Multi-energy load three-stage pattern mining method and system

Abstract

The application provides a multi-energy load three-stage mode mining method and system, which comprises the following steps: in the first stage, multi-energy load measurement data is acquired, a cascade Markov model is constructed, the measurement data is encoded, load time subsegments are generated and matched, and the time mode of the multi-energy load is obtained; in the second stage, the multi-energy load with the same time mode is acquired, a day trend similarity calculation method is adopted, a day trend similarity matrix is constructed, an improved near neighbor propagation model adaptive clustering is input, multi-energy load clusters are generated in the day time scale, and the daily mode of the multi-energy load is obtained; and in the third stage, the curve contour similarity between the load daily modes is used to construct a curve contour similarity matrix, the improved near neighbor propagation model adaptive clustering is input, and the typical contour mode of the multi-energy load is generated. The application comprehensively mines the law of the multi-energy load in three stages, improves the precision of the multi-energy load mode mining, and has practicability.

Description

Technical Field

[0001] This invention relates to the field of integrated energy system technology, specifically to a method and system for mining three-stage patterns of multi-energy loads. Background Technology

[0002] Energy is fundamental to human survival and development. In recent years, due to the impact of the pandemic and extreme weather, many countries have faced energy crises. Simultaneously, with the development of competition in the electricity market, energy suppliers are required to provide high-quality load demand services. Multi-energy systems encompass various loads, including cooling, heating, and electricity, and these loads can be flexibly coordinated through energy storage and conversion equipment. Therefore, understanding the operating modes of multi-energy loads is crucial for better adapting to the ever-changing energy environment and achieving effective energy management. However, multi-energy loads are multi-dimensional time series; not only do the load patterns themselves change, but there are also relative changes and coupling relationships between loads. Furthermore, the load changes over time, and this complex state variation of multi-energy loads presents challenges for pattern discovery.

[0003] Prior to this invention, traditional load pattern mining was mainly achieved through clustering, and it was usually only applicable to single-dimensional electrical loads. Therefore, traditional load pattern mining techniques have many shortcomings, specifically as follows: (1) Load pattern mining lacks a complete framework. Current technologies mainly focus on the research of clustering algorithms, and the mining patterns are relatively simple, lacking comprehensive mining and analysis applications for loads; (2) The load pattern mining methods cannot be fully applied to multi-dimensional multi-energy loads. In addition to the changes in their own loads, multi-energy loads also need to consider the relative changes between multi-dimensional loads; (3) Existing load pattern mining methods are not accurate enough. When mining load patterns, they cannot effectively measure the similarity of time series, thus causing deviations in the pattern mining results; (4) The current load pattern mining process requires manually defining clustering parameters, which is impractical in actual multi-energy load big data. Based on these reasons, the results of traditional load pattern mining are not accurate enough and cannot be widely applied to energy systems. Summary of the Invention

[0004] In view of the above-mentioned shortcomings of the prior art, the present invention provides a three-stage pattern mining method and system for multi-energy loads. This method fully mines the patterns of multi-energy loads by considering the characteristics of multi-energy loads in time series, daily trends and curve profiles, and effectively solves the problem that the traditional load pattern mining results are not accurate enough.

[0005] In a first aspect, the present invention provides a three-stage mode mining method for multi-energy loads, the method comprising the following steps:

[0006] Acquire multi-energy load measurement data as the sample set for the first stage of time pattern mining;

[0007] A cascaded Markov model is constructed to encode the multi-energy load measurement data, obtain the encoding cost of the multi-energy load, and generate load time sub-segments.

[0008] Matching load time segments with the same encoding yields the time pattern of the multi-energy load;

[0009] Obtain multi-energy loads with the same time pattern as a sample set for the second stage of daily pattern mining;

[0010] A daily trend similarity calculation method is used to construct a daily trend similarity matrix, which is then input into an improved nearest neighbor propagation model for adaptive clustering to generate multi-energy daily load clusters.

[0011] The average value of the daily load cluster with the largest number is used to generate the daily pattern of the multi-energy load.

[0012] The daily patterns of the multi-energy loads are obtained as a sample set for the third stage of typical profile pattern mining;

[0013] Based on the curve contour similarity between daily load patterns, a curve contour similarity matrix is ​​constructed, and the improved nearest neighbor propagation model is input for adaptive clustering to generate typical contour patterns of the multi-energy load.

[0014] Preferably, the acquisition of multi-energy load measurement data as the sample set for the first stage of time pattern mining specifically includes:

[0015] Acquire multi-energy load measurement data for the whole year, wherein the multi-energy load measurement data is multi-dimensional time series data, including hourly cooling load, heating load and electrical load data;

[0016] The multi-energy load measurement data have the same magnitude after normalization.

[0017] Preferably, the construction of the cascaded Markov model, encoding the multi-energy load measurement data, obtaining the encoding cost of the multi-energy load, and generating load time sub-segments specifically includes:

[0018] A cascaded Markov model was constructed to discover the state changes of multi-energy loads in the time series based on the inherent laws of the load and the relative changes between loads, and the load data was compressed and encoded.

[0019] The cascaded Markov model, consisting of multiple hidden Markov models, is used to calculate the state probabilities and state transition probabilities of the load over time. The calculation parameters are as follows:

[0020]

[0021] in, Let be the state probability of the load over time. This represents the total number of state patterns of the load over time. This is the state mode transition probability matrix;

[0022] The encoding cost of multi-energy loads is obtained by using the encoding principle of minimum description length. The encoding method with the lowest cost is selected, and the time series composed of the multi-energy load measurement data is divided to generate load time sub-segments.

[0023] The minimum description length is obtained automatically by minimizing the encoding cost of the load time series to obtain the optimal time series segmentation point. The specific formula for calculating the encoding cost of the load time series is as follows:

[0024]

[0025] in, For the encoding cost of the cascaded Markov model, To encode the load using cascaded Markov Encoding cost, For the load dimension, For load time sub-segments, The cost coefficient is encoded for floating-point numbers. The transition probability of the cascaded Markov state pattern.

[0026] Preferably, the time pattern of the multi-energy load includes: the number of load state categories, the time sub-segments contained in each category, and the start and end times of each sub-segment.

[0027] Preferably, the method of employing daily trend similarity calculation to construct a daily trend similarity matrix, inputting an improved nearest neighbor propagation model for adaptive clustering, and generating multi-energy daily load clusters specifically includes:

[0028] The daily trend similarity calculation method combines Euclidean distance and cosine similarity, which can simultaneously measure the distance and directional similarity of load time series points. The specific calculation formula is as follows:

[0029]

[0030] in, , Let ED be the value of the measured load at time t, ED be the Euclidean distance, and CD be the line connecting the load at every two time intervals. and The cosine similarity is given by , while TSD is the desired trend similarity.

[0031] The daily trend similarity matrix is ​​composed of the daily trend similarity distance of the load. The nearest neighbor propagation model uses clustering evaluation index as an adaptive clustering method. The clustering process is completed when the clustering effect is optimal.

[0032] The obtained clustering results can distinguish between rare loads and daily loads, classifying the load clusters with fewer daily loads as rare loads and the daily load clusters with the most daily loads as daily loads.

[0033] Preferably, the step of constructing a curve contour similarity matrix based on the curve contour similarity between daily load patterns, inputting it into an improved nearest neighbor propagation model for adaptive clustering, and generating typical contour patterns of the multi-energy load specifically includes:

[0034] The curve contour similarity calculation method is a windowed dynamic time bending distance, which can measure the shape similarity of the load time series. The specific calculation formula is as follows:

[0035]

[0036]

[0037] in, , For the load data to be measured, and It is a load and The maximum time point, To calculate the cumulative distance using dynamic time bending, the dynamic time bending process is set... The range of values ​​for is specifically [ , ],in The width is the sum of the cumulative distances when the sum is minimized. When the minimum is reached, the obtained For curve contour similarity;

[0038] The curve contour similarity matrix is ​​composed of the contour similarity between daily load patterns. The nearest neighbor propagation model uses clustering evaluation index as an adaptive clustering method. When the clustering effect is optimal, the clustering process is completed. The obtained clustering results can distinguish the contour shape of the load and obtain the typical contour pattern of the load.

[0039] Preferably, the input improved nearest neighbor propagation model adaptive clustering specifically includes:

[0040] The improved nearest neighbor propagation model uses clustering evaluation metrics to adaptively select clustering results. The DB clustering evaluation metric is used as the selection method for the adaptive clustering results of the nearest neighbor propagation model, which can effectively measure the density within clusters with similar workloads and the separation between different clusters. The calculation method for the DB metric is as follows:

[0041]

[0042] in, For the number of cluster categories, For the first The distance from the load sample within a cluster to the load at the cluster center. For the first For the first From the cluster center to the first The distance between cluster centers;

[0043] The smaller the DB index, the higher the density within clusters and the higher the separation between clusters, indicating a better clustering effect.

[0044] Secondly, the present invention provides a three-stage mode mining system for multi-energy loads, the system comprising:

[0045] The time pattern mining module is used to mine the patterns of multi-energy loads in multi-dimensional time series, specifically including:

[0046] The first acquisition unit is used to acquire multi-energy load measurement data as a sample set for the first stage of time pattern mining;

[0047] The first processing unit is used to construct a cascaded Markov model, encode the multi-energy load measurement data, obtain the encoding cost of the multi-energy load, and generate load time sub-segments.

[0048] First generation unit: used to match the load time segments with the same encoding to obtain the time pattern of the multi-energy load;

[0049] The daily pattern mining module is used to uncover patterns in the daily operation of multiple energy loads, specifically including:

[0050] The second acquisition unit is used to acquire multiple energy loads with the same time pattern as a sample set for the second stage of daily pattern mining;

[0051] The second processing unit is used to construct a daily trend similarity matrix using a daily trend similarity calculation method, input the improved nearest neighbor propagation model for adaptive clustering, and generate multi-energy daily load clusters.

[0052] The second generation unit is used to take the average of the most numerous daily load clusters to generate the daily pattern of the multi-energy load.

[0053] The typical profile pattern mining module is used to uncover the patterns of multi-energy loads in terms of curve profiles, specifically including:

[0054] The third acquisition unit is used to acquire the daily patterns of the multi-energy load as a sample set for the third stage of typical profile pattern mining.

[0055] The third processing unit is used to construct a curve contour similarity matrix based on the curve contour similarity between daily load patterns, and input it into the improved nearest neighbor propagation model for adaptive clustering.

[0056] The third generation unit is used to generate typical profile patterns of the multi-energy load.

[0057] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the three-stage mode mining method for multi-energy loads as described in the first aspect of the present invention.

[0058] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the three-stage pattern mining method for multi-energy loads as described in the first aspect of the present invention.

[0059] This invention provides a three-stage multi-energy load pattern mining method and system, establishing a complete framework for multi-energy load pattern mining, including load pattern discovery, segmentation, matching, clustering, and analysis. By leveraging the characteristics and patterns of load data in terms of time series, daily trends, and shape profiles, it achieves the mining of time patterns, daily patterns, and typical patterns of multi-energy loads. The load pattern mining method is applicable to multi-dimensional load data, fully reflecting the relative changes between multi-dimensional loads over time. The proposed load similarity measurement method and improved adaptive nearest neighbor propagation model can improve the accuracy of load clustering results and enhance the precision of multi-energy load pattern mining. Furthermore, the multi-energy load pattern mining process is adaptive without human intervention, relying solely on the inherent characteristics and patterns of the multi-energy load data itself, thus improving the objectivity of the pattern mining results. By analyzing the application of multi-energy load patterns in energy systems, it is possible to fully utilize the characteristics of multi-energy loads to provide guidance for multi-energy systems in load management and scheduling planning, demonstrating practicality. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a flowchart of a three-stage pattern mining method for multi-energy loads provided in an embodiment of the present invention;

[0062] Figure 2 This is a framework diagram for mining multi-energy load patterns in an embodiment of the present invention;

[0063] Figure 3 This is a multi-energy load state transition diagram based on a cascaded Markov model in an embodiment of the present invention;

[0064] Figure 4 This is a diagram illustrating the iterative process of mining multi-energy load time patterns in an embodiment of the present invention;

[0065] Figure 5 This is a graph showing the results of the seasonal energy load pattern in an embodiment of the present invention;

[0066] Figure 6 This is a graph showing the energy load time pattern results in an embodiment of the present invention;

[0067] Figure 7 This is a clustering flowchart of the improved adaptive nearest neighbor propagation model in this embodiment of the invention;

[0068] Figure 8 This is a clustering result diagram of daily load trends in an embodiment of the present invention;

[0069] Figure 9 This is a comparison chart of the results of directly taking the daily average load in this embodiment of the invention and the results of the normal mode;

[0070] Figure 10 This is a diagram showing the compression results of a multi-energy load scenario throughout the year in an embodiment of the present invention;

[0071] Figure 11 This is a typical profile pattern result diagram of multi-energy load in an embodiment of the present invention;

[0072] Figure 12 This is a schematic diagram of adaptive parameters for clustering of typical contour patterns in an embodiment of the present invention;

[0073] Figure 13 This is a structural block diagram of a three-stage mode mining system for multi-energy loads provided in an embodiment of the present invention;

[0074] Figure 14 This is a structural block diagram of the time pattern mining module in an embodiment of the present invention;

[0075] Figure 15 This is a structural block diagram of the daily pattern mining module in an embodiment of the present invention;

[0076] Figure 16This is a block diagram of a typical contour pattern mining module in an embodiment of the present invention;

[0077] Figure 17 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0078] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be further described clearly and completely below with reference to the accompanying drawings of the embodiments of this invention. It should be noted that the described embodiments are merely some embodiments of this invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0079] In the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0080] The terms "first" and "second" used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a system, product, or device that includes a series of components or units is not limited to the listed components or units, but may optionally include unlisted components or units, or may optionally include other components or units inherent to such products or devices. In the description of this application, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0081] Multi-energy systems encompass various loads, including cooling, heating, and electricity, which can be flexibly coordinated through energy storage and conversion devices. Therefore, understanding the operating modes of multi-energy loads is crucial for adapting to the ever-changing energy environment and achieving effective energy management. However, multi-energy loads are multi-dimensional time series; not only do the individual load patterns change, but there are also relative changes and coupling relationships between loads. Furthermore, the load changes over time, posing a challenge to pattern mining. Traditional load pattern mining primarily relies on clustering and typically only targets single-dimensional electrical loads, resulting in insufficient accuracy and limited applicability to energy systems.

[0082] This invention provides a three-stage pattern mining method and system for multi-energy loads. This method fully mines the patterns of multi-energy loads by considering their characteristics in time series, daily trends, and shape profiles, thereby improving the accuracy and practicality of load pattern mining.

[0083] According to a first aspect of the present invention, a method for mining three-stage patterns of multi-energy loads is provided.

[0084] Figure 1 This is a flowchart of a three-stage mode mining method for multi-energy load according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0085] S1. Obtain multi-energy load measurement data as the sample set for the first stage of time pattern mining;

[0086] S2. Construct a cascaded Markov model to encode the multi-energy load measurement data, obtain the encoding cost of the multi-energy load, and generate load time sub-segments;

[0087] S3. Match the load time segments with the same encoding to obtain the time pattern of the multi-energy load;

[0088] S4. Obtain multi-energy loads with the same time pattern as a sample set for the second stage of daily pattern mining;

[0089] S5. Using the daily trend similarity calculation method, construct the daily trend similarity matrix, input the improved nearest neighbor propagation model for adaptive clustering, and generate multi-energy daily load clusters;

[0090] S6. Take the average value of the daily load cluster with the largest number to generate the daily pattern of the multi-energy load;

[0091] S7. Obtain the daily patterns of the multi-energy loads as a sample set for the third stage of typical profile pattern mining;

[0092] S8. Based on the curve contour similarity between daily load patterns, construct a curve contour similarity matrix, input the improved nearest neighbor propagation model for adaptive clustering, and generate typical contour patterns of the multi-energy load.

[0093] Figure 2 This is a framework diagram for multi-energy load pattern mining in one embodiment of the present invention, such as... Figure 2 As shown, the multi-energy load pattern mining framework specifically includes:

[0094] Multi-energy load time pattern mining. Annual multi-energy load data is acquired, including hourly active power for cooling, heating, and electrical loads throughout the year. After normalization, a cascaded Markov model is used to mine the load time series data. Pattern discovery of multi-energy loads is achieved on the time series, with pattern segmentation and matching to obtain the time patterns of multi-energy loads. Its application in load scenario discovery and load forecasting is analyzed.

[0095] Mining of daily patterns of multi-energy loads. Based on time patterns, similar load sub-segments are obtained. The load data within each sub-segment is divided into weekday load and rest day load by date. After normalization, an adaptive nearest neighbor propagation model with trend similarity measure is used to cluster common daily loads and uncommon daily loads. The average of common loads is taken to generate daily patterns of multi-energy loads. The application of this in load scenario reduction and abnormal load behavior detection is analyzed.

[0096] Mining of typical patterns of multi-energy loads. Based on the annual daily patterns of multi-energy loads, the daily patterns are normalized on a daily basis to generate a daily load sample set. An adaptive nearest neighbor propagation model with contour similarity metric is used to cluster contour-similar daily loads to obtain typical patterns of multi-energy loads and analyze their application in demand response.

[0097] Based on the above flowchart and framework diagram, the specific steps of the three-stage multi-energy load pattern mining method in this embodiment of the invention are as follows:

[0098] S1. Obtain multi-energy load measurement data as the sample set for the first stage of time pattern mining;

[0099] Acquire annual data on multi-energy loads. The data is multi-dimensional time series data, including the active power of cooling load, heating load, and electrical load per hour throughout the year.

[0100] In this embodiment of the invention, the data needs to be preprocessed by normalization. By unifying the reference benchmark and range of multi-energy load data, the normalized time series data is generated and is unified in terms of dimensions and magnitude, thus making it comparable.

[0101] Specifically, time pattern mining uses the standard normalization method, and the normalization object is the annual multi-energy load data. Daily pattern mining uses the maximum and minimum value normalization method, and the normalization object is the load time pattern sub-segment data of weekdays and rest days. Typical pattern mining uses the maximum and minimum value normalization method, and the normalization object is the daily load data of daily patterns.

[0102] The calculation formula for the standard normalization method is as follows:

[0103]

[0104] in, The load data is normalized. For raw data of multi-energy load, This is the average load. This represents the standard deviation of the load.

[0105] The calculation formula for the maximum-minimum normalization method is as follows:

[0106]

[0107] in, The load data is normalized. For raw data of multi-energy load, This is the minimum load value. This represents the maximum load value.

[0108] S2. Construct a cascaded Markov model to encode the multi-energy load measurement data, obtain the encoding cost of the multi-energy load, and generate load time sub-segments;

[0109] After obtaining and normalizing the annual multi-energy load measurement data, a joint Markov model is constructed. Data is divided according to the time of load state changes, generating load time segments. Load time segments with similar states are categorized to obtain the time patterns of multi-energy loads, specifically including:

[0110] S21. Construct a cascaded Markov model, discover the state changes of multi-energy loads in the time series based on the load's own laws and the relative changes between loads, and compress and encode the load data.

[0111] In this embodiment of the invention, load states are identified and classified using a cascaded Markov model. Figure 3 This is a multi-energy load state transition diagram based on a cascaded Markov model in an embodiment of the present invention, such as... Figure 3 As shown in the figure, there are two load modes. Mode 1 includes two multi-energy load states, meaning that the cooling, heating, and electrical loads in Mode 1 involve two different relative positional changes. Mode 2 includes three multi-energy load states, among which... Represents the state in pattern r Transition to state The probability, Representative pattern Transfer to mode The probability of.

[0112] The cascaded Markov model, composed of multiple hidden Markov models, identifies the state changes of multi-energy loads over time and calculates the state probabilities of the loads over time. With state transition probability The parameters for cascaded Markov calculations are shown below:

[0113]

[0114] in, Let be the state probability of the load over time. This represents the total number of state patterns of the load over time. This is the state mode transition probability matrix.

[0115] S22. Obtain the encoding cost of multi-energy loads through the encoding principle of minimum description length, select the encoding method with the lowest cost, divide the time series of the multi-energy loads, and generate load time sub-segments.

[0116] The minimum description length automatically obtains the optimal split point of the time series by minimizing the encoding cost of the load time series. The encoding cost calculation method for the load time series is as follows:

[0117]

[0118] in, For the encoding cost of the cascaded Markov model, To encode the load using cascaded Markov Encoding cost, For the load dimension, For load time sub-segments, The cost coefficient is encoded for floating-point numbers. The transition probability of the cascaded Markov state pattern.

[0119] S3. Match the load time segments with the same encoding to obtain the time pattern of the multi-energy load;

[0120] Based on cascaded Markov parameters, load time segments with consistent classification parameters and similar states are identified as time patterns for multi-energy loads. Specifically, a multi-energy load time pattern includes: the number of load state categories, the time segments contained within each category, and the start and end times of each segment.

[0121] In this embodiment of the invention, to verify the effectiveness of the time pattern, the load seasonal pattern using seasonal dates is compared with the load time pattern obtained through data mining, and the average load variance, cooling-to-electricity ratio, and heating-to-electricity ratio of the two patterns are calculated.

[0122] The formula for calculating the average load variance is as follows:

[0123]

[0124] Where V represents the load variance, which reflects the degree of load fluctuation, and the superscripts e, c, and h represent the electrical, cooling, and heating loads, respectively. For the number of time patterns, Represents the time pattern Total load hours, Representative at the Hours Load power data in time mode This represents the average load under the j-th mode. This represents the total number of load hours across all modes.

[0125] The calculation formulas for the cold power ratio and the hot power ratio are as follows:

[0126]

[0127] and These are the load cooling-to-electricity ratio and the heating-to-electricity ratio, respectively. As the average load, the cooling-to-electricity ratio and the heating-to-electricity ratio can reflect the relative proportions of the loads.

[0128] Figure 4 The figure shows an iterative process diagram for mining time patterns of multi-energy loads according to an embodiment of the present invention. As the number of time patterns of loads increases, the encoding cost changes. When the number of time patterns is 6, the multi-energy load is divided into 11 segments. At this time, the encoding cost is the minimum of 36179, which achieves the minimum description length principle. At this time, the segmentation and matching results of the time pattern discovery of multi-energy loads are optimal.

[0129] Figure 5 This is a graph showing the results of the seasonal energy load model in an embodiment of the present invention. Figure 6 This is a graph showing the energy load time pattern results in an embodiment of the present invention, for comparison. Figure 5 and Figure 6 It was found that under the seasonal model, the load was divided into 4 segments, representing 4 different modes, each representing a different season. However, the relative quantities of cooling, heating, and electrical loads changed within the same mode. For example, in seasonal mode 1, the cooling load showed significant changes, leading to instability in the relative values ​​of the multi-energy load (cooling, heating, and electricity). Under the time model, the load was divided into 11 segments, representing 6 modes. Within the same time mode, both the cooling, heating, and electrical loads themselves and their relative values ​​remained relatively stable without significant changes. As shown in Table 1, the average variance of the load was smallest when divided according to the time mode, especially the variance of the cooling load, which was three times smaller than that of the seasonal model. This indicates that loads with similar patterns and fluctuations were grouped into the same time mode.

[0130] Table 1

[0131]

[0132] Table 2

[0133]

[0134] Table 2 shows the cooling-to-electricity ratio and the heating-to-electricity ratio. It can be seen that the cooling-to-electricity ratio and the heating-to-electricity ratio under the time pattern are wider in range than those under the seasonal pattern, which are [0.699, 3.424] and [0.294, 0.428], respectively. This indicates that the time pattern can more finely divide the relative changes in the quantity of multi-energy loads and can provide richer information on the multi-energy load ratio. The above results verify the effectiveness of the time pattern in load pattern discovery, segmentation and matching.

[0135] Multi-energy load time patterns can be applied to load scenario discovery and load forecasting in multi-energy systems. Since multi-energy load patterns can change with load energy consumption behavior, this algorithm, based on the principle of minimum description length, mines the temporal variation patterns of multi-energy load data without human intervention. When the state of multi-energy loads changes, it can effectively and automatically discover new load patterns and scenarios. Furthermore, the time pattern matches and categorizes loads with similar variation patterns, thus it can be used to select days with similar times in multi-energy load forecasting. Moreover, within the sub-segments divided by the time pattern, the coupling relationship of multi-energy loads is relatively stable, allowing for a more detailed mining of multi-energy load variation patterns and, consequently, prediction of future trends in multi-energy loads.

[0136] S4. Obtain multi-energy loads with the same time pattern as a sample set for the second stage of daily pattern mining;

[0137] Based on time pattern mining, the maximum-minimum normalization method is used to normalize the load time pattern sub-segments of weekdays and rest days, and obtain multi-energy loads with the same time pattern. These are used as the sample set for the second stage of daily pattern mining to further explore the daily patterns of multi-energy loads.

[0138] S5. Using a daily trend similarity calculation method, a daily trend similarity matrix is ​​constructed. This matrix is ​​then input into an improved nearest neighbor propagation model for adaptive clustering, generating multi-energy daily load clusters. Specifically, this includes:

[0139] S51. Calculate the similarity of daily load trends for multiple energy sources under the same time pattern, and construct a load trend similarity matrix;

[0140] S52. Construct an improved adaptive nearest neighbor propagation model, input the similarity matrix to cluster the load, and generate multi-energy daily load clusters.

[0141] Figure 7 This is a flowchart of the clustering process of the improved adaptive nearest neighbor propagation model in this embodiment of the invention, as shown below. Figure 7As shown, after inputting the similarity matrix S, the representativeness matrix R and the effectiveness matrix A are initialized. Matrices R and A are iteratively updated using nearest neighbor propagation. When matrices R and A are no longer updated, the candidate clustering results are obtained. The diagonal elements u of the similarity matrix are updated with a certain step size, where u ranges from [min...]. , max The representativeness matrix R and validity matrix A are repeatedly calculated iteratively to obtain candidate clustering results. The DB index of the selected clustering results is calculated, and the best clustering result is adaptively selected.

[0142] The formula for calculating the daily load trend similarity is as follows:

[0143]

[0144] in, , Let ED be the load value measured at time t, ED be the Euclidean distance, and CD be the line connecting the load at every two time intervals. and The cosine similarity is given by , and the trend similarity is given by TSD.

[0145] The calculation formula for nearest neighbor propagation is as follows:

[0146]

[0147]

[0148]

[0149]

[0150] Where S is the similarity matrix, R is the representation matrix, A is the validity matrix, the superscript "new" represents the updated matrix, and "old" represents the matrix from the previous iteration. is the damping coefficient, which is usually set to [0.5, 1]. Its function is to enhance the stability of the nearest neighbor propagation model. c represents the load cluster center, that is, the load with the greatest representativeness and effectiveness is selected as the cluster center.

[0151] The DB index is calculated using the following formula:

[0152]

[0153] in, For the number of cluster categories, For the first The distance from the load sample within a cluster to the load at the cluster center. For the first For the first From the cluster center to the first The distance between cluster centers. The smaller the DB index, the higher the density within clusters, the higher the separation between clusters, and the better the clustering effect.

[0154] S6. Take the average value of the daily load cluster with the largest number to generate the daily pattern of the multi-energy load;

[0155] The cluster with the most clusters is designated as the common load cluster, and the average value of these clusters is taken as the daily pattern of multi-energy load. Clusters with fewer clusters are designated as rare load clusters, and their daily load variation trend characteristics are analyzed.

[0156] Figure 8 This is a clustering result diagram of the daily trend of the load in an embodiment of the present invention, such as... Figure 8 As shown, cluster 1 has the largest number of loads and is a common load, while clusters 2, 3, and 4 have fewer loads and are rare loads. Furthermore, the rare loads show a large deviation in daily load values, which is not similar to the trend of the common loads.

[0157] In this embodiment of the invention, the average load value from the above clustering results is taken as the daily load pattern. To verify the effectiveness of the daily load pattern, the control group does not cluster daily loads and directly takes the average load value. Specifically, Figure 9 This is a comparison chart of the results of directly averaging the load on a daily basis in this embodiment of the invention and the results of the normal mode, as shown in the figure. Figure 9 As shown, the daily pattern is similar to the shape and trend of common loads, which can effectively reflect the daily load energy consumption. However, the daily load that is directly taken as the average is affected by rare loads, with a lower initial load value and lower load values ​​at 9 o'clock and 12 o'clock, which cannot truly reflect the daily load energy consumption.

[0158] The daily load model for multi-energy systems can be applied to load scenario compression and abnormal load energy consumption behavior detection in multi-energy systems. The daily model can realistically reflect the common energy consumption status and trends of the load, and can compress the annual energy consumption of the load based on the time pattern division.

[0159] Figure 10 This is a compressed image of the multi-energy load scenario throughout the year in an embodiment of the present invention, such as... Figure 10 As shown, (a) represents the compression of 11 weekday load scenarios, (b) represents the compression of 11 weekend load scenarios, and (c) represents the duration of the compressed scenarios. These results can provide refined load scenarios for annual load planning and daily scheduling, which conform to the actual daily energy consumption patterns of the load. In addition, during the clustering process of similar daily load trends, the detected rare loads have a guiding role in investigating abnormal energy consumption behavior of the load.

[0160] S7. Obtain the daily patterns of the multi-energy loads as a sample set for the third stage of typical profile pattern mining;

[0161] Based on the daily pattern mining, the maximum and minimum value normalization method is used to normalize the daily load data and use it as the sample set for the third stage of typical profile pattern mining to further explore the typical patterns of multi-energy loads.

[0162] S8. Based on the curve contour similarity among daily load patterns, construct a curve contour similarity matrix, input it into the improved nearest neighbor propagation model for adaptive clustering, and generate typical contour patterns of the multi-energy loads, specifically including:

[0163] S81. Calculate the contour similarity of daily patterns of multi-energy loads and construct a load contour similarity matrix;

[0164] S82. Construct an improved adaptive nearest neighbor propagation pattern, input the similarity matrix to cluster the load, and use the clustering results as a typical profile pattern of multi-energy loads.

[0165] S83. Based on the peak and valley values ​​of typical profiles of multi-energy loads, analyze the characteristics of peak and valley changes in energy loads, calculate the bucket entropy of load peak and valley values, and provide guidance for load demand response.

[0166] like Figure 7 The diagram shows the clustering flowchart of the adaptive nearest neighbor propagation model used in the data mining process. After inputting the similarity matrix S, the representativeness matrix R and the effectiveness matrix A are initialized. Matrix R and A are iteratively updated through nearest neighbor propagation. When matrix R and A are no longer updated, the candidate clustering results are obtained. The diagonal elements u of the similarity matrix are updated with a certain step size, and the value range of u is [min...]. , max The representativeness matrix R and validity matrix A are repeatedly calculated iteratively to obtain candidate clustering results. The DB index of the selected clustering results is calculated, and the best clustering result is adaptively selected.

[0167] The formula for calculating the load profile similarity is as follows:

[0168]

[0169]

[0170] in, , For the load data to be measured, and It is a load and The maximum time point, To calculate the cumulative distance using dynamic time bending, the range of values ​​for j in dynamic time bending is set, specifically [ , ], where w is the window width, and the sum of cumulative distances is minimized. When the minimum value is reached, the obtained WDTW is the curve contour similarity.

[0171] The calculation formula for nearest neighbor propagation is as follows:

[0172]

[0173]

[0174]

[0175]

[0176] Where S is the similarity matrix, R is the representation matrix, A is the validity matrix, the superscript "new" represents the updated matrix, and "old" represents the matrix from the previous iteration. is the damping coefficient, which is usually set to [0.5, 1]. Its function is to enhance the stability of the nearest neighbor propagation model. c represents the load cluster center, that is, the load with the greatest representativeness and effectiveness is selected as the cluster center.

[0177] The DB index is calculated using the following formula:

[0178]

[0179] in, For the number of cluster categories, For the first The distance from the load sample within a cluster to the load at the cluster center. For the first For the first From the cluster center to the first The distance between cluster centers. The smaller the DB index, the higher the density within clusters, the higher the separation between clusters, and the better the clustering effect.

[0180] The formula for calculating the bucket entropy is as follows:

[0181]

[0182] Where BE represents bucket entropy, X is the common load value, the superscripts e, c, and h represent electrical, cooling, and heating loads respectively, and the subscripts p and v represent the peak and valley values ​​of the load respectively. Taking the bucket entropy calculation of the load peak as an example, the range of common load peak values ​​is [ , Divide it into equal parts One interval, That is, the number of buckets. This represents the probability distribution of the load peak value in the k-th interval. Similarly, the bucket entropy calculation method for the load trough value is the same.

[0183] In this embodiment of the invention, the daily patterns of cooling, heating, and electrical loads are clustered respectively. Figure 11 These are typical profile patterns of multi-energy loads in embodiments of the present invention. The clustering results of electrical loads are (a), (b), and (c), representing typical single-peak, double-peak, and triple-peak loads, respectively. The clustering results of cooling loads are (d), (e), and (f). The peak value of cluster 1 is around 3-5 PM, and the valley value is around 7 AM. The peak value of cluster 2 is around 10 AM, and the valley value is around 7 PM. The peak value of cluster 3 lasts for a longer period, from 9 AM to around 5 PM, and the valley value is around 8 PM. The clustering results of heat loads are (g) and (h). Cluster 1 has two typical peak values ​​and one valley value in the morning and evening within a day, while cluster 2 has one peak and one valley.

[0184] Typical profile patterns of multi-energy loads can be applied to load demand response analysis. Taking the weekday routine pattern under time pattern 1 in this embodiment as an example, the demand response of multi-energy loads is analyzed. The typical profile patterns of the above-mentioned loads (electricity, cooling, and heat) are (a), (d), and (h). The peak times of the electricity load and the cooling load are close, but the peak times of the electricity and cooling loads are close to the valley times of the heat load. The first peak time of the heat load corresponds to the valley time of the electricity and cooling loads, indicating that the heat load and the cooling and electricity loads operate in a staggered manner, which is beneficial to the peak shaving and valley filling of the total load.

[0185] also, Figure 12 This is a schematic diagram of adaptive parameters for typical contour pattern clustering in an embodiment of the present invention. When the u-parameters of the electric, cooling, and heating loads reach -0.67, -1.13, and -0.85 respectively through a step size of 0.01, the clustering DB index corresponding to the multi-energy load reaches its minimum, thereby adaptively obtaining the best contour clustering result.

[0186] In this embodiment of the invention, using 1kW as the bucket benchmark, the peak-valley bucket entropy of multi-energy load is calculated, and the results are shown in Table 3.

[0187] Table 3

[0188]

[0189] Among them, the peak-valley entropy of the cooling load is the largest, indicating that its peak-valley distribution is relatively uniform and its uncertainty is relatively high, making it suitable for demand response under energy price regulation. Furthermore, the cooling load needs to maintain a certain margin during operation; therefore, cooling energy is suitable as a flexible energy conversion resource for load demand response, and corresponding cooling energy conversion equipment and cold storage equipment can be equipped to ensure the energy supply and consumption balance of the multi-energy system. In addition, the peak-valley entropy of the heat load is the smallest, indicating that its peak-valley distribution has a relatively fixed distribution within the load, its certainty is relatively high, and it represents rigid demand, making it suitable for demand response driven by policy incentives.

[0190] Based on the three-stage pattern mining method for multi-energy loads in the above embodiments, the second aspect of the present invention provides a three-stage pattern mining system for multi-energy loads.

[0191] Figure 13 This is a structural block diagram of a three-stage mode mining system for multi-energy loads provided in an embodiment of the present invention, as shown below. Figure 13 As shown, the system includes:

[0192] The time pattern mining module 1310 is used to mine the patterns of multi-energy loads in multi-dimensional time series. Figure 14 This is a structural block diagram of the time pattern mining module in an embodiment of the present invention, such as... Figure 14 As shown, the time pattern mining module 1310 specifically includes:

[0193] The first acquisition unit 1311 is used to acquire multi-energy load measurement data as a sample set for the first stage of time pattern mining;

[0194] The first processing unit 1312 is used to construct a cascaded Markov model, encode the multi-energy load measurement data, obtain the encoding cost of the multi-energy load, and generate load time sub-segments.

[0195] First generation unit 1313: used to match the same encoded load time segments to obtain the time pattern of the multi-energy load;

[0196] The daily mode mining module 1320 is used to mine the patterns of energy load in daily operation. Figure 15 This is a structural block diagram of the daily pattern mining module in an embodiment of the present invention, such as... Figure 15 As shown, the daily pattern mining module 1320 specifically includes:

[0197] The second acquisition unit 1321 is used to acquire multiple energy loads with the same time pattern as a sample set for the second stage of daily pattern mining.

[0198] The second processing unit 1322 is used to construct a daily trend similarity matrix by using a daily trend similarity calculation method, input the improved nearest neighbor propagation model for adaptive clustering, and generate multi-energy daily load clusters;

[0199] The second generation unit 1323 is used to take the average of the most numerous daily load clusters to generate the daily pattern of the multi-energy load.

[0200] The Typical Profile Pattern Mining Module 1330 is used to mine the patterns of multiple energy loads in terms of curve profiles. Figure 16 This is a typical contour pattern mining module structure block diagram in an embodiment of the present invention, such as... Figure 16 As shown, the typical contour pattern mining module 1330 specifically includes:

[0201] The third acquisition unit 1331 is used to acquire the daily patterns of the multi-energy load as a sample set for the third stage of typical profile pattern mining.

[0202] The third processing unit 1332 is used to construct a curve contour similarity matrix based on the curve contour similarity between daily load patterns and input it into the improved nearest neighbor propagation model for adaptive clustering.

[0203] The third generation unit 1333 is used to generate a typical profile pattern of the multi-energy load.

[0204] Based on the same concept, embodiments of the present invention also provide a schematic diagram of an electronic device structure, such as... Figure 17 As shown, the server may include a processor 1710, a communications interface 1720, a memory 730, and a communication bus 1740. The processor 1710, communications interface 1720, and memory 1730 communicate with each other via the communication bus 1740. The processor 1710 can call logical instructions in the memory 1730 to execute the steps of the three-stage multi-energy load pattern mining method described in the above embodiments. For example, this includes:

[0205] S1. Obtain multi-energy load measurement data as the sample set for the first stage of time pattern mining;

[0206] S2. Construct a cascaded Markov model to encode the multi-energy load measurement data, obtain the encoding cost of the multi-energy load, and generate load time sub-segments;

[0207] S3. Match the load time segments with the same encoding to obtain the time pattern of the multi-energy load;

[0208] S4. Obtain multi-energy loads with the same time pattern as a sample set for the second stage of daily pattern mining;

[0209] S5. Using the daily trend similarity calculation method, construct the daily trend similarity matrix, input the improved nearest neighbor propagation model for adaptive clustering, and generate multi-energy daily load clusters;

[0210] S6. Take the average value of the daily load cluster with the largest number to generate the daily pattern of the multi-energy load;

[0211] S7. Obtain the daily patterns of the multi-energy loads as a sample set for the third stage of typical profile pattern mining;

[0212] S8. Based on the curve contour similarity between daily load patterns, construct a curve contour similarity matrix, input the improved nearest neighbor propagation model for adaptive clustering, and generate typical contour patterns of the multi-energy load.

[0213] The processor 1710 can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations thereof.

[0214] Furthermore, the logical instructions in the aforementioned memory 1730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0215] The memory 1730 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0216] Based on the same concept, embodiments of the present invention also provide a non-transitory computer-readable storage medium storing a computer program containing at least one piece of code executable by a master control device to control the master control device to implement the steps of the multi-energy load three-stage mode mining method as described in the above embodiments. For example, it includes:

[0217] S1. Obtain multi-energy load measurement data as the sample set for the first stage of time pattern mining;

[0218] S2. Construct a cascaded Markov model to encode the multi-energy load measurement data, obtain the encoding cost of the multi-energy load, and generate load time sub-segments;

[0219] S3. Match the load time segments with the same encoding to obtain the time pattern of the multi-energy load;

[0220] S4. Obtain multi-energy loads with the same time pattern as a sample set for the second stage of daily pattern mining;

[0221] S5. Using the daily trend similarity calculation method, construct the daily trend similarity matrix, input the improved nearest neighbor propagation model for adaptive clustering, and generate multi-energy daily load clusters;

[0222] S6. Take the average value of the daily load cluster with the largest number to generate the daily pattern of the multi-energy load;

[0223] S7. Obtain the daily patterns of the multi-energy loads as a sample set for the third stage of typical profile pattern mining;

[0224] S8. Based on the curve contour similarity between daily load patterns, construct a curve contour similarity matrix, input the improved nearest neighbor propagation model for adaptive clustering, and generate typical contour patterns of the multi-energy load.

[0225] Based on the same technical concept, this application also provides a computer program, which, when executed by a main control device, is used to implement the above-described method embodiments.

[0226] The program may be stored, in whole or in part, on a storage medium packaged with the processor, or in part or in whole on a memory not packaged with the processor.

[0227] Based on the same technical concept, this application also provides a processor for implementing the above-described method embodiments. The processor can be a chip.

[0228] In summary, the embodiments of the present invention provide a method and system for mining three-stage patterns of multi-energy loads, which have the following beneficial effects:

[0229] (1) Establish a complete multi-energy load pattern mining framework, including load pattern discovery, segmentation, matching, clustering and analysis; realize the mining of multi-energy load time patterns, daily patterns and typical patterns through the characteristics and patterns of load data in terms of time series, daily trends and shape contours.

[0230] (2) The load pattern mining method is applicable to multidimensional load data and can fully reflect the relative changes of multidimensional loads over time. The proposed load similarity measurement method and the improved adaptive nearest neighbor propagation model can improve the accuracy of load clustering results and improve the accuracy of multi-energy load pattern mining.

[0231] (3) The multi-energy load pattern mining process is self-adaptive without human intervention, relying only on the characteristics and patterns of the multi-energy load data itself, thus improving the objectivity of the pattern mining results.

[0232] (4) Analyzing the application of multi-energy load patterns in energy systems can fully utilize the characteristics of multi-energy loads to provide guidance for multi-energy systems in load management and scheduling planning, and is practical.

[0233] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0234] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0235] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-energy load three-stage pattern mining method, characterized in that, The method includes the following steps: Acquire multi-energy load measurement data as the sample set for the first stage of time pattern mining; A cascaded Markov model is constructed to encode the multi-energy load measurement data, obtain the encoding cost of the multi-energy load, and generate load time sub-segments. Matching load time segments with the same encoding yields the time pattern of the multi-energy load; Obtain multi-energy loads with the same time pattern as a sample set for the second stage of daily pattern mining; A daily trend similarity calculation method is used to construct a daily trend similarity matrix, which is then input into an improved nearest neighbor propagation model for adaptive clustering to generate multi-energy daily load clusters. The average value of the daily load cluster with the largest number is used to generate the daily pattern of the multi-energy load. The daily patterns of the multi-energy loads are obtained as a sample set for the third stage of typical profile pattern mining; Based on the curve contour similarity between daily load patterns, a curve contour similarity matrix is ​​constructed, and the improved nearest neighbor propagation model is input for adaptive clustering to generate typical contour patterns of the multi-energy load.

2. The three-stage mode mining method for multi-energy loads according to claim 1, characterized in that, The acquisition of multi-energy load measurement data, used as the sample set for the first stage of time pattern mining, specifically includes: Acquire multi-energy load measurement data for the whole year, wherein the multi-energy load measurement data is multi-dimensional time series data, including: hourly cooling load, heating load, and electrical load data; The multi-energy load measurement data have the same magnitude after normalization.

3. The method of claim 1, wherein the three-phase pattern mining method is applied to a multi-energy load. The construction of the cascaded Markov model, encoding the multi-energy load measurement data, obtaining the encoding cost of the multi-energy load, and generating load time sub-segments specifically includes: A cascaded Markov model was constructed to discover the state changes of multi-energy loads in the time series based on the inherent laws of the load and the relative changes between loads, and the load data was compressed and encoded. The cascaded Markov model, consisting of multiple hidden Markov models connected together, is used to calculate the state probabilities and state transition probabilities of the load over time. The calculation parameters are as follows: wherein, is the state probability of the load over the time series, r is the total number of state patterns of the load over the time series, is the state pattern transition probability matrix; The encoding cost of multi-energy loads is obtained by using the encoding principle of minimum description length. The encoding method with the lowest cost is selected, and the time series composed of the multi-energy load measurement data is divided to generate load time sub-segments. The minimum description length is obtained automatically by minimizing the encoding cost of the load time series to obtain the optimal time series segmentation point. The specific formula for calculating the encoding cost of the load time series is as follows: in, For the encoding cost of the cascaded Markov model, To encode the load using cascaded Markov Encoding cost, For multi-energy load length, For the load dimension, This represents the number of load time segments. Let be the number of hidden Markov states. For load time sub-segments, The cost coefficient is encoded for floating-point numbers. The transition probability of the cascaded Markov state pattern.

4. The method of claim 1, wherein the three-phase pattern mining method is applied to a multi-energy load. The time pattern of the multi-energy load includes: the number of load state categories, the time sub-segments contained in each category, and the start and end times of each sub-segment.

5. The method of claim 1, wherein the three-phase pattern mining method is applied to a multi-energy load. The method employs a daily trend similarity calculation to construct a daily trend similarity matrix, inputs it into an improved nearest neighbor propagation model for adaptive clustering, and generates multi-energy daily load clusters, specifically including: The daily trend similarity calculation method combines Euclidean distance and cosine similarity, which can simultaneously measure the distance and directional similarity of load time series points. The specific calculation formula is as follows: in, , Let ED be the value of the measured load at time t, ED be the Euclidean distance, and CD be the line connecting the load at every two time intervals. and The cosine similarity is given by , while TSD is the desired trend similarity. The daily trend similarity matrix is ​​composed of the daily trend similarity distance of the load. The nearest neighbor propagation model uses clustering evaluation index as an adaptive clustering method. The clustering process is completed when the clustering effect is optimal. The obtained clustering results can distinguish between rare loads and daily loads, classifying the load clusters with fewer daily loads as rare loads and the daily load clusters with the most daily loads as daily loads.

6. The method of claim 1, wherein the three-phase pattern mining method is applied to a multi-energy load. The process involves constructing a curve contour similarity matrix based on the curve contour similarity between daily load patterns, inputting it into an improved nearest neighbor propagation model for adaptive clustering, and generating typical contour patterns for the multi-energy loads. Specifically, this includes: The curve contour similarity calculation method is a windowed dynamic time bending distance, which can measure the shape similarity of the load time series. The specific calculation formula is as follows: in, , For the load data to be measured, and It is a load and The maximum time point, To calculate the cumulative distance using dynamic time bending, the dynamic time bending process is set... The range of values ​​for is specifically [ , ],in The width is the sum of the cumulative distances when the sum is minimized. When the minimum is reached, the obtained For curve contour similarity; The curve contour similarity matrix is ​​composed of the contour similarity between daily load patterns. The nearest neighbor propagation model uses clustering evaluation index as an adaptive clustering method. When the clustering effect is optimal, the clustering process is completed. The obtained clustering results can distinguish the contour shape of the load and obtain the typical contour pattern of the load.

7. The method of claim 5 or 6, wherein the three-phase pattern mining method is applied to a multi-energy load. The input improved nearest neighbor propagation model adaptive clustering specifically includes: The improved nearest neighbor propagation model uses clustering evaluation metrics to adaptively select clustering results. The DB clustering evaluation metric is used as the selection method for the adaptive clustering results of the nearest neighbor propagation model, which can effectively measure the density within clusters with similar workloads and the separation between different clusters. The calculation method for the DB metric is as follows: in, For the number of cluster categories, For the first The distance from the load sample within a cluster to the load at the cluster center. For the first The distance from the load sample within a cluster to the load at the cluster center. For the first From the cluster center to the first The distance between cluster centers; The smaller the DB index, the higher the density within clusters and the higher the separation between clusters, indicating a better clustering effect.

8. A three-stage excavation system with multi-energy load, characterized in that, The system includes: The time pattern mining module is used to mine the patterns of multi-energy loads in multi-dimensional time series, specifically including: The first acquisition unit is used to acquire multi-energy load measurement data as a sample set for the first stage of time pattern mining; The first processing unit is used to construct a cascaded Markov model, encode the multi-energy load measurement data, obtain the encoding cost of the multi-energy load, and generate load time sub-segments. First generation unit: used to match the load time segments with the same encoding to obtain the time pattern of the multi-energy load; The daily pattern mining module is used to uncover patterns in the daily operation of multiple energy loads, specifically including: The second acquisition unit is used to acquire multiple energy loads with the same time pattern as a sample set for the second stage of daily pattern mining; The second processing unit is used to construct a daily trend similarity matrix using a daily trend similarity calculation method, input the improved nearest neighbor propagation model for adaptive clustering, and generate multi-energy daily load clusters. The second generation unit is used to take the average of the most numerous daily load clusters to generate the daily pattern of the multi-energy load. The typical profile pattern mining module is used to extract patterns in the curve profiles of multiple energy loads, specifically including: The third acquisition unit is used to acquire the daily patterns of the multi-energy load as a sample set for the third stage of typical profile pattern mining. The third processing unit is configured to construct a curve profile similarity matrix according to the curve profile similarity between the load daily patterns, and input the improved near neighbor propagation model for adaptive clustering. The third generating unit is configured to generate a typical profile pattern of the multi-energy load.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the multi-energy load three-stage pattern mining method according to any one of claims 1 to 7 when executing the program. 10.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program implements the steps of the multi-energy load three-stage pattern mining method according to any one of claims 1 to 7 when executed by the processor.

Images