Electric power system intelligent scheduling method considering new energy output uncertainty

Abstract

The invention relates to a power system intelligent scheduling method considering new energy output uncertainty. The method comprises the following steps: (1) data acquisition and processing; (2) predicting new energy output by using the composite prediction model; (3) calculating an objective function value by utilizing a multi-objective optimization model and combining constraint conditions; and (4) generating a scheduling strategy. The method is high in prediction accuracy, adapts to various scenes, and realizes efficient configuration of complex power system resources.

Description

Technical Field

[0001] The invention relates to an intelligent dispatching method for a power system taking into account the uncertainty of output of new energy sources. Background Art

[0002] In the context of global energy transformation, the large-scale access of new energy (such as wind power and solar energy) has brought profound changes to the operation mode of the power system. New energy is clean and renewable, but its randomness, intermittency and volatility significantly increase the complexity of power grid operation and put forward higher requirements for traditional dispatching strategies.

[0003] In recent years, with the expansion of power grid scale and improvement of power market system, the number of dispatch objects and users has exploded, and intelligent dispatch technology has been widely studied. However, existing research and technical methods have the following shortcomings: poor applicability, limited ability to deal with the uncertainty of new energy processing, and complex calculations. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent dispatching method for an electric power system which has high prediction accuracy and strong adaptability and takes into account the uncertainty of output of new energy sources.

[0005] The present invention adopts the following technical solution:

[0006] An intelligent dispatching method for a power system considering the uncertainty of output of new energy sources comprises the following steps:

[0007] (1) Data collection and processing;

[0008] (2) Use a composite prediction model to predict the output of new energy;

[0009] (3) Calculate the objective function value using a multi-objective optimization model combined with constraints;

[0010] (4) Generate a scheduling strategy.

[0011] Furthermore, the data described in step (1) includes: historical output data, meteorological data, load demand data and system operating parameters.

[0012] Furthermore, in step (1), the output history data includes: actual wind power generation per hour, actual photovoltaic power generation per hour, actual thermal power generation per hour, and actual charge and discharge power of energy storage equipment per hour.

[0013] Furthermore, meteorological data include: temperature, air pressure, humidity, wind speed, and wind direction.

[0014] Furthermore, the system operating parameters include: maximum wind power output, maximum photovoltaic output, minimum and maximum thermal power output, minimum and maximum output of energy storage batteries, thermal power start and stop costs, thermal power generation costs, and penalties for wind and solar power abandonment.

[0015] Furthermore, in step (1), the high-frequency noise is removed by smoothing and averaging the hourly wind power and photovoltaic meteorological data; XGBoost is used to reduce the dimension of the data and extract the time features and meteorological features.

[0016] Furthermore, in step (2), the composite prediction model is as follows:

[0017] Y′=p ARIMA ·Y ARIMA +p MLP ·Y MLP

[0018] Among them, Y′ represents the composite prediction result; p ARIMA Indicates the probability of selecting the ARIMA model for prediction; Y ARIMA represents the new energy output value of the ARIMA model prediction period T; p MLP Indicates the probability of selecting the MLP model for prediction; Y MLP Represents the new energy output value of the MLP prediction model in the prediction period T.

[0019] Furthermore, the probability of selecting the ARIMA model or the MLP model for prediction is calculated by the following formula:

[0020]

[0021] Among them, p i It represents the probability of selecting ARIMA model or MLP model for prediction, and RMSE represents the root mean square error.

[0022] Furthermore, the root mean square error RMSE is calculated by the following formula:

[0023]

[0024] Among them, y i is the predicted output value, y i is the actual output value, and N is the total number of samples.

[0025] Furthermore, in step (3), the objective function of the multi-objective optimization model is:

[0026]

[0027] Among them, T is a scheduling cycle; S u ,P u , Respectively represent the start / stop status of the unit, unit output and the amount of wind and solar power abandoned; W t (·), C t (·) and θ represent the unit start / stop cost, unit output cost and wind and solar power curtailment penalty factor in period t, respectively.

[0028] Furthermore, in step (3), the constraints are:

[0029]

[0030] Where t is the index period; P t fire For thermal power unit i fire Output; P t wind For wind turbine i wind Output; P t solar Photoelectric motor group i solar Output; P t dis For battery pack dis Output; d t , are the power demand of the load and the battery unit i dis power storage needs.

[0031] Furthermore, the objective function value of the multi-objective optimization model is calculated by using the improved particle swarm algorithm.

[0032] The beneficial effects of the present invention are:

[0033] (1) Improve the new energy consumption capacity and prediction accuracy. The present invention uses a composite prediction model (a combination of the ARIMA prediction model and the MLP prediction model) to effectively deal with the randomness and intermittency of new energy output and achieve high-precision new energy output prediction. The results of accurate prediction enable power grid dispatching to more efficiently utilize new energy power generation resources, reduce wind and solar power abandonment, and improve the new energy consumption rate.

[0034] (2) Optimize the operating costs and environmental benefits of the power system.

[0035] The multi-objective optimization model of the present invention comprehensively considers economy (minimizing operating costs), environmental protection (minimizing carbon emissions) and system safety (load balance). The optimized scheduling scheme significantly reduces system operating costs while ensuring stable system operation, and performs well in meeting carbon emission reduction targets.

[0036] (3) Enhance the real-time and robustness of the system.

[0037] The improved particle swarm algorithm improves the global search capability and convergence speed of the algorithm through dynamic parameter adjustment and hybrid optimization strategy, so that the scheduling scheme can quickly respond to changes in load demand and fluctuations in new energy sources, ensuring the real-time performance of the system. Even in extreme fluctuation scenarios, the scheduling system still has high robustness.

[0038] (4) Improve system adaptability and multi-scenario adaptation capabilities.

[0039] The dispatching method proposed in this invention is not only applicable to a single power system, but can also be extended to multiple scenarios such as wind-solar-thermal-storage systems and microgrids, and has strong adaptability and versatility. Its modular design also facilitates deployment in different power grid architectures.

[0040] (5) Efficient computing and easy deployment.

[0041] The present invention optimizes the computational complexity of the algorithm and reduces the time required to solve large-scale dispatch problems. Combining the real-time simulation module and the distributed computing framework ensures that the dispatch strategy can be quickly deployed and applied in the actual power grid, thus improving the efficiency and reliability of the overall dispatch system. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is the architecture diagram of the composite prediction model.

[0043] Figure 2 These are the autocorrelation diagram and partial autocorrelation diagram of wind power samples.

[0044] Figure 3 Autocorrelation diagram and partial autocorrelation diagram of photoelectric samples.

[0045] Figure 4 This is the prediction result of the wind power ARIMA model.

[0046] Figure 5 This is the prediction result of the optoelectronic ARIMA model.

[0047] Figure 6 is the importance of each feature of wind power samples.

[0048] Figure 7 is the importance of each feature of the photoelectric sample.

[0049] Figure 8 This is the prediction result of the wind power MLP model.

[0050] Fig. 9 is the prediction result of the optoelectronic MLP model.

[0051] Fig.10 It is the actual value and predicted value of photovoltaic power generation output and wind power generation output.

[0052] Fig.11 This is a graph showing how generator output and load change over time.

[0053] Fig.12 This is a graph showing the change in the output ratio of each energy source over time. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is described in detail below using embodiments and drawings.

[0055] Example 1

[0056] 1. Data Collection and Processing

[0057] Collect historical data on generator output, meteorological data, load demand data, and system operating parameters. Preprocess the data, including denoising and feature extraction.

[0058] Among them, the historical output data includes: wind power: the actual power generation of wind power per hour; photovoltaic: the actual power generation of photovoltaic per hour; thermal power: the actual power generation of thermal power per hour; energy storage: the actual charging and discharging power of energy storage equipment per hour.

[0059] Among them, meteorological data include: temperature: temperature data recorded by the meteorological station; air pressure: atmospheric pressure recorded by the meteorological station; humidity: relative humidity of the air; wind speed: wind speed per unit time; wind direction: wind direction; load demand data: hourly electricity demand data of the target area.

[0060] Among them, the system operation parameters include: generator set parameters and cost parameters.

[0061] Generator set parameters: Maximum wind power output: maximum power of wind turbine set; Maximum photovoltaic output: maximum power of photovoltaic power station; Minimum and maximum thermal power output: minimum and maximum power generation of thermal power set; Minimum and maximum output of energy storage battery: charging and discharging power range of energy storage equipment.

[0062] Cost parameters: Thermal power start-up and shutdown costs: the costs of starting and stopping thermal power units; Thermal power generation cost (yuan / kWh): the power generation cost of thermal power units; Penalty for wind and solar power abandonment (yuan / kWh): the penalty cost when wind energy and photovoltaic energy are not used.

[0063] By smoothing and averaging the hourly wind power and photovoltaic meteorological data, high-frequency noise is removed. XGBoost is used to reduce the dimension of the data and extract time features, including hours, seasons, etc., and meteorological features, such as temperature, humidity, wind speed, etc., which are related to power generation.

[0064] 2. Using the composite prediction model to predict the output of new energy

[0065] The composite prediction model architecture is as follows Figure 1 shown.

[0066] (1) Using the ARIMA model to predict the output of renewable energy in period T

[0067] The ARIMA model is a prediction method based on time series data analysis. Its basic idea is to use the historical information of the data itself to predict the future. It is mainly composed of three parts, namely the autoregressive model (AR), the difference process (I) and the moving average model (MA). Considering the significant differences in the output of the region in different time periods, it is assumed that the electricity consumption characteristics of each time period are independent of each other, and the time series are constructed with the output data of each time period, and the ARIMA model is established as follows:

[0068]

[0069] Among them, Y t is the output data at time t; parameters p, q, d represent the autoregressive order, moving average order, and difference order respectively; L is the lag operator; ε is the error term; and θ i are the parameters of the autoregressive process and the moving average process, respectively.

[0070] After the difference change, W is satisfied t =ΔY t , and then its specific expansion form is as follows:

[0071]

[0072] (2) Using the MLP model to predict the output of renewable energy in period T

[0073] The MLP model continuously adjusts the weights and threshold parameters in the network through training sample data so that the model's prediction results are increasingly close to the expected target output. The sample training process can be transformed into a minimization problem to achieve the minimum mapping loss value.

[0074] The objective function of the mapping is:

[0075]

[0076] Among them, F(.) is the mapping function.

[0077] The first layer is the input layer, which is the influencing factors of renewable energy output, such as wind speed, wind direction, temperature, etc., represented by V i Let x1 represent the input vector, then we get:

[0078] x1=V i (4)

[0079] Then, above the input layer is a stack of fully connected hidden layers that are able to capture the nonlinear dependencies between latent factors. Formally, the hidden layers are defined as:

[0080] y l =f(W l ·x l +b l ), l=1,...,d-1 (5)

[0081] Among them, y l represents the output vector of the lth layer; f represents the activation function; W l is the weight vector; b l is the bias term. As for the activation function, the sigmoid function is selected, that is,

[0082]

[0083] The last layer is the prediction layer, which outputs the mapped new energy output:

[0084] Y MLP =f(W d ·y d-1 +b d ) (7)

[0085] The above formula can be implemented through MLP to predict the output of new energy.

[0086] (3) Calculate the root mean square error of each prediction method as the evaluation index of the prediction model.

[0087] The root mean square error is calculated as follows:

[0088]

[0089] Among them, y i is the predicted output value, y i is the actual output value, and N is the total number of samples.

[0090] (4) The roulette method is used to predict the output of renewable energy in each time period of period T. The probability of each method being used in each time period is:

[0091]

[0092] Among them, p i Indicates the probability of selecting the ARIMA model or the MLP model for prediction.

[0093] (5) The final composite prediction result is:

[0094] Y′=p ARIMA ·Y ARIMA+p MLP ·Y MLP (10)

[0095] 3. Using the multi-objective optimization model and combining constraints to calculate the objective function value

[0096] (1) Objective function

[0097] The basic optimization goal of the system is to minimize the economic cost and maximize the consumption of new energy. The objective function here mainly considers the operating cost of thermal power generating units and the penalty for wind and solar power abandonment, and its expression is as follows:

[0098]

[0099] Among them, T is a scheduling cycle; S u ,P u , Respectively represent the start / stop status of the unit, unit output and the amount of wind and solar power abandoned; W t (·), C t (·) and θ represent the unit start / stop cost, unit output cost and wind and solar power abandonment penalty factor in period t, respectively. Considering the carbon emission reduction target of the power system, carbon emission cost is often used as part of the objective function. Generally, carbon emission cost is expressed as a function of thermal power unit output.

[0100] (2) Constraints

[0101] 1) Power balance constraint, that is, in any period of time, the system's active power generation is equal to the active power consumption. The constraint expression is as follows:

[0102]

[0103] Where t is the index period; P t fire For thermal power unit i fire Output; P t wind For wind turbine i wind Output; P t solar Photoelectric motor group i solar Output; P t dis For battery pack dis Output; d t , are the power demand of the load and the battery unit i dis power storage needs.

[0104] 2) Output constraints of each generator set.

[0105] The output constraints of the thermal unit are:

[0106]

[0107] In the formula, and It is the minimum and maximum output of thermal power unit.

[0108] The output constraint of wind turbine is:

[0109]

[0110] In the formula, is the maximum output of the wind turbine.

[0111] The output constraint of the photovoltaic group is:

[0112]

[0113] In the formula, is the maximum output of the photovoltaic group.

[0114] The output constraint of the battery unit is:

[0115]

[0116] In the formula, It is the maximum output of the battery unit, and the battery unit cannot discharge and store electricity at the same time in any period of time.

[0117] (3) Solve the objective function value through the improved particle swarm algorithm

[0118] Based on the traditional particle swarm algorithm, the particle swarm algorithm is improved in terms of inertia weight factor and learning factor.

[0119]

[0120] Where: IT is the current iteration number; MI is the total iteration number; w s and w e is the initial and final value of the inertia weight factor; c 1s and c 1e is the initial and stop value of c1, c 1s Greater than c 1e ;c 2s and c 2e is the initial and stop value of c2, c 2s Less than c 2e .

[0121] 4. Generate Scheduling Strategy

[0122] Based on the output results of each generator at each moment, the power generation planning and scheduling plan is given.

[0123] Example 2

[0124] Referring to the specific steps of Example 1, this embodiment provides an intelligent dispatching method for a power system taking into account the uncertainty of the output of new energy sources. It includes two steps: 1. Using the historical data of wind power and photovoltaic power to construct their composite output prediction model; 2. Referring to the constructed multi-objective planning model to calculate the optimal dispatching method of the embodiment data.

[0125] Step 1: Use historical data of wind power and photovoltaic power to build their composite output prediction model.

[0126] Step 1.1, smooth the historical meteorological data of wind power and photovoltaic power respectively.

[0127] Step 1.2, construct the ARIMA model for wind power and photovoltaic power, and determine the appropriate p, d, and q values.

[0128] Observe the ACF diagram of the wind power sample ( Figure 2 ), we cannot intuitively get the characteristics of truncation. Then we observe the PACF diagram of the sequence and find that the process of decaying to zero is relatively continuous and slow. In summary, according to the principle of ARIMA model order determination, we preliminarily determine that p=9, q=2. The original sequence is non-stationary, and the first-order difference is required for model fitting, so d is 1. In summary, we set the ARIMA model parameters p, q, and d of the wind power sample to 9, 1, and 2 respectively.

[0129] Observe the ACF diagram of the photoelectric sample ( Figure 3 ), we can find that it roughly presents the characteristics of 20th order truncation. Then we observe the PACF diagram of the sequence and find that the process of decaying to zero is relatively continuous and slow. In summary, according to the principle of ARIMA model order determination, we preliminarily determine that p=20, q=2. The original sequence is stable, and the first-order difference is not required when fitting the model, so d is 0. In summary, we set the ARIMA model parameters p, q, and d of the optoelectronic sample to 20, 0, and 2 respectively.

[0130] In order to test the fitting degree of the model, the real data of the embodiment is compared with the model fitting value. The prediction results of the ARIMA model are as follows: Figure 4 and Figure 5 shown.

[0131] Among them, the blue curve is the original sequence value, and the red curve is the ARIMA model fitting value. It can be seen that the two are highly consistent, indicating that the model has a good degree of fit.

[0132] Step 1.3, extract features from the meteorological data of wind power and photovoltaic power, use XGBoost to reduce the dimension of the data, extract important factors and build a prediction model. The importance of each influencing factor in the data sample is as follows: Figure 6 and Figure 7 shown.

[0133] It can be seen from the figure that the main influencing characteristics of wind power are temperature, air pressure, humidity, wind speed 10m above the wind tower and wind direction 10m above the wind tower; the main influencing characteristics of photovoltaic power are component temperature, temperature, air pressure, humidity, scattered radiation and total radiation.

[0134] Step 1.4, construct the MLP prediction model. According to the important features obtained, the data of these features are selected as the input vector, and the actual power is used as the output vector to construct the MLP prediction model. Among them, 20% of the embodiment data are selected as the test set, and 80% of the embodiment data are selected as the training set; the number of hidden layers is 3, and the hidden nodes of each layer are 128, 64, and 32 respectively; the learning rate is 0.001, and the number of iterations is 400. The model prediction results are as follows Figure 8 and Fig. 9 shown.

[0135] Among them, the blue curve is the actual power generation, and the orange curve is the predicted power generation. It can be seen that the two are highly consistent, indicating that the model has a good degree of fitting.

[0136] Step 1.5, the composite prediction model is constructed by integrating the ARIMA model and the MLP model. The RMSE values ​​of the ARIMA model and the MLP model trained with the embodiment data are shown in Table 1.

[0137] Table 1 RMSE of different methods for predicting renewable energy output

[0138]

[0139] According to the results in Table 1 and formula 10, we can get the composite prediction results of wind power and photovoltaic output. In order to match the results of subsequent experiments, the data of a certain day in the sample data are selected for composite prediction. The degree of fit between the prediction results and the actual values ​​is shown in Fig.10 shown.

[0140] Step 2, referring to the constructed multi-objective programming model to calculate the optimal scheduling method for the implementation example data.

[0141] Step 2.1, randomly select one day's data from the embodiment data as experimental data.

[0142] Step 2.2, based on the multi-objective programming model constructed in the previous article, the improved particle swarm algorithm is used to solve it. The parameter settings are as follows: the population size is 100, the archive size is 100, the number of grids in each dimension is 7, the inflation rate is 0.1, the leader selection pressure is 2, the deletion selection pressure is 2, the mutation rate is 0.1, and the number of iterations is 100. The output distribution of each generator is obtained by solving the particle swarm algorithm as follows: Fig.11 shown.

[0143] It can be seen from the figure that the embodiment has the lowest power consumption in the period 0:00-6:00, because social production and life are in the least active state during this period, and the corresponding power consumption is the least. The power consumption is the highest in the period 19:00-22:00, because the demand for lighting during this period greatly increases the power consumption. In addition, the remaining periods show a fluctuating trend, which is closely related to the changes in factors such as people's daily routines, natural weather temperature, etc.

[0144] Step 2.3, based on the output results of each generator at each moment, a power generation planning and scheduling scheme of the embodiment is given.

[0145] from Fig.12 It can be seen that in the dispatching process, thermal power generation is the main energy supply mode throughout the dispatching process, which indicates that thermal generators may play the role of base load power generation in the entire system and continuously provide reliable power supply for the load. Especially in the time period when photovoltaic and wind power generation are low, thermal generators are particularly important, such as the time period of 1:00-5:00. Photovoltaic power generation gradually increases and reaches a peak in the time period of 9:00-14:00, with a maximum output of 13.3MW, and then gradually decreases to around 17:00. This is consistent with the typical solar radiation distribution, and photovoltaic power generation usually reaches its peak around noon. The output of wind power generation is volatile. For example, wind power generation between the time period of 1:00-8:00 is very low, with almost no obvious output; but the output in the time period of 11:00-14:00 and other time periods has increased, even reaching nearly 5MW. This fluctuation is related to the instability of wind speed on the day of the experimental data. The output of the battery is positive and negative, which indicates that the system is in a charging state from 23:00 to 9:00, and is discharging in other time periods. The operating state of the battery shows a complementary role to other power generation resources. For example, when the output of other generators is low, it tends to discharge to supplement the power demand.

Claims

1. An intelligent dispatching method for a power system considering the uncertainty of renewable energy output, characterized in that: It includes the following steps: (1) Data collection and processing; (2) Use a composite prediction model to predict the output of new energy; (3) Calculate the objective function value using a multi-objective optimization model combined with constraints; (4) Generate a scheduling strategy.

2. The intelligent dispatching method for power system considering the uncertainty of new energy output according to claim 1 is characterized in that: The data described in step (1) include: historical output data, meteorological data, load demand data and system operating parameters.

3. The intelligent dispatching method for power system considering the uncertainty of new energy output according to claim 2 is characterized in that: In step (1), the output history data includes: actual wind power generation power per hour, actual photovoltaic power generation power per hour, actual thermal power generation power per hour, and actual charge and discharge power of energy storage equipment per hour; Meteorological data include: temperature, air pressure, humidity, wind speed and wind direction; The system operating parameters include: maximum wind power output, maximum photovoltaic output, minimum and maximum thermal power output, minimum and maximum output of energy storage batteries, thermal power start and stop costs, thermal power generation costs, and penalties for wind and solar power abandonment.

4. The intelligent dispatching method for power system considering the uncertainty of new energy output according to claim 3 is characterized in that: In step (1), the high-frequency noise is removed by smoothing and averaging the wind power and photovoltaic meteorological data for each hour; XGBoost is used to reduce the dimension of the data and extract the time characteristics and meteorological characteristics.

5. The intelligent dispatching method for power system considering the uncertainty of new energy output according to claim 4 is characterized in that: In step (2), the composite prediction model is as follows: And′=p ARIMA ·AND ARIMA +p MLP ·AND MLP Among them, Y′ represents the composite prediction result; p ARIMA Indicates the probability of selecting the ARIMA model for prediction; Y ARIMA represents the new energy output value of the ARIMA model prediction period T; p MLP Indicates the probability of selecting the MLP model for prediction; Y MLP Represents the new energy output value of the MLP prediction model in the prediction period T.

6. The intelligent dispatching method for power system considering the uncertainty of new energy output according to claim 5 is characterized in that: The probability of selecting the ARIMA model or the MLP model for prediction is calculated by the following formula: Among them, p i It represents the probability of selecting ARIMA model or MLP model for prediction, and RMSE represents the root mean square error.

7. The intelligent dispatching method for power system considering the uncertainty of new energy output according to claim 6 is characterized in that: The root mean square error RMSE is calculated by the following formula: Among them, y i is the predicted output value, y i is the actual output value, and N is the total number of samples.

8. The intelligent dispatching method for power system considering the uncertainty of new energy output according to claim 7 is characterized in that: In step (3), the objective function of the multi-objective optimization model is: Among them, T is a scheduling cycle; S u ,P u , Respectively represent the start / stop status of the unit, unit output and the amount of wind and solar power abandoned; W t (·), C t (·) and θ represent the unit start / stop cost, unit output cost and wind and solar power curtailment penalty factor in period t, respectively.

9. The intelligent dispatching method for power system considering the uncertainty of new energy output according to claim 8 is characterized in that: In step (3), the constraints are: Where t is the index period; P t fire For thermal power unit i fire Output; P t wind For wind turbine i wind Output; P t solar Photoelectric motor group i solar Output; P t dis For battery pack dis Output; d t , are the power demand of the load and the battery unit i dis power storage needs.

10. The intelligent dispatching method for power system considering the uncertainty of new energy output according to claim 9 is characterized in that: The objective function value of the multi-objective optimization model is calculated using an improved particle swarm algorithm.

Images