A load composition analysis method, device and equipment

Abstract

The application discloses a load composition analysis method, device and equipment, and the method comprises the following steps: arranging the state change information of the acquired power grid load node into a state time sequence, wherein the state change information comprises voltage and power; and performing component prediction analysis on the state time sequence by using a preset load composition analysis model to obtain a load composition analysis result, wherein the preset load composition analysis model is obtained by training through a DDPG algorithm. Since the action mechanism and the evaluation mechanism exist in the DDPG algorithm, the model can be optimized into an analyzer that is more suitable for the actual load change condition. The application can solve the technical problems that the prior art has the problems of high complexity and poor analysis matching, and the analysis process is slow in calculation speed and low in accuracy.

Description

Technical Field

[0001] This application relates to the field of load analysis technology, and in particular to a load composition analysis method, apparatus and equipment. Background Technology

[0002] With the development and construction of power systems, the scale of power grids is gradually increasing, the types of loads are becoming more diverse, and the time-varying nature of load composition is increasing. The enhanced time-varying nature of loads brings greater difficulties to power system load modeling. How to quickly and accurately identify the components of loads has become a key issue in power system load modeling research.

[0003] Traditional load composition analysis methods mainly include offline and online approaches. On the one hand, comprehensive statistical methods are typically used to survey and analyze the load composition of the power system. This method is mainly conducted through manual surveys, which is labor-intensive, and the data from offline surveys often differs from the actual load data of the power grid, resulting in poor matching. On the other hand, power system operators also analyze the load composition of the power system through overall measurement and fault simulation methods. This method relies on actual measurements of the power grid, and the calculated load composition results can match the actual power grid operating data well. However, when performing load composition analysis, the diversity of actual load composition and the nonlinearity of the load model lead to a large computational load, and the computational efficiency and accuracy of the results are difficult to meet the requirements of online load composition analysis. Summary of the Invention

[0004] This application provides a method, apparatus, and equipment for load composition analysis, which addresses the technical problems of existing technologies having high complexity and poor analysis matching, resulting in slow calculation speed and low accuracy in the analysis process.

[0005] In view of this, the first aspect of this application provides a method for load composition analysis, comprising:

[0006] The acquired state change information of the power grid load nodes is organized into a state time series, and the state change information includes voltage and power.

[0007] The state time series is subjected to component prediction analysis using a preset load composition analysis model, and the load composition analysis results are obtained. The preset load composition analysis model is trained by the DDPG algorithm.

[0008] Preferably, the step of organizing the acquired state change information of the power grid load nodes into a state time series, wherein the state change information includes voltage and power, and further includes:

[0009] Information on the status changes of power grid load nodes within a preset time period is collected based on a preset frequency.

[0010] Preferably, the step of organizing the acquired state change information of the power grid load nodes into a state time series, wherein the state change information includes voltage and power, and further includes:

[0011] The state change information is normalized based on a preset normalization formula.

[0012] Preferably, the step of performing component prediction analysis on the state time series using a preset load composition analysis model to obtain load composition analysis results, wherein the preset load composition analysis model is trained using the DDPG algorithm, and further includes:

[0013] An initial load composition analysis model was constructed based on a convolutional neural network.

[0014] The initial load composition analysis model is trained by the analysis effect evaluation model built based on the DDPG algorithm to obtain the preset load composition analysis model.

[0015] Preferably, the step of training the initial load composition analysis model using an analysis effect evaluation model constructed based on the DDPG algorithm to obtain a preset load composition analysis model includes:

[0016] The preset training dataset is input into the initial load composition analysis model to perform load analysis and obtain the load ratio adjustment amount.

[0017] The analysis effect evaluation model based on the DDPG algorithm generates load ratio adjustment evaluation results based on the load ratio adjustment amount;

[0018] Based on the load ratio adjustment amount and the load ratio adjustment evaluation results, the model is optimized and trained to generate a preset load composition analysis model.

[0019] A second aspect of this application provides a load composition analysis apparatus, comprising:

[0020] The sequence extraction module is used to organize the acquired state change information of the power grid load nodes into a state time series, wherein the state change information includes voltage and power.

[0021] The load analysis module is used to perform component prediction analysis on the state time series using a preset load composition analysis model to obtain load composition analysis results. The preset load composition analysis model is trained using the DDPG algorithm.

[0022] Preferably, it further includes:

[0023] The information acquisition module is used to collect information on the status changes of power grid load nodes within a preset time period based on a preset frequency.

[0024] Preferably, it further includes:

[0025] The model building module is used to build an initial load composition analysis model based on a convolutional neural network.

[0026] The model training module is used to train the initial load composition analysis model using an analysis effect evaluation model built based on the DDPG algorithm, so as to obtain a preset load composition analysis model.

[0027] Preferably, the model training module is specifically used for:

[0028] The preset training dataset is input into the initial load composition analysis model to perform load analysis and obtain the load ratio adjustment amount.

[0029] The analysis effect evaluation model based on the DDPG algorithm generates load ratio adjustment evaluation results based on the load ratio adjustment amount;

[0030] Based on the load ratio adjustment amount and the load ratio adjustment evaluation results, the model is optimized and trained to generate a preset load composition analysis model.

[0031] A third aspect of this application provides a load composition analysis device, the device including a processor and a memory;

[0032] The memory is used to store program code and transmit the program code to the processor;

[0033] The processor is used to execute the load composition analysis method described in the first aspect according to the instructions in the program code.

[0034] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0035] This application provides a load composition analysis method, including: organizing the acquired state change information of power grid load nodes into a state time series, the state change information including voltage and power; performing component prediction analysis on the state time series using a preset load composition analysis model to obtain the load composition analysis result, the preset load composition analysis model being trained by the DDPG algorithm.

[0036] The load composition analysis method provided in this application studies and analyzes the state change information of power grid load nodes, and uses a pre-set load composition analysis model trained by the DDPG algorithm for load composition analysis. Because the DDPG algorithm has action and evaluation mechanisms, it ensures that the model is optimized into an analyzer that better reflects actual load changes. Furthermore, the end-to-end processing of the model is simpler, guaranteeing the timeliness and accuracy of the load composition analysis task. Therefore, this application can solve the technical problems of existing technologies, such as high complexity and poor analysis matching, leading to slow calculation speed and low accuracy in the analysis process. Attached Figure Description

[0037] Figure 1 A schematic flowchart illustrating a load composition analysis method provided in an embodiment of this application;

[0038] Figure 2 This is a schematic diagram of the structure of a load composition analysis device provided in an embodiment of this application;

[0039] Figure 3 This is a schematic diagram of the network structure of the preset load composition analysis model provided in the embodiments of this application;

[0040] Figure 4 The training result curve of the load composition analysis model provided for the application example of this application;

[0041] Figure 5 A schematic diagram of the load composition analysis system module provided for the application example of this application. Detailed Implementation

[0042] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0043] For easier understanding, please refer to Figure 1 An embodiment of a load composition analysis method provided in this application includes:

[0044] Step 101: Organize the obtained state change information of the power grid load nodes into a state time series. The state change information includes voltage and power.

[0045] State change information mainly refers to voltage and power. Obtaining state change information of power grid load nodes for load analysis provides a clearer picture of load changes, resulting in more accurate and reliable analysis results. Organizing state change information into a state time series facilitates input into the analysis model.

[0046] Furthermore, step 101, preceding the following, also includes:

[0047] Information on the status changes of power grid load nodes within a preset time period is collected based on a preset frequency.

[0048] In this embodiment, the preset frequency is 0.02Hz, and the preset sampling time period is 5000 seconds, that is, the state change information is collected within a time period of 5000 seconds. In addition, the sampling time interval in this embodiment is set to 50 seconds, and 100 data points of voltage and power change are collected to form state change information.

[0049] Furthermore, step 101, preceding the following, also includes:

[0050] The state change information is normalized based on a preset normalization formula.

[0051] Considering the numerical differences in power grid status change information, this information is normalized to facilitate subsequent research and analysis. The preset normalization formula in this embodiment is expressed as follows:

[0052]

[0053] in, Let i be the measured value of a certain state change information i at time t. This is the rated value for the state change information i. The state change information can be voltage, power, or other power-related parameters that can reflect load changes; there are no restrictions here.

[0054] Step 102: Perform component prediction analysis on the state time series using a preset load composition analysis model to obtain the load composition analysis results. The preset load composition analysis model is trained using the DDPG algorithm.

[0055] The preset load composition analysis model is a pre-trained model that can be trained offline. The training dataset includes state time series data with labels for load composition ratios. The model is trained using time series data such as voltage and power, as well as currently collected load composition ratios, enabling it to learn the analysis of load-related data and achieve good analytical results. In this embodiment, the load composition mainly considers static loads, motor loads, and distributed power supply loads; therefore, the load composition analysis results obtained based on the model are the calculated composition ratios of these three types of loads.

[0056] The DDPG algorithm, or Deep Deterministic Policy Gradient (DDPG), mainly includes two types of networks: Actor and Critic. These two networks continuously optimize their own network parameters during the iterative optimization process. However, their optimization objective functions are different, and their optimization results can influence each other, thus making the model generated after optimization training more accurate and reliable.

[0057] Furthermore, step 102, preceding the following, also includes:

[0058] An initial load composition analysis model was constructed based on a convolutional neural network.

[0059] The initial load composition analysis model was trained by the analysis effect evaluation model built based on the DDPG algorithm to obtain the preset load composition analysis model.

[0060] Furthermore, the initial load composition analysis model is trained using an analysis effect evaluation model built based on the DDPG algorithm to obtain a preset load composition analysis model, including:

[0061] Input the preset training dataset into the initial load composition analysis model to perform load analysis and obtain the load ratio adjustment amount;

[0062] The analysis effect evaluation model based on the DDPG algorithm generates load ratio adjustment evaluation results based on the load ratio adjustment amount;

[0063] Based on the load ratio adjustment amount and the load ratio adjustment evaluation results, the model is optimized and trained to generate a preset load composition analysis model.

[0064] It should be noted that you should refer to [link / reference]. Figure 3 The DDPG algorithm can be used to construct an analysis and evaluation model of load composition, and the constructed initial load composition analysis model can be optimized and trained. The output of the load composition analysis and evaluation model is the evaluation result of the adjustment of the current load composition ratio; the output of the initial load composition analysis model is the load ratio adjustment amount of the load composition.

[0065] In this embodiment, the proportional parameter adjustment amounts of the static load, motor load, and distributed power supply load, i.e., the load proportional adjustment amounts, are used as the output action space of the load composition analysis model. In the DDPG action space design, the action output of the load composition analysis model is defined as the correction amount of the load composition result. The parameter correction process is expressed as follows:

[0066]

[0067] in, Let i be the scaling factor of model i at step k+1. This is the correction amount for the proportional coefficient at step k+1 calculated by the model.

[0068] The purpose of load composition analysis is to find the optimal load composition result, so that the power system simulation results are close to the actual measurement results. Therefore, the effectiveness of the load composition analysis model can be evaluated by comparing the differences between the power system simulation results and the actual results, thus obtaining the load ratio adjustment evaluation result. The calculation process for the evaluation is expressed as follows:

[0069]

[0070] Where m and n are the types of state change information and the length of each type of data, respectively. The state change information i is the power system model simulation value at time t.

[0071] In this embodiment, the Actor module in the DDPG algorithm is based on the state matrix S. t (The matrix composed of input features) directly outputs the action matrix A. t :

[0072] A t =π(S) t )

[0073] The Critic model, on the other hand, determines the state S. t Next, execute action A t The long-term return Q is expressed as:

[0074] Q = Q(S) t A t )

[0075] Long-term return Q is determined by the Bellman equation:

[0076]

[0077] Where γ is the discount rate, a hyperparameter with a value range of [0, 1]. Its purpose is to discount future rewards to the current time step. For finite-length decision sequences, it can be set to 1; for infinite-length decision sequences, it takes a value less than 1 to prevent Q(S) from being discounted. t A t (differences to infinity.) R (·) indicates state S t+τ Next action A t+τ The reward function obtained.

[0078] Under the pattern to which the action matrix belongs, the long-term benefit Q can be iterated as:

[0079]

[0080] Among them, R t+1 D is the reward obtained at time t+1. t This is the action termination marker, and π(·) represents state S. t+1 The action policy function is given below. The above formula is the training objective of the Q-network, which is to continuously draw experience (S) from the experience pool. t A t ,S t+1 ,R t+1 D t The training of the π network is performed by π; while the training objective of the π network is to obtain the maximum long-term benefit Q.

[0081]

[0082] To avoid excessive fluctuations and difficulty in convergence during network model training, the DDPG algorithm employs a soft-update training method. At the start of training, the Q-network and π-network are copied to obtain the value target network Q. T and Value Target Network π T Two new networks are trained, and during training, the two target networks are updated only with minimal increments at regular intervals. The training objective of the Q-network is calculated using the target networks, thereby improving training stability. The loss function for training the two networks is expressed as:

[0083]

[0084]

[0085] Here, Batch is the empirical dataset extracted each time.

[0086] Then, gradient descent penalty is used to update the midweight parameters θ of the Q-network and π-network. Q and θ π Continuously reduce the value of the above loss function:

[0087]

[0088]

[0089] Where, θ Q θ π The weight parameters for the Q-network and π-network are lr and lr, respectively. Q ,lr π All are learning rates. The loss function in the Q network is θ Q The partial derivatives of the loss function in the π network with respect to θ π The partial derivatives. The update method for the weight parameters in the two target networks is expressed as:

[0090]

[0091]

[0092] Where, τ Q τ π These are all soft update rates, generally much less than 1.

[0093] For ease of understanding, this application uses a three-machine nine-node system as the test system. In the PSASP software, static load, motor load, distributed power supply model and their composition ratio values ​​are set to test the proposed load composition analysis scheme based on deep reinforcement learning. Table 1 shows the load composition ratio and its numerical distribution.

[0094] Table 1 Example of load composition ratio distribution

[0095] Load type Typical values ​​of the proportion Average distribution interval Search range static load 0.8 [0,1] [0,1] Motor load 0.1 [0,1] [0,1] Distributed power 0.1 [0,1] [0,1]

[0096] The DDPG algorithm was used for training, with a total of 4000 training epochs. Figure 4 The training effect evaluation index and the change of average reward value with the number of training rounds are given. As can be seen from the figure, the effect is more obvious in the initial stage of training, the training error index decreases faster, and the average reward value increases faster.

[0097] Based on the load composition analysis model generated through training, a set of test scenarios was randomly selected to test the accuracy of the model. The results are shown in Table 2.

[0098] Table 2 Example of results from randomly selected test groups

[0099] Load type actual value Identification value relative error rate static load 0.583 0.559 4.11％ Motor load 0.212 0.234 10.38％ Distributed power 0.205 0.207 0.98％

[0100] As shown in Table 2, the load composition analysis model trained based on the DDPG algorithm can effectively identify the proportions of load components in the system. Furthermore, it should be noted that the calculation time of the load composition analysis model for the tested scenario is only 453.25 milliseconds, indicating that its calculation speed meets the requirements for online load composition analysis, thus ensuring both timeliness and accuracy.

[0101] It should be noted that, in practice, the method of this application can be used to build a system, mainly including a measurement data stage, a data caching stage, an offline training stage, and an application stage. For details, please refer to [link / reference needed]. Figure 5The measurement data stage collects and records data on changes in the power grid's operating status and generates sample data for the data caching stage. The data caching stage stores the data samples needed to train the load composition analysis model. The offline training stage, through interaction with the data storage module, enables offline training of the load composition analysis model. The application stage relies on the measured state change data in the power grid to estimate the load composition and output the load composition results. The offline training stage requires state characteristics and action reward results from the data caching stage as input to correct the model parameters of the load composition analysis model, thus achieving offline training. Based on the offline training model of load composition analysis, the online load composition analysis model can be generated / updated.

[0102] The load composition analysis method provided in this application analyzes the state change information of power grid load nodes and uses a preset load composition analysis model trained by the DDPG algorithm for load composition analysis. Because the DDPG algorithm has action and evaluation mechanisms, it ensures that the model is optimized into an analyzer that better reflects actual load changes. Furthermore, the end-to-end processing of the model is simpler, guaranteeing the timeliness and accuracy of the load composition analysis task. Therefore, this application solves the technical problems of existing technologies, such as high complexity and poor analysis matching, leading to slow calculation speed and low accuracy in the analysis process.

[0103] For easier understanding, please refer to Figure 2 This application provides an embodiment of a load composition analysis device, comprising:

[0104] The sequence extraction module 201 is used to organize the acquired state change information of the power grid load node into a state time series, and the state change information includes voltage and power.

[0105] The load analysis module 202 is used to perform component prediction analysis on the state time series using a preset load composition analysis model to obtain the load composition analysis results. The preset load composition analysis model is trained by the DDPG algorithm.

[0106] Furthermore, it also includes:

[0107] The information acquisition module 203 is used to collect information on the status changes of power grid load nodes within a preset time period based on a preset frequency.

[0108] Furthermore, it also includes:

[0109] Model building module 204 is used to build an initial load composition analysis model based on a convolutional neural network;

[0110] The model training module 205 is used to train the initial load composition analysis model using the analysis effect evaluation model built based on the DDPG algorithm, so as to obtain the preset load composition analysis model.

[0111] Furthermore, the model training module 202 is specifically used for:

[0112] Input the preset training dataset into the initial load composition analysis model to perform load analysis and obtain the load ratio adjustment amount;

[0113] The analysis effect evaluation model based on the DDPG algorithm generates load ratio adjustment evaluation results based on the load ratio adjustment amount;

[0114] Based on the load ratio adjustment amount and the load ratio adjustment evaluation results, the model is optimized and trained to generate a preset load composition analysis model.

[0115] This application also provides a load composition analysis device, which includes a processor and a memory;

[0116] The memory is used to store program code and transfer the program code to the processor;

[0117] The processor is used to execute the load composition analysis method in the above method embodiments according to the instructions in the program code.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0120] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0121] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the methods described in the various embodiments of this application through a computer device (which may be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0122] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for analyzing load composition, characterized in that, include: The acquired state change information of the power grid load nodes is organized into a state time series, and the state change information includes voltage and power. An initial load composition analysis model was constructed based on a convolutional neural network. The initial load composition analysis model is trained using an analysis effect evaluation model built based on the DDPG algorithm to obtain a preset load composition analysis model, which specifically includes: The preset training dataset is input into the initial load composition analysis model to perform load analysis and obtain the load ratio adjustment amount. The analysis effect evaluation model based on the DDPG algorithm generates load ratio adjustment evaluation results based on the load ratio adjustment amount; Based on the load ratio adjustment amount and the load ratio adjustment evaluation results, the model is optimized and trained to generate a preset load composition analysis model. The preset load composition analysis model is used to perform component prediction analysis on the state time series to obtain the load composition analysis results. The preset load composition analysis model is trained by the DDPG algorithm.

2. The load composition analysis method according to claim 1, characterized in that, The acquired state change information of the power grid load nodes is organized into a state time series. The state change information includes voltage and power, and the preceding steps also include: Information on the status changes of power grid load nodes within a preset time period is collected based on a preset frequency.

3. The load composition analysis method according to claim 1, characterized in that, The acquired state change information of the power grid load nodes is organized into a state time series. The state change information includes voltage and power, and the preceding steps also include: The state change information is normalized based on a preset normalization formula.

4. A load composition analysis device, characterized in that, include: The sequence extraction module is used to organize the acquired state change information of the power grid load nodes into a state time series, wherein the state change information includes voltage and power. The model building module is used to build an initial load composition analysis model based on a convolutional neural network. The model training module is used to train the initial load composition analysis model using an analysis effect evaluation model built based on the DDPG algorithm, to obtain a preset load composition analysis model. Specifically, the model training module is used for: The preset training dataset is input into the initial load composition analysis model to perform load analysis and obtain the load ratio adjustment amount. The analysis effect evaluation model based on the DDPG algorithm generates load ratio adjustment evaluation results based on the load ratio adjustment amount; Based on the load ratio adjustment amount and the load ratio adjustment evaluation results, the model is optimized and trained to generate a preset load composition analysis model. The load analysis module is used to perform component prediction analysis on the state time series using the preset load composition analysis model to obtain the load composition analysis results. The preset load composition analysis model is trained by the DDPG algorithm.

5. The load composition analysis device according to claim 4, characterized in that, Also includes: The information acquisition module is used to collect information on the status changes of power grid load nodes within a preset time period based on a preset frequency.

6. A load composition analysis device, characterized in that, The device includes a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the load composition analysis method according to any one of claims 1-3 according to the instructions in the program code.

Images