Novel power prediction method and system for aggregated resources of power distribution network

Abstract

The invention relates to the technical field of power distribution management, and discloses a novel power distribution network aggregation resource power prediction method and system, and the system comprises a meteorological data unit, a mutual information analysis unit, a load data unit, a load analysis unit, and a neural unit. According to the novel power prediction method and system for the aggregated resources of the power distribution network, various types of weather forecast data are obtained through a weather data unit, mutual information between actually measured wind power and weather factors and mutual information between different weather factors are calculated through a mutual information analysis unit, and then feature sorting is carried out through an incremental search method; a load data unit obtains load data and communication people flow data, space-time grid clustering and space-time coupling are carried out, user load classification, load grading and communication people flow and load relevance are analyzed, and a neural unit learns and predicts resident user loads and resident area distribution transformer loads; a load increase and decrease target is formulated, power prediction is more accurate, and the utilization rate of aggregation resources is high.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution management, and in particular to a novel power prediction method and system for aggregated resources of a power distribution network. Background Art

[0002] Load aggregators use professional technology to evaluate the demand response potential of users and integrate scattered demand response resources to participate in the operation of the distribution system. As an intermediary, load aggregators can integrate user demand response resources and introduce them into market transactions, so that idle load resources can be put into use. At the same time, it also brings benefits to other participants in the distribution system. On the other hand, load aggregators can fully explore the demand response potential of loads through professional technical means, form a scientific way of electricity use, and improve the electricity efficiency of terminal power equipment. Active load aggregation is mainly aimed at power loads with certain adjustment capabilities, such as air conditioners, refrigerators, electric vehicles, etc. Short-term switching or changing control parameters will not have a significant negative impact on power equipment, and can ensure the user's comfort requirements. Before each system dispatch, the load aggregator obtains the allocated system power shortage from the superior dispatching department and formulates a load increase target or a load reduction target. According to the mathematical model and control method of the load, the demand response potential of each load is determined, and the loads within the jurisdiction are optimized and aggregated, so that the final load output is as close to the load increase target or the load reduction target as possible, and the electricity efficiency of terminal power equipment is improved.

[0003] At present, the scale of distributed renewable energy access is growing explosively, and a high proportion of distributed renewable energy access will become a typical feature of the distribution system. The natural and social factors on the energy supply side and the energy consumption side cause the "source-load" to present more extreme randomness and uncertainty. The traditional distribution system scheduling method, operation control strategy, and management means are difficult to support the construction of a new low-carbon distribution system. In view of the problem of inaccurate matching and coupling on both sides of the "source-load", it is necessary to conduct in-depth research on power prediction technology for regional aggregated resources of new distribution systems. Summary of the invention

[0004] 1. Technical issues to be solved

[0005] In view of the shortcomings of the prior art, the present invention provides a new power prediction method and system for aggregated resources of distribution network, which has the advantages of multi-dimensional analysis of data correlation and high utilization rate of aggregated resources, and solves the problems of low load prediction accuracy and low utilization rate of aggregated resources.

[0006] (II) Technical solution

[0007] To achieve the above object, the present invention provides the following technical solution: a new power prediction method for distribution network aggregated resources, characterized in that it includes the following steps:

[0008] Step 1: Obtain various types of meteorological forecast data through the weather forecast NWP model and classify them into TMF sets;

[0009] Step 2: Based on the TMF set, use the mRMR feature screening algorithm to calculate the measured wind power p and meteorological factors x i The mutual information between different meteorological factors x i , x j The mutual information between them is then used to sort the features using the incremental search method and form an MCF set;

[0010] Step 3: Connect the distribution network through the network to obtain the load data of residential users at different time points, and perform spatiotemporal grid clustering to obtain the typical load classification and load classification of residential users;

[0011] Step 4: Obtain communication traffic data through network connection to the distribution network, and then combine it with the residential user load data for spatiotemporal coupling to analyze the temporal and spatial correlation between the communication traffic data and the residential user load data;

[0012] Step 5: Use neural units to learn communication traffic data, residential user load data and weather forecast data to predict residential user load and residential distribution transformer load.

[0013] A novel power prediction system for aggregated resources of distribution network, comprising a meteorological data unit, a mutual information analysis unit, a load data unit, a load analysis unit and a neural unit;

[0014] The meteorological data unit is provided with a weather forecast NWP model for acquiring various types of meteorological forecast data and classifying and forming a TMF set;

[0015] The mutual information analysis unit is provided with an mRMR feature screening algorithm to calculate the measured wind power p and the meteorological factor x i The mutual information between different meteorological factors x i , x j The mutual information between them is then used to sort the features using the incremental search method and form an MCF set;

[0016] The load data unit is connected to the distribution network through the network, obtains the load data of the residential users at different time points, and performs spatiotemporal grid clustering to obtain the typical load classification and load classification of the residential users;

[0017] The load analysis unit obtains communication traffic data by connecting to the distribution network through the network, and then combines it with the residential user load data for time-space coupling to analyze the temporal and spatial correlation between the communication traffic data and the residential user load data;

[0018] The neural unit is used to learn communication traffic data, residential user load data and weather forecast data, and predict residential user load and residential distribution transformer load.

[0019] Compared with the prior art, the present invention provides a new power prediction method and system for aggregated resources in a distribution network, which has the following beneficial effects:

[0020] 1. The present invention sets a weather forecast NWP model through a meteorological data unit, obtains various types of meteorological forecast data, and classifies them into a TMF set. The meteorological factors in the TMF set include but are not limited to temperature, humidity and wind speed. Various meteorological factor data can improve the accuracy of power prediction, and this improvement mainly depends on the continuous training and iteration of numerical weather forecast and power model. The mutual information analysis unit is provided with an mRMR feature screening algorithm to calculate the measured wind power p and the meteorological factor x i The mutual information between different meteorological factors x i , x j The mutual information between them is used to sort the features through the incremental search method, and the MCF set is formed to analyze the data correlation in multiple dimensions.

[0021] 2. The present invention connects the distribution network through a load data unit network, obtains the load data of residential users at different time points, and performs spatiotemporal grid clustering to obtain typical load classification and load grading of residential users. The load analysis unit network is connected to the distribution network to obtain communication personnel flow data, and then combines the residential user load data for spatiotemporal coupling to analyze the temporal and spatial correlation between the communication personnel flow data and the residential user load data. The neural unit learns the communication personnel flow data, the residential user load data and the meteorological forecast data, predicts the residential user load and the residential substation distribution transformer load, and formulates the load increase target or load reduction target at the existing time point, thereby improving the accuracy of power system power prediction, effectively ensuring the safe and stable operation of the power system and the reliable supply of power resources, and high utilization of aggregated resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a step diagram of the method of the present invention;

[0023] Figure 2 It is a schematic diagram of the system flow of the present invention;

[0024] Figure 3 It is the system load forecasting framework of the present invention;

[0025] Figure 4 This is a schematic diagram of the classification and grading of residential user loads according to the present invention. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] Due to the explosive growth in the scale of distributed renewable energy access, a high proportion of distributed renewable energy access will become a typical feature of the distribution system. The natural and social factors on the energy supply and consumption sides have caused the "source-load" to present more extreme randomness and uncertainty. The traditional distribution system scheduling method, operation control strategy, and management means are difficult to support the construction of a new low-carbon distribution system. In view of the problem of inaccurate matching and coupling on both sides of the "source-load", it is necessary to conduct in-depth research on the power prediction technology for regional aggregated resources of the new distribution system. Therefore, a new power prediction method and system for aggregated resources of the distribution network is provided. Please refer to Figure 1 and Figure 3 , a new power prediction method for distribution network aggregated resources, comprising the following steps:

[0028] Step 1: Obtain various types of meteorological forecast data through the weather forecast NWP model and classify them into TMF sets. The expression of TMF set is G n ={x i ,i=1,2,3,...,n}, n represents the meteorological factor x The number of types of i. The meteorological factors in the TMF set include but are not limited to temperature, humidity and wind speed. Multiple meteorological factor data can improve the accuracy of power forecasting, and this improvement mainly depends on the continuous training and iteration of numerical weather forecasting and power models;

[0029] Step 2: Based on the TMF set, use the mRMR feature screening algorithm to calculate the measured wind power p and meteorological factors x i The mutual information between different meteorological factors x i , x j The mutual information between them is used, and then the features are sorted by incremental search method, and the MCF set is formed to analyze the data correlation in multiple dimensions;

[0030] The mutual information calculation formula is as follows:

[0031]

[0032] In the formula, H(·) represents the information entropy calculation function, I(·) represents the mutual information calculation function between two variables, where x i ,x j ∈G n , and i≠j;

[0033] The expression of the MCF set is S m ={x i ,i=1,2,...,m}, where m≤n, The mRMR principle is composed of the maximum correlation principle and the minimum redundancy principle. The calculation formula is as follows:

[0034]

[0035] In the formula, maxD(S m ,p) represents the maximum correlation calculation principle, which represents the maximum average mutual information between the meteorological factors in MCF and p, minR(S m ) is the minimum redundancy calculation principle, which indicates that the average value of mutual information between different meteorological factors in MCF is the minimum, and N(·) is the calculation function of the number of meteorological factor types in MCF;

[0036]

[0037] In the formula, represents the operator in the mRMR calculation principle, and D and R both represent the mutual information value;

[0038] According to the mutual information value with p in TMF, extract the largest meteorological factor x k As S m The first MCF within the range, and the remaining meteorological factors are recorded as G n =G n -x k , and then calculate G n The mutual information between internal meteorological factors and p satisfies:

[0039]

[0040] In the formula, represents the operator increment, Indicates that G n Internal meteorological factor x i The mutual information between G and p minus G n The difference between the average mutual information of the remaining meteorological factors is the evaluation index for the ranking of meteorological factor characteristics. In each screening process, select The meteorological factors with the largest values ​​are classified into the subset S of MCF. m , thus obtaining the feature ranking of TMF, and then selecting meteorological factors with operator increment greater than 0 to form MCF;

[0041] Step 3: Connect to the distribution network through the network to obtain the load data of residential users at different time points, and perform spatiotemporal grid clustering to obtain the typical load classification and load classification of residential users. The calculation process is as follows:

[0042] In each grid, the residential user load density Z t The expression is:

[0043]

[0044] In the formula, n a represents the number of residential users, B ti represents the load of resident user i in the spatial dimension, V ti represents the average load of residential user i, A ti represents the time dimension load density of residential user i;

[0045]

[0046] In the formula, Z ti represents the maximum load of residential user i, n b represents the number of grid dimensions, β ti represents the weight coefficient of resident user i for the grid dimension, V ti represents the average load of residential user i;

[0047] See also Figure 4 The typical load classification of residential users includes periodic load and non-periodic load. According to different typical load classifications, the load classification of residential users is analyzed. The calculation formula is as follows:

[0048]

[0049] In the formula, I a Indicates the residential electricity load classification, t c represents the load sampling time period, ω a represents the average value of the load cycle, I mai represents the periodic load of residential user i, I nbi represents the non-periodic load of residential user i;

[0050] Step 4: Obtain communication traffic data by connecting to the distribution network through the network, and then combine it with the residential user load data for time-space coupling, and analyze the temporal and spatial correlation between the communication traffic data and the residential user load data. The calculation process is as follows:

[0051] Scale series of spatiotemporal coupling in residential areas C al The expression is:

[0052]

[0053] In the formula, t d The number of moments representing the spatiotemporally coupled time periods;

[0054] Time scale series of communication traffic in residential areas D al for:

[0055]

[0056] C al With D al Fitting is performed and the generated sequence E c for:

[0057] E c =f(C al ,D al )

[0058] Coupling standard G a for:

[0059]

[0060] In the formula, σ a represents the coupling weight, λ a represents the data residual when the communication traffic data is spatiotemporally coupled, n d Indicates the number of couplings;

[0061] Fit the generated sequence E c for:

[0062]

[0063] In the formula, σ aj represents the coupling weight at time j, C alj represents the time scale series of residential area load at time j, D alj The time scale of the communication flow at time j is deeply coupled with multi-dimensional data, which helps to reduce hardware costs, increase the computing power elasticity of subsequent neural unit algorithm models, and is also conducive to the isolation and decoupling of the underlying data collection and computing power infrastructure from the upper-level prediction models and applications;

[0064] Step 5: Use neural units to learn communication traffic data, residential user load data and weather forecast data to predict residential user load and residential distribution transformer load, and achieve high aggregate resource utilization;

[0065] The neural unit is composed of input, output and storage gates. When the neural unit is learning, the communication flow data, residential user load data and weather forecast data are input at time t-1. At time t, the memory weight value of the first neural unit is passed to the next neural unit. The storage gate determines the proportion of the data input at time t-1 to the data saved at time t. The storage gate R a The calculation formula is as follows:

[0066] R a =Ψ(σ e [E t-1 ,Q t-1 ])

[0067] In the formula, Ψ represents the activation function, σ e Indicates the weight coefficient of the residential users of the storage gate. The weight coefficient value is set according to the characteristics of the regional residential area. t-1 represents the coupled data of residential user load and communication traffic, Q t-1 Represents weather forecast data;

[0068] Input data status J a for:

[0069]

[0070] In the formula, tanh represents the memory state function, represents the weight coefficient of residential users;

[0071] Output data status R c for:

[0072]

[0073] In the formula, represents the input activation function, R b Represents the input gate.

[0074] Preferably, in step 5, the load forecast of the distribution transformer in the residential area adds economic data on the basis of the residential load forecast data, and corrects the deviation of the residential load forecast data. When the neural unit predicts the load of the distribution transformer in the residential area, the storage gate R at for:

[0075]

[0076] In the formula, n k represents the number of users who need to distribute and transform load in the area where the residential user i is located, λ represents the activation function, σ ati represents the coupling weight of the transformer in the area where the residential user i is located, E di represents the coupled data of the input load of residential user i and the communication traffic, Q di represents the weather forecast data of the area where the residential user i is located, H d Indicates economic data;

[0077] Output status data R ct for:

[0078] R ct =R at×tanh(R c )×J a

[0079] In the formula, R c Represents the output data of residential user load forecast;

[0080] In actual use, based on the forecast results of residential user load and residential distribution transformer load, it is possible to formulate load increase targets or load reduction targets at the current time point, thereby improving the accuracy of power system power forecasting and effectively ensuring the safe and stable operation of the power system and the reliable supply of power resources.

[0081] See also Figure 2 , a novel power prediction system for aggregated resources in distribution networks, including a meteorological data unit, a mutual information analysis unit, a load data unit, a load analysis unit and a neural unit;

[0082] The meteorological data unit is equipped with a weather forecast NWP model, which is used to obtain various types of meteorological forecast data and classify them into TMF sets;

[0083] The mutual information analysis unit is equipped with an mRMR feature screening algorithm to calculate the measured wind power p and meteorological factors x i The mutual information between different meteorological factors x i , x j The mutual information between them is used, and then the features are sorted by incremental search method, and the MCF set is formed to analyze the data correlation in multiple dimensions;

[0084] The load data unit is connected to the distribution network through the network to obtain the load data of residential users at different time points, and performs spatiotemporal grid clustering to obtain the typical load classification and load classification of residential users;

[0085] The load analysis unit obtains communication traffic data through network connection to the distribution network, and then combines it with the residential user load data for time-space coupling, and analyzes the temporal and spatial correlation between the communication traffic data and the residential user load data, which is helpful for subsequent optimization of resource allocation;

[0086] The neural unit is used to learn communication traffic data, residential user load data and weather forecast data, predict residential user load and residential distribution transformer load, and aggregate resource utilization rate is high.

[0087] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A new power prediction method for aggregated resources in distribution network, characterized in that: The following steps are involved: Step 1: Obtain various types of meteorological forecast data through the weather forecast NWP model and classify them into TMF sets; Step 2: Based on the TMF set, use the mRMR feature screening algorithm to calculate the measured wind power p and meteorological factors x i The mutual information between different meteorological factors x i With x j The mutual information between them is then used to sort the features using the incremental search method and form an MCF set; Step 3: Connect the distribution network through the network to obtain the residential user load data at different time points, and perform spatiotemporal grid clustering to obtain the typical load classification and load classification of the residential users; Step 4: Obtain communication traffic data through network connection to the distribution network, and then combine it with the residential user load data for spatiotemporal coupling to analyze the temporal and spatial correlation between the communication traffic data and the residential user load data; Step 5: Use neural units to learn communication traffic data, residential user load data and weather forecast data to predict residential user load and residential distribution transformer load.

2. The power prediction method of a novel distribution network aggregated resource according to claim 1 is characterized by: In step 1, the expression of the TMF set is G n ={x i ,i=1,2,3,...,n}, n represents the meteorological factor x i The number of types.

3. The power prediction method of a novel distribution network aggregated resource according to claim 2 is characterized by: In the step 2, According to the mutual information value between the TMF set and the measured wind power p, the largest meteorological factor x is extracted. k As S m The first MCF within the range, and the remaining meteorological factors are recorded as G n =G n -x k , and then calculate G n The mutual information between the internal meteorological factors and p is G n Internal meteorological factor x i The mutual information between G and p minus G n The difference between the average mutual information of the remaining meteorological factors is the evaluation index for the ranking of meteorological factor characteristics.

4. The power prediction method of a novel distribution network aggregated resource according to claim 3 is characterized by: In the step 2, during each screening process, select The meteorological factors with the largest values ​​are classified into the subset S of MCF. m , thus obtaining the feature ranking of TMF, and then selecting meteorological factors with operator increment greater than 0 to form MCF.

5. The power prediction method of a novel distribution network aggregated resource according to claim 4 is characterized by: In the step three, the typical load classification of residential users includes periodically changing loads and non-periodically changing loads, and then the load classification of residential users is analyzed according to different typical load classifications.

6. The power prediction method of a novel distribution network aggregated resource according to claim 5 is characterized by: In step five, the neural unit is composed of input, output and storage gates. When the neural unit is learning, communication traffic data, residential user load data and weather forecast data are input at time t-1. At time t, the memory weight value of the first neural unit is passed to the next neural unit.

7. The power prediction method of a novel distribution network aggregated resource according to claim 6 is characterized by: in, The storage gate determines the proportion of data input at time t-1 that is saved at time t.

8. The power prediction method of a novel distribution network aggregated resource according to claim 7 is characterized by: In the step five, the predicted distribution transformer load in the residential area includes the residential load prediction data and the economic data; the deviation of the residential load prediction data is corrected, and the neural unit predicts the distribution transformer load in the residential area.

9. A new power prediction system for aggregated resources of a distribution network, applied to a new power prediction method for aggregated resources of a distribution network as claimed in any one of claims 1 to 8, characterized in that: It includes a meteorological data unit, a mutual information analysis unit, a load data unit, a load analysis unit and a neural unit; The meteorological data unit is provided with a weather forecast NWP model for acquiring various types of meteorological forecast data and classifying and forming a TMF set; The mutual information analysis unit is provided with an mRMR feature screening algorithm to calculate the measured wind power p and the meteorological factor x i The mutual information between different meteorological factors x i , x j The mutual information between them is then used to sort the features using the incremental search method and form an MCF set; The load data unit is connected to the distribution network through the network, obtains the load data of the residential users at different time points, and performs spatiotemporal grid clustering to obtain the typical load classification and load classification of the residential users; The load analysis unit obtains communication traffic data by connecting to the distribution network through the network, and then combines it with the residential user load data for time-space coupling to analyze the temporal and spatial correlation between the communication traffic data and the residential user load data; The neural unit is used to learn communication traffic data, residential user load data and weather forecast data, and predict residential user load and residential distribution transformer load.

Images