Power load prediction method and system

Abstract

The invention relates to a power load prediction method and system, and the method comprises the following steps: collecting power load related data, and carrying out the preprocessing of the data; extracting multi-source features from the preprocessed power load related data and fusing the multi-source features to obtain a comprehensive feature vector; training the power load prediction model by taking the comprehensive feature vector as the input of the power load prediction model and taking a power load true value at a historical moment as a label value; and predicting the power load by using the trained power load prediction model.

Description

Technical Field

[0001] The present application relates to the technical field of power load forecasting, and more specifically, to a power load forecasting method and system. Background Art

[0002] Power load forecasting is crucial in the stable operation and efficient management of power systems. Accurate forecasting can help power companies optimize power generation plans, reasonably arrange grid dispatching, effectively reduce operating costs and improve power supply reliability. Traditional forecasting methods are often limited to a single data source. For example, relying solely on historical power load data, it is difficult to fully reflect the complex factors that affect load changes, resulting in poor forecasting accuracy. With the continuous expansion of the scale of power systems and the increasing diversity of power consumption scenarios, factors such as meteorological fluctuations and social and economic development have increasingly significant effects on power loads. For example, the demand for electricity for refrigeration equipment surges in hot weather, and the growth of industrial output value will drive up production electricity consumption. Therefore, the power load forecasting method based on comprehensive multi-source data has become a research hotspot, but the existing technology still has room for improvement in data fusion, model construction, and training optimization. New methods are urgently needed to improve forecasting accuracy and stability to meet the needs of modern power system operation.

[0003] Prior art, such as the Chinese patent application with publication number "CN117856240A", discloses a power grid load forecasting method based on multi-source data fusion, including the following steps: Step S1: Acquire multi-source data and perform fusion processing; Step S2: Perform data decomposition and preprocessing on the fused multi-source data; Step S3: Based on the preprocessed data, feature engineering is used to extract key features and construct a load history data set; Step S4: Construct a multi-layer Encoder-Decoder structure model, combine the self-attention mechanism and frequency domain analysis, and train based on the load history data set to obtain a power grid load forecasting model; Step S5: Preprocess the multi-source data acquired in real time and input it into the power grid load forecasting model to obtain a load fluctuation prediction value. This invention can more comprehensively capture the various factors that affect the change of power load and improve the accurate prediction of future load fluctuations in the power system.

[0004] The problem with the above-mentioned existing technology is that, in the data processing flow, this method decomposes and preprocesses after fusion, which may increase data complexity and the risk of information loss; in addition, the multi-layer Encoder-Decoder structure combined with the self-attention mechanism and frequency domain analysis has certain defects in efficiency and effectiveness in capturing long sequence dependencies and complex factor associations. Summary of the invention

[0005] In order to solve the above technical problems, the present invention proposes a method and system for predicting power load.

[0006] The technical solution of the present invention is as follows:

[0007] The present invention provides a method for predicting power load, comprising the following steps:

[0008] Step S1, collecting power load related data and performing preprocessing;

[0009] Step S2, extracting and fusing multi-source features from the pre-processed power load related data to obtain a comprehensive feature vector;

[0010] Step S3, using the comprehensive feature vector as the input of the power load forecasting model and the real value of the power load at the historical moment as the label value to train the power load forecasting model;

[0011] Step S4, using the trained power load prediction model to predict the power load.

[0012] As a preferred implementation manner, the power load related data include: power system operation data, meteorological data and socio-economic data; the power system operation data include: historical power load data, power grid topology data and power generation equipment operation status data; meteorological data include: temperature, humidity, wind speed and sunshine time; socio-economic data include: regional gross domestic product, population size, industrial output value.

[0013] As a preferred implementation, the multi-source features are extracted and fused from the pre-processed power load related data to obtain a comprehensive feature vector, the expression of which is as follows:

[0014] F(t)=[X elec (t) T ,X weather (t) T ,X socior (t) T ];

[0015] Where, X elec (t) is the n1 features extracted from the power system operation data; X weather (t) is the n2 features extracted from the meteorological data; X socior (t) is n3 features extracted from socioeconomic data; the dimension of the comprehensive feature vector of F(t) is n, n=n1+n2+n3.

[0016] As a preferred implementation, the power load forecasting model adopts an improved Transformer model, which includes improvements to position encoding, improvements to the multi-head attention mechanism, and addition of residual connections and layer normalization; wherein:

[0017] The improved position encoding expression is:

[0018]

[0019] Where: PE pos,i is the value of the improved position encoding at position pos and in the i-th dimension; α is the learnable parameter weight, which is used to balance the learnable parameter matrix W pos and original position encoding; W pos,i is the learnable parameter matrix of the i-th dimension, which is used to adjust the representation of position encoding; For the improved position encoding PE pos,i the extent of contribution;

[0020] Improvements to the multi-head attention mechanism include increasing the number of attention heads; and using an adaptive weight assignment method to calculate the weight ω for each attention head r r , the specific formula is as follows:

[0021]

[0022] Where: r is the weight of the rth attention head; β r is the learnable parameter of the rth attention head; Score(head r ) is the score of the rth attention head; h ′ is the number of attention heads after improvement; β b is the learnable parameter of the bth attention head; Score(head b ) is the score of the b-th attention head;

[0023] Add residual connections and layer normalization. In each layer, the input x passes through the multi-head attention mechanism and the feedforward network to obtain the output y. The residual connection is:

[0024] y res =x+y;

[0025] The layer normalization formula is:

[0026]

[0027] Where: γ, β are learnable parameters; σ 2 , μ are y res The variance and mean of ; ε is a constant to prevent the denominator from being zero.

[0028] As a preferred implementation, the Transformer model uses mean square error as the loss function, specifically:

[0029]

[0030] Where: L is the mean square error function value; N is the number of samples; v is the sample index; t v The power load forecast value of the Transformer model at time L(t v ) is t v The actual value of the power load at the moment.

[0031] As a preferred implementation, during the training process of the power load forecasting model, it also includes using an Adam optimizer to update the learnable parameters in the power load forecasting model. The specific formula is as follows:

[0032]

[0033] Where: θ s is the value of the learnable parameter θ at the sth iteration; ∈ is the learning rate; is the bias-corrected first-order moment estimate; is the bias-corrected second-order moment estimate; τ is a constant to prevent the denominator from being zero.

[0034] On the other hand, the present invention also provides a power load forecasting system, comprising:

[0035] Data collection and preprocessing module, collects power load related data and performs preprocessing;

[0036] The feature extraction and fusion module extracts and fuses multi-source features from the pre-processed power load-related data to obtain a comprehensive feature vector;

[0037] The model training module takes the comprehensive feature vector as the input of the power load forecasting model, predicts the real value of the power load at the historical moment as the output, and trains the power load forecasting model;

[0038] The model prediction module uses the trained power load prediction model to predict the power load.

[0039] As a preferred implementation, in the data acquisition and preprocessing module, the power load related data include: power system operation data, meteorological data and socio-economic data; the power system operation data include: historical power load data, power grid topology data and power generation equipment operation status data; meteorological data include: temperature, humidity, wind speed and sunshine time; socio-economic data include: regional gross domestic product, population size, industrial output value.

[0040] On the other hand, the present invention further provides an electronic device having a computer program stored thereon, wherein when the computer program is executed by a processor, a method for predicting electric load as described in any embodiment of the present invention is implemented.

[0041] On the other hand, the present invention also provides a computer-readable medium for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement a power load forecasting method as described in any embodiment of the present invention.

[0042] The present invention has the following beneficial effects: In data processing, it collects multi-source data such as power system operation, meteorology, and social economy and pre-processes them reasonably, comprehensively covering various factors affecting the load, and laying the foundation for accurate prediction. In the feature extraction and fusion link, multi-source features are innovatively extracted and scientifically spliced ​​into comprehensive feature vectors to fully explore the intrinsic correlation of the data. The improved Transformer model greatly improves the ability to capture long sequence dependencies and the efficiency and stability of model training by optimizing position encoding, multi-head attention mechanism, and adding residual connections and layer normalization. At the same time, the use of mean square error as the loss function and combined with the Adam optimizer can effectively adjust the model parameters, improve prediction accuracy, enhance the generalization ability of the model, and provide strong technical support for the reliable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] It should be understood that the step numbers used in this document are only for convenience of description and are not intended to limit the order in which the steps are executed.

[0047] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0048] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0049] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0050] Embodiment 1:

[0051] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following will be combined with the specific embodiments of the present application and refer to the attached Figure 1 , clearly and completely describe the technical solution of the present invention.

[0052] To solve the problems in the prior art, the present invention provides a method for predicting power load, comprising the following steps:

[0053] Step S1, collecting power load related data and performing preprocessing;

[0054] The power load related data include: power system operation data, meteorological data and socio-economic data; the power system operation data include: historical power load data, power grid topology data and power generation equipment operation status data; meteorological data include: temperature, humidity, wind speed and sunshine time; socio-economic data include: regional gross domestic product, population size, industrial output value.

[0055] In this embodiment, take the power system operation data as an example:

[0056] Step S101, collect historical power load data A(t), where t is a time point; grid topology data G, which can be represented as a graph structure G=(E,V), where V is a node set (representing generators, transformers, load nodes, etc.), and E is an edge set (representing transmission lines); power generation equipment operating status data E(t), including the active power P of the generator g (t), reactive power Q g (t), power generation efficiency η g (t) etc.;

[0057] Step S102, cleaning the outliers in the data collected by using the 3σ principle; filling the missing values ​​by using the linear interpolation method; and normalizing the data by using the minimum-maximum normalization method.

[0058] Step S2, extracting and fusing multi-source features from the pre-processed power load related data to obtain a comprehensive feature vector;

[0059] Extracting multi-source features includes: time series features, meteorological features and socio-economic features; among them:

[0060] Time series characteristics:

[0061] Extract daily, weekly, and seasonal cycle characteristics. For example, calculate the daily load change rate: (Assuming the time interval is 1 hour); Weekly load change rate: The seasonal load change rate can be similarly calculated based on seasonal divisions.

[0062] Meteorological characteristics:

[0063] Extract meteorological characteristics according to the daily variation and monthly mean. For example, calculate the daily variation of temperature: ΔT day (t) = max c∈|t-24,t| T(c)-min c∈|t-24,t| T(c); Monthly average humidity: d is the number of days in the month;

[0064] Socio-economic characteristics: Calculation of regional GDP and industrial electricity consumption can be estimated through the relationship between industrial output value and electricity consumption per unit of industrial output value.

[0065] The extracted features are concatenated by column to form a comprehensive feature vector; the expression of the comprehensive feature vector is as follows:

[0066] F(t)=[X elec (t) T ,X weather (t) T ,X socior (t) T ];

[0067] Where, X elec (t) is the n1 features extracted from the power system operation data; X weather (t) is the n2 features extracted from the meteorological data; X socior (t) is n3 features extracted from socioeconomic data; the dimension of the comprehensive feature vector of F(t) is n, n=n1+n2+n3.

[0068] Step S3, using the comprehensive feature vector as the input of the power load forecasting model and the real value of the power load at the historical moment as the label value to train the power load forecasting model;

[0069] The power load forecasting model adopts an improved Transformer model, which includes improvements in position encoding, improvements in multi-head attention mechanism, and addition of residual connection and layer normalization;

[0070] In the traditional position encoding, we added learnable parameters so that the model can better capture the position information in the time series data. The improved position encoding is expressed as:

[0071]

[0072] Where: PE pos,i is the value of the improved position encoding at position pos and in the i-th dimension; α is the learnable parameter weight, which is used to balance the learnable parameter matrix W pos and original position encoding; W pos,i is the learnable parameter matrix of the i-th dimension, which is used to adjust the representation of position encoding; For the improved position encoding PE pos,i the extent of contribution;

[0073] Improvements to the multi-head attention mechanism include increasing the number of attention heads; and using an adaptive weight assignment method to calculate the weight ω for each attention head r r , so that the model pays more attention to the important information at different time steps and improves the ability to capture long sequence dependencies. The specific weight calculation formula is as follows:

[0074]

[0075] The final output of the improved multi-head attention mechanism is as follows:

[0076]

[0077] Where: r is the weight of the rth attention head; β r is the learnable parameter of the rth attention head; Score(head r ) is the score of the rth attention head; h ′ is the number of attention heads after improvement; β b is the learnable parameter of the bth attention head; Score(head b ) is the score of the b-th attention head;

[0078] Add residual connections and layer normalization. In each layer of the model, residual connections and layer normalization operations are added to improve the training efficiency and stability of the model. Specifically, we add the input data to the output of the model to form a residual connection, and then perform layer normalization on the result of the residual connection to improve the training efficiency and stability of the model. In each layer, the input x passes through the multi-head attention mechanism and the feedforward network to obtain the output y. The residual connection is:

[0079] y res =x+y;

[0080] The layer normalization formula is:

[0081]

[0082] Where: γ, β are learnable parameters; σ 2 , μ are y res The variance and mean of ; ε is a constant to prevent the denominator from being zero.

[0083] The Transformer model uses mean square error as the loss function, specifically:

[0084]

[0085] Where: L is the mean square error function value; N is the number of samples; v is the sample index; t v The power load forecast value of the Transformer model at time L(t v ) is t v The actual value of the power load at the moment.

[0086] The training process of the power load forecasting model also includes using the Adam optimizer to update the learnable parameters in the power load forecasting model. The specific formula is as follows:

[0087]

[0088] Where: θ s is the value of the learnable parameter θ at the sth iteration; ∈ is the learning rate; is the bias-corrected first-order moment estimate; is the bias-corrected second-order moment estimate; τ is a constant to prevent the denominator from being zero.

[0089] in:

[0090] m s =β1m s-1 +(1-β1)g s ;

[0091]

[0092] Where: β1, β2 are the decay rates of the first-order and second-order matrix estimates respectively; m s is the first-order moment estimate of the gradient; v s is the second-order moment estimate of the gradient; g s is the gradient vector of the parameter θ at the sth iteration.

[0093] Step S4, using the trained power load prediction model to predict the power load.

[0094] Embodiment 2:

[0095] This embodiment provides a power load forecasting system, including:

[0096] Data collection and preprocessing module, collects power load related data and performs preprocessing;

[0097] The feature extraction and fusion module extracts and fuses multi-source features from the pre-processed power load-related data to obtain a comprehensive feature vector;

[0098] The model training module uses the comprehensive feature vector as the input of the power load forecasting model and the real value of the power load at the historical moment as the label value to train the power load forecasting model;

[0099] The model prediction module uses the trained power load prediction model to predict the power load.

[0100] Embodiment three:

[0101] This embodiment provides an electronic device having a computer program stored thereon, and when the computer program is executed by a processor, a method for predicting electric load as described in any embodiment of the present invention is implemented.

[0102] Embodiment 4:

[0103] This embodiment provides a computer-readable medium for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement a power load forecasting method as described in any embodiment of the present invention.

[0104] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0105] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0106] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0107] In several embodiments provided in the present application, if any function 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, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.

[0108] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting power load, characterized in that: The following steps are involved: Step S1, collecting power load related data and performing preprocessing; Step S2, extracting and fusing multi-source features from the pre-processed power load related data to obtain a comprehensive feature vector; Step S3, using the comprehensive feature vector as the input of the power load forecasting model and the real value of the power load at the historical moment as the label value to train the power load forecasting model; Step S4, using the trained power load prediction model to predict the power load.

2. A method for predicting power load according to claim 1, characterized in that: The power load related data include: power system operation data, meteorological data and socio-economic data; the power system operation data include: historical power load data, power grid topology data and power generation equipment operation status data; meteorological data include: temperature, humidity, wind speed and sunshine time; socio-economic data include: regional gross domestic product, population size, industrial output value.

3. A method for predicting power load according to claim 1, characterized in that: The multi-source features are extracted and fused from the pre-processed power load related data to obtain a comprehensive feature vector, the expression of which is as follows: F(t)=[X elec (t) T ,X weather (t) T ,X socior (t) T ]; Where, X elec (t) is the n1 features extracted from the power system operation data; X weather (t) is the n2 features extracted from the meteorological data; X socior (t) is n3 features extracted from socioeconomic data; The dimension of the comprehensive feature vector of F(t) is n, n=n1+n2+n3.

4. A method for predicting power load according to claim 1, characterized in that: The power load forecasting model adopts an improved Transformer model, which includes improvements in position encoding, improvements in the multi-head attention mechanism, and addition of residual connections and layer normalization; wherein: The improved position encoding expression is: Where: PE pos,i is the value of the improved position encoding at position pos and in the i-th dimension; α is the learnable parameter weight, which is used to balance the learnable parameter matrix W pos and original position encoding; W pos,i is the learnable parameter matrix of the i-th dimension, which is used to adjust the representation of position encoding; For the improved position encoding PE pos,i the extent of contribution; Improvements to the multi-head attention mechanism include increasing the number of attention heads; and using an adaptive weight assignment method to calculate the weight ω for each attention head r r , the specific formula is as follows: Where: r is the weight of the rth attention head; β r is the learnable parameter of the rth attention head; Score(head r ) is the score of the rth attention head; h ′ is the number of attention heads after improvement; β b is the learnable parameter of the bth attention head; Score(head b ) is the score of the b-th attention head; Add residual connections and layer normalization. In each layer, the input x passes through the multi-head attention mechanism and the feedforward network to obtain the output y. The residual connection is: and res =x+y; The layer normalization formula is: Where: γ, β are learnable parameters; σ 2 , μ are y res The variance and mean of ; ε is a constant to prevent the denominator from being zero.

5. A method for predicting power load according to claim 4, characterized in that: The Transformer model uses mean square error as the loss function, specifically: Where: L is the mean square error function value; N is the number of samples; v is the sample index; t v The power load forecast value of the Transformer model at time L(t v ) is t v The actual value of the power load at the moment.

6. A method for predicting power load according to claim 1, characterized in that: The training process of the power load forecasting model also includes using an Adam optimizer to update the learnable parameters in the power load forecasting model. The specific formula is as follows: Where: θ s is the value of the learnable parameter θ at the sth iteration; ∈ is the learning rate; is the bias-corrected first-order moment estimate; is the bias-corrected second-order moment estimate; τ is a constant to prevent the denominator from being zero.

7. A power load forecasting system, characterized in that: include: Data collection and preprocessing module, collects power load related data and performs preprocessing; The feature extraction and fusion module extracts and fuses multi-source features from the pre-processed power load-related data to obtain a comprehensive feature vector; The model training module takes the comprehensive feature vector as the input of the power load forecasting model, predicts the real value of the power load at the historical moment as the output, and trains the power load forecasting model; The model prediction module uses the trained power load prediction model to predict the power load.

8. A power load forecasting system according to claim 7, characterized in that: In the data acquisition and preprocessing module, the power load related data include: power system operation data, meteorological data and socio-economic data; the power system operation data include: historical power load data, power grid topology data and power generation equipment operation status data; meteorological data include: temperature, humidity, wind speed and sunshine time; socio-economic data include: regional gross domestic product, population size, industrial output value.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, an electric power load forecasting method as described in any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, an electric power load forecasting method as described in any one of claims 1 to 6 is implemented.

Images