A regional power grid load side resource refinement identification method, device and medium

Abstract

This application discloses a method, device, and medium for refined identification of load-side resources in a regional power grid. The method includes the following steps: establishing a physical driving model, extracting the inherent attributes of the regional load, and physically modeling the regional power grid load as a set of different types of loads; obtaining the active power fluctuation of the power grid load when facing random events; decomposing the regional power grid load fluctuation sequence into its trend component, periodic component, and random component, and extracting the impact of random events on load demand; associating the characteristics of random events in the regional power grid with load-side resources to obtain multi-dimensional key information on event-load resources; constructing a refined identification model of the regional power grid load, outputting the proportion of regional load composition, and correcting the physical driving model based on the output results. This invention achieves refined identification of power grid load-side resources through in-depth analysis of regional power grid events.

Description

Technical Field

[0001] This invention relates to a method, device, and medium for refined identification of load-side resources in a regional power grid, belonging to the field of intelligent operation and maintenance technology for power systems. Background Technology

[0002] The refined identification of grid load-side resources is crucial in modern energy management and power system optimization. It helps improve the efficiency, reliability, and sustainability of the power system, providing vital support for its future development, thereby meeting growing electricity demand and addressing the challenges of climate change.

[0003] Current refined identification of regional power grid load-side resources is mostly based on routine research. Existing technical literature, such as "Research on User Load Classification Based on Comprehensive Electricity Consumption Behavior Indicators," proposes a method to classify users by comprehensively considering their actual information (such as income level, house size, energy conservation and environmental awareness) and electricity load information. Background technical literature, "Short-Term Load Forecasting of Power Systems Combining Recurrent Neural Networks with Restricted Boltzmann Machines," combines the unique advantages of recurrent neural networks in analyzing time-series data with the powerful unsupervised learning capabilities of restricted Boltzmann machines to analyze the periodicity and intra-period characteristics of load forecast data. Background technical literature, "Short-Term Power Load Forecasting Based on Bayesian Optimization using CNN-GRU," constructs a CNN-GRU prediction model to address the temporal and nonlinear characteristics of load data and uses Bayesian optimization to find the optimal hyperparameter combination for hyperparameter optimization. However, the aforementioned background technical literature rarely focuses on random events. Under random events such as sudden temperature drops, electricity user loads can change significantly. Continuing to use routine regional power grid load-side resource refined identification methods would be detrimental to the accuracy of the results. Summary of the Invention

[0004] This application provides a method, device, and medium for refined identification of load-side resources in a regional power grid, aiming to address the shortcomings of existing technologies and achieve accurate identification of load-side resources in a regional power grid.

[0005] The technical solution of this application is as follows:

[0006] On the one hand, this application provides a method for refined identification of regional power grid load-side resources, including the following steps:

[0007] Step 1: Establish a physical driving model, extract the inherent attributes of the regional load, and physically model the regional power grid load as a set of different types of loads.

[0008] The active power fluctuation of the regional power grid load when faced with random events is obtained and denoted as load fluctuation;

[0009] Step 2: Decompose the regional power grid load fluctuation sequence into its trend component, periodic component, and random component, and extract the impact of random events on load demand;

[0010] Step 3: Correlate the characteristics of random events in the regional power grid with load-side resources to obtain multi-dimensional key information on event-load resources;

[0011] Step 4: Construct a refined identification model of regional power grid load based on multi-dimensional key information of event-load resources. The output is the regional load composition ratio. The physical driving model is corrected based on the output results to achieve refined identification of the regional power grid.

[0012] Furthermore, in step 1, the regional power grid load is physically modeled as a set of different types of loads, as shown below:

[0013]

[0014] in, These are the total regional load, electric vehicle load, lighting load, and air conditioning load, respectively; α1, α2, and α3 are the weights of the loads.

[0015] Furthermore, in step 1, a hybrid driving method combining data and physical dual-drive is used to analyze random events, and the active power fluctuation ΔP of the power grid load when the regional power grid faces random events is analyzed and extracted using a data-driven method.

[0016] Furthermore, in step 2, the CEEMDAN method is used to decompose the power load sequence into its trend component, periodic component, and random component; the random component represents the impact of random events on the regional power grid load; after decomposition using the CEEMDAN method, the loads of different frequency components are effectively decomposed into their corresponding intrinsic mode components, and the random component can be extracted; the final decomposition result is:

[0017]

[0018] in, r is the k-th intrinsic mode component; K (t) represents the residual signal.

[0019] Furthermore, in step 3, a two-stage decision model based on RF-BWM and RF-MARCOS under dual-drive mode is adopted to associate random event characteristics with load-side resources.

[0020] Furthermore, in step 3, in the two-stage decision model based on RF-BWM and RF-MARCOS, the RF-BWM method is used to identify the weight of the impact of random event characteristics on the load. After the attribute weights are determined, the random event characteristics are associated with the load-side resources by using the RF-MARCOS method to obtain multi-dimensional key information on event-load resources.

[0021] Furthermore, the characteristics of the random event include temperature, humidity, light intensity, and whether it is a weekday.

[0022] Furthermore, in step 4, an LSTM neural network from the hybrid multi-attribute intelligent identification method is used to construct a refined identification model of the regional power grid load. The LSTM network includes a forget gate, an input gate, and an output gate. The calculation formula for the LSTM network is as follows:

[0023] f t =σ(W f ×[h t-1 ,x t ]+B f (3)

[0024] i t =σ(W i ×[h t-1 ,x t ]+B i (4)

[0025] O t =σ(W o ×[h t-1 ,x t ]+B o (5)

[0026] Among them, f t i t O t These are the forget gate, input gate, and output gate, respectively. f W i W o Let B be the weight matrix. f B i B o Here, σ is the bias term, and x is the change in the sigmoid function. t For input, h t-1 This is the output from the previous time step.

[0027] On the other hand, this application provides a device for fine-grained identification of regional power grid load-side resources, including a memory and a processor. The memory stores a computer program, and when the computer program is called and executed by the processor, it implements the fine-grained identification method for regional power grid load-side resources as described above.

[0028] In another aspect, this application provides a computer-readable medium storing a computer program, which, when called and executed by a computer, implements the refined identification method for regional power grid load-side resources as described above.

[0029] The advantages of this application are as follows:

[0030] 1) This method combines a hybrid multi-attribute driving approach that integrates data-driven and physical-driven approaches to accurately identify regional power grid load-side resources.

[0031] 2) This method uses a two-stage decision model based on RF-BWM and RF-MARCOS to associate random event characteristics with load-side resources, which can handle multi-attribute decision problems. It also combines fuzzy set theory and rough set theory to fully consider the uncertainty and fuzziness in decision problems.

[0032] 3) This method uses the LSTM neural network in the hybrid multi-attribute intelligent identification method to construct a refined identification model of regional power grid load, and uses deep learning algorithms to output the regional load composition ratio, effectively solving the gradient explosion and gradient vanishing problems. Attached Figure Description

[0033] Figure 1 This is a flowchart of the method for refined identification of regional power grid load-side resources based on random event driving according to the present invention. Detailed Implementation

[0034] The present application will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be construed as limiting the scope of protection of the present application.

[0035] Appendix Figure 1 This is a flowchart of a method for refined identification of regional power grid load-side resources according to a specific embodiment of the present invention. The method includes steps 1 to 4.

[0036] Step 1: Use data-driven methods to analyze and extract the active power fluctuation of the power grid load ΔP when the regional power grid faces random events, denoted as load fluctuation;

[0037] A physical driving model is established, and the inherent attributes of the regional load are extracted based on the physical driving model. The regional power grid load is equivalent to a set of different types of loads for physical modeling, as shown below.

[0038]

[0039] in, These are the total regional load, electric vehicle load, lighting load, and air conditioning load, respectively; α1, α2, and α3 are the weights of the loads.

[0040] Step 2: To further refine the impact of random events on load demand, this invention utilizes the fully integrated empirical mode decomposition method (CEEMDAN method) to decompose the power load sequence into its trend component, periodic component, and random component. The random component characterizes the impact of random events on the regional power grid load, which is helpful for subsequent refined identification of regional power grid load-side resources.

[0041] The CEEMDAN method obtains intrinsic mode functions by introducing white noise with a specific frequency band at each decomposition stage. First, CEEMDAN decomposes the original data x(t) + ε0No... i (i = 1, 2, ..., I) undergo I experiments, with white noise added according to a standard normal distribution. The first intrinsic mode component obtained through empirical mode decomposition is:

[0042]

[0043] In the first stage, k=1, the signal is removed from the original signal. Modal components and residual signal calculation:

[0044]

[0045] Construct the integrated residual signal r1(t)+ε1E1(No) i (t)), (i=1,2,…,I), and decompose the integrated signal to obtain

[0046]

[0047] For k = 2, 3, ..., the calculation process is similar. First, calculate the k-th residual signal r. k (t); then, calculate k+1 intrinsic mode components.

[0048]

[0049]

[0050] When the number of extreme points in the residual signal is less than 3, the obtained residual signal cannot be further decomposed, and the algorithm terminates. Thus, a total of K intrinsic mode components are obtained, and the final decomposition result is:

[0051]

[0052] After decomposition using the CEEMDAN method, the complexity of the load data is greatly reduced. Loads of different frequency components are effectively decomposed into their corresponding intrinsic mode components, and random components can be extracted.

[0053] Step 3: A two-stage decision model based on RF-BWM and RF-MARCOS under dual-drive mode is used to associate random event characteristics with load-side resources to obtain multi-dimensional key information on event-load resources. This invention uses the Rough-Fuzzy BWM method (RF-BWM) to identify the weights of factors such as temperature, humidity, light intensity, and whether it is a working day in random events on the load.

[0054] First, regarding the evaluation and assessment system C, which includes both qualitative and quantitative attributes... j (j = 1, 2, ..., n), the optimal (most important) attribute and the worst (least important) attribute are determined by the decision-making group, denoted as C respectively. B and C W .

[0055] Next, a group semantic preference vector is constructed, and the preference comparisons include: pairwise comparisons of the best attribute with other attributes and pairwise comparisons of the worst attribute with other attributes.

[0056] make It was decided by decision-maker D k Provided C B Compared to fuzzy preference vectors of other attributes, It was decided by decision-maker D k Other attributes provided relative to the worst attribute C W The fuzzy preference vector, and It has the following forms:

[0057]

[0058]

[0059] in,

[0060] By using K semantic vectors provided by all decision-makers Integrate them into a matrix to form C B Group semantic preference vector relative to other attributes As shown below:

[0061]

[0062] Similarly, other attributes relative to the worst attribute C W Group semantic preference vector as follows:

[0063]

[0064] Then, the group semantic vector is transformed into a coarse fuzzy vector.

[0065] matrix and Each column in the table is a set of group semantic variables used to describe relative importance. and These are respectively converted into the form of the corresponding coarse fuzzy vectors, i.e. and As shown below:

[0066]

[0067]

[0068] Secondly, determine the coarse fuzzy weights for each indicator. Let... For indicator C j The optimal coarse fuzzy weights are then The following conditions must be met:

[0069]

[0070]

[0071] in, and These represent the optimal attributes. and worst attribute The above two conditions mean that the optimal weight for each indicator should remain consistent between the best comparison with other and the worst comparison with other, that is... and Considering the properties of the coarse fuzzy number and the constraints of the BWM model, the six boundaries in each coarse fuzzy weight should satisfy the following constraints:

[0072]

[0073] Therefore, the following coarse fuzzy optimization model was constructed to determine the optimal weights of the indicators:

[0074]

[0075] Finally, by solving the above model, the optimal rough fuzzy weight set of the index is obtained. It has been confirmed.

[0076] After the attribute weights are determined, the RF-MARCOS method is used to associate the random event characteristics (temperature, humidity, light intensity, whether it is a working day, etc.) with the load-side resources to obtain multi-dimensional key information on event-load resources.

[0077] The decision-making process of the RF-MARCOS method is as follows:

[0078] Construct a fuzzy initial decision matrix.

[0079] Let A = {A1, A2, ..., A} q Let} be a finite set of solutions, where A i (i = 1, 2, ..., q) represents the i-th option. The set of decision-makers (E) consists of experts. p (p = 1, 2, ..., P) requires the use of the semantic variables provided in Table 1 for the n attributes (C) of these schemes. j The evaluation is performed using the formula (j = 1, 2, ..., n). Here, TFN represents the triangular fuzzy number. Then, based on the mapping relationship in Table 1, the decision-maker's semantic variables are converted into triangular fuzzy numbers. These are then evaluated by decision-maker E. p Provided fuzzy initial decision matrix As shown below:

[0080]

[0081] in, This indicates that for option A i Attribute C j The triangular fuzzy information.

[0082] Table 1. Semantic variables and corresponding fuzzy numbers of the evaluation scheme

[0083]

[0084] Construct a fuzzy decision matrix for the group.

[0085] By aggregating the fuzzy initial decision matrices of P decision-makers, a group fuzzy decision matrix can be obtained. As shown below:

[0086]

[0087] in, or and

[0088] Construct a coarse fuzzy decision matrix.

[0089] Converted to coarse and blurry numbers Right now:

[0090]

[0091] Therefore, the matrix It can be transformed into a coarse fuzzy decision matrix. Right now:

[0092]

[0093] Establish an extended coarse fuzzy decision matrix.

[0094] By defining the coarse fuzzy ideal solution and the coarse fuzzy non-ideal solution (denoted as AI and AAI, respectively), an extended coarse fuzzy decision matrix is ​​formed. Right now

[0095]

[0096] in, This represents the sequence of the worst-performing alternatives for each metric; conversely, This represents the sequence of the best performing alternatives for each metric. and The identification is as follows:

[0097]

[0098]

[0099] Where B and C represent the sets of benefit-type and cost-type indicators, respectively; ∨ and ∧ represent the maximum and minimum operations, respectively.

[0100] Normalized extended coarse fuzzy decision matrix.

[0101] Extended coarse fuzzy decision matrix Normalized to

[0102]

[0103] in,

[0104] Determine the weighted rough fuzzy decision matrix.

[0105] Weighted coarse fuzzy decision matrix The calculation is as follows:

[0106]

[0107] in, The index weights are obtained based on the RF-BWM method.

[0108] Determine the utility of the plan.

[0109] Option A i The rough and fuzzy utility definitions relative to AAI and AI are as follows:

[0110]

[0111]

[0112] in, For weighted rough fuzzy matrix The sum of the elements in each row of the array is:

[0113]

[0114] Thus, we obtain the rough fuzzy utility set associated with the anti-ideal solution and the rough fuzzy utility set associated with the ideal solution: and Furthermore, determine the rough fuzzy utility. and The clear result, namely and

[0115] Determine the utility function of the proposed solution.

[0116] Utility function f(K) i f(K) represents a compromise between the proposed solution and both the anti-ideal and ideal solutions. i The calculation is as follows:

[0117]

[0118]

[0119]

[0120] in, and These represent the utility functions related to the anti-ideal solution and the ideal solution, respectively. Finally, the impact of random events on the load is comprehensively evaluated based on the utility function values, and the utility function values ​​are used to guide the refined identification of regional power grid load-side resources.

[0121] Step 4: Use the LSTM neural network in the hybrid multi-attribute intelligent identification method to construct a refined identification model of regional power grid load. The input is regional random events, load fluctuations and historical load data, and the output is the regional load composition ratio, thereby correcting the physical driving model and realizing refined identification of the regional power grid.

[0122] LSTM is a special type of RNN whose hidden layers consist of one or more memory units, each including a forget gate, an input gate, and an output gate. The forget gate, input gate, and output gate are the fundamental components of an LSTM network. The forget gate determines whether the memory from the previous time step should be forgotten or retained in the current time step. The input gate determines which part of the input information from the current time step should be added to the network's memory. The output gate determines which part of the information from the current time step is passed to the next part of the network. The calculation formula is:

[0123] f t =σ(W f ×[h t-1 ,x t ]+B f (43)

[0124] i t =σ(W i ×[h t-1 ,x t ]+B i (44)

[0125] O t =σ(W o ×[h t-1 ,x t ]+B o (45)

[0126] Among them, f t i t O t These are the forget gate, input gate, and output gate, respectively. f W i W o Let B be the weight matrix. f B i B o Here, σ is the bias term, and x is the change in the sigmoid function. t For input, h t-1 This is the output from the previous time step.

[0127] This example uses real-time smart meter electricity consumption data from a city throughout 2021, collecting electricity consumption data for each user every 30 minutes to obtain a massive user load curve. Based on physical-driven extraction of inherent regional load attributes, the regional power grid load is physically modeled as a set of different load types: 50% lighting load + 30% air conditioning load + 20% electric vehicle load. When the city suddenly experiences a sharp drop in temperature, a two-stage decision model using RF-BWM and RF-MARCOS under dual-drive mode correlates random event characteristics with load-side resources, obtaining multi-dimensional key information on event-load resources. An LSTM neural network is then used to construct a refined regional power grid load identification model, ultimately resulting in a regional load composition of 38% lighting load + 48% air conditioning load + 14% electric vehicle load.

[0128] Another specific embodiment of this application provides a regional power grid load-side resource fine identification device, including a memory and a processor. The memory stores a computer program. When the computer program is called and executed by the processor, it implements the regional power grid load-side resource fine identification method as described above.

[0129] In another specific embodiment of this application, a computer-readable medium is provided, which stores a computer program. When the computer program is called and executed by a computer, it implements the refined identification method for regional power grid load-side resources as described above.

[0130] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for refined identification of load-side resources in a regional power grid, characterized in that, The method includes the following steps: Step 1: Establish a physical driving model, extract the inherent attributes of the regional load, and physically model the regional power grid load as a set of different types of loads, as shown below: in, These represent the total regional load, electric vehicle load, lighting load, and air conditioning load, respectively; α1, α2, and α3 represent the weights of each load. The active power fluctuation of the regional power grid load when facing random events is obtained and denoted as load fluctuation; a hybrid driving method integrating data and physical dual drive is used to analyze random events, and the active power fluctuation of the regional power grid load when facing random events is extracted by data-driven method; Step 2: Decompose the regional power grid load fluctuation sequence into its trend component, periodic component, and random component, and extract the impact of random events on load demand; use the CEEMDAN method to decompose the power load sequence into its trend component, periodic component, and random component; after decomposition using the CEEMDAN method, the load of different frequency components is effectively decomposed into the corresponding intrinsic mode components, and the random component is extracted; the final decomposition result is: in, r is the k-th intrinsic mode component; K (t) represents the residual signal; Step 3: Correlate the characteristics of random events in the regional power grid with load-side resources to obtain multi-dimensional key information on event-load resources; adopt a two-stage decision model based on RF-BWM and RF-MARCOS under dual-drive mode to correlate random event characteristics with load-side resources; in the two-stage decision model, the RF-BWM method is used to identify the weight of the impact of random event characteristics on the load. After the attribute weights are determined, the RF-MARCOS method is used to correlate random event characteristics with load-side resources to obtain multi-dimensional key information on event-load resources. Step 4: Construct a refined identification model of regional power grid load based on multi-dimensional key information of event-load resources. The output is the regional load composition ratio. The physical driving model is corrected based on the output results to achieve refined identification of the regional power grid.

2. The method for refined identification of regional power grid load-side resources according to claim 1, characterized in that: The characteristics of the random events include temperature, humidity, light intensity, and whether it is a working day.

3. The method for refined identification of regional power grid load-side resources according to claim 1, characterized in that: Step 4: Construct a refined identification model of regional power grid load using the LSTM neural network in the hybrid multi-attribute intelligent identification method; the LSTM network includes a forget gate, an input gate, and an output gate, and the LSTM network calculation formula is as follows: f t =σ(W f ×[h t-1 ,x t ]+B f ) (3) i t =σ(W i ×[h t-1 ,x t ]+B i ) (4) The t =σ(W o ×[h t-1 ,x t ]+B o ) (5) Among them, f t i t O t These are the forget gate, input gate, and output gate, respectively. f W i W o Let B be the weight matrix. f B i B o Here, σ is the bias term, and x is the change in the sigmoid function. t For input, h t-1 This is the output from the previous time step.

4. A device for refined identification of load-side resources in a regional power grid, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is called and executed by the processor, it implements the regional power grid load-side resource fine identification method as described in any one of claims 1-3.

5. A computer-readable medium, characterized in that, The computer-readable medium stores a computer program, which, when called and executed by a computer, implements the regional power grid load-side resource refinement identification method as described in any one of claims 1-3.

Images