Multi-source data fusion method of power system and real-time monitoring device

Abstract

The invention discloses a multi-source data fusion method of a power system and a real-time monitoring device, and the method employs a self-adaptive DTW to align the multi-source data of the power system, and solves the problems of time scale difference and time delay between different data sources. And then a Transform model is adopted to perform feature extraction on the aligned data, and a complex time sequence dependency relationship in the data is captured, so that the validity of the data is further enhanced. And finally, quantifying the contribution degrees of different data sources through an entropy weight method, and optimizing the weight distribution of each data source, thereby improving the overall quality of the fused data. Through the method, information from multiple data sources such as wind power can be effectively fused, so that the accuracy of state estimation of the power system and the stability of real-time monitoring are improved. Moreover, the method can improve the application effect of the multi-source data in the power system, and avoids the estimation deviation caused by the heterogeneity of the data sources.

Description

Technical Field

[0001] The present application relates to the technical field of power distribution networks, and in particular to a multi-source data fusion method and a real-time monitoring device for a power system. Background Art

[0002] With the rapid development of my country's economy, the power system, a critical infrastructure of modern society, continues to expand in scale. Distribution networks, in particular, have become more extensive, more complex, and operate in increasingly diverse modes. Traditional power system monitoring and management methods are no longer sufficient to meet current needs, and intelligent and automated upgrades to the power system are urgently needed. To ensure the safe, stable, and efficient operation of the power system, real-time monitoring and accurate state estimation technologies are crucial.

[0003] Multi-source data fusion technology plays a crucial role in the development of power systems. Power system operations generate massive amounts of data, covering photovoltaic power generation, load demand, energy storage system charge and discharge power, and various electrical parameters. This data reflects the power system's operating status from various perspectives. Effectively integrating this data can provide comprehensive, accurate, and real-time operational status information. By integrating this multi-source data, not only can we gain in-depth insights into the power system's operating patterns, but we can also promptly identify potential problems, enabling targeted measures to improve grid reliability and stability and ensure the quality of power supply.

[0004] However, in practical applications, multi-source data fusion in power systems faces numerous challenges. First, the diversity of data sources leads to uneven data quality. Different data acquisition devices, different acquisition environments, and interference during transmission often result in errors, missing values, or outliers in the collected multi-source data. Second, traditional data fusion methods have significant flaws. Regarding data alignment, due to the inconsistent time scales and time delays of different data sources, traditional methods struggle to achieve efficient and accurate alignment. Taking the DTW method as an example, the traditional DTW alignment accuracy is limited, making it inflexible in complex and changing data environments. This leads to poor alignment results, which in turn impacts subsequent data processing and analysis. Regarding feature extraction, traditional methods often only extract superficial data features, making it difficult to explore deeper dependencies between data and fully utilize the effective information within the data. In the weight assignment process, traditional methods lack a scientific and rational basis and often rely on subjective weighting. This results in irrational weight assignments that fail to truly reflect the contribution of each data source to the final result, severely limiting the effectiveness of data fusion and the accuracy of real-time monitoring.

[0005] While the Transformer model has emerged as a leader in time series data processing thanks to its powerful feature extraction capabilities, effectively capturing deep dependencies in multi-source data and extracting effective features, the challenge of properly assigning weights to each data source during multi-source data fusion to ensure the fused data is more accurate and reliable remains a pressing issue. While the entropy weight method, a weight assignment method based on information theory, can optimize weight assignment by measuring the information contribution of each data source, further research and exploration is needed to organically combine it with other technologies to achieve efficient fusion of multi-source data in power systems. Summary of the Invention

[0006] The embodiments of the present application provide a multi-source data fusion method and a real-time monitoring device for an electric power system, which can improve the accuracy and stability of real-time monitoring and state estimation of the electric power system.

[0007] In order to achieve the above-mentioned purpose, the technical solution of the embodiment of the present invention is:

[0008] In a first aspect, an embodiment of the present invention provides a multi-source data fusion method for an electric power system, comprising: obtaining multi-source data of the electric power system, the multi-source data including but not limited to photovoltaic power generation power, load demand, and charge and discharge power of the energy storage system; preprocessing the multi-source data, including data cleaning, time alignment based on an adaptive DTW algorithm, and normalization processing; wherein, based on the adaptive DTW algorithm, time series alignment of the multi-source data is achieved by dynamically adjusting the time window and weighted matching strategy; the preprocessed multi-source data is input into a Transformer model, and time series features are extracted through a multi-layer self-attention mechanism; the entropy value of each data source in the multi-source data is calculated through an entropy weight method, and weights are dynamically assigned to quantify the information contribution of each data source; based on the assigned weights, the time series features are weightedly fused to obtain real-time monitoring data of the electric power system.

[0009] In some possible implementations, multi-source data of the power system is obtained, including but not limited to photovoltaic power generation power, load demand, charge and discharge power of the energy storage system, and electrical parameters; the multi-source data is preprocessed, including data cleaning, time alignment based on an adaptive DTW algorithm, and normalization processing; wherein, based on the adaptive DTW algorithm, time series alignment of the multi-source data is achieved by dynamically adjusting the time window and weighted matching strategy; the preprocessed multi-source data is input into a Transformer model, and time series features are extracted through a multi-layer self-attention mechanism; the entropy value of each data source in the multi-source data is calculated using an entropy weight method, and weights are dynamically assigned to quantify the information contribution of each data source;

[0010] Based on the assigned weights, the time series features are weightedly fused to obtain real-time monitoring data of the power system.

[0011] In some possible implementations, dynamically adjusting the time window size according to data volatility is achieved through the following formula:

[0012]

[0013] Among them, D(i,j) is the distance from the starting point to the reference data set D r Time step i and the dataset to be aligned D i The minimum cumulative distance of time step j, dist(D r [i],D i [j]) is the reference dataset D r and the dataset to be aligned D i The point-to-point distance between time steps i and j, W(i,j) is the preset variable window size, D r [i] is the reference dataset D r The data value at time step i, D i [j] is the dataset D to be aligned i The data value at time step j.

[0014] In some possible implementations, the weighted DTW distance is calculated as:

[0015]

[0016] Where M(i,j) is the number of data points per pair D r [i] and D i The weight coefficient of [j] is calculated based on the volatility or rate of change of the data.

[0017] In some possible implementations, the calculation formula of the self-attention mechanism in the Transformer model is expressed as:

[0018]

[0019] Among them, Q is the query matrix, K is the key matrix, V is the value matrix, d k is the dimension of the key vector, Attention() represents the calculation formula of the self-attention mechanism, and T is the inverse matrix.

[0020] In some possible implementations, the entropy value of each data source in the multi-source data is calculated by an entropy weight method, and a weight is dynamically assigned to quantify the information contribution of each data source, including: calculating the probability distribution of the corresponding indicator of each data in the preprocessed multi-source data, expressed as:

[0021]

[0022] Among them, P ijis the normalized value of the i-th solution on the j-th indicator, Z′ ij For each indicator, S j is the sum of the data for each indicator; the information entropy of each indicator is calculated according to the probability distribution of each indicator, expressed as:

[0023]

[0024] Among them, E j is the information entropy of the j-th indicator, Is a normalization constant used to ensure that the information entropy is between [0,1]. The weight of each indicator is calculated based on the information entropy, expressed as:

[0025]

[0026] Among them, E j is the information entropy value of the jth indicator, n is the total number of indicators; 1-E j It represents the comparison value of information entropy. The smaller the value, the more uncertain the information of the indicator is and the smaller the weight is.

[0027] In some possible implementations, after dynamically allocating weights using the entropy weight method, the method further includes:

[0028] Calculate the comprehensive score of each solution using the following formula:

[0029]

[0030] All solutions are ranked according to the comprehensive scores.

[0031] In a second aspect, an embodiment of the present invention provides a real-time monitoring device for a power system based on multi-source data fusion, comprising:

[0032] A data acquisition module is used to obtain multi-source data of the power system, including but not limited to photovoltaic power generation power, load demand, charge and discharge power of the energy storage system, and electrical parameters;

[0033] The preprocessing module is used to preprocess multi-source data, including data cleaning, time alignment based on the adaptive DTW algorithm, and normalization. The adaptive DTW algorithm dynamically adjusts the time window and weighted matching strategy to achieve time series alignment of multi-source data.

[0034] The feature extraction module integrates the Transformer model to extract long-term temporal dependency features of multi-source data;

[0035] A weight allocation module is used to dynamically allocate the weights of each data source by adopting the entropy weight method;

[0036] Data fusion and decision-making module: used to weight and fuse time series features, generate real-time monitoring results and output corresponding decision instructions.

[0037] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages:

[0038] In an embodiment of the present invention, adaptive DTW is used to align multi-source data of the power system, solving the time scale differences and time delay problems between different data sources. The Transformer model is then used to extract features from the aligned data to capture the complex temporal dependencies in the data, further enhancing the validity of the data. Finally, the contribution of different data sources is quantified using the entropy weight method, and the weight distribution of each data source is optimized, thereby improving the overall quality of the fused data. Through this method, information from multiple data sources such as wind power can be effectively fused, thereby improving the accuracy of power system state estimation and the stability of real-time monitoring. In addition, this method can improve the application effect of multi-source data in the power system and avoid estimation bias caused by the heterogeneity of data sources. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 A schematic flow chart of an embodiment of a multi-source data fusion method for a power system provided for the implementation of the present invention;

[0041] Figure 2 A schematic flow chart of another embodiment of a multi-source data fusion method for a power system provided by an embodiment of the present invention;

[0042] Figure 3 is a feature weight distribution diagram of the entropy weight method in an embodiment of the present invention;

[0043] Figure 4 Schematic diagram of the correlation between the comprehensive score of the entropy weight method and the target variable in an embodiment of the present invention;

[0044] Figure 5 Schematic diagram of the comparison between the comprehensive score of the entropy weight method and the target variable in an embodiment of the present invention;

[0045] Figure 6 Schematic diagram of the structure of a real-time monitoring device for a power system based on multi-source data fusion in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0047] In the relevant description of this embodiment, the terms "including, containing, having" and the like are open terms, and are generally understood to include but not be limited to; the term "at least one" is generally understood to mean one or more, where "plurality" refers to two or more; the term "at least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items, for example, "at least one of a, b or c", or "at least one of a, b and c", can all represent: a, b, c, ab (i.e., a and b), ac, bc, or abc, where a, b, c can be single or multiple respectively; the symbol "A / B" is used to describe the selection relationship of associated objects, generally indicating an "or" relationship before and after.

[0048] In the following description of the present embodiment, the terms used in the embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. The singular forms "a", "an" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0049] Those skilled in the art should understand that in the following description of the embodiments of the present application, the order of serial numbers does not mean the order of execution, some or all of the steps can be executed in parallel or sequentially, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0050] It will be understood by those skilled in the art that the numerical ranges in the examples of the present application are to be understood as also specifically disclosing each intermediate value between the upper and lower limits of the ranges. Each smaller range between the intermediate value in any stated value or stated range and any other stated value or intermediate value in the range is also included in the present invention. The upper and lower limits of these smaller ranges may be independently included or excluded in the scope.

[0051] Unless otherwise indicated, the technical / scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which this application belongs. Although this application describes only preferred methods and materials, any methods and materials similar or equivalent to those herein may also be used in the implementation or testing of this application. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods and / or materials related to the documents. In the event of any conflict with any incorporated document, the content of this specification shall prevail.

[0052] In order to illustrate the technical solution of the present invention, specific embodiments are provided below.

[0053] Multi-source data fusion technology plays a crucial role in the development of power systems. Power system operations generate massive amounts of data, covering photovoltaic power generation, load demand, energy storage system charge and discharge power, and various electrical parameters. This data reflects the power system's operating status from various perspectives. Effectively integrating this data can provide comprehensive, accurate, and real-time operational status information. By integrating this multi-source data, not only can we gain in-depth insights into the power system's operating patterns, but we can also promptly identify potential problems, enabling targeted measures to improve grid reliability and stability and ensure the quality of power supply.

[0054] In practical applications, multi-source data fusion in power systems faces numerous challenges. For one thing, the diversity of data sources results in uneven data quality. Different data acquisition devices, collection environments, and interference factors during transmission often lead to errors, missing values, or outliers in the collected multi-source data. For example, in some harsh natural environments, the collection of photovoltaic power generation may be affected by factors such as unstable light intensity and equipment failures, resulting in data deviations. Furthermore, load demand data may also be partially missing during the collection and transmission process due to network failures. These issues seriously affect the availability and accuracy of data, making data fusion extremely difficult.

[0055] On the other hand, traditional data fusion methods have obvious flaws. In terms of data alignment, due to the inconsistencies in the time scales and time delays of different data sources, traditional methods find it difficult to achieve efficient and accurate alignment. Taking the DTW method as an example, the alignment accuracy of traditional DTW is limited. When faced with complex and changing data environments, it cannot be flexibly adjusted, resulting in poor alignment results, which in turn affects subsequent data processing and analysis. In terms of feature extraction, traditional methods can often only extract shallow features of the data, making it difficult to explore the deep dependencies between data and make full use of the effective information in the data. In the weight allocation link, traditional methods lack a scientific and reasonable basis and often use subjective weight setting methods, resulting in unreasonable weight distribution and an inability to truly reflect the contribution of each data source to the final result, which seriously restricts the effect of data fusion and the accuracy of real-time monitoring.

[0056] Based on this, an embodiment of the present invention provides a multi-source data fusion method for an electric power system, which can improve the accuracy and stability of real-time monitoring and state estimation of the electric power system.

[0057] Figure 1 A schematic diagram of an embodiment of a multi-source data fusion method for a power system provided by the present invention is shown in FIG. Figure 1 As shown, the above method may include:

[0058] S101, acquiring multi-source data of the power system;

[0059] In some embodiments, representative multi-source power system data can be selected for experimentation to verify the effectiveness and applicability of the method. The dataset used can include typical data on various power system operating conditions, including but not limited to electrical parameters such as photovoltaic power generation, load demand, and energy storage system charge and discharge power. This data can comprehensively reflect the operating status of the power system.

[0060] Among them, photovoltaic power generation power is a key indicator of clean energy generation. Its numerical changes reflect the utilization of solar energy resources and the operating efficiency of photovoltaic equipment. Affected by factors such as light intensity and weather conditions, it varies significantly at different times and under different circumstances. Load demand data directly reflects the energy consumption of various users and has obvious time characteristics. For example, the load curves on weekdays and holidays, daytime and nighttime vary greatly. In addition, the load demand patterns of different industries and user types are also different. The charge and discharge power of the energy storage system demonstrates the role of energy storage equipment in balancing power supply and demand and stabilizing system operation. When there is excess power, the energy storage system charges, and the power is positive; when there is insufficient power, the power is negative. In addition to the above data, multi-source data can also include various electrical parameters such as voltage, current, power factor, etc. Any slight change in these data may indicate a change in the system's operating status.

[0061] Furthermore, net load integrates the impact of photovoltaic power generation, load demand, and energy storage systems to characterize the real-time load status of the power system. By incorporating random perturbations and noise into these data, we can simulate the volatility and uncertainty that can occur in actual power systems. By analyzing these characteristic data using the entropy weight method, we further validate the effectiveness of multidimensional feature fusion for power system status assessment.

[0062] S102, preprocessing the multi-source data, including data cleaning, time alignment based on the adaptive DTW algorithm, and normalization processing;

[0063] Among them, the time series alignment of multi-source data is achieved by dynamically adjusting the time window and weighted matching strategy based on the adaptive DTW algorithm;

[0064] In this embodiment of the present invention, raw data is cleaned, aligned, and normalized to improve data quality and consistency. By optimizing raw data through a series of steps, noise can be removed, missing information can be restored, and differences in time and dimensions between data sources can be addressed.

[0065] It's important to note that when cleaning raw multi-source data, outliers often appear. These are often caused by measurement errors or data entry errors. Outliers can lead to poor model predictions and even hinder the training process.

[0066] In some embodiments, a standard score method (Z-Score) may be used to detect outliers.

[0067] For example, for a certain feature, if its value deviates too much from the mean (for example, more than 3 times the standard deviation), it is considered an outlier. The specific calculation method can be expressed as:

[0068]

[0069] Where X is the data point, μ is the mean, and σ is the standard deviation. For most data, the Z-Score is between -3 and 3. If the Z-Score is greater than 3 or less than -3, the data point can be considered an outlier and removed, thereby performing data cleaning and ensuring data consistency.

[0070] In some embodiments, time alignment based on an adaptive DTW algorithm includes:

[0071] S1021, determining a reference dataset from the datasets of the multi-source data according to a preset condition, and aligning the other datasets based on the timestamp of the reference dataset;

[0072] S1022: For each of the datasets to be aligned, the time window size is dynamically adjusted according to the data volatility. When the data fluctuates violently, the window is expanded. At the same time, a weighted matching strategy is introduced to dynamically adjust the weights based on the importance, volatility, or anomalies of the data points, and calculate the weighted DTW distance.

[0073] Specifically, let's assume there are a set of different time series datasets, each with different timestamps and possibly varying sampling frequencies. To select an alignment target, you first need to choose a reference dataset. Prerequisites might include selecting a dataset with a higher sampling frequency or more complete timestamps. This dataset will serve as the alignment baseline, and the other datasets will be aligned based on its timestamps.

[0074] For example, select dataset D r As the reference dataset, D r ={(t1,x1),(t2,x2)…,(t n ,x n )} represents the observation value of the reference dataset at n time points, while other datasets D1, D2, ..., D A It is the time series that needs to be aligned with the reference dataset. For each pair of time series datasets D i and the reference dataset D r , adaptive DTW can be used to calculate the matching degree between them. Based on traditional DTW, more flexibility is introduced, and the adaptive ability of the matching process is enhanced through methods such as weighted matching and time window adjustment. The specific steps include dynamically adjusting the time window and weighted matching strategy.

[0075] It's important to note that the purpose of adjusting the time window is to control the constraints in the matching process based on data volatility or sampling frequency, ensuring that the DTW algorithm can better adapt to data changes in areas with drastic temporal variations or high noise. Specifically, when data is highly variable, the matching time window can be expanded to allow for more time steps to be aligned, thus avoiding over-constraints in areas of high variability.

[0076] In some embodiments, in the standard DTW example formula, there is a recursive relationship expressed as:

[0077]

[0078] Among them, D(i,j) is the distance from the starting point to the reference data set D r Time step i and the dataset to be aligned D i The minimum cumulative distance of time step j, dist(D r [i],D i [j]) is the reference dataset D r and the dataset to be aligned D i The point-to-point distance between time steps i and j.

[0079] In adaptive DTW, time window adjustment introduces a variable window size W(i,j) in the distance calculation process. For example, in some high-fluctuation areas, the window may be enlarged to allow more points to participate in the matching.

[0080] In the embodiment of the present invention, the time window size is dynamically adjusted according to data volatility by the following formula:

[0081]

[0082] Among them, W(i,j) is the preset variable window size, D r [i] is the reference dataset D r The data value at time step i, D i [j] is the dataset D to be aligned i The data value at time step j.

[0083] Using weighted matching, we adjust the matching strategy by assigning different weights to different data points. The weights of data points can be dynamically adjusted based on their importance, volatility, or anomalies. In the DTW calculation with weighted matching, the weight matrix affects the point-to-point distance calculation. Traditional DTW distance calculation is based on the Euclidean distance between the original data points, while in weighted matching, the distance calculation formula becomes:

[0084]

[0085] Where M(i,j) is the number of data points per pair D r [i] and D i The weight coefficient of [j] is calculated based on the volatility or rate of change of the data.

[0086] After aligning the time series using the aforementioned dynamic time window adjustment and weighted matching strategy, the data can initially achieve accurate correspondence in the temporal dimension. However, because the multi-source data in the power system involves a variety of different physical quantities, such as photovoltaic power generation measured in kilowatts (kW), load demand calculated in megawatts (MW), energy storage system charge and discharge power also has its own specific power units, and electrical parameters such as voltage in volts (V) and current in amperes (A), these different physical quantities have different dimensions, which can adversely affect subsequent data processing and model training.

[0087] Therefore, in embodiments of the present invention, normalization can also be used to convert data. The core purpose of normalization is to convert the values of different physical quantities into a unified range and eliminate dimensional differences. There are many common normalization methods, such as Min-Max Normalization.

[0088] By normalizing multi-source data, all data on different physical quantities are unified into the same numerical range. During the data processing process, the unified numerical range enables data from different data sources to be directly compared and integrated, avoiding erroneous analysis caused by dimensional differences. In terms of model training, normalized data can make model training more stable. Because if the scale differences of the data are large, the model may be more inclined to features with larger values during training, thereby affecting the model's learning effect on other features and resulting in a decrease in the model's generalization ability. After normalization, all features are in a relatively equal position during the model training process, which helps the model better learn the patterns in the data, improves the model's training efficiency and prediction accuracy, and thus provides more reliable support for tasks such as power system state estimation, fault diagnosis and prediction.

[0089] S103, input the preprocessed multi-source data into the Transformer model, and extract temporal features through a multi-layer self-attention mechanism;

[0090] In this embodiment of the present invention, the Transformer model is used to extract features from normalized and aligned multi-source data, capturing long-term dependencies in the data and optimizing feature representation. The model can also model important features in the data through a multi-layer self-attention mechanism, enhancing data validity.

[0091] In some embodiments, the core formula of the Transformer model, that is, the calculation formula of the self-attention mechanism, can be expressed as:

[0092]

[0093] Among them, Q is the query matrix, K is the key matrix, V is the value matrix, d k is the dimension of the key vector, Attention() represents the calculation formula of the self-attention mechanism, and T is the inverse matrix.

[0094] The preprocessed data is then fed into the Transformer model for training. The Transformer's self-attention mechanism allows the model to weight the data at each time step, capturing temporal correlations and complex features within the data. The self-attention mechanism enables the model to process data at each time step, not just focusing on local information, but also simultaneously attending to other time steps in the data sequence, thereby capturing long-term dependencies between data. For example, load demand data in power systems typically exhibits certain periodicity and trends. The self-attention mechanism effectively identifies correlations between load demand at different time points. When analyzing daily load curves, the model not only considers load variations at different times of the day but also incorporates load data from similar historical dates (such as the same day last week or the same week of the previous month). This approach allows the model to uncover patterns in load demand across different timescales, such as weekly, monthly, and even seasonal cycles.

[0095] Photovoltaic power generation data is affected by multiple factors, such as light intensity and weather conditions, and its fluctuations are complex. The self-attention mechanism can comprehensively analyze information such as light conditions, weather changes, and the operating status of photovoltaic power generation equipment at different times from a time series perspective, accurately capturing the fluctuation characteristics of photovoltaic power generation.

[0096] When processing energy storage system charge and discharge power data, the self-attention mechanism can combine the power system's load demand, photovoltaic power generation, and the energy storage system's own charge and discharge history data to analyze the rationality of the energy storage system's charge and discharge strategies at different time points, as well as the synergistic relationship between charge and discharge power and other power data. For example, when load demand suddenly increases and photovoltaic power generation is insufficient, the energy storage system needs to discharge in a timely manner to meet power demand. The self-attention mechanism can discover the inherent connection between the energy storage system's charge and discharge power and other data in this situation.

[0097] During the training process, the Transformer model can optimize data features through the self-attention mechanism and automatically learn the most important time series features, thereby improving its effectiveness in subsequent data fusion.

[0098] S104, calculating the entropy value of each data source in the multi-source data by using an entropy weight method, and dynamically assigning weights to quantify the information contribution of each data source;

[0099] The entropy weighting method, an objective weighting method based on information theory, focuses on determining the relative importance of each data source in the data fusion process by measuring the uncertainty of the information contained in each data source. In the multi-source data environment of power systems, different types of data, such as photovoltaic power generation, load demand, energy storage system charge and discharge power, and various electrical parameters, contribute to varying degrees to power system status assessment and analysis. The entropy weighting method effectively exploits these differences, assigning appropriate weights to each data source and thus enabling scientific fusion of multi-source data.

[0100] Specifically, the weight allocation in step S104 may include:

[0101] First, calculations are performed using the data preprocessed in step S102. The first step is to calculate the probability distribution of each indicator (that is, the characteristic indicator corresponding to each data source). Taking data containing m scenarios as an example (the scenario here can be represented as a combination of multi-source data at different time points or different operating conditions), for each indicator, the sum of its data is calculated, expressed as:

[0102]

[0103] Among them, S j is the sum of each indicator data, Z′ ij For each indicator, m is the number of solutions.

[0104] After that, the probability distribution of the corresponding indicators of each data in the preprocessed multi-source data is calculated, which is expressed as:

[0105]

[0106] Among them, P ij is the normalized value of the i-th solution on the j-th indicator.

[0107] Then the information entropy of each indicator is calculated according to the probability distribution of each indicator, which is expressed as:

[0108]

[0109] Among them, E j is the information entropy of the j-th indicator, Is a normalization constant used to ensure that the information entropy is between [0,1];

[0110] Finally, the weight of each indicator is calculated by information entropy. The larger the information entropy value, the more dispersed the information of the indicator, and the smaller its weight; the smaller the information entropy value, the more concentrated the information of the indicator, and the larger its weight. The weight of each indicator can be expressed as:

[0111]

[0112] Among them, w j is the weight of each indicator, E j is the information entropy value of the jth indicator, n is the total number of indicators; 1-E j It represents the comparison value of information entropy. The smaller the value, the more uncertain the information of the indicator is and the smaller the weight is.

[0113] For example, in an actual power system analysis scenario, assuming that within a specific time period, the stability assessment of the power system involves three key data sources: photovoltaic power generation, load demand, and energy storage system discharge power. Using the entropy weighting method, it is found that the information entropy of load demand data is relatively low, meaning that its data is relatively concentrated and its variation patterns are relatively stable, which has a greater impact on the power system stability assessment. Therefore, it is given a higher weight in the data fusion process. On the other hand, photovoltaic power generation data fluctuates greatly due to the influence of various uncertain factors such as weather, and its information entropy is relatively high, resulting in a relatively low weight. The weight of the energy storage system discharge power is determined to a reasonable level using the entropy weighting method based on its correlation with the power system stability within the time period.

[0114] By dynamically assigning weights, the entropy weight method can accurately quantify the contribution of different data sources to the final prediction results or power system state assessment based on their information characteristics. This quantification method avoids the arbitrariness and one-sidedness of traditional subjective weight setting methods, allowing multi-source data to more rationally play their respective roles in the fusion process, further improving the accuracy and reliability of data fusion, and thus providing more scientific and accurate support for tasks such as power system state estimation, fault diagnosis, and prediction.

[0115] In some embodiments, after dynamically allocating weights using the entropy weight method, the method may further include:

[0116] Calculate the comprehensive score of each solution using the following formula:

[0117]

[0118] All solutions are ranked according to the comprehensive scores.

[0119] Among them, the comprehensive score of the plan reflects the overall performance of the corresponding plan under all indicators. Sorting all plans according to the calculated comprehensive score can intuitively compare the advantages and disadvantages of different plans. The higher the score, the better it is in the comprehensive evaluation. In the actual application scenarios of the power system, these plans can represent different power operation states, different power dispatch strategies, etc. Through sorting, power system managers can quickly find the optimal or better plan. When allocating power resources, plans that can make power supply and demand more balanced and system operation more stable can be selected based on the ranking; when assessing potential risks of the power system, plans with lower rankings may mean higher risks and require special attention and corresponding optimization measures.

[0120] S105 , performing weighted fusion on the time series features based on the assigned weights to obtain real-time monitoring data of the power system.

[0121] Understandably, the time series features extracted by the Transformer model contain a variety of power system operational information. These features can reflect the temporal variations and inherent connections of the power system from different perspectives. However, different time series features have varying representational capabilities and importance for the real-time state of the power system. Therefore, they need to be properly integrated using the weights assigned by the entropy weight method in S104.

[0122] Take, for example, the time series characteristics of photovoltaic power generation, load demand, and energy storage system charge and discharge power, which are common in power systems. The time series characteristics of photovoltaic power generation reflect the utilization of solar energy at different times, and are affected by factors such as light intensity and weather changes, showing obvious daily and seasonal variations. The time series characteristics of load demand reflect the changes in electricity consumption behavior of various users over time, with typical peaks and valleys during the day and night, as well as differences between weekdays and holidays. The time series characteristics of energy storage system charge and discharge power record the energy regulation of energy storage equipment in balancing power supply and demand, and are closely related to photovoltaic power generation and load demand. During the weighted fusion process, these different time series characteristics are specifically weighted according to the weights determined by the entropy weight method.

[0123] For example, assume that the entropy weight method calculates the time series characteristic weight of the load demand as w1, the time series characteristic weight of the photovoltaic power generation as w2, and the time series characteristic weight of the energy storage system charging and discharging power as w3. For a specific time t, the corresponding time series characteristic value of the load demand f1, the time series characteristic value of the photovoltaic power generation f2, and the time series characteristic value of the energy storage system charging and discharging power f3 are obtained respectively. Then, the fused characteristic value at that time can be calculated by weighted summation, expressed as:

[0124] f(t)=w1×f1+w2×f2+w2×f2;

[0125] This weighted fusion approach considers the information contribution of each data source and highlights the role of time series features that significantly impact the real-time state of the power system. Assuming that at a given moment, load demand has the greatest impact on power system stability, its corresponding weight w1 will be relatively large, and in the fusion calculation, f1 will contribute more to the final fused feature value. By performing this weighted fusion of time series features at each moment, a series of fused feature data reflecting the real-time operating state of the power system can be generated. This data integrates the advantages of multiple data sources to comprehensively and accurately describe the operating status of the power system at different points in time. It not only includes key information such as power production (photovoltaic power generation related characteristics), consumption (load demand related characteristics), and storage (energy storage system charge and discharge power related characteristics), but also uses weight adjustment to ensure that this information is appropriately reflected in the fusion process.

[0126] Further collation and analysis of this fused feature data yields real-time monitoring data for the power system. This real-time monitoring data intuitively and accurately presents the real-time state of the power system, providing a precise data foundation for power system state estimation. Subsequent state estimation can directly utilize this fused real-time monitoring data through corresponding models to quickly and accurately calculate estimated values of electrical parameters such as voltage, current, and power at each node in the power system, thereby promptly identifying potential operational risks. For fault diagnosis, real-time monitoring data helps the diagnostic system detect abnormal changes in the power system's operating state. By comparing and analyzing the feature data under normal operating conditions, the fault's location and type can be quickly located, effectively supporting the timely implementation of fault remediation measures. In power system forecasting tasks, this data, combined with corresponding prediction algorithms, can accurately predict future load demand and power generation changes in the power system, providing a scientific basis for power dispatch and optimal resource allocation, thereby ensuring the safe, stable, and economical operation of the power system.

[0127] In an embodiment of the present invention, adaptive DTW is used to align the multi-source data of the power system, solving the time scale differences and time delay problems between different data sources; then, the Transformer model is used to extract deep features, fully explore the potential information in the data, capture the complex temporal dependencies in the data, and further enhance the validity of the data; finally, the contribution of different data sources is quantified through the entropy weight method, and the weight distribution of each data source is optimized, thereby improving the overall quality of the fused data. Through this method, information from multiple data sources such as wind power can be effectively integrated, thereby improving the accuracy of power system state estimation and the stability of real-time monitoring. In addition, this method can improve the application effect of multi-source data in the power system and avoid estimation bias caused by the heterogeneity of data sources.

[0128] The following is a specific example to illustrate the technical solution of the embodiment of the present invention:

[0129] Figure 2 This is a flow chart of another embodiment of a multi-source data fusion method for a power system provided by an embodiment of the present invention. Figure 2 As shown, first, representative power data sets such as photovoltaic power generation power, load demand, and charging and discharging power of energy storage systems are input, and the original data are cleaned, aligned, and normalized. Among them, the Z-Score method is used for data cleaning to eliminate outliers, and the DTW algorithm is used for time alignment. Among them, the time alignment of multi-source data specifically includes: selecting a data set with a higher sampling frequency or a more complete timestamp from the data set of multi-source data to determine it as the reference data set as the alignment target, and the others as the data sets to be aligned. Then the time window is adjusted, and the adjustment of the time window size is achieved by the following formula:

[0130]

[0131] Then the weighted DTW distance is calculated by the following formula for weighted matching, which is expressed as:

[0132]

[0133] After that, the preprocessed multi-source data is input into the Transformer model to extract the temporal features of the multi-source data. In the Transformer model, the calculation formula of the self-attention mechanism is expressed as:

[0134]

[0135] According to the extracted time series features, the entropy weight method is used to assign weights. First, the probability distribution of the corresponding indicators of each data in the preprocessed multi-source data is calculated, which is expressed as:

[0136]

[0137] Then the information entropy of each indicator is calculated according to the probability distribution of each indicator, which is expressed as:

[0138]

[0139] The weight of each indicator is calculated according to the information entropy, which is expressed as:

[0140]

[0141] Finally, the comprehensive score of each solution is calculated and expressed by the following formula:

[0142]

[0143] According to the final comprehensive score, multi-source data is fused through weighted fusion, and further sorting and analysis are carried out to obtain the final real-time monitoring data of the power system.

[0144] Furthermore, the present invention also tests the fusion method of multi-source data. Figures 3 to 5 As shown. Among them, Figure 3 This is a feature weight distribution diagram for the entropy weight method in an embodiment of the present invention, which shows the weights assigned to different features in the entropy weight method. This diagram allows you to intuitively see the importance of each feature to the target variable. The higher the weight value, the greater the contribution of the feature to the target variable. Figure 4 Schematic diagram of the correlation between the comprehensive score of the entropy weight method and the target variable in the embodiment of the present invention. Figure 4 As shown, by calculating the Pearson correlation coefficient, Figure 4 The linear relationship between the comprehensive score and the target variable is revealed. The correlation coefficient reflects the degree of matching between the comprehensive score and the target variable. It can be seen that the correlation coefficient is 0.9044, indicating that the data fusion effect is good. Figure 5 This is a schematic diagram comparing the comprehensive score of the entropy weight method and the target variable in the embodiment of the present invention, see Figure 5 As shown in the figure, the comprehensive score calculated by the entropy weight method is highly consistent with the trend change of the target variable, indicating that the entropy weight method effectively integrates and evaluates multidimensional features.

[0145] In an embodiment of the present invention, adaptive DTW is used to align multi-source data of the power system, solving the time scale differences and time delay problems between different data sources. The Transformer model is then used to extract features from the aligned data to capture the complex temporal dependencies in the data, further enhancing the validity of the data. Finally, the contribution of different data sources is quantified using the entropy weight method, and the weight distribution of each data source is optimized, thereby improving the overall quality of the fused data. Through this method, information from multiple data sources such as wind power can be effectively fused, thereby improving the accuracy of power system state estimation and the stability of real-time monitoring. In addition, this method can improve the application effect of multi-source data in the power system and avoid estimation bias caused by the heterogeneity of data sources.

[0146] Based on the same inventive concept, an embodiment of the present application also provides a real-time monitoring device for a power system based on multi-source data fusion. Figure 6 This is a schematic diagram of the structure of a real-time monitoring device for a power system based on multi-source data fusion according to an embodiment of the present invention. Figure 6 As shown, the power system real-time monitoring device 600 based on multi-source data fusion may include:

[0147] The data acquisition module 601 is used to obtain multi-source data of the power system, including but not limited to photovoltaic power generation power, load demand, charge and discharge power of the energy storage system, and electrical parameters;

[0148] A preprocessing module 602 is configured to preprocess the multi-source data, including data cleaning, time alignment based on an adaptive DTW algorithm, and normalization processing; wherein the adaptive DTW algorithm achieves time series alignment of the multi-source data by dynamically adjusting the time window and weighted matching strategy;

[0149] Feature extraction module 603, integrating the Transformer model to extract long-term temporal dependency features of multi-source data;

[0150] The weight allocation module 604 is used to dynamically allocate the weight of each data source by adopting the entropy weight method;

[0151] The data fusion and decision module 605 is used to weight and fuse time series features, generate real-time monitoring results and output corresponding decision instructions.

[0152] Among them, the specific monitoring data acquisition method can be as shown in the multi-source data fusion method of the above-mentioned power system. After obtaining the real-time monitoring data of the power system through the above-mentioned method, the corresponding decision-making instructions can be output based on these results. For example, when it is monitored that the photovoltaic power generation power is in excess and the load demand is low, the module will issue an instruction to adjust the charging strategy of the energy storage system, increase the charging power, and store the excess electricity to improve energy utilization; and when the load demand suddenly increases and the photovoltaic power generation power and the energy storage system discharge power cannot meet the demand, the module will issue a startup instruction to other backup power sources to ensure the stability of the power supply. When a fault occurs in the power system, the module can quickly output the corresponding fault handling instructions according to the type and severity of the fault, such as cutting off the fault line, starting the backup equipment, etc., to minimize the impact of the fault on the operation of the power system.

[0153] In some possible implementations, time alignment based on the adaptive DTW algorithm includes:

[0154] Determine a reference data set from the data sets of the multi-source data according to preset conditions, and align the other data sets based on the timestamp of the reference data set;

[0155] For each of the datasets to be aligned, the time window size is dynamically adjusted according to the data volatility. When the data fluctuates violently, the window is expanded. At the same time, a weighted matching strategy is introduced to dynamically adjust the weights based on the importance, volatility or anomalies of the data points and calculate the weighted DTW distance.

[0156] In some possible implementations, dynamically adjusting the time window size based on data volatility is achieved through the following formula:

[0157]

[0158] Among them, D(i,j) is the distance from the starting point to the reference data set D r Time step i and the dataset to be aligned D i The minimum cumulative distance of time step j, dist(D r [i],D i [j]) is the reference dataset D r and the dataset to be aligned D i The point-to-point distance between time steps i and j, W(i,j) is the preset variable window size, D r [i] is the reference dataset D r The data value at time step i, D i [j] is the dataset D to be aligned i The data value at time step j.

[0159] In some possible implementations, the weighted DTW distance is calculated as:

[0160]

[0161] Where M(i,j) is the number of data points per pair D r [i] and D i The weight coefficient of [j] is calculated based on the volatility or rate of change of the data.

[0162] In some possible implementations, in the Transformer model, the calculation formula of the self-attention mechanism is expressed as:

[0163]

[0164] Among them, Q is the query matrix, K is the key matrix, V is the value matrix, d k is the dimension of the key vector, Attention(Q,K,V) is the calculation formula of the self-attention mechanism, and T is the inverse matrix.

[0165] In some possible implementations, the entropy value of each data source in the multi-source data is calculated using an entropy weight method, and a weight is dynamically assigned to quantify the information contribution of each data source, including:

[0166] The probability distribution of the corresponding indicators of each data in the preprocessed multi-source data is calculated and expressed as:

[0167]

[0168] Among them, Pij is the normalized value of the i-th solution on the j-th indicator, Z′ ij For each indicator, S j The sum of the data for each indicator;

[0169] The information entropy of each indicator is calculated according to the probability distribution of each indicator, which is expressed as:

[0170]

[0171] Among them, E j is the information entropy of the j-th indicator, Is a normalization constant used to ensure that the information entropy is between [0,1];

[0172] The weight of each indicator is calculated according to the information entropy, which is expressed as:

[0173]

[0174] Among them, E j is the information entropy value of the jth indicator, n is the total number of indicators; 1-E j It represents the comparison value of information entropy. The smaller the value, the more uncertain the information of the indicator is and the smaller the weight is.

[0175] In some possible implementations, after dynamically allocating weights using the entropy weight method, the method further includes:

[0176] Calculate the comprehensive score of each solution using the following formula:

[0177]

[0178] All solutions are ranked according to the comprehensive scores.

[0179] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other. Each embodiment focuses on the differences from other embodiments.

[0180] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some or all of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present application.

Claims

1. A multi-source data fusion method for a power system, characterized in that: include: Acquiring multi-source data of the power system, including but not limited to photovoltaic power generation, load demand, and charge and discharge power of the energy storage system; Preprocessing the multi-source data includes data cleaning, time alignment based on an adaptive DTW algorithm, and normalization processing; wherein the adaptive DTW algorithm achieves time series alignment of the multi-source data by dynamically adjusting the time window and weighted matching strategy; The preprocessed multi-source data is input into the Transformer model, and temporal features are extracted through a multi-layer self-attention mechanism; Calculating the entropy value of each data source in the multi-source data by an entropy weight method, and dynamically assigning weights to quantify the information contribution of each data source; Based on the assigned weights, the time series features are weightedly fused to obtain real-time monitoring data of the power system.

2. The method according to claim 1, characterized in that The time alignment based on the adaptive DTW algorithm includes: Determine a reference data set from the multi-source data set according to preset conditions, and align the other data sets based on the timestamp of the reference data set; For each of the datasets to be aligned, the time window size is dynamically adjusted according to the data volatility. When the data fluctuates violently, the window is expanded. At the same time, a weighted matching strategy is introduced to dynamically adjust the weights based on the importance, volatility or anomalies of the data points and calculate the weighted DTW distance.

3. The method according to claim 2, characterized in that The dynamic adjustment of the time window size according to data volatility is achieved by the following formula: Among them, D(i,j) is the distance from the starting point to the reference data set D r Time step i and the dataset to be aligned D i The minimum cumulative distance of time step j, dist(D r [i],D i [j]) is the reference dataset D r and the dataset to be aligned D i The point-to-point distance between time steps i and j, W(i,j) is the preset variable window size, D r [i] is the reference dataset D r The data value at time step i, D i [j] is the dataset D to be aligned i The data value at time step j.

4. The method according to claim 3, characterized in that The weighted DTW distance is calculated as: Among them, M(i,j) is each pair of data points D r [i] and D i The weight coefficient of [j] is calculated based on the volatility or rate of change of the data.

5. The method according to claim 4, characterized in that In the Transformer model, the calculation formula of the self-attention mechanism is expressed as: Among them, Q is the query matrix, K is the key matrix, V is the value matrix, d k is the dimension of the key vector, Attention() represents the calculation formula of the self-attention mechanism, and T is the inverse matrix.

6. The method according to claim 5, characterized in that The entropy value of each data source in the multi-source data is calculated by the entropy weight method, and the weight is dynamically assigned to quantify the information contribution of each data source, including: The probability distribution of the corresponding indicators of each data in the preprocessed multi-source data is calculated and expressed as: Among them, P ij is the normalized value of the i-th solution on the j-th indicator, Z′ ij For each indicator, S j The sum of the data for each indicator; The information entropy of each indicator is calculated according to the probability distribution of each indicator, which is expressed as: Among them, E j is the information entropy of the j-th indicator, Is a normalization constant used to ensure that the information entropy is between [0,1]; The weight of each indicator is calculated according to the information entropy, which is expressed as: Among them, E j is the information entropy value of the jth indicator, n is the total number of indicators; 1-E j It represents the comparison value of information entropy. The smaller the value, the more uncertain the information of the indicator is and the smaller the weight is.

7. The method according to claim 6, characterized in that After dynamically allocating weights using the entropy weight method, the method further includes: Calculate the comprehensive score of each solution using the following formula: All solutions are ranked according to the comprehensive scores.

8. A real-time monitoring device for power systems based on multi-source data fusion, characterized in that: include: A data acquisition module for acquiring multi-source data of the power system, including but not limited to photovoltaic power generation power, load demand, charge and discharge power of the energy storage system, and electrical parameters; A preprocessing module, configured to preprocess the multi-source data, including data cleaning, time alignment based on an adaptive DTW algorithm, and normalization processing; wherein the adaptive DTW algorithm achieves time series alignment of the multi-source data by dynamically adjusting the time window and weighted matching strategy; The feature extraction module integrates the Transformer model to extract long-term temporal dependency features of multi-source data; A weight allocation module is used to dynamically allocate the weights of each data source by adopting the entropy weight method; The data fusion and decision module is used to weight and fuse time series features, generate real-time monitoring results and output corresponding decision instructions.

Images