A load forecasting method and related device based on information fusion

Abstract

The present application discloses a load forecasting method and related device based on information fusion. The method includes: obtaining load data and text information in a distribution network automation system, where the text information includes holidays, production cycles, maintenance plans, and emergency repair tasks; performing feature calculations on the load data based on a factorization algorithm and a parallel branch calculation method to obtain initial data features; performing vectorization conversion on the text information based on a one-hot encoding algorithm to obtain text vector features; splicing the initial data features and the text vector features and inputting them into a preset prediction network model for prediction to obtain a load forecasting result, and the preset prediction network model includes a CNN layer, an LSTM layer, and a preset residual connection structure. The present application solves the technical problem that the existing load forecasting methods lack research on text information and detailed optimization of the prediction model, resulting in the lack of accuracy and reliability of the load forecasting results.

Description

Technical Field

[0001] This application relates to the technical field of load forecasting, and in particular, to a load forecasting method and related devices based on information fusion. Background Art

[0002] Short-term load forecasting is the basis for realizing the safe and economic operation and scientific management of power systems. The size of the forecasting error is directly related to the subsequent safety checking and analysis of the power grid. Therefore, load forecasting is of great significance for power grid situation awareness, load dispatching, and distribution network emergency repair. Currently, the main methods for short-term load forecasting are divided into two categories: artificial intelligence-based methods and statistics-based methods. Among them, the statistics-based methods include linear regression, autoregression, and multiple linear regression, etc. This method has good stability, but the data processing scale is small, the requirements for original data are high, and it is difficult to apply to the current situation of emerging massive data. The artificial intelligence-based methods include artificial neural networks, grey relational algorithms, bionic intelligent algorithms, etc. The prediction models of this type of method are constructed on the basis of complex theories. Therefore, the selection of some parameters or weights in the model will affect the stability of the algorithm.

[0003] The current research focus of the methods lies in data processing, that is, the load data level. However, the load fluctuation is greatly affected by the environment, such as factors like weather and holidays. Moreover, the models in the artificial intelligence-based methods lack detail optimization to a certain extent, and the stable reliability of the prediction model cannot be ensured. Summary of the Invention

[0004] This application provides a load forecasting method and related devices based on information fusion, which are used to solve the technical problems that the existing load forecasting methods lack research on text information and detail optimization of the forecasting model, resulting in the lack of accuracy and reliability of the load forecasting results.

[0005] In view of this, the first aspect of this application provides a load forecasting method based on information fusion, including:

[0006] Obtain load data and text information in the distribution network automation system, where the text information includes holidays, production cycles, maintenance plans, and emergency repair tasks;

[0007] Perform feature calculation on the load data based on the factorization algorithm and the parallel branch calculation method to obtain initial data features;

[0008] Perform vector quantization conversion on the text information based on the one-hot encoding algorithm to obtain text vector features;

[0009] Concatenate the initial data features and the text vector features and input them into a preset prediction network model for prediction to obtain a load forecasting result, where the preset prediction network model includes a CNN layer, an LSTM layer, and a preset residual connection structure.

[0010] Preferably, the load data is subjected to feature calculation based on the factorization algorithm and the parallel branch calculation method to obtain initial data features, including:

[0011] Randomly decompose the load data into multiple matrices based on the factorization algorithm, and perform matrix filling processing on each matrix respectively to obtain a load matrix;

[0012] Perform feature mining calculation on the load matrix respectively based on the parallel branch calculation method to obtain branch features;

[0013] Integrate all the branch features to obtain initial data features.

[0014] Preferably, the text information is vectorized and transformed based on the hot encoding algorithm to obtain text vector features, including:

[0015] Map the text information to text discrete values;

[0016] Expand the text discrete values to the Euclidean space based on the hot encoding algorithm to obtain text vector features.

[0017] Preferably, before inputting the initial data features and the text vector features into a preset prediction network model for prediction to obtain a load prediction result, it further includes:

[0018] Construct an initial prediction network model based on a CNN network, an LSTM network, and a preset residual connection structure;

[0019] Add a preset step size information factor to the initial prediction network model to obtain a preset prediction network model.

[0020] The second aspect of the present application provides a load prediction device based on information fusion, including:

[0021] An information acquisition module, configured to acquire load data and text information in a distribution network automation system, where the text information includes holidays, production cycles, maintenance plans, and emergency repair tasks;

[0022] A feature calculation module, configured to perform feature calculation on the load data based on the factorization algorithm and the parallel branch calculation method to obtain initial data features;

[0023] A feature conversion module, configured to perform vectorization conversion on the text information based on the hot encoding algorithm to obtain text vector features;

[0024] A load prediction module, configured to splice the initial data features and the text vector features and input them into a preset prediction network model for prediction to obtain a load prediction result. The preset prediction network model includes a CNN layer, an LSTM layer, and a preset residual connection structure.

[0025] Preferably, the feature calculation module is specifically configured to:

[0026] Randomly decompose the load data into multiple matrices based on a factorization algorithm, and perform matrix filling processing on each of them to obtain a load matrix;

[0027] Perform feature mining calculations on the load matrix respectively based on a parallel branch calculation method to obtain branch features;

[0028] Integrate all the branch features to obtain initial data features.

[0029] Preferably, the feature conversion module is specifically configured to:

[0030] Map the text information to text discrete values;

[0031] Expand the text discrete values to the Euclidean space based on a one-hot encoding algorithm to obtain text vector features.

[0032] Preferably, it further includes:

[0033] A model construction module, configured to construct an initial prediction network model based on a CNN network, an LSTM network, and a preset residual connection structure;

[0034] A model optimization module, configured to add a preset step size information factor to the initial prediction network model to obtain a preset prediction network model.

[0035] The third aspect of the present application provides a load prediction device based on information fusion. The device includes a processor and a memory;

[0036] The memory is used to store program codes and transmit the program codes to the processor;

[0037] The processor is used to execute the load prediction method based on information fusion described in the first aspect according to the instructions in the program codes.

[0038] The fourth aspect of the present application provides a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the load prediction method based on information fusion described in the first aspect.

[0039] From the above technical solutions, it can be seen that the embodiments of the present application have the following advantages:

[0040] In this application, a load forecasting method based on information fusion is provided, including: obtaining load data and text information in the distribution network automation system, where the text information includes holidays, production cycles, maintenance plans, and emergency repair tasks; calculating the feature of the load data based on the factorization algorithm and the parallel branch calculation method to obtain the initial data features; performing vectorization conversion on the text information based on the one-hot encoding algorithm to obtain the text vector features; splicing the initial data features and the text vector features and inputting them into a preset prediction network model for prediction to obtain the load forecasting result, and the preset prediction network model includes a CNN layer, an LSTM layer, and a preset residual connection structure.

[0041] The load forecasting method based on information fusion provided by this application takes into account both the load data and the text information related to load forecasting, such as holidays, production cycles, etc.; analyzes from the environmental factors affecting the forecasting result to ensure that the forecasting result is more in line with the actual situation and more accurate and reliable; and in the preset prediction network model, there is not only a CNN layer for data feature analysis and an LSTM layer for time series analysis of text information, but also a preset residual connection structure is added for model optimization, making the model more targeted, and thus double guaranteeing the reliability of the load forecasting result. Therefore, this application can solve the technical problems that the existing load forecasting methods lack research on text information and detailed optimization of the forecasting model, resulting in the lack of accuracy and reliability of the load forecasting result. Description of the Drawings

[0042] Figure 1 It is a schematic flow chart of a load forecasting method based on information fusion provided by an embodiment of this application;

[0043] Figure 2 It is a schematic structural diagram of a load forecasting device based on information fusion provided by an embodiment of this application;

[0044] Figure 3 It is a schematic flow chart of the initial mining process of load data provided by an embodiment of this application;

[0045] Figure 4 It is a schematic structural diagram of the preset prediction network model provided by an embodiment of this application;

[0046] Figure 5 It is a schematic diagram of the improved structure of the LSTM network provided by an embodiment of this application. Detailed Embodiments

[0047] To enable those skilled in the art to better understand the solution of this application, the following will clearly and completely describe the technical solution in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

[0048] For ease of understanding, please refer to Figure 1 , an embodiment of a load forecasting method based on information fusion provided by this application, includes:

[0049] Step 101, obtain load data and text information in the distribution network automation system, where the text information includes holidays, production cycles, maintenance plans, and emergency repair tasks.

[0050] When obtaining load data and text information from the distribution network automation system, it should be noted that the data dimensions should be kept the same during the acquisition of different types of data, and all information has a recording time for subsequent research and analysis.

[0051] The load data generally includes: historical load, current load, electricity price, power flow, meteorological data, etc., and the text information includes: holidays, production cycles, maintenance plans, and emergency repair tasks, etc. Specifically, other types of data can be added according to the actual situation, and specific limitations are not made.

[0052] Step 102, perform feature calculation on the load data based on the factorization algorithm and the parallel branch calculation method to obtain initial data features.

[0053] Further, step 102 includes:

[0054] Randomly decompose the load data into multiple matrices based on the factorization algorithm, and perform matrix filling processing on each matrix respectively to obtain a load matrix;

[0055] Perform feature mining calculations on the load matrix respectively based on the parallel branch calculation method to obtain branch features;

[0056] Integrate all the branch features to obtain initial data features.

[0057] In order to enable the initial data features to maintain the model optimization gradient and rate after being input into the model, the factorization algorithm and the parallel branch calculation method are used here to perform initialization feature mining processing on the load data. Factorization is a processing operation based on the idea of matrix decomposition, which can embed large-dimensional feature vectors into a low-dimensional vector space, reducing the complexity of data and the model, and can effectively improve the prediction accuracy. The parallel branch calculation method refers to the process of separately mining feature information from the decomposed spatial vectors.

[0058] Please refer to Figure 3 , the load data can be factorized into three different sets of sub - data. The data form of these sub - data is a matrix. After factorization, matrix filling can be appropriately performed to keep the matrix form consistent. The factorization process is random, as long as it is ensured that the information after matrix superposition is consistent with the initial load data. The three sets of sub - data are calculated using different kernels to extract features at different levels of the load data, namely branch features. According to the superposition integration strategy, the parallel branch features are comprehensively considered to obtain the initial data features. Generally, the splicing method can be used for integration, and specifically, it can also be designed according to needs.

[0059] Step 103: Perform vector quantization transformation on the text information based on the one - hot encoding algorithm to obtain text vector features.

[0060] Further, step 103 includes:

[0061] Map the text information to text discrete values;

[0062] Based on the one - hot encoding algorithm, expand the text discrete values to the Euclidean space to obtain text vector features.

[0063] Since the amount of information contained in text information is often not in one dimension, it is difficult to fuse with digital information for load prediction analysis. In this embodiment, the text information is vectorized based on the one - hot encoding algorithm, unifying the text information and digital information to a specific level, which is also convenient for subsequent model prediction operations. Moreover, the one - hot encoding method considers both the discrimination effect and the computational complexity, and has good characteristics.

[0064] In this embodiment, an example is given. If the text information is a holiday, then the holiday text information of the device can actually be represented as D type ={working day, weekend, holiday, ……}, and the vector quantization process of the corresponding one - hot encoding algorithm is:

[0065]

[0066] where, len(D type ) represents a dimension of the matrix D type . The text elements are encoded as row vectors, and the type of time is correspondingly represented by row vectors; or, the text state of the power distribution line can be vectorized. For example, the description "very poor insulation condition, large historical load change" can be represented by a row vector. It can be understood that after the text data is vectorized, the dimension needs to be further expanded to the same dimension as the initial data features.

[0067] For each piece of text information, if there are m values, after being processed by the one-hot encoding algorithm, they can be converted into m binary features, and each binary feature has a corresponding encoding. For example, for the three indicators of good, medium, and poor of a device, the conversions are 100, 010, and 001 respectively; and the above holidays can also be represented in this way. In addition, multiple features are expressed using concatenated encoding. For example, "device feature + date feature" is expressed as "100010", which can be parsed as "device number + weekend".

[0068] Step 104: Concatenate the initial data features and the text vector features and input them into the preset prediction network model for prediction to obtain the load prediction result. The preset prediction network model includes a CNN layer, an LSTM layer, and a preset residual connection structure.

[0069] Please refer to Figure 4 , the CNN layer in the preset prediction network model has a good effect on analyzing the initial data features, while the LSTM layer has a good effect on analyzing the temporal correlation of the text vectors. In addition, to solve the problem of model gradient diffusion, in this embodiment, a bypass residual connection structure, that is, a preset residual connection structure, is added to the preset prediction network model to optimize the model.

[0070] The initial data features and the text vector features can be concatenated into a matrix and then input into the preset prediction network model for feature analysis and load prediction.

[0071] Furthermore, before step 104, it also includes:

[0072] Construct an initial prediction network model based on the CNN network, the LSTM network, and the preset residual connection structure;

[0073] Add a preset step size information factor to the initial prediction network model to obtain the preset prediction network model.

[0074] Please refer to Figure 4 , which includes a multi-layer network structure, Z is the network output, Y is the output of the third layer of the network, and the specific feature processing relationship can be expressed as:

[0075] Y = z n-1 + x1 + x2

[0076] Among them, Z n-1 is the output of the (n - 1)th layer, and x1 and x2 respectively represent the initial data features and the text vector features. If the loss function of the network model in this embodiment is denoted as L(·), then the derivatives of the loss function with respect to x1 and x2 can be obtained:

[0077]

[0078]

[0079] Figure 4 The outputs of the CNN layer and the network state are as follows:

[0080]

[0081] s t 、s t-1 are the outputs of the neurons in the t-th layer and the (t - 1)-th layer respectively, and w ss 、w sx 、w ys are all convolutional layer weight matrices, x t-1 is the input of the (t - 1)-th layer, b s 、b y are all bias matrices, g(·) is the activation function, and y t is the output of the CNN layer.

[0082] In the LSTM layer, the selection unit controls the accumulation of internal information in the recurrent structure, can selectively forget information to prevent algorithm overload. For the specific framework, please refer to Figure 5 . The preset step size information factor can be interpreted as follows: When the output of the forget gate in the LSTM layer is 0, it means all the information of the previous state is discarded; when the output of the forget gate is 1, it means all the information of the previous state is retained, enabling the network to reconsider the previous information in later calculations. This is a kind of information across time steps, which can alleviate the problem of model gradient disappearance and use LSTM to strengthen the correlation analysis of historical data to improve the prediction accuracy. In addition, it should be noted that: The forget gate determines which part of the data at the previous moment needs to be forgotten, the input gate determines which part of the current input needs to be retained in the state, and the output gate determines the system input at the current moment and the input at the previous moment. The specific expressions are as follows:

[0083] i t =σ(W xi x t +W hi h t-1 +W ci c t-1 +b i )

[0084] f t =σ(W xf x t +W hf h t-1 +W cf c t-1 +b f )

[0085] c t =f t c t-1 +i t tanh(Wxc x t +W hc h t-1 +b c )

[0086] o t =σ(W xo x t +W ho h t-1 +W co c t-1 +b o )

[0087] h t =o t tanh(c t )

[0088] Among them, i t , f t , c t , c t-1 , o t are process variables (output of sigmod), forget gate, cell states at time t and t-1, and output gate respectively; W · are all weights of the network layer, where "·" are different subscripts, b · are all bias values, σ(·) is the sigmod function, x t is the input of the LSTM layer, and h · is the cell output.

[0089] The load forecasting method based on information fusion provided by the embodiments of the present application takes into account both load data and text information related to load forecasting, such as holidays, production cycles, etc.; analyzes from environmental factors affecting the forecasting results to ensure that the forecasting results are more in line with the actual situation and more accurate and reliable; and the pre-set forecasting network model not only has a CNN layer for data feature analysis and an LSTM layer for time series analysis of text information, but also adds a pre-set residual connection structure for model optimization, making the model more targeted, and thus doubling the reliability of the load forecasting results. Therefore, the embodiments of the present application can solve the technical problems that the existing load forecasting methods lack research on text information and detailed optimization of the forecasting model, resulting in inaccurate and unreliable load forecasting results.

[0090] For ease of understanding, please refer to Figure 2 , and the embodiments of the present application provide a load forecasting device based on information fusion, including:

[0091] An information acquisition module 201, configured to acquire load data and text information in a distribution network automation system, where the text information includes holidays, production cycles, maintenance plans, and emergency repair tasks;

[0092] A feature calculation module 202, configured to perform feature calculation on load data based on a factorization algorithm and a parallel branch calculation method to obtain initial data features;

[0093] A feature conversion module 203, configured to perform vector quantization conversion on text information based on a one-hot encoding algorithm to obtain text vector features;

[0094] A load prediction module 204, configured to splice the initial data features and the text vector features and input them into a preset prediction network model for prediction to obtain a load prediction result, where the preset prediction network model includes a CNN layer, an LSTM layer, and a preset residual connection structure.

[0095] Further, the feature calculation module 202 is specifically configured to:

[0096] Randomly decompose the load data into multiple matrices based on a factorization algorithm, and perform matrix filling processing on each of them to obtain load matrices;

[0097] Perform feature mining calculations on the load matrices respectively based on a parallel branch calculation method to obtain branch features;

[0098] Integrate all the branch features to obtain initial data features.

[0099] Further, the feature conversion module 203 is specifically configured to:

[0100] Map the text information to text discrete values;

[0101] Expand the text discrete values to the Euclidean space based on a one-hot encoding algorithm to obtain text vector features.

[0102] Further, it further includes:

[0103] A model construction module 205, configured to construct an initial prediction network model based on a CNN network, an LSTM network, and a preset residual connection structure;

[0104] A model optimization module 206, configured to add a preset step size information factor to the initial prediction network model to obtain a preset prediction network model.

[0105] This application also provides a load prediction device based on information fusion, where the device includes a processor and a memory;

[0106] The memory is used to store program codes and transmit the program codes to the processor;

[0107] The processor is used to execute the load prediction method based on information fusion in the above method embodiments according to the instructions in the program codes.

[0108] The present application also provides a computer-readable storage medium for storing program code for executing the load prediction method based on information fusion in the above method embodiments.

[0109] In several embodiments provided by the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the apparatus or unit can be in an electrical, mechanical or other form.

[0110] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0111] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit exists physically alone, or two or more units are integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0112] If the above integrated unit is implemented in the form of 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 the present 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. The 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 each embodiment of the present application by a computer device (which can be a personal computer, a server, or a network device, etc.). The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (English full name: Read-Only Memory, English abbreviation: ROM), random access memories (English full name: Random Access Memory, English abbreviation: RAM), magnetic disks or optical discs that can store program code.

[0113] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A load forecasting method based on information fusion, characterized in that, Including: Obtain load data and text information in the distribution network automation system, where the text information includes holidays, production cycles, maintenance plans, and emergency repair tasks; Perform feature calculation on the load data based on the factorization algorithm and the parallel branch calculation method to obtain initial data features. The specific process is as follows: Randomly decompose the load data into multiple matrices based on the factorization algorithm, and perform matrix filling processing on each matrix respectively to obtain a load matrix; Perform feature mining calculations on the load matrix respectively based on the parallel branch calculation method to obtain branch features; Integrate all the branch features to obtain initial data features; Perform vector quantization conversion on the text information based on the one-hot encoding algorithm to obtain text vector features. The specific process is as follows: Map the text information to text discrete values; Expand the text discrete values to the Euclidean space based on the one-hot encoding algorithm to obtain text vector features; Concatenate the initial data features and the text vector features and input them into a pre-set prediction network model for prediction to obtain a load prediction result. The pre-set prediction network model includes a CNN layer, an LSTM layer, and a pre-set residual connection structure.

2. The load prediction method based on information fusion according to claim 1, wherein Before the step of concatenating the initial data features and the text vector features and inputting them into a pre-set prediction network model for prediction to obtain a load prediction result, it further includes: Construct an initial prediction network model based on a CNN network, an LSTM network, and a pre-set residual connection structure; Add a pre-set step size information factor to the initial prediction network model to obtain a pre-set prediction network model.

3. A load forecasting device based on information fusion, characterized in that Including: An information acquisition module for obtaining load data and text information in the distribution network automation system, where the text information includes holidays, production cycles, maintenance plans, and emergency repair tasks; A feature calculation module for performing feature calculation on the load data based on the factorization algorithm and the parallel branch calculation method to obtain initial data features. The feature calculation module is specifically used for: Randomly decompose the load data into multiple matrices based on the factorization algorithm, and perform matrix filling processing on each matrix respectively to obtain a load matrix; Perform feature mining calculations on the load matrix respectively based on the parallel branch calculation method to obtain branch features; Integrate all the branch features to obtain initial data features; A feature conversion module for performing vector quantization conversion on the text information based on the one-hot encoding algorithm to obtain text vector features. The feature conversion module is specifically used for: Map the text information to text discrete values; Expand the text discrete values to the Euclidean space based on the one-hot encoding algorithm to obtain text vector features; A load prediction module for concatenating the initial data features and the text vector features and inputting them into a pre-set prediction network model for prediction to obtain a load prediction result. The pre-set prediction network model includes a CNN layer, an LSTM layer, and a pre-set residual connection structure.

4. The load prediction device based on information fusion according to claim 3, characterized in that, It further includes: A model construction module for constructing an initial prediction network model based on a CNN network, an LSTM network, and a pre-set residual connection structure; A model optimization module for adding a pre-set step size information factor to the initial prediction network model to obtain a pre-set prediction network model.

5. A load forecasting device based on information fusion, characterized in that, The device includes a processor and a memory; The memory is used for storing program code and transmitting the program code to the processor; The processor is used for executing the information fusion-based load forecasting method according to any one of claims 1-2 based on the instructions in the program code.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used for storing program code, and the program code is used for executing the information fusion-based load forecasting method according to any one of claims 1-2.

Images