database according to the real-time status data and each scheduling objective function;

[0010] The real-time status data and the second prediction data are screened in the multi-objective dynamic switching scheduling database, and the target virtual power plant is scheduled with the target scheduling scheme obtained after screening.

[0011] As one of the preferred solutions, the real-time status data includes internal status data and external status data, the internal status data includes wind power output data, photovoltaic output data, energy storage charge status data, and adjustable load status data, and the external status data includes grid power data, grid voltage data, and grid frequency data;

[0012] The real-time status data of the target virtual power plant is obtained based on the wind turbine output model, photovoltaic unit output model, energy storage model and adjustable load model of the target virtual power plant respectively constructed, including:

[0013] Analyzing the resource types in the target virtual power plant and the response characteristics corresponding to the resource types based on the wind turbine output model, the photovoltaic unit output model, the energy storage model and the adjustable load model respectively;

[0014] According to the analysis results, determining the real-time status data of the target virtual power plant that needs to be monitored;

[0015] The real-time status data is collected through multiple types of sensors and power grid detection equipment.

[0016] As one of the preferred solutions, the real-time status data is analyzed by a multi-model integration method based on a big tree mechanism, including:

[0017] Inputting the real-time status data and the acquired environmental data into a pre-built time series model for prediction to obtain a first prediction result;

[0018] Inputting the real-time status data and the environmental data into a pre-built Bayesian neural network model for prediction to obtain a second prediction result;

[0019] Inputting the real-time status data and the environmental data into a pre-built support vector machine model for prediction to obtain a third prediction result;

[0020] The first prediction result, the second prediction result and the third prediction result are weightedly fused through a multi-model integration method based on a big tree mechanism to obtain first prediction data.

[0021] As one of the preferred solutions, the step of adaptively correcting the first prediction data according to the improved Kalman filter algorithm to obtain the second prediction data includes:

[0022] Initialize the parameters of the Kalman filter algorithm, the parameters including at least a state transfer matrix, an observation matrix and a covariance matrix;

[0023] Perform multi-level prediction on the first prediction data according to the multi-level state transfer mechanism and the state transfer matrix to obtain a first intermediate result;

[0024] Based on a multi-scale covariance adjustment method, adjusting the covariance matrix;

[0025] According to the adjusted covariance matrix and the observation matrix, the first intermediate result is dynamically adaptively smoothed to obtain second prediction data.

[0026] As one of the preferred solutions, the multi-level prediction of the first prediction data is performed according to the multi-layer state transfer mechanism and the state transfer matrix, which is expressed as:

[0027]

[0028]

[0029]

[0030] in, is the predicted power at time k, is the state transfer matrix after adaptive adjustment, For external control input, is a multi-layer adjustment item, L represents the number of adjustment layers, is the weight coefficient of each layer, To control the adaptive adjustment term of noise, is the observed value at the current time k, is the observation matrix, Indicates the first layer of multiple adjustment items or feedback layers. layer.

[0031] As one of the preferred solutions, the covariance matrix is ​​adjusted based on the multi-scale covariance adjustment method, which is expressed as:

[0032]

[0033]

[0034] in, represents the number of multi-scale layers, is the multi-scale feedback coefficient, is the process noise covariance, Indicates time The prediction covariance matrix of is the state transition matrix, Representation Matrix The transposed matrix of Indicates the scale or level of the current layer. represents the index representation of the accumulated term in the denominator, Indicates time The variance of the observation noise.

[0035] As one of the preferred solutions, the first intermediate result is dynamically adaptively smoothed according to the adjusted covariance matrix and the observation matrix to obtain second prediction data, which is expressed as:

[0036]

[0037]

[0038] in, Indicates time The prediction covariance matrix of represents the Kalman gain, is the observation noise covariance matrix, is the observation matrix, is the second prediction data.

[0039] As one of the preferred solutions, the dispatching requirements include grid stability, maximum utilization of renewable energy, maximum load flexibility, maximum safety margin and maximum user comfort;

[0040] The step of constructing a scheduling objective function corresponding to each of the scheduling requirements based on the multiple scheduling requirements of the target virtual power plant includes:

[0041] With the goal of minimizing grid frequency fluctuations and voltage deviations, the grid stability objective function is constructed by combining the constraints of grid voltage data and grid frequency data.

[0042] With the goal of maximizing the output of wind power and photovoltaic power generation, the objective function of maximizing renewable energy utilization is constructed by combining the constraints of wind power output data and photovoltaic output data;

[0043] Taking the maximum output of the adjustable load as the goal, combined with the constraints of the adjustable load state data, the load flexibility maximization objective function is constructed;

[0044] Taking the maximum redundant power of the target virtual power plant as the goal, combining the wind power output data, the photovoltaic output data and the energy storage charge state data, constructing a safety margin maximization objective function;

[0045] Taking the user's ideal room temperature and ideal lighting power as the goals, combined with the constraints of temperature, lighting power and adjustable load state data, the objective function of maximizing user comfort is constructed.

[0046] As one preferred solution, before screening in the multi-objective dynamic switching scheduling database based on the real-time status data and the second prediction data, the method further includes:

[0047] Construct a fusion evaluation model, expressed as:

[0048]

[0049] in, and are the weights of the real-time status data and the second prediction data, and satisfy =1; , They are real-time status data and second prediction data respectively;

[0050] Inputting the real-time status data and the second prediction data into the fusion evaluation model to obtain comprehensive evaluation data;

[0051] Screening is performed in the multi-objective dynamic switching scheduling database based on the comprehensive evaluation data.

[0052] Another embodiment of the present application provides a multi-objective dynamic switching optimization scheduling system based on a virtual power plant, including:

[0053] An acquisition module, used to acquire real-time status data of the target virtual power plant based on the wind turbine output model, photovoltaic unit output model, energy storage model and adjustable load model of the constructed target virtual power plant;

[0054] An analysis module, used to analyze the real-time status data through a multi-model integration method based on a big tree mechanism, and predict and obtain first prediction data of the target virtual power plant;

[0055] A correction module, configured to adaptively correct the first prediction data according to an improved Kalman filter algorithm to obtain second prediction data, wherein the adaptive correction is designed to reduce the impact of uncertainty caused by energy fluctuations through a multi-layer state transition mechanism;

[0056] A construction module, for constructing a scheduling objective function corresponding to each scheduling requirement based on multiple scheduling requirements of the target virtual power plant, and constructing a multi-objective dynamic switching scheduling database according to the real-time status data and each scheduling objective function;

[0057] A scheduling module is used to filter the multi-objective dynamic switching scheduling database based on the real-time status data and the second prediction data, and to schedule the target virtual power plant with the target scheduling plan obtained after screening.

[0058] Compared with the prior art, the beneficial effects of the embodiments of the present application are at least one of the following:

[0059] 1) By building a multi-objective dynamic switching scheduling database, flexible switching and optimization between different scheduling objectives can be achieved to meet the scheduling needs in different scenarios.

[0060] 2) Through precise prediction and adaptive correction, the real-time status and future trends of the virtual power plant can be more accurately grasped, so as to formulate more reasonable scheduling plans, reduce operating costs and improve economic benefits.

[0061] 3) By comprehensively considering multiple dispatch objectives and the uncertainty of energy fluctuations, a more robust dispatch plan can be formulated to enhance the stability of the virtual power plant and the entire power system.

[0062] 4) By optimizing scheduling, we can better coordinate the output of new energy sources such as wind power and photovoltaics, improve the utilization rate and consumption level of new energy, and help promote the transformation and upgrading of the energy structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of a multi-objective dynamic switching optimization scheduling method based on a virtual power plant in one embodiment of the present application;

[0064] Figure 2 It is a scheduling diagram based on a virtual power plant in one embodiment of the present application;

[0065] Figure 3 This is a schematic diagram of a multi-objective dynamic switching optimization scheduling system based on a virtual power plant in one of the embodiments of the present application. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0067] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0068] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used in this article are only for illustrative purposes, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on this application. The term "and / or" used in this article includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0069] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this application have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0070] An embodiment of the present application provides a multi-objective dynamic switching optimization scheduling method based on a virtual power plant. For details, see Figure 1 , Figure 1 The flowchart of the multi-objective dynamic switching optimization scheduling method based on a virtual power plant in one embodiment of the present application is shown, wherein it includes steps S1-S5:

[0071] S1: acquiring real-time status data of the target virtual power plant based on the wind turbine output model, photovoltaic unit output model, energy storage model and adjustable load model of the constructed target virtual power plant;

[0072] S2: Analyze the real-time status data through a multi-model integration method based on a big tree mechanism, and predict and obtain first prediction data of the target virtual power plant;

[0073] S3: Adaptively correcting the first prediction data according to an improved Kalman filter algorithm to obtain second prediction data, wherein the adaptive correction is designed to reduce the impact of uncertainty caused by energy fluctuations through a multi-layer state transition mechanism;

[0074] S4: Based on multiple scheduling requirements of the target virtual power plant, respectively construct a scheduling objective function corresponding to each scheduling requirement, and construct a multi-objective dynamic switching scheduling database according to the real-time status data and each scheduling objective function;

[0075] S5: Based on the real-time status data and the second prediction data, the multi-objective dynamic switching scheduling database is screened, and the target virtual power plant is scheduled with the target scheduling plan obtained after screening.

[0076] Existing virtual power plant scheduling methods usually have the following defects:

[0077] Insufficient real-time dispatch response: Traditional dispatch methods are difficult to respond quickly to real-time changes in the power grid. When faced with power grid load fluctuations and changes in market electricity prices, the dispatch system reacts slowly, affecting the overall system operating efficiency.

[0078] Lack of multi-objective optimization switching capabilities: Existing methods usually only target a single optimization objective (such as cost minimization or revenue maximization), ignoring the diversified needs of the power system. They are unable to flexibly switch scheduling objectives according to grid status and market dynamics, making it difficult to achieve a balance between operational stability, economic benefits, and environmental sustainability.

[0079] Limitations of optimization algorithms: Current scheduling methods mostly rely on rules and heuristic algorithms. When dealing with complex nonlinear problems and multiple constraints, the optimization capabilities are limited and it is difficult to find the global optimal solution, resulting in suboptimal scheduling results and low resource utilization.

[0080] Insufficient coordinated dispatching capabilities: The dispatching of wind power, photovoltaic and energy storage equipment failed to fully combine their respective characteristics for comprehensive optimization, resulting in insufficient utilization of resources and the flexibility and efficiency of the dispatching system failing to meet expectations.

[0081] To solve the above problems, this application proposes a virtual power plant optimization scheduling method with real-time multi-objective switching. Through precise environmental modeling and global optimization scheduling, the virtual power plant can achieve flexible switching between multiple objectives in a complex power system, ensure efficient coordinated scheduling of wind power, photovoltaic and energy storage equipment, and achieve more efficient resource utilization and grid stability.

[0082] Preferably, in one embodiment of the present application, the real-time status data includes internal status data and external status data, the internal status data includes wind power output data, photovoltaic output data, energy storage charge status data, and adjustable load status data, and the external status data includes grid power data, grid voltage data, and grid frequency data;

[0083] The real-time status data of the target virtual power plant is obtained based on the wind turbine output model, photovoltaic unit output model, energy storage model and adjustable load model of the target virtual power plant respectively constructed, including:

[0084] Analyzing the resource types in the target virtual power plant and the response characteristics corresponding to the resource types based on the wind turbine output model, the photovoltaic unit output model, the energy storage model and the adjustable load model respectively;

[0085] According to the analysis results, determining the real-time status data of the target virtual power plant that needs to be monitored;

[0086] The real-time status data is collected through multiple types of sensors and power grid detection equipment.

[0087] Specifically, the construction of the virtual power plant resource model includes:

[0088] 1) Wind power output equipment;

[0089] Virtual power plants exist Wind power output equipment, its total output power for:

[0090]

[0091] in, For the t period The output of distributed wind power units, =1,2,......, .

[0092] Wind power equipment output is controlled by the virtual power plant dispatch center

[0093]

[0094] in: It represents the maximum output power of all wind power output equipment in period t.

[0095] 2) Photovoltaic output equipment:

[0096] Assuming the existence of a virtual power plant The total output power of photovoltaic devices for:

[0097]

[0098] in, For the t period The output of distributed photovoltaic units, =1,2,......, .

[0099] The output of photovoltaic equipment is controlled by the virtual power plant dispatch center

[0100]

[0101] in: Indicates the maximum output power of all photovoltaic output devices in time period t.

[0102] 3) Energy storage equipment:

[0103] Assume that there are k energy storage units in the virtual power plant, and their total charging and discharging power is for:

[0104]

[0105] in, is the charging and discharging power of the kth energy storage unit in time period t, k=1,2,......,K.

[0106] Assume that there are k energy storage units in the virtual power plant, and their total state of charge is for:

[0107]

[0108]

[0109] in, is the charge state of the kth energy storage unit in time period t, k=1,2,......,K.

[0110] The following corresponding relationship exists between the charging and discharging power, capacity and state of charge of the energy storage unit:

[0111]

[0112]

[0113]

[0114] in, and are the maximum and minimum states of charge that can be executed; , is the maximum and minimum value of the ratio of charge and discharge energy to total energy storage capacity, , is the total energy storage capacity.

[0115] 4) Adjustable load:

[0116] The power of the adjustable load is controlled by the virtual power plant dispatching center. units, the total adjustable load power is:

[0117]

[0118] Its output is determined by the current load of the equipment and user demand, and is adjusted in real time by the dispatch center. The power constraint of the adjustable load is:

[0119] .

[0120] Next, the resource types in the target virtual power plant and the corresponding response characteristics of the resource types are analyzed. Specifically, the internal state of the virtual power plant is determined by the operation of its core components - wind power generation, photovoltaic power generation, energy storage units and adjustable loads. Through sensor and smart device monitoring, the system can obtain the following state parameters in real time:

[0121] 1) Wind power equipment status:

[0122] Output power: The actual output power of the wind turbine at the current moment .

[0123] 2) Photovoltaic power generation equipment status:

[0124] Output power: The actual output power of the photovoltaic generator at the current moment .

[0125] 3) Energy storage device status:

[0126] State of charge: The state of charge of the energy storage at the current moment .

[0127] 4) Adjustable load state:

[0128] Adjustable power: The output power that the adjustable load can adjust at the current moment .

[0129] By integrating these internal state information together, the state set of the virtual power plant is formed, which is represented by a multidimensional vector:

[0130] =

, ,

[0131] This state set is used in scheduling decisions to evaluate the resource availability and scheduling flexibility of the virtual power plant.

[0132] The state of the external power grid directly affects the dispatch strategy of the virtual power plant. The state of the external power grid has a direct impact on the dispatch strategy of the virtual power plant. By monitoring the operation of the power grid in real time, pay attention to the following three core indicators:

[0133] Grid power: To ensure the stable operation of the grid, the grid power must be kept within a reasonable range. Power sensors are usually used to obtain the current grid power data in real time. .

[0134] Grid voltage: The grid voltage level has an important impact on dispatch, especially when the load increases or decreases. The voltage level must be kept within the specified safety range, and voltage sensors are usually used to obtain the current grid voltage in real time. .

[0135] Grid frequency: Grid frequency reflects the supply and demand balance in the grid. Real-time frequency data is an important parameter for evaluating grid stability. The current grid frequency can be measured with the help of professional instruments. .

[0136] =【 】.

[0137] Specifically, by establishing a multi-level equipment monitoring and data processing architecture, the collection frequency of internal equipment in the virtual power plant can be dynamically adjusted to improve monitoring efficiency, achieve resource optimization and real-time response.

[0138] The device layer uses an adaptive collection frequency mechanism to dynamically adjust the collection frequency to improve monitoring efficiency.

[0139] 1) Wind power generation equipment:

[0140] Sensor installation: Install wind speed sensors, power sensors and health monitoring sensors at key locations of each wind turbine (such as generators and pitch systems).

[0141] Collection frequency control: Dynamically adjust the collection frequency according to the rate of change of wind speed and power output. When the wind speed changes greatly, the collection frequency automatically increases; when the wind speed is stable, the collection frequency is reduced.

[0142] Adaptive acquisition formula:

[0143]

[0144] in, is the adaptive acquisition frequency; is the basic acquisition frequency; is the adjustment coefficient; is the wind speed change, is the maximum wind speed.

[0145] 2) Photovoltaic power generation equipment:

[0146] Sensor installation: Solar radiation sensors, temperature sensors and power output sensors are installed on the photovoltaic panels.

[0147] Dynamic collection formula:

[0148] in, is the current light intensity; is the light intensity at the previous moment; is the adjustment coefficient; is the maximum light intensity.

[0149] Sensor types: optical radiation sensor, temperature sensor, electrical power meter.

[0150] 3) Energy storage equipment:

[0151] Sensor installation: Install SOC monitoring sensors and power sensors on each energy storage unit.

[0152] Collection frequency control: Based on the change of the state of charge (SOC) of the energy storage unit and the charge and discharge power rate, the collection frequency is flexibly adjusted. When the SOC fluctuates violently or the power changes greatly, the collection frequency is increased.

[0153]

[0154] in, is the adjustment coefficient; is the change in SOC.

[0155] Sensor type: SOC sensor, charge and discharge power meter.

[0156] 4) Adjustable load:

[0157] Sensor installation: Each adjustable load unit is installed with a load power sensor and a load response sensor.

[0158] Collection frequency control: Adjust the collection frequency according to the load response status and current load power. When the load fluctuates greatly, increase the collection frequency, and when the load is stable, reduce the frequency.

[0159]

[0160] in, is the adjustment coefficient; is the maximum value of the load.

[0161] Sensor type: power sensor, load response condition monitoring device.

[0162] 5) Voltage and frequency monitoring:

[0163] Installation location: Install voltage and frequency sensors at key points where the virtual power plant is connected to the grid.

[0164] Collection frequency control: dynamically adjust the collection frequency according to the fluctuation of voltage and frequency. When the power grid status fluctuates greatly, increase the collection frequency to maintain high real-time performance.

[0165]

[0166] in, is the adjustment coefficient; and are the changes of voltage and frequency respectively; and are the maximum voltage and frequency values ​​respectively.

[0167] Sensor type: voltage sensor, frequency sensor.

[0168] After data collection is completed, the collected status data is first preprocessed in the data processing layer to ensure the accuracy and stability of the data.

[0169] Preferably, in one embodiment of the present application, the real-time status data is analyzed by a multi-model integration method based on a big tree mechanism, including:

[0170] Inputting the real-time status data and the acquired environmental data into a pre-built time series model for prediction to obtain a first prediction result;

[0171] Inputting the real-time status data and the environmental data into a pre-built Bayesian neural network model for prediction to obtain a second prediction result;

[0172] Inputting the real-time status data and the environmental data into a pre-built support vector machine model for prediction to obtain a third prediction result;

[0173] The first prediction result, the second prediction result and the third prediction result are weightedly fused through a multi-model integration method based on a big tree mechanism to obtain first prediction data.

[0174] Specifically, the virtual power plant output prediction based on the improved Kalman filter uses multi-model integration and adaptive adjustment, and the prediction object is mainly the future output of various energy sources in the virtual power plant, reducing the uncertainty caused by energy fluctuations. This step uses the real-time feedback mechanism of the Kalman filter to accurately predict and adjust the output of each distributed energy source according to dynamic changes in electricity demand, weather, etc., to ensure that the system can still operate efficiently under the fluctuations of wind power, photovoltaics and energy storage equipment, and reduce the output uncertainty caused by the characteristics of new energy itself. In addition, by adaptively adjusting the noise covariance and handling nonlinear problems, the robustness and flexibility of virtual power plant scheduling are enhanced, providing an accurate basis for switching scheduling targets. The specific optimization steps are as follows:

[0175] Firstly, a multi-model integration method based on big tree mechanism is proposed to reduce the overall prediction error of the system by integrating multiple prediction models.

[0176] +

[0177]

[0178] in, It is the forecast result of the time series model, which mainly captures the time correlation of power output; It is the prediction result of the Bayesian neural network model, which is used to deal with nonlinear characteristics and is suitable for complex environments; It is the prediction result of the support vector machine. The input of the three prediction results is the internal and external information data of the monitoring. It has good generalization ability and is suitable for load demand prediction. , , are the weight coefficients of each model.

[0179] In order to further improve the weight optimization efficiency of the binary method, it can be improved by combining it with the roulette algorithm. By combining the binary method with the roulette method, the optimization process not only relies on the convergence speed of the binary method, but also can select the optimal solution according to the probability of the error size, increase the global search capability, and prevent the local optimal trap. Specifically,

[0180] First, the objective function is defined as the root mean square error of the prediction error, which represents the error between the predicted value and the true value. In each iteration, a roulette wheel selection mechanism is introduced to dynamically adjust the exploration direction of the dichotomy.

[0181]

[0182] in: For time The actual power output; is the total number of time steps.

[0183] In the binary optimization process, each group Each weight combination generates a corresponding prediction value and calculates the corresponding RMSE. This RMSE value is used as the basis of the roulette algorithm, by giving a higher selection probability to combinations with lower errors.

[0184] The roulette algorithm assigns the probability of selecting a weighted combination inversely proportional to the current error.

[0185] Use weighted combinations Calculate RMSE. The fitness value is the inverse of RMSE and is defined as:

[0186]

[0187] The search process begins by randomly generating multiple weight combinations, and normalizing the fitness values ​​of all current weight combinations to form a selection probability:

[0188]

[0189] in, For the The probability of selecting a weighted combination is is the number of candidate combinations in the current weight interval.

[0190] Through the roulette wheel selection algorithm, according to the probability A weight combination is randomly selected for the next optimization step, and a weight combination with high fitness has a higher probability of being selected.

[0191] Specifically, the weight optimization steps of the combination of adaptive dichotomy and roulette include:

[0192] (1) Initialize the search interval and weight combination:

[0193] The initial weight search interval , The range is set to , and guarantee .

[0194] (2) Calculate the midpoint using the binary method:

[0195] ,

[0196] ,

[0197] - .

[0198] (3) Calculate RMSE and update the weight combination:

[0199] Use Current The RMSE value of the integrated model is calculated by the combination of the current weight combination and the RMSE of other possible combinations is compared to generate a fitness value.

[0200] (4) Roulette wheel selection of optimal weights:

[0201] A roulette wheel selection mechanism is used to assign selection probabilities inversely proportional to the RMSE of each combination. The starting combination for the next iteration is selected by a random number.

[0202] (5) Dynamically adjust the search range:

[0203] Use binary search to narrow the weight search interval: If the selected combination , Provides a lower RMSE value, then the interval of this combination is used as the new optimization interval. Then update , , Right now:

[0204] , ,

[0205] , .

[0206] (6) Repeated iteration:

[0207] Repeat steps (2)-(5) until the convergence condition is met, that is, when the decrease in RMSE is less than a certain threshold Stop when.

[0208] Preferably, in one embodiment of the present application, the adaptively correcting the first prediction data according to the improved Kalman filter algorithm to obtain the second prediction data includes:

[0209] Initializing parameters of the Kalman filter algorithm, the parameters including at least a state transfer matrix, an observation matrix, and a covariance matrix;

[0210] Perform multi-level prediction on the first prediction data according to the multi-level state transfer mechanism and the state transfer matrix to obtain a first intermediate result;

[0211] Based on a multi-scale covariance adjustment method, adjusting the covariance matrix;

[0212] According to the adjusted covariance matrix and the observation matrix, the first intermediate result is dynamically adaptively smoothed to obtain second prediction data.

[0213] Specifically, based on the integration of multiple models, an improved Kalman filter algorithm is introduced to further correct the prediction results. In order to enhance the adaptive ability of the filter, a multi-layer state transfer mechanism is introduced. That is, in different time steps, the state transfer matrix is ​​dynamically adjusted according to the feedback, noise, and environmental factors of the system, so that the state prediction at each moment can change adaptively. The Kalman filter dynamically adjusts the output of the prediction model through real-time monitoring data and feedback of the prediction results, so that the dispatching system can adaptively adjust the power output of each distributed energy source to ensure the accuracy of real-time dispatching.

[0214] Among them, the state transfer matrix: defines the transfer relationship between the system state at different time steps, providing a basis for the prediction of future states. The observation matrix: maps the state space of the system to the observation space in order to calculate the error between the predicted state and the actual observed value. The covariance matrix: is used to describe the initial uncertainty of the system state and evaluate the initial conditions of the noise level and state accuracy.

[0215] In order to enhance the filter's adaptive capability, a multi-layer state transfer mechanism is introduced. That is, in different time steps, the state transfer matrix is ​​dynamically adjusted according to the system's feedback, noise, and environmental factors, so that the state prediction at each moment can change adaptively. This step improves the flexibility and adaptability of the prediction value under different environmental conditions.

[0216] Preferably, in one embodiment of the present application, the multi-level prediction of the first prediction data is performed according to the multi-layer state transfer mechanism and the state transfer matrix, which is expressed as:

[0217]

[0218]

[0219]

[0220] in, is the predicted power at time k; is the state transfer matrix after adaptive adjustment, depending on the previous moment The state estimation error and observation error are dynamically adjusted. , so that the system responds quickly when the state changes drastically, and reduces the adjustment force when the state is stable; For external control input; It is a multi-layer adjustment item, which performs multiple feedback calculations on the current observation error through nonlinear functions to generate adjustment values ​​at different levels; L represents the number of adjustment layers, which is set to 3; is the weight coefficient of each layer, which is adjusted according to the size and change rate of the error; To control the adaptive adjustment term for noise, the rate of change of the observation error is taken into account. is the observed value at the current time k, the actual measurement data obtained by sensors and other means; is the observation matrix, which maps the state space vector to the observation space vector for error calculation; Indicates the first layer of multiple adjustment items or feedback layers. Layers are used to perform weight calculations at different feedback levels. For example, For the first The feedback weight of a layer indicates the contribution of the error of that layer to the overall adjustment. j is an index used for traversal, which is used to accumulate the observed error values ​​of each layer to form the sum of the denominators.

[0221] In order to enhance the adaptability of covariance, a multi-scale covariance adjustment mechanism is introduced. The covariance matrix is ​​processed at multiple scales, each scale representing a different uncertainty in the system state. This hierarchical covariance update mechanism enables the system to cope with different degrees of noise and uncertainty at the same time, enhancing its ability to adapt to complex environments. This process further clarifies the uncertainty range of the system state at the current moment.

[0222] Preferably, in one embodiment of the present application, the covariance matrix is ​​adjusted based on the multi-scale covariance adjustment method, which is expressed as:

[0223]

[0224]

[0225] in, It indicates the number of multi-scale layers 3, each layer reflects the system state changes of different granularity; It is a multi-scale feedback coefficient, which is adaptively adjusted according to the state changes of each layer. By superimposing feedback information at different scales, the system can adaptively adjust the covariance under multiple dynamic feedbacks and enhance its response ability to uncertainty. is the process noise covariance, which is used to adjust the internal noise level of the system. Indicates time The prediction covariance matrix is ​​used to represent the uncertainty of the system state at the current moment; is the state transition matrix, Representation Matrix The transposed matrix of Indicates the scale or level of the current layer; Indicates the index of the term to be added in the denominator. Indicates time The variance of the observation noise.

[0226] The dynamic adaptive smoothing algorithm enhances the update step of the standard Kalman filter, and adjusts and corrects the observation and prediction errors in multiple dimensions through a multi-level dynamic feedback mechanism. The introduction of a multi-level feedback correction mechanism enables the smoothing process to respond more flexibly to system uncertainties and adaptively adjusts the degree of dependence on observation and prediction data, thereby achieving a more accurate final prediction result. The multi-level error smoothing adjustment improves the stability and accuracy of the final prediction result and meets the prediction requirements in different environments.

[0227] Preferably, in one embodiment of the present application, the first intermediate result is dynamically adaptively smoothed according to the adjusted covariance matrix and the observation matrix to obtain second prediction data, which is expressed as:

[0228]

[0229]

[0230] in, Indicates time The prediction covariance matrix of represents the Kalman gain, is the observation noise covariance matrix, is the observation matrix, is the second prediction data.

[0231] Finally, in the above formula, the Kalman gain K is calculated based on the adjustment result of the covariance Pk-1. The final prediction is the first intermediate result Based on the covariance The multi-scale adjustment results are updated. The adjustment directly affects the Kalman gain , which in turn affects the observation error weight in the smoothing step, making the system more dependent on the predicted value when the noise level is large.

[0232] Preferably, in one embodiment of the present application, the dispatching requirements include grid stability, maximum utilization of renewable energy, maximum load flexibility, maximum safety margin, and maximum user comfort;

[0233] The step of constructing a scheduling objective function corresponding to each of the scheduling requirements based on the multiple scheduling requirements of the target virtual power plant includes:

[0234] With the goal of minimizing grid frequency fluctuations and voltage deviations, the grid stability objective function is constructed by combining the constraints of grid voltage data and grid frequency data.

[0235] With the goal of maximizing the output of wind power and photovoltaic power generation, the objective function of maximizing renewable energy utilization is constructed by combining the constraints of wind power output data and photovoltaic output data;

[0236] Taking the maximum output of the adjustable load as the goal, combined with the constraints of the adjustable load state data, the load flexibility maximization objective function is constructed;

[0237] Taking the maximum redundant power of the target virtual power plant as the goal, combining the wind power output data, the photovoltaic output data and the energy storage charge state data, constructing a safety margin maximization objective function;

[0238] Taking the user's ideal room temperature and ideal lighting power as the goals, combined with the constraints of temperature, lighting power and adjustable load state data, the objective function of maximizing user comfort is constructed.

[0239] The following is the specific objective function:

[0240] (1) Grid stability objectives:

[0241] The stability of the power grid is mainly achieved by controlling voltage and frequency fluctuations:

[0242]

[0243] Constraints:

[0244] (Voltage Constraint)

[0245] (Frequency Constraint)

[0246] + + * = (Power balance constraint)

[0247] in is the grid stability objective function; is the current grid voltage; is the target voltage of the grid; The target frequency of the grid.

[0248] (2) Goal of maximizing the use of renewable energy:

[0249] To maximize the use of renewable energy, it is necessary to ensure that the output of wind power and photovoltaic power generation is maximized.

[0250]

[0251] in Maximize the objective function for renewable energy.

[0252] Constraints: Wind power generation equipment and photovoltaic power generation equipment only need to meet their own output constraints.

[0253] (3) Load flexibility maximization goal:

[0254] Adjust available load to reduce peak load pressure and improve system response flexibility.

[0255]

[0256] in is the load flexibility objective function; For the The output power of an adjustable load; A coefficient that indicates load adjustability.

[0257] Constraints: The load can be called to meet its own output constraints.

[0258] (4) Safety margin maximization goal:

[0259] 1) Resource redundancy power calculation:

[0260] The resource redundancy of a virtual power plant can be expressed as the sum of the redundancy on the power generation side and the redundancy on the energy storage side.

[0261]

[0262]

[0263] in It means taking a non-negative value, that is, when the power on the generating side is greater than the grid demand, the difference is taken; otherwise, 0 is taken.

[0264] 2) Safety margin calculation:

[0265] The safety margin can be calculated based on the redundant power and a certain safety threshold

[0266] S

[0267] in, is the set safety redundancy threshold.

[0268] 3) Optimization objectives:

[0269] Safety margin maximization goal It can be expressed as maximizing the safety margin, that is

[0270]

[0271] Constraint conditions: The output of power generation equipment and energy storage equipment must meet their own output constraints.

[0272] S

[0273] in is the safety margin threshold.

[0274] (5) User comfort maximization goal:

[0275] In order to meet the user's comfort requirements for temperature, lighting, etc., the present invention prioritizes adjusting the load to improve the user experience when the user's preferences are not met. The goal is achieved by adjusting the load power, which is specifically expressed as:

[0276]

[0277] in is the current room temperature, The ideal room temperature set by the user, is the current lighting power, Ideal lighting level set for the user.

[0278] Constraints:

[0279] (temperature constraint),

[0280] (Lighting Constraints),

[0281] (Load regulation constraints).

[0282] In order to more comprehensively and accurately capture the current state and future dynamics of the power grid, a comprehensive evaluation method is adopted, which is a weighted sum calculation that integrates real-time monitoring data and forecast data. This method effectively reduces the limitations caused by relying on a single data source by combining the immediate feedback of real-time monitoring with the forward-looking nature of forecast data, providing a more solid and reliable information foundation for dispatching decisions.

[0283] Preferably, in an embodiment of the present application, before screening in the multi-objective dynamic switching scheduling database based on the real-time status data and the second prediction data, the method further includes:

[0284] Construct a fusion evaluation model, expressed as:

[0285]

[0286] in, and are the weights of the real-time status data and the second prediction data, and satisfy =1; , They are real-time status data and second prediction data respectively;

[0287] Inputting the real-time status data and the second prediction data into the fusion evaluation model to obtain comprehensive evaluation data;

[0288] Screening is performed in the multi-objective dynamic switching scheduling database based on the comprehensive evaluation data.

[0289] After calculating the weighted sum of each monitoring variable, a comprehensive threshold is preset as the basis for switching the dispatch target. When the weighted sum of a monitoring variable exceeds this threshold, the corresponding dispatch target switching logic will be triggered. The design of the switching logic needs to comprehensively consider multiple factors such as grid stability, renewable energy utilization, load flexibility, resource redundancy, and user comfort. Figure 2 As shown, Figure 2 The diagram shows a scheduling diagram based on a virtual power plant provided in one embodiment of the present application. The specific target switching logic is described as follows.

[0290] 1) Grid stability priority trigger:

[0291] When the grid voltage or frequency exceeds the predetermined safety threshold: , immediately switching to the grid stability target and prioritizing power output.

[0292] 2) Renewable energy utilization priority trigger:

[0293] When the amount of electricity generated by renewable energy is high and the grid load is low, / and When the system is switched to the renewable energy source, it is prioritized to maximize the use of renewable energy and reduce the use of conventional energy.

[0294] 3) Load flexibility priority trigger:

[0295] When the grid load pressure is too large and the energy storage unit power is low: and , switch to the goal of maximizing load flexibility and reduce load pressure by adjusting the available load.

[0296] 4) Resource redundancy priority triggering:

[0297] If the weather forecast shows that wind and solar power generation fluctuates greatly and the grid load is high, that is, the wind speed / light forecast shows a large fluctuation range and the current grid load When approaching the critical value, the resource redundancy strategy is initiated in advance to ensure that the energy storage equipment has sufficient buffering capacity when power generation fluctuates.

[0298] 5) User comfort priority trigger:

[0299]

[0300] When the system detects that the comfort conditions set by the user (such as temperature, lighting, etc.) are not met: or The system automatically switches to the user comfort target and prioritizes load adjustment to ensure user experience.

[0301] In the scheduling scheme of the virtual power plant, each state node corresponds to an objective function and related decision-making actions. During the scheduling process, different goals will have different power scheduling requirements, and the implementation of these actions is driven by monitoring data. When the system monitors different system states, it will lock the scheduling target and execute the corresponding scheduling action based on the preset multi-objective dynamic switching scheduling database. The details are as follows.

[0302] (1) Grid stability objectives.

[0303] 1) Scheduling actions:

[0304] Wind power dispatching: Increase wind power output to the maximum available output to help ease grid load. If the load exceeds 10%, wind power output is increased to 90% of the maximum available power.

[0305]

[0306] in To dispatch wind power, The maximum available power of the wind turbine.

[0307] Photovoltaic dispatch capacity: Photovoltaic output increased to 90% of the maximum available power.

[0308]

[0309] in To dispatch wind power, is the maximum available power of the optoelectronic device.

[0310] Energy storage dispatch capacity: If the energy storage charge state is sufficient, release the energy storage power to reduce the grid load. The discharge power is 10-20% to cope with additional load.

[0311]

[0312] 2) Jump condition: Normal load -> Switch to maximize renewable energy, helping to reduce carbon emissions and save costs.

[0313] (2) The goal of maximizing load flexibility.

[0314] 1) Scheduling actions:

[0315] Wind power dispatching: Maintain wind power output, partially use it to support loads and reduce reliance on energy storage.

[0316] .

[0317] Photovoltaic dispatching capacity: maximize photovoltaic output, part of which is used for load regulation, and 30% is used to alleviate grid load.

[0318] .

[0319] Available load dispatch: Cut some available loads to ease grid load and reduce energy storage discharge requirements.

[0320] .

[0321] in is the percentage of load reduction.

[0322] Jump condition: If the energy storage is fully charged -> switch to maximizing renewable energy and continue to promote the development of green electricity.

[0323] (3) Renewable energy maximization goal.

[0324] 1) Scheduling actions:

[0325] Wind power dispatching: maximize wind power to meet grid load.

[0326] .

[0327] Photovoltaic dispatching capacity: maintain the photovoltaic output to the maximum value.

[0328] .

[0329] Energy storage dispatching capacity: Excess wind and solar power is charged into the energy storage system.

[0330] .

[0331] 2) Jump condition: If the load increases or the energy storage decreases -> switch to maximize load flexibility. If the load pressure increases or the energy storage decreases, the system needs to adjust to the load flexibility target to prevent excessive load pressure on the grid.

[0332] (4) User comfort priority trigger.

[0333] 1) Scheduling actions:

[0334] Available load scheduling: Increase or decrease the scheduling of air conditioning, lighting and other loads according to demand.

[0335] .

[0336] Energy storage dispatching capacity: If the grid load is high, discharge is carried out through energy storage equipment to meet user load requirements.

[0337] .

[0338] 2) Jump condition: If the user's comfort level meets the standard --> switch to maximizing renewable energy. Once the user's comfort needs are met, the system will return to maximizing renewable energy utilization to continue to promote energy optimization.

[0339] (5) Maximize resource redundancy and safety margin.

[0340] 1) Scheduling actions:

[0341] Energy storage dispatch: Maintain the charging level of the energy storage system to at least 50% of the total capacity.

[0342] .

[0343] Wind power dispatch: Reduce wind power by 20% to reserve room for fluctuations and increase energy storage charging.

[0344] .

[0345] 2) Jump condition: If the resource redundancy is sufficient -> switch to maximizing renewable energy. When the resource redundancy reaches the predetermined safety margin, the system can switch back to maximizing renewable energy and continue to give priority to wind and solar power.

[0346] The initial dispatch action is based on real-time monitored grid voltage, frequency and load data, as well as the forecast model's estimate of future power demand and supply. The initial dispatch can be expressed as follows:

[0347] -

[0348] The initial dispatching power output is used to meet the grid load demand and ensure the maximum utilization of wind and solar power.

[0349] After the dispatch is executed, the system continuously monitors the key parameters of the power grid through sensors, mainly including voltage, frequency and load power. The monitoring formula is as follows:

[0350] , , ,

[0351] in The current grid voltage monitored; The current grid frequency monitored; is the monitored current load power; is the monitored energy storage charge state.

[0352] After each dispatch action is executed, the system will obtain real-time feedback data and use this data to adjust power changes to ensure that various dispatch goals are continuously met. This mechanism uses feedback equations to adjust and correct the system output:

[0353]

[0354] in is the dispatching power after feedback, indicating the output power of the system after adjustment; is the feedback factor, which determines the adjustment strength. If the monitored voltage or frequency deviates from the target value, the system will adjust the power generation or load through feedback adjustment to reduce the deviation.

[0355] Evaluate the difference between the power output after feedback adjustment and the preset target to determine whether the system has reached the optimal state. Less than a preset tolerance ,Right now The system is considered to have reached the target state. The target confirmation formula is as follows:

[0356] + +

[0357] in, is the gap from the target state; is the target power of the load;

[0358] Jump logic: If the scheduling has reached the target state, it will jump according to the preset jump logic. Otherwise, the scheduling action will continue until the target state is met.

[0359] Compared with the prior art, the beneficial effects of the embodiments of the present application are at least one of the following:

[0360] 1) Based on the wind turbine output model, photovoltaic unit output model, energy storage model and adjustable load model of the target virtual power plant, the real-time status data of the target virtual power plant is obtained. These models can reflect the real-time operating status of various energy sources and loads in the virtual power plant, which is the basis for subsequent analysis and scheduling.

[0361] 2) Through the multi-model integration method based on the big tree mechanism, the real-time status data is analyzed to predict the first prediction data of the target virtual power plant. The multi-model integration method of the big tree mechanism can combine the advantages of multiple models and improve the accuracy and robustness of the prediction.

[0362] 3) Adaptively modify the first prediction data according to the improved Kalman filter algorithm to obtain the second prediction data. The adaptive modification is designed to reduce the impact of uncertainty caused by energy fluctuations through a multi-layer state transition mechanism. The Kalman filter algorithm can dynamically adjust the prediction results according to the new observation data to improve the accuracy of the prediction.

[0363] 4) Based on the multiple dispatch requirements of the target virtual power plant, the dispatch objective functions corresponding to each dispatch requirement are constructed. These dispatch objective functions may include multiple aspects such as economy, reliability, and environmental protection; based on the real-time status data and each dispatch objective function, a multi-objective dynamic switching dispatch database is constructed. This database can store various possible dispatch schemes and their corresponding benefits, providing a basis for subsequent screening.

[0364] 5) Based on the real-time status data and the second prediction data, the multi-objective dynamic switching scheduling database is screened, and the target virtual power plant is dispatched with the target scheduling scheme obtained after screening. This screening process can comprehensively consider multiple scheduling objectives and realize dynamic switching and optimization between multiple objectives.

[0365] Another embodiment of the present application provides a multi-objective dynamic switching optimization scheduling system based on a virtual power plant. For details, please refer to Figure 3 , Figure 3 The figure shows a schematic diagram of a multi-objective dynamic switching optimization scheduling system based on a virtual power plant provided in one embodiment of the present application, which includes an acquisition module 11, an analysis module 12, a correction module 13, a construction module 14, and a scheduling module 15, wherein:

[0366] An acquisition module 11 is used to acquire real-time status data of the target virtual power plant based on the wind turbine output model, photovoltaic unit output model, energy storage model and adjustable load model of the constructed target virtual power plant;

[0367] An analysis module 12 is used to analyze the real-time status data through a multi-model integration method based on a big tree mechanism, and predict and obtain first prediction data of the target virtual power plant;

[0368] A correction module 13, configured to adaptively correct the first prediction data according to an improved Kalman filter algorithm to obtain second prediction data, wherein the adaptive correction is designed to reduce the impact of uncertainty caused by energy fluctuations through a multi-layer state transition mechanism;

[0369] A construction module 14 is used to construct a scheduling objective function corresponding to each scheduling requirement based on multiple scheduling requirements of the target virtual power plant, and to construct a multi-objective dynamic switching scheduling database according to the real-time status data and each scheduling objective function;

[0370] The scheduling module 15 is used to filter the multi-objective dynamic switching scheduling database based on the real-time status data and the second prediction data, and schedule the target virtual power plant with the target scheduling plan obtained after screening.

[0371] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A multi-objective dynamic switching optimization scheduling method based on virtual power plant, characterized in that: include: Based on the constructed wind turbine output model, photovoltaic unit output model, energy storage model and adjustable load model of the target virtual power plant, real-time status data of the target virtual power plant is obtained; By using a multi-model integration method based on a big tree mechanism, the real-time status data is analyzed to predict and obtain first prediction data of the target virtual power plant; The real-time status data is analyzed by a multi-model integration method based on a large tree mechanism, including: Inputting the real-time status data and the acquired environmental data into a pre-built time series model for prediction to obtain a first prediction result; Inputting the real-time status data and the environmental data into a pre-built Bayesian neural network model for prediction to obtain a second prediction result; Inputting the real-time status data and the environmental data into a pre-built support vector machine model for prediction to obtain a third prediction result; The first prediction result, the second prediction result and the third prediction result are weightedly integrated by a multi-model integration method based on a big tree mechanism to obtain first prediction data; Adaptively correcting the first prediction data according to an improved Kalman filter algorithm to obtain second prediction data, wherein the adaptive correction is designed to reduce the impact of energy fluctuations through a multi-layer state transition mechanism; Adaptively correcting the first prediction data according to the improved Kalman filter algorithm to obtain the second prediction data includes: Initializing parameters of the Kalman filter algorithm, the parameters including at least a state transfer matrix, an observation matrix, and a covariance matrix; Perform multi-level prediction on the first prediction data according to the multi-level state transfer mechanism and the state transfer matrix to obtain a first intermediate result; Based on a multi-scale covariance adjustment method, adjusting the covariance matrix; According to the adjusted covariance matrix and the observation matrix, dynamically adaptively smoothing the first intermediate result to obtain second prediction data; Based on multiple scheduling requirements of the target virtual power plant, respectively construct a scheduling objective function corresponding to each scheduling requirement, and construct a multi-objective dynamic switching scheduling database according to the real-time status data and each scheduling objective function; The real-time status data and the second prediction data are screened in the multi-objective dynamic switching scheduling database, and the target virtual power plant is scheduled with the target scheduling scheme obtained after screening.

2. The multi-objective dynamic switching optimization scheduling method based on virtual power plant according to claim 1 is characterized in that: The real-time status data includes internal status data and external status data. The internal status data includes wind power output data, photovoltaic output data, energy storage charge status data, and adjustable load status data. The external status data includes grid power data, grid voltage data, and grid frequency data. The real-time status data of the target virtual power plant is obtained based on the wind turbine output model, photovoltaic unit output model, energy storage model and adjustable load model of the target virtual power plant respectively constructed, including: Analyzing the resource types in the target virtual power plant and the response characteristics corresponding to the resource types based on the wind turbine output model, the photovoltaic unit output model, the energy storage model and the adjustable load model respectively; According to the analysis results, determining the real-time status data of the target virtual power plant that needs to be monitored; The real-time status data is collected through multiple types of sensors and power grid detection equipment.

3. The multi-objective dynamic switching optimization scheduling method based on virtual power plant according to claim 1 is characterized in that: The multi-level prediction of the first prediction data according to the multi-level state transfer mechanism and the state transfer matrix is ​​expressed as: in, is the predicted power at time k, is the state transfer matrix after adaptive adjustment, For external control input, is a multi-layer adjustment item, L represents the number of adjustment layers, is the weight coefficient of each layer, To control the adaptive adjustment term of noise, is the observed value at the current time k, is the observation matrix, Indicates the first layer of multiple adjustment items or feedback layers. layer.

4. The multi-objective dynamic switching optimization scheduling method based on virtual power plant according to claim 1 is characterized in that: The multi-scale covariance adjustment method is based on which the covariance matrix is ​​adjusted, which is expressed as: in, represents the number of multi-scale layers, is the multi-scale feedback coefficient, is the process noise covariance, Indicates time The prediction covariance matrix of is the state transition matrix, Representation Matrix The transposed matrix of Indicates the scale or level of the current layer. represents the index representation of the accumulated term in the denominator, Indicates time The variance of the observation noise.

5. The multi-objective dynamic switching optimization scheduling method based on virtual power plant according to claim 1 is characterized in that: According to the adjusted covariance matrix and the observation matrix, the first intermediate result is dynamically adaptively smoothed to obtain second prediction data, which is expressed as: in, Indicates time The prediction covariance matrix of represents the Kalman gain, is the observation noise covariance matrix, is the observation matrix, is the second prediction data.

6. The multi-objective dynamic switching optimization scheduling method based on virtual power plant according to claim 1 is characterized in that: The dispatch requirements include grid stability, maximum utilization of renewable energy, maximum load flexibility, maximum safety margin, and maximum user comfort; The step of constructing a scheduling objective function corresponding to each of the scheduling requirements based on the multiple scheduling requirements of the target virtual power plant includes: With the goal of minimizing grid frequency fluctuations and voltage deviations, the grid stability objective function is constructed by combining the constraints of grid voltage data and grid frequency data. With the goal of maximizing the output of wind power and photovoltaic power generation, the objective function of maximizing renewable energy utilization is constructed by combining the constraints of wind power output data and photovoltaic output data; Taking the maximum output of the adjustable load as the goal, combined with the constraints of the adjustable load state data, the load flexibility maximization objective function is constructed; Taking the maximum redundant power of the target virtual power plant as the goal, combining the wind power output data, the photovoltaic output data and the energy storage charge state data, constructing a safety margin maximization objective function; Taking the user's ideal room temperature and ideal lighting power as the goals, combined with the constraints of temperature, lighting power and adjustable load state data, the objective function of maximizing user comfort is constructed.

7. The multi-objective dynamic switching optimization scheduling method based on virtual power plant according to claim 1 is characterized in that: Before the screening in the multi-objective dynamic switching scheduling database based on the real-time status data and the second prediction data, the method further includes: Construct a fusion evaluation model, expressed as: in, and are the weights of the real-time status data and the second prediction data, and satisfy =1; , They are real-time status data and second prediction data respectively; Inputting the real-time status data and the second prediction data into the fusion evaluation model to obtain comprehensive evaluation data; Screening is performed in the multi-objective dynamic switching scheduling database based on the comprehensive evaluation data.

8. A multi-objective dynamic switching optimization scheduling system based on virtual power plant, characterized in that: include: An acquisition module, used to acquire real-time status data of the target virtual power plant based on the wind turbine output model, photovoltaic unit output model, energy storage model and adjustable load model of the constructed target virtual power plant; An analysis module, used to analyze the real-time status data through a multi-model integration method based on a big tree mechanism, and predict and obtain first prediction data of the target virtual power plant; The real-time status data is analyzed by a multi-model integration method based on a large tree mechanism, including: Inputting the real-time status data and the acquired environmental data into a pre-built time series model for prediction to obtain a first prediction result; Inputting the real-time status data and the environmental data into a pre-built Bayesian neural network model for prediction to obtain a second prediction result; Inputting the real-time status data and the environmental data into a pre-built support vector machine model for prediction to obtain a third prediction result; The first prediction result, the second prediction result and the third prediction result are weightedly integrated by a multi-model integration method based on a big tree mechanism to obtain first prediction data; A correction module is used to adaptively correct the first prediction data according to an improved Kalman filter algorithm to obtain second prediction data, wherein the adaptive correction is designed to reduce the impact caused by energy fluctuations through a multi-layer state transition mechanism; the adaptive correction of the first prediction data according to the improved Kalman filter algorithm to obtain the second prediction data includes: Initializing parameters of the Kalman filter algorithm, the parameters including at least a state transfer matrix, an observation matrix, and a covariance matrix; Perform multi-level prediction on the first prediction data according to the multi-level state transfer mechanism and the state transfer matrix to obtain a first intermediate result; Based on a multi-scale covariance adjustment method, adjusting the covariance matrix; According to the adjusted covariance matrix and the observation matrix, dynamically adaptively smoothing the first intermediate result to obtain second prediction data; A construction module, for constructing a scheduling objective function corresponding to each scheduling requirement based on multiple scheduling requirements of the target virtual power plant, and constructing a multi-objective dynamic switching scheduling database according to the real-time status data and each scheduling objective function; A scheduling module is used to filter the multi-objective dynamic switching scheduling database based on the real-time status data and the second prediction data, and to schedule the target virtual power plant with the target scheduling plan obtained after screening.

Images