Virtual power plant load prediction method and system based on deep learning

Abstract

The invention provides a virtual power plant load prediction method and system based on deep learning, and the method comprises the steps: firstly carrying out the collection of power load data, meteorological data, population data and energy storage data of each load power user of a virtual power plant, and carrying out the preprocessing; extracting feature parameters from the preprocessed data based on a deep learning algorithm; and then a power load prediction model is constructed and trained by taking the historical load coefficient, the meteorological coefficient, the population number and the energy storage coefficient as inputs and taking the corresponding power load value as an output. And finally, predicting a future power load, calculating a power prediction deviation value between a prediction result of the power load and an actual power load, and judging whether to optimize the power load prediction model based on a power prediction deviation threshold value. According to the method, multi-aspect influences of effective key parameters can be considered, so that a more accurate power load prediction model is established, and an effective support is provided for subsequently deploying power resources of a virtual power plant in advance.

Description

Technical Field

[0001] The present invention belongs to the field of power forecasting, and specifically relates to a virtual power plant load forecasting method and system based on deep learning. Background Art

[0002] Virtual power plant is an innovative power management concept, the core of which is to achieve the aggregation and coordinated optimization of distributed energy through advanced information and communication technologies and software systems.

[0003] With the rapid development of electronic devices and internet technology, smart grids have become the clear direction for future power grid development. Smart grids aim to achieve transparent, instantaneous, and bidirectional transmission of energy and information, enabling users to intuitively and in real time understand current power load usage and make economic decisions based on load conditions. Load forecasting is a key technology that uses historical data on power load, combined with factors such as economic and social development and meteorological factors, to scientifically predict future power loads.

[0004] The load forecasting methods of traditional power grids are often affected by various factors such as external weather, population size, energy storage technology, etc., which leads to limited prediction accuracy of the load forecasting models of traditional power grids. Summary of the Invention

[0005] In response to the problems existing in the prior art, the present invention provides a virtual power plant load forecasting method and system based on deep learning, which can improve the accuracy of power load forecasting.

[0006] To solve the above technical problems, the present invention provides the following technical solution: a virtual power plant load forecasting method based on deep learning, comprising the following steps:

[0007] S1. Construct a power plant load characteristic parameter dataset based on meteorological data, population data, and energy storage data, including: historical power load values, load growth rate, load peak and valley values, temperature, wind speed, rainfall, population, and energy storage data;

[0008] S2, constructing and training a power load forecasting model using historical load factors, meteorological factors, population, and energy storage factors as inputs and corresponding power load values ​​as outputs;

[0009] S3. Predict the future power load using the power load forecasting model, then calculate the power forecast deviation between the predicted value and the actual power load, compare the power forecast deviation with a preset power forecast deviation threshold, and determine whether the power forecast deviation is greater than the preset power forecast deviation threshold. If so, issue an early warning and optimize and update the power load forecasting model to obtain an optimized power load forecasting model. Otherwise, execute step S4.

[0010] S4. Return to step S3.

[0011] Furthermore, the aforementioned virtual power plant load forecasting method based on deep learning is characterized in that it also includes preprocessing of meteorological data, population data and energy storage data, including data cleaning, data denoising and filling in missing data values.

[0012] Furthermore, in the aforementioned step S2, the power load forecasting model is as follows:

[0013]

[0014] Where, F λ is the power load forecast value of the λth load power user, X λ is the historical load factor, Y is the meteorological factor, Z λ is the population size, b0, b1, b2, b3 are weight coefficients, and p is the energy storage coefficient.

[0015] Furthermore, in the aforementioned step S2, the historical load factor X λ Obtain it as follows:

[0016] Establish a coordinate system with time as the horizontal axis and power load as the vertical axis; fit and draw a power load curve that changes over time within N preset periods for each power load user; draw upper and lower limit straight lines for the power load based on historical data, with periods above the upper limit representing peak periods and periods below the lower limit representing valley periods, and obtain the power load change curve x(t) for each peak period or valley period;

[0017] The historical load factor X of the current power load user can be calculated by the following formula λ :

[0018]

[0019] In the formula For the Preset cycles, is the impact factor corresponding to each preset period, is the reference value, E is the total number of peak hours and off-peak hours, x max is the maximum value of power load within a preset period, x min is the minimum value of the power load in a preset period, ξ1 and ξ2 are weight factors, Q is the total number of peak and valley values ​​of the power load in a preset period, t i , t i+1 The two time endpoints of any preset period.

[0020] Furthermore, in the aforementioned step S2, the meteorological coefficient is obtained as follows: Substitute the temperature T, wind speed V, and rainfall H into the following formula to calculate the meteorological coefficient Y:

[0021]

[0022] In the formula, when T∈preset temperature range [T cd , T cg ], f(T)=1; when T<T cd hour, When T>T cg hour, D is the light intensity, Dc is the standard value of light intensity, Vc is the standard value of wind speed, H(t) is the curve of rainfall changing with time, μ T 、μ D 、μ V 、μ H is the preset proportional coefficient, φ σ is the weight coefficient corresponding to the σth power generation user, and n is the number of power generation users.

[0023] Furthermore, in the aforementioned step S2, the energy storage coefficient is calculated as follows:

[0024]

[0025] Where Θ1, Θ2, Θ3, Θ4 are adjustment coefficients, η is the energy storage efficiency, η max is the maximum value of energy storage efficiency, θ is the energy storage capacity, is the reference energy storage capacity, δ is the energy storage power, δ max , δ min are the maximum and minimum values ​​of energy storage power respectively. is the energy storage density, are the maximum and minimum values ​​of energy storage density, respectively.

[0026] Furthermore, in the aforementioned step S3, the power prediction deviation value is calculated as follows:

[0027]

[0028] Where ΔUF is the power forecast deviation value of the current load power user, FS λ is the actual value of the power load of the current power user, FS λmax , FS λmin They are respectively the actual maximum and minimum values ​​of the power load of the current power user within a preset period.

[0029] Furthermore, in the aforementioned step S2, the characteristic parameters of the power plant load are extracted based on the convolutional neural network CNN or the autoencoder Autoencoder to construct a data set of characteristic parameters of the power plant load.

[0030] The present invention also provides a virtual power plant load forecasting system based on deep learning, comprising:

[0031] The data collection and preprocessing module is used to collect power load data, meteorological data, population data, and energy storage data from each load power user of the virtual power plant, and preprocess the collected data, including cleaning, denoising, and missing value filling;

[0032] The feature extraction module is used to extract features from power load data, meteorological data, population data and energy storage data to obtain historical power load values, load growth rates, load peak and valley values, temperature, wind speed, rainfall, population size and energy storage efficiency.

[0033] The model construction and training module is used to construct and train a power load forecasting model using historical load factors, meteorological factors, population, and energy storage factors as inputs and corresponding power load values ​​as outputs;

[0034] The power forecasting and model optimization module is used to predict future power loads using the power load forecasting model, calculate the power forecast deviation between the predicted value and the actual power load, compare the power forecast deviation with a preset power forecast deviation threshold, and determine whether the power forecast deviation is greater than the preset power forecast deviation threshold. If so, an early warning is issued and the power load forecasting model is optimized and updated to obtain an optimized power load forecasting model.

[0035] Compared with the prior art, the beneficial technical effects of the present invention using the above technical solution are as follows:

[0036] The present invention extracts and screens multiple key parameters that affect power load forecasting, and then constructs a power load forecasting model based on the above parameters. Compared with the existing technology that only conducts a single analysis of historical power load data and ignores the influence of other factors, resulting in low accuracy of power load forecasting, the present invention can take into account the multiple influences of effective key parameters, thereby establishing a more accurate power load forecasting model, providing effective support for the subsequent early allocation of power resources of virtual power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of the logical steps of the method of the present invention. DETAILED DESCRIPTION

[0038] In order to better understand the technical content of the present invention, specific embodiments are given and described below with reference to the accompanying drawings.

[0039] Various aspects of the present invention are described herein with reference to the accompanying drawings, which show a number of illustrative embodiments. The embodiments of the present invention are not limited to those described in the accompanying drawings. It should be understood that the present invention can be implemented by any of the various concepts and embodiments described above, as well as the concepts and implementations described in detail below, because the concepts and embodiments disclosed herein are not limited to any particular implementation. In addition, some aspects disclosed herein may be used alone or in any appropriate combination with other aspects disclosed herein.

[0040] refer to Figure 1 The present invention provides a virtual power plant load forecasting method based on deep learning, comprising the following steps: S1, constructing a power plant load characteristic parameter data set based on meteorological data, population data and energy storage data, including: historical power load values, load growth rate, load peak and valley values, temperature, wind speed, rainfall, population size and energy storage data;

[0041] S2, constructing and training a power load forecasting model using historical load factors, meteorological factors, population, and energy storage factors as inputs and corresponding power load values ​​as outputs;

[0042] S3. Predict the future power load using the power load forecasting model, then calculate the power forecast deviation between the predicted value and the actual power load, compare the power forecast deviation with a preset power forecast deviation threshold, and determine whether the power forecast deviation is greater than the preset power forecast deviation threshold. If so, issue an early warning and optimize and update the power load forecasting model to obtain an optimized power load forecasting model. Otherwise, execute step S4.

[0043] S4. Return to step S3.

[0044] As a preferred embodiment of the present invention, it also includes collecting meteorological data, population data and energy storage data, and preprocessing the collected data, including cleaning, denoising and missing value filling.

[0045] As a preferred embodiment of the present invention, characteristic parameters are extracted from the preprocessed data based on a deep learning algorithm, including historical load values, load growth rates, load peak and valley values, temperature, wind speed, rainfall, population size and energy storage efficiency; the deep learning algorithm is one of a convolutional neural network (CNN) or an autoencoder.

[0046] As a preferred embodiment of the present invention, the historical load factor X λ Obtain it as follows:

[0047] Establish a coordinate system with time as the horizontal axis and power load as the vertical axis; fit and draw a power load curve that changes over time within N preset periods for each power load user; draw upper and lower limit straight lines for the power load based on historical data, with periods above the upper limit representing peak periods and periods below the lower limit representing valley periods, and obtain the power load change curve x(t) for each peak period or valley period;

[0048] By analyzing the power load change curve of the current power load user in multiple preset cycles, the power load curve is divided by two straight lines of upper limit and lower limit, so as to quickly identify the peak period and off-peak period in a preset cycle, and then the cumulative change of the power load is calculated by integrating the peak period and off-peak period. Obviously, if the cumulative change is The larger the value is, the greater the ratio between the power load change of the power load user and the actual reference value is. At the same time, (x max -x min ) Q It is obvious that if the difference between the maximum and minimum values ​​in a preset period is larger, it means that the fluctuation range of the power load of the current circuit load user in the preset period is larger. The number of peak periods and low peak periods in a preset period can be expressed in the form of an index to more sensitively reflect the effect of Q on x. max -x min The extent of the impact; therefore The larger the (x max -x min ) Q The larger the historical load factor X is, the t The larger the value, the more accurate the historical load factor of the current power load user can be calculated, thereby improving the accuracy of subsequent power load forecasts.

[0049] The historical load factor X of the current power load user can be calculated by the following formula λ :

[0050]

[0051] In the formula For the Preset cycles, is the impact factor corresponding to each preset period, is the reference value, E is the total number of peak hours and off-peak hours, x max is the maximum value of power load within a preset period, x min is the minimum value of the power load in a preset period, ξ1 and ξ2 are weight factors, Q is the total number of peak and valley values ​​of the power load in a preset period, t i , ti+1 The two time endpoints of any preset period.

[0052] As a preferred embodiment of the present invention, the meteorological coefficient is obtained as follows: Substitute the temperature T, wind speed V, and rainfall H into the following formula to calculate the meteorological coefficient Y:

[0053]

[0054] In the formula, when T∈preset temperature range [T cd , T cg ], f(T)=1; when T<T cd hour, When T>T cg hour, D is the light intensity, Dc is the standard value of light intensity, Vc is the standard value of wind speed, H(t) is the curve of rainfall changing with time, μ T 、μ D 、μ V 、μ H is the preset proportional coefficient, φ σ is the weight coefficient corresponding to the σth power generation user, and n is the number of power generation users.

[0055] Through the above formula μ T f(T) shows that when the temperature is too high or too low, the energy consumption will increase. For example, in the hot summer, due to the widespread use of cooling equipment such as air conditioners, the power load will increase significantly, so μ T The larger f(T) is, the smaller the meteorological coefficient Y is. It can be seen that the impact of wind speed on the power load of the virtual power plant is mainly reflected in wind power generation, the impact of light intensity on the power load of the virtual power plant is mainly reflected in solar power generation, and the impact of rainfall on the power load of the virtual power plant is mainly reflected in hydropower generation; obviously, when the wind speed is high, the output power of the wind turbine will increase, thereby increasing the power supply of the virtual power plant, and the power load will decrease, then the meteorological coefficient Y will be smaller; conversely, when the wind speed is low, the output power of the wind turbine will decrease, the power supply of the virtual power plant will also decrease accordingly, the power load will increase, then the meteorological coefficient Y will be larger; the same can be said for light intensity and rainfall, the difference is that rainfall is reflected in the cumulative change, and will not affect hydropower generation in a short time, so through the integration method Calculations are performed to comprehensively achieve accurate meteorological assessments, providing data support for accurate predictions of power loads.

[0056] As a preferred embodiment of the present invention, the energy storage coefficient is calculated as follows:

[0057]

[0058] Where Θ1, Θ2, Θ3, Θ4 are adjustment coefficients, η is the energy storage efficiency, η max is the maximum value of energy storage efficiency, θ is the energy storage capacity, θ ref is the reference energy storage capacity, δ is the energy storage power, δ max , δ min are the maximum and minimum values ​​of energy storage power respectively. is the energy storage density, are the maximum and minimum values ​​of the energy storage density, respectively. Obviously, the greater the actual energy storage efficiency, capacity, power, and density, the greater the energy storage coefficient p, indicating a stronger energy storage capacity. Therefore, it can store excess, high-quality electricity and release it when needed, thereby balancing electricity supply and demand and playing a stronger role in stabilizing the power load. Conversely, the weaker the effect.

[0059] As a preferred embodiment of the present invention, in step S2, the power load forecasting model is as follows:

[0060]

[0061] Where, F λ is the power load forecast value of the λth load power user, X λ is the historical load factor, Y is the meteorological factor, Z λ is the population size, b0, b1, b2, b3 are weight coefficients, and p is the energy storage coefficient.

[0062] As a preferred embodiment of the present invention, in step S3, the power prediction deviation value is calculated as follows:

[0063]

[0064] Where ΔUF is the power forecast deviation value of the current load power user, FS λ is the actual value of the power load of the current power user, FS λmax , FS λmin They are respectively the actual maximum and minimum values ​​of the power load of the current power user within a preset period.

[0065] The invention calculates the difference between the actual power load value and the predicted value to obtain the deviation value of the power load prediction model, and then calculates the fluctuation amplitude of the actual value through the actual maximum and minimum values ​​of the power load within a preset period, so as to obtain the actual power prediction deviation value. Obviously, the larger the power prediction deviation value, the more accurate the power load prediction model is, and vice versa, the greater the error of the power load prediction model is, which provides a judgment basis for the subsequent selective execution model optimization, reduces the resource occupation problem caused by frequent model optimization, and reduces the problem of affecting the normal use of the power load model.

[0066] Another aspect of the present invention provides a virtual power plant load forecasting system based on deep learning, comprising:

[0067] The data collection and preprocessing module is used to collect power load data, meteorological data, population data, and energy storage data from each load power user of the virtual power plant, and preprocess the collected data, including cleaning, denoising, and missing value filling;

[0068] The feature extraction module is used to extract features from power load data, meteorological data, population data and energy storage data to obtain historical power load values, load growth rates, load peak and valley values, temperature, wind speed, rainfall, population size and energy storage efficiency.

[0069] The model construction and training module is used to construct and train a power load forecasting model using historical load factors, meteorological factors, population, and energy storage factors as inputs and corresponding power load values ​​as outputs;

[0070] The power forecasting and model optimization module is used to predict future power loads using the power load forecasting model, calculate the power forecast deviation between the predicted value and the actual power load, compare the power forecast deviation with a preset power forecast deviation threshold, and determine whether the power forecast deviation is greater than the preset power forecast deviation threshold. If so, an early warning is issued and the power load forecasting model is optimized and updated to obtain an optimized power load forecasting model.

[0071] While the present invention has been described above with reference to preferred embodiments, this is not intended to limit the present invention. Persons skilled in the art will readily appreciate that various modifications and variations can be made without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A virtual power plant load forecasting method based on deep learning, characterized in that: The following steps are involved: S1. Construct a power plant load characteristic parameter dataset based on meteorological data, population data, and energy storage data, including: historical power load values, load growth rate, load peak and valley values, temperature, wind speed, rainfall, population, and energy storage data; S2. Build and train a power load forecasting model using historical load factors, meteorological factors, population, and energy storage factors as inputs and corresponding power load values ​​as outputs; S3. Predict the future power load using the power load forecasting model, then calculate the power forecast deviation between the predicted value and the actual power load, compare the power forecast deviation with a preset power forecast deviation threshold, and determine whether the power forecast deviation is greater than the preset power forecast deviation threshold. If so, issue an early warning and optimize and update the power load forecasting model to obtain an optimized power load forecasting model. Otherwise, execute step S4. S4. Return to step S3.

2. A virtual power plant load forecasting method based on deep learning according to claim 1, characterized in that: It also includes the preprocessing of meteorological data, population data and energy storage data, including data cleaning, data denoising and filling in missing data values.

3. The method for virtual power plant load forecasting based on deep learning according to claim 1, characterized in that: In step S2, the power load forecasting model is as follows: Where, F λ is the power load forecast value of the λth load power user, X λ is the historical load factor, Y is the meteorological factor, Z λ is the population size, b0, b1, b2, b3 are weight coefficients, and p is the energy storage coefficient.

4. A virtual power plant load forecasting method based on deep learning according to claim 3, characterized in that: In step S2, the historical load factor X λ Obtain it as follows: Establish a coordinate system with time as the horizontal axis and power load as the vertical axis; draw a power load curve that changes with time within N preset periods for each power load user; Based on historical data, the upper and lower limit straight lines of the power load are drawn. The period above the upper limit is the peak period, and the period below the lower limit is the valley period. The power load change curve x(t) of each peak period or valley period is obtained. The historical load factor X of the current power load user can be calculated by the following formula λ : In the formula For the Preset cycles, is the impact factor corresponding to each preset period, is the reference value, E is the total number of peak hours and off-peak hours, x max is the maximum value of power load within a preset period, x min is the minimum value of the power load in a preset period, ξ1 and ξ2 are weight factors, Q is the total number of peak and valley values ​​of the power load in a preset period, t i , t i+1 The two time endpoints of any preset period.

5. The method for virtual power plant load forecasting based on deep learning according to claim 3, characterized in that: In step S2, the meteorological coefficient is obtained as follows: Substitute the temperature T, wind speed V, and rainfall H into the following formula to calculate the meteorological coefficient Y: In the formula, when T∈preset temperature range [T cd , T cg ], f(T)=1; when T<T cd hour, When T>T cg hour, D is the light intensity, Dc is the standard value of light intensity, Vc is the standard value of wind speed, H(t) is the curve of rainfall changing with time, μ T 、μ D 、μ V 、μ H is the preset proportional coefficient, φ σ is the weight coefficient corresponding to the σth power generation user, and n is the number of power generation users.

6. The method for virtual power plant load forecasting based on deep learning according to claim 3, characterized in that: In step S2, the energy storage coefficient is calculated as follows: Where Θ1, Θ2, Θ3, Θ4 are adjustment coefficients, η is the energy storage efficiency, η max is the maximum value of energy storage efficiency, θ is the energy storage capacity, θ ref is the reference energy storage capacity, δ is the energy storage power, δ max , δ min are the maximum and minimum values ​​of energy storage power respectively. is the energy storage density, are the maximum and minimum values ​​of energy storage density, respectively.

7. The method for virtual power plant load forecasting based on deep learning according to claim 3, characterized in that: In step S3, the power forecast deviation value is calculated as follows: Where, ΔUF is the power forecast deviation value of the current load power user, FS λ is the actual value of the power load of the current power user, FS λmax , FS λmin They are respectively the actual maximum and minimum values ​​of the power load of the current power user within a preset period.

8. The method for virtual power plant load forecasting based on deep learning according to claim 1, characterized in that: In step S2, the characteristic parameters of the power plant load are extracted based on the convolutional neural network CNN or the autoencoder Autoencoder to construct a data set of the characteristic parameters of the power plant load.

9. A virtual power plant load forecasting system based on deep learning, characterized in that: include: The data collection and preprocessing module is used to collect power load data, meteorological data, population data, and energy storage data from each load power user of the virtual power plant, and preprocess the collected data, including cleaning, denoising, and missing value filling; The feature extraction module is used to extract features from power load data, meteorological data, population data and energy storage data to obtain historical power load values, load growth rates, load peak and valley values, temperature, wind speed, rainfall, population size and energy storage efficiency. The model construction and training module is used to construct and train a power load forecasting model using historical load factors, meteorological factors, population, and energy storage factors as inputs and corresponding power load values ​​as outputs; The power forecasting and model optimization module is used to predict future power loads using the power load forecasting model, calculate the power forecast deviation between the predicted value and the actual power load, compare the power forecast deviation with a preset power forecast deviation threshold, and determine whether the power forecast deviation is greater than the preset power forecast deviation threshold. If so, an early warning is issued and the power load forecasting model is optimized and updated to obtain an optimized power load forecasting model.

Images