An adjustable load clustering method and system based on longitudinal trend prediction

Abstract

A kind of adjustable load clustering method and system based on longitudinal trend prediction, the adjustable load historical data is divided into different output level equal-probability state interval, the longitudinal load data matrix constructed is converted into state variable matrix;The state transition matrix of different daily load at each time is calculated, the predicted output curve of different load is obtained, the adjustable rate curve of corresponding load is calculated according to predicted load and historical load average;For the adjustable rate curve of different adjustable load obtained based on longitudinal trend prediction, clustering algorithm is used to cluster it, and the final clustering result is output.This application characterizes the adjustable ability of different adjustable load at different levels by adjustable rate and clusters according to it, effectively classifies adjustable load objects with similar adjustable ability, thereby providing a solid foundation for hierarchical zoning aggregation and control scheduling of adjustable load, and promoting efficient use of regional adjustable load resources.

Description

Technical Field

[0001] This invention belongs to the field of energy and power service technology, and more specifically, relates to an adjustable load clustering method and system based on longitudinal trend prediction. Background Technology

[0002] With the widespread integration of diverse loads such as distributed power sources and electric vehicles into the power system, the challenges of load forecasting and dispatch control are gradually increasing. Understanding the electricity consumption behavior of users with different regulation potentials is crucial for load forecasting, demand-side management, and electricity pricing. User electricity load exhibits significant uncertainty; daily load curves reflect a user's electricity consumption behavior throughout the day, representing the horizontal characteristics of the load. However, daily load curves can also differ over a period, such as a week or a month, representing the vertical characteristics of the load. The degree of vertical variation among different users is significant. By studying the vertical trend changes of adjustable loads, we can measure the regulation capacity of different types of adjustable loads, which helps in the reasonable classification of users with varying regulation capabilities.

[0003] Existing research on load clustering mostly focuses on the lateral characteristics of loads. Preprocessing of daily load curves over multiple days often involves simply averaging or selecting a specific day as the typical load day after removing outliers, lacking consideration for the longitudinal characteristics of loads. Furthermore, the preprocessing process can lead to the loss of some useful information. When users' daily load curves vary significantly, making it difficult to characterize their electricity consumption behavior with a single typical load curve, it is necessary to consider the longitudinal fluctuations of the load. By studying longitudinal trends, corresponding adjustability indicators can be established to measure the adjustment capabilities of different adjustable loads. Appropriate clustering methods can then effectively classify adjustable loads with similar adjustment capabilities within a region, promoting the efficient utilization of regional adjustable load resources.

[0004] Chinese patent CN 112884077 A, "Short-term load forecasting method for parks based on shape-based dynamic time-integration clustering", discloses a short-term load forecasting method with the aim of improving forecasting accuracy; however, it does not involve the analysis of load adjustment capacity. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for clustering adjustable loads based on longitudinal trend prediction. The method involves grouping historical data of different adjustable loads at a specific time scale according to different days at the same time. Through normalization, equal-probability state intervals with different output levels are defined, and the constructed longitudinal load data matrix is ​​transformed into a state variable matrix. The state transition matrix of the load at each time point on different days is calculated. Based on the initial state and the transition matrix, the predicted load value at each time point on the target day is calculated, resulting in predicted output curves for different loads. The adjustability rate curves of the corresponding loads are calculated based on the predicted load and the historical load average. For the adjustability rate curves of different adjustable loads obtained based on longitudinal trend prediction, the Canopy-Kmeans clustering algorithm is used to cluster them, outputting the final clustering results. This invention fully considers the longitudinal trend changes of different adjustable loads, characterizes the different levels of adjustability of different adjustable loads through adjustability rate, and clusters them accordingly. Adjustable load objects with similar adjustability capabilities are aggregated, thus providing a solid foundation for hierarchical and partitioned aggregation and control scheduling of adjustable loads.

[0006] The present invention adopts the following technical solution.

[0007] An adjustable load clustering method based on load longitudinal trend prediction specifically includes the following steps:

[0008] Step 1: Select historical adjustable load data from users in different industries, group the historical load data according to different days at the same time, and construct a longitudinal load data matrix;

[0009] Step 2: Normalize the historical load data, divide it into equal probability state intervals for different output levels, and transform the constructed longitudinal load data matrix into a state variable matrix.

[0010] Step 3: Calculate the state transition matrix of the load under different days at each moment. Calculate the load forecast value at each moment of the target day based on the initial state and the transition matrix to obtain the predicted output curve of different loads. Calculate the adjustability curve of the corresponding load based on the predicted load and the historical load average.

[0011] Step 4: For the adjustability rate curves of different adjustable loads obtained based on longitudinal trend prediction, a clustering algorithm is used to cluster them, and the clustering results of typical users under different adjustability levels are output.

[0012] Preferably, in step 1, historical adjustable load data with a time scale of m days is selected, and the daily historical load data is divided into n time points at equal time intervals. A longitudinal load data matrix L is constructed based on the divided load data sequence, and the longitudinal load data matrix L is shown in the following formula:

[0013]

[0014] L i =[L i ,…L i,j …L i,m ],j=1,2,3,…,m

[0015] Among them, L i Let L be a data vector consisting of historical load values ​​from day m at time i; i,j This represents the historical load value at the i-th time point on day j.

[0016] Preferably, in step 2, the load data sequence is first subjected to outlier detection and processing. Methods include, but are not limited to, the 3σ principle: if the load values ​​exceed the range (μ-3σ, μ+3σ), they are considered outliers and removed. Here, μ is the mean of the load data sequence, and σ is the standard deviation of the load data sequence. This processing is to prevent outliers from affecting subsequent state classification, thereby impacting the accuracy of the prediction.

[0017] The load data after outlier processing is normalized and defined as follows:

[0018]

[0019] Among them, L i,min L is the minimum historical load value among the historical load values ​​of day m at time i; i,max It represents the maximum historical load value from day m at time i.

[0020] The load value after normalization satisfies L′ i,j ∈(0,1), divide this range into K state intervals with equal probability, and the interval length is The resulting state interval S is shown below:

[0021] S = (S1,S2,…,S) k ,…,S K )

[0022]

[0023] Where S represents the total state interval that is equally divided into adjustable load outputs; S k To divide and obtain the kth sub-state interval.

[0024] Each historical load value is assigned to a corresponding output state interval according to its magnitude, and the longitudinal load data matrix is ​​transformed into a state matrix E, as shown below:

[0025]

[0026]

[0027] E i,j ∈S

[0028] Among them, E i,j Let i be the load status at time i on day j.

[0029] Preferably, in step 3, the state matrix E at time i on day m is... i There are h different output states [E′1, E′2, E′3, ..., E′] h ] Calculate the state transition probability matrix P i The formula is shown below:

[0030]

[0031]

[0032] E′ a ,E′ b ∈[E′1,E′2,E′3,…,E′ h ]

[0033] Among them, P a,b For state E′ a To state E′ b The probability of transition; N(E′) a →E′ b Let E′ be the state in the state matrix at time i on day m. a To state E′ b The transition statistic, N(E′) a ) represents state E′ a The statistics.

[0034] Convert the load state at time i on day m into a load state probability matrix. As the initial state, the load state probability matrix π at the corresponding time point on the target day m+1 is calculated based on the initial state and the transition matrix. 1 The formula is shown below:

[0035]

[0036] π i,1 ,π i,2 ,…,π i,h ∈[0,1]

[0037]

[0038] The output state with the highest probability in the load state probability matrix at time i on day m+1 is taken as the output state at that point on the prediction day, and the median of the state interval to which it belongs is taken as the output value L′. i,m+1 Furthermore, the predicted output curve for this load on day m+1 is obtained;

[0039] Based on the load forecast value L′ at time i on day m+1 i,m+1 Compared with the historical average load over the past m′ days Calculate the corresponding adjustability λ i,m+1 Thus, the adjustability curve for the target load day is obtained.

[0040]

[0041]

[0042] Where, λ i,m+1 Let λ be the adjustability rate of the load at time i on day m+1 to characterize the adjustment capability of that load point. i,m+1 >1 indicates that the load at that point has an upward adjustable capability, when λ i,m+1 The larger the difference of -1, the stronger the adjustability at that point in time; when 0 < λ i,m+1 <1 indicates that the load at this point has downward adjustable capability, when 1-λ i,m+1 The larger the difference, the stronger its adjustability.

[0043] Preferably, step 4 employs the Canopy-Kmeans clustering algorithm, specifically including the following steps:

[0044] Step 4.1, adjust the target daily adjustability data W for different users. N Perform random permutations and set initial datasets W1, W2, ... W1 respectively. N A center point P is selected from the initial cluster sample set based on three indicators: rate of change, peak-to-trough difference, and average adjustment rate. N ,in,

[0045] Rate of change:

[0046]

[0047] in, G is the average value of the adjustable rate curve. m This represents the maximum value of the adjustability curve;

[0048] Peak-to-valley difference:

[0049] Δ=G m -G n

[0050] Among them, Gn This represents the minimum value of the adjustable rate curve;

[0051] Step 4.2: Select the distance closest to the center point as the distance threshold T. 2-N The distance from the center point to the farthest point is the distance threshold T. 1-N And T 1-N >T 2-N ;

[0052] Step 4.3, P N Point P is designated as the cluster center of the first cluster. N From the initial cluster sample set W N Remove from;

[0053] Step 4.4, from the remaining data sample set W N Randomly select a point Q N Calculate Q N Consider the distances to all known cluster centers, and examine the minimum distance D. N If T 2-N ≤D N ≤T 1-N Then a weak tag is used to record Q. N , representing point Q N Belonging to this cluster, Q N Add it to it; if D N ≤T 2-N Then a strongly marked record point Q is used. N , representing point Q N Belonging to this cluster, Q N From the data sample set S N Delete; if D N >T 1-N Then Q N Form a new cluster, Q N From the data sample set W N Delete;

[0054] Step 4.5, repeat step 4.4 until set W. N The number of elements in it is zero;

[0055] Step 4.6, Generate K N Cluster centers y1, y2, ..., y KN The rate of change, peak-to-valley difference, and average adjustment rate were selected as cluster evaluation indicators.

[0056] Step 4.7: Calculate the similarity between the adjustable rate curve of each user and the cluster evaluation index of the cluster center, add the user to the cluster with the highest similarity to the center, and update the cluster center;

[0057] Step 4.8 iteratively executes step 4.7 until the iteration count reaches 500. This yields typical user clustering results under different adjustable levels.

[0058] An adjustable load clustering system based on load longitudinal trend prediction includes a user historical load data collection module, an adjustable load prediction module, an adjustable rate calculation module, and a cluster analysis module.

[0059] The user historical load data collection module selects m days of historical adjustable load data, groups them according to different days at a unified time, and constructs a longitudinal load data matrix.

[0060] The adjustable load forecasting module normalizes historical adjustable load data, divides it into equal probability intervals for different output levels, and transforms the constructed longitudinal load data matrix into a state variable matrix.

[0061] The adjustability calculation module calculates the state transition matrix of the load under different days at each moment, calculates the load forecast value at each moment of the target day based on the initial state and the transition matrix, obtains the predicted output curve of different loads, and calculates the adjustability curve of the corresponding load based on the predicted load and the historical load average.

[0062] The clustering analysis module uses a clustering algorithm to cluster the adjustability rate curves of different adjustable loads obtained based on longitudinal trend prediction, and outputs the clustering results of typical users under different adjustability levels.

[0063] The beneficial effects of this invention are compared with those of the prior art:

[0064] 1. Clustering algorithms were used to classify user data from different industries in terms of adjustment level and time scale, providing a basis for analyzing the adjustment potential of different types of users. Furthermore, the obtained typical user curves can serve as a basis for calculating the adjustment capacity of this type of load in the future.

[0065] 2. The adjustable load prediction method in this invention can effectively handle the instability that may occur in the input data and effectively reduce the impact of data instability on the prediction results.

[0066] 3. This method uses the maximum probability output and the average historical output to establish an adjustable capacity index, without the need to establish a complex equipment load model, and effectively and quickly measures the adjustable load adjustment capacity. Attached Figure Description

[0067] Figure 1 This is a flowchart of an adjustable load clustering method based on load longitudinal trend prediction according to the present invention.

[0068] Figure 2This is a flowchart of the adjustable load forecasting process in this invention;

[0069] Figure 3 Here is a flowchart of the Canopy-Kmeans clustering algorithm;

[0070] Figure 4 This is a schematic diagram of the original longitudinal load data of user 1 over 31 days in the example.

[0071] Figure 5 This is a schematic diagram of the normalized longitudinal load data in the example.

[0072] Figure 6 This is a schematic diagram of the longitudinal load prediction curve for user 1 at time 0 in the embodiment.

[0073] Figure 7 This is a schematic diagram of the adjustability curves for 24 user prediction days in the example;

[0074] Figure 8 This is a schematic diagram of initializing user curve cluster centers in the embodiment;

[0075] Figure 9 This is a schematic diagram of a typical user curve after clustering in the example. Detailed Implementation

[0076] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, other embodiments obtained by those skilled in the art without creative effort are all within the protection scope of this invention.

[0077] Embodiment 1 of the present invention provides an adjustable load clustering method based on load longitudinal trend prediction, specifically including the following steps:

[0078] Step 1: Select historical adjustable load data from users in different industries, group the historical load data according to different days at the same time, and construct a longitudinal load data matrix;

[0079] In step 1, historical adjustable load data with a time scale of m days is selected. The daily historical load data is divided into n time points at equal time intervals. Based on the divided load data sequence, a longitudinal load data matrix L is constructed. The longitudinal load data matrix L is shown in the following formula:

[0080]

[0081] L i =[L i ,…Li,j …L i,m ],j=1,2,3,…,m

[0082] Among them, L i Let L be a data vector consisting of historical load values ​​from day m at time i; i,j This represents the historical load value at the i-th time point on day j.

[0083] Step 2: Normalize the historical load data, divide it into equal probability state intervals for different output levels, and transform the constructed longitudinal load data matrix into a state variable matrix.

[0084] In step 2, the load data sequence is first subjected to outlier detection and processing. Methods include, but are not limited to, the 3σ principle: if the load values ​​exceed the range (μ-3σ, μ+3σ), they are considered outliers and removed. Here, μ is the mean of the load data sequence, and σ is the standard deviation of the load data sequence. This processing is to prevent outliers from affecting subsequent state classification, thereby impacting the accuracy of the prediction.

[0085] The load data after outlier processing is normalized and defined as follows:

[0086]

[0087] Among them, L i,min L is the minimum historical load value among the historical load values ​​of day m at time i; i,max It represents the maximum historical load value from day m at time i.

[0088] The load value after normalization satisfies L′ i,j ∈(0,1), divide this range into K state intervals with equal probability, and the interval length is The resulting state interval S is shown below:

[0089] S = (S1,S2,…,S) k ,…,S K )

[0090]

[0091] Where S represents the total state interval that is equally divided into adjustable load outputs; S k To divide and obtain the kth sub-state interval.

[0092] Each historical load value is assigned to a corresponding output state interval according to its magnitude, and the longitudinal load data matrix is ​​transformed into a state matrix E, as shown below:

[0093]

[0094] E i =[E i,1 …E i,j …E i,m ],j=1,2,3,…,m

[0095] E i,j ∈S

[0096] Among them, E i,j Let i be the load status at time i on day j.

[0097] Step 3: Calculate the state transition matrix of the load under different days at each moment. Calculate the load forecast value at each moment of the target day based on the initial state and the transition matrix to obtain the predicted output curve of different loads. Calculate the adjustability curve of the corresponding load based on the predicted load and the historical load average.

[0098] In step 3, the state matrix E at time i on day m is... i There are h different output states [E′1, E′2, E′3, ..., E′] h ] Calculate the state transition probability matrix P i The formula is shown below:

[0099]

[0100]

[0101] E′ a ,E′ b ∈[E′1,E′2,E′3,…,E′ h ]

[0102] Among them, P a,b For state E′ a To state E′ b The probability of transition; N(E′) a →E′ b Let E′ be the state in the state matrix at time i on day m. a To state E′ b The transition statistic, N(E′) a ) represents state E′ a The statistics.

[0103] The load state at time i on day m is transformed into a load state probability matrix. As the initial state, the load state probability matrix π at the corresponding time point on the target day m+1 is calculated based on the initial state and the transition matrix. 1 The formula is shown below:

[0104]

[0105] π i,1 ,π i,2 ,…,π i,h ∈[0,1]

[0106]

[0107] The output state with the highest probability in the load state probability matrix at time i on day m+1 is taken as the output state at that point on the prediction day, and the median of the state interval to which it belongs is taken as the output value L′. i,m+1 Furthermore, the predicted values ​​for all time points on that day are obtained, thus yielding the predicted output curve for that load on day m+1.

[0108] Based on the load forecast value L′ at time i on day m+1 i,m+1 Compared with the historical average load over the past m′ days Calculate the corresponding adjustability λ i,m+1 Thus, the adjustability curve for the target load day is obtained.

[0109]

[0110]

[0111] Where, λ i,m+1 Let λ be the adjustability rate of the load at time i on day m+1 to characterize the adjustment capability of that load point. i,m+1 >1 indicates that the load at that point has an upward adjustable capability, when λ i,m+1 The larger the difference of -1, the stronger the adjustability at that point in time; when 0 < λ i,m+1 <1 indicates that the load at this point has downward adjustable capability, when 1-λ i,m+1 The larger the difference, the stronger its adjustability.

[0112] Step 4: For the adjustability rate curves of different adjustable loads obtained based on longitudinal trend prediction, the Canopy-Kmeans clustering algorithm is used to cluster them, and the typical user clustering results under different adjustability levels are output.

[0113] It is worth noting that the clustering algorithm here can be implemented in various forms, including K-means, hierarchical clustering, DBSCAN, etc. To achieve automatic determination of the number of clusters and cluster centers, and to achieve faster speed and more convenient calculation, the preferred embodiment of this invention uses the Canopy-Kmeans clustering algorithm as the implementation method. However, this is only a preferred but non-limiting implementation method. Those skilled in the art can obtain clustering results in any other form within the spirit of this invention, and all such results fall within the protection scope of this invention.

[0114] More specifically, in the preferred but non-limiting embodiments of the present invention

[0115] Step 4.1, adjust the target daily adjustability data W for different users. N Perform random permutations and set initial datasets W1, W2, ... W1 respectively. N A center point P is selected from the initial cluster sample set based on three indicators: rate of change, peak-to-trough difference, and average adjustment rate. N ,in,

[0116] Rate of change:

[0117]

[0118] in, G is the average value of the adjustable rate curve. m This represents the maximum value of the adjustability curve;

[0119] Peak-to-valley difference:

[0120] Δ=G m -G n

[0121] Among them, G n This represents the minimum value of the adjustable rate curve;

[0122] Step 4.2: Select the distance closest to the center point as the distance threshold T. 2-N The distance from the center point to the farthest point is the distance threshold T. 1-N And T 1-N >T 2-N ;

[0123] Step 4.3, P N Point P is designated as the cluster center of the first cluster. N From the initial cluster sample set W N Remove from;

[0124] Step 4.4, from the remaining data sample set W N Randomly select a point Q N Calculate Q N Consider the distances to all known cluster centers, and examine the minimum distance D. N If T 2-N ≤D N ≤T 1-N Then a weak tag is used to record Q. N , representing point Q N Belonging to this cluster, Q N Add it to it; if D N ≤T 2-N Then a strongly marked record point Q is used. N, representing point Q N Belonging to this cluster, Q N From the data sample set S N Delete; if D N >T 1-N Then Q N Form a new cluster, Q N From the data sample set W N Delete;

[0125] Step 4.5, repeat step 4.4 until set W. N The number of elements in it is zero;

[0126] Step 4.6, Generate K N Cluster centers y1, y2, ..., y KN The rate of change, peak-to-valley difference, and average adjustment rate were selected as cluster evaluation indicators.

[0127] Step 4.7: Calculate the similarity between the adjustable rate curve of each user and the cluster evaluation index of the cluster center, add the user to the cluster with the highest similarity to the center, and update the cluster center;

[0128] Step 4.8 is repeated, iterating through step 4.7 until the iteration count reaches 500. This yields typical user clustering results under different adjustable levels.

[0129] In a preferred but non-limiting embodiment of the present invention, the adjustable loads of users in different industries in a certain area are clustered based on load sample data of 24 users in a certain area; wherein, there are 24 users in the area, and the sampling points are 24 per day.

[0130] Figure 4 The original longitudinal load data of user 1 over 24 hours and 31 days in the best embodiment shows that the longitudinal load trend of user 1 is roughly similar at different times, and the longitudinal load curve is relatively stable on weekdays, while non-weekdays are the low period of longitudinal load. Figure 5 The figure shows the normalized longitudinal load data, and it can be seen from the figure that the similarity of the longitudinal load curves is more significant after normalization. Figure 6 The longitudinal load prediction curve for user 1 at time 0 shows that the prediction results are close to the actual values, indicating good prediction performance. Figure 7 Adjustability curves for 24 users on the predicted day are shown. The curves show that the adjustability of these 24 users is between 0.85 and 1.2. The maximum adjustability of these 24 users is mainly distributed between 6:00 and 8:00, that is, the adjustability of these users reaches its peak during this period.

[0131] Figure 8 To initialize the user curve cluster centers, Figure 9These are typical user curves after clustering. Based on curve analysis, Type 1 user curves have one typical curve, exhibiting a clear "bimodal" characteristic, with a load adjustability rate greater than 1, indicating upward adjustability, and a adjustment margin within 12.5%. The maximum adjustment capacity is distributed between 7:00-8:00 and 16:00-17:00, while the minimum adjustment capacity occurs around 20:00. Type 2 user curves have two typical curves, one of which exhibits a "single-peak" characteristic with an adjustability rate greater than 1, showing upward adjustability, and an adjustment margin... Within 15%, the maximum regulating capacity is distributed between 7:00 and 8:00, while the minimum regulating capacity occurs between 2:00 and 3:00 and between 18:00 and 19:00. Another curve exhibits a more "flat" characteristic with a regulateability rate less than 1, indicating a downward adjustable load with a regulating margin within 12%. The maximum regulating capacity is distributed between 6:00 and 7:00 and between 17:00 and 20:00, while the minimum regulating capacity occurs between 3:00 and 4:00 and between 10:00 and 12:00. This method can be used to classify the adjustable potential of different users in a region, providing technical support for power demand-side management and load dispatching.

[0132] Embodiment 2 of the present invention provides an adjustable load clustering system based on load longitudinal trend prediction, which runs an adjustable load clustering method based on load longitudinal trend prediction as described in Embodiment 1, and includes a user historical load data collection module, an adjustable load prediction module, an adjustable rate calculation module, and a cluster analysis module.

[0133] The user historical load data collection module selects m days of historical adjustable load data, groups them according to different days at a unified time, and constructs a longitudinal load data matrix.

[0134] The adjustable load forecasting module normalizes historical adjustable load data, divides it into equal probability intervals for different output levels, and transforms the constructed longitudinal load data matrix into a state variable matrix.

[0135] The adjustability calculation module calculates the state transition matrix of the load under different days at each moment, calculates the load forecast value at each moment of the target day based on the initial state and the transition matrix, obtains the predicted output curve of different loads, and calculates the adjustability curve of the corresponding load based on the predicted load and the historical load average.

[0136] The clustering analysis module uses a clustering algorithm to cluster the adjustability rate curves of different adjustable loads obtained based on longitudinal trend prediction, and outputs the clustering results of typical users under different adjustability levels.

[0137] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0138] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0139] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0140] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0141] 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 it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. An adjustable load clustering method based on longitudinal trend prediction, characterized in that, Includes the following steps: Step 1: Select historical adjustable load data from users in different industries, group the historical adjustable load data according to different days at the same time, and construct a longitudinal load data matrix; Step 2: Normalize the historical adjustable load data, divide it into equal probability intervals for different output levels, and transform the constructed longitudinal load data matrix into a state variable matrix. Step 3: Calculate the state transition matrix of the load under different days at each moment. Calculate the load forecast value at each moment of the target day based on the initial state and the transition matrix to obtain the predicted output curve of different loads. Calculate the adjustability curve of the corresponding load based on the predicted load and the historical load average. Step 4: For the adjustability rate curves of different adjustable loads obtained based on longitudinal trend prediction, a clustering algorithm is used to cluster them to obtain the typical user clustering results under different adjustability levels. Then, based on the adjustment capability of different adjustable loads, hierarchical and partitioned aggregation and control scheduling are performed for adjustable loads. The Canopy-Kmeans clustering method was used. The Canopy-Kmeans clustering algorithm includes: selecting a center point from the initial clustering sample set based on three indicators: rate of change, peak-to-trough difference, and average adjustment rate; and generating... After identifying cluster centers, the rate of change, peak-to-valley difference, and average adjustment rate are selected as cluster evaluation indicators. The similarity between the adjustment rate curve of each user and the cluster evaluation indicators of the cluster center is calculated. Users are added to the cluster with the highest similarity to the center, the cluster center is updated, and the iteration continues until convergence, resulting in typical user clustering results under different adjustment levels.

2. The adjustable load clustering method based on longitudinal trend prediction according to claim 1, characterized in that: In step 1, historical adjustable load data with a time scale of m days is selected. The daily historical load data is divided into n time points at equal time intervals. Based on the divided load data sequence, a longitudinal load data matrix L is constructed. The longitudinal load data matrix L is shown in the following formula: in, A data vector consisting of historical load values ​​from day m at time i; This represents the historical load value at the i-th time point on day j.

3. The adjustable load clustering method based on longitudinal trend prediction according to claim 1, characterized in that: In step 2, outlier detection and processing are performed on the load data sequence from step 1. When the load value distribution exceeds... This range is considered an outlier and is removed. The mean of the load data series. is the standard deviation of the load data sequence.

4. The adjustable load clustering method based on longitudinal trend prediction according to claim 3, characterized in that: In step 2, the normalized load value range is divided into K state intervals with equal probability, and the interval length is... The resulting state interval S is: in, The total state interval is divided equally probabilistically for adjustable load output; To obtain the k-th sub-state interval; Each historical load value is assigned to a corresponding output state interval according to its magnitude, and the longitudinal load data matrix is ​​transformed into a state matrix E, which is: in, Let i be the load status at time i on day j.

5. The adjustable load clustering method based on longitudinal trend prediction according to claim 1, characterized in that: In step 3, the state matrix at time i on day m. There are h different output states. Calculate the state transition probability matrix The formula is shown below: in, For state To state The probability of transition; The state matrix at time i on day m contains the state To state Statistics on transfers For state The statistics.

6. The adjustable load clustering method based on longitudinal trend prediction according to claim 5, characterized in that: Convert the load state at time i on day m into a load state probability matrix. As the initial state, the load state probability matrix at the corresponding time point on the target day m+1 is calculated based on the initial state and the transition matrix. The formula is shown below: 。 7. The adjustable load clustering method based on longitudinal trend prediction according to claim 6, characterized in that: The output state with the highest probability in the load state probability matrix at time i on day m+1 is taken as the output state at that point on the prediction day, and the median of the state interval to which it belongs is taken as the output value. Furthermore, the predicted output curve for this load on day m+1 is obtained; Based on the load forecast value at time i on day m+1 With near Daily historical average load Calculate the corresponding adjustability Thus, the adjustability curve for the target load day is obtained; in, Let the adjustability rate of the load at time i on day m+1 be used to characterize the adjustment capability of that load point. This indicates that the load at this point has an upward adjustable capability, when The larger the difference, the stronger the adjustability at that point in time; when This indicates that the load at this point has downward adjustment capability, when The larger the difference, the stronger its adjustability.

8. An adjustable load clustering system based on longitudinal trend prediction, used to implement the adjustable load clustering method based on longitudinal trend prediction as described in any one of claims 1-7, comprising a user historical load data collection module, an adjustable load prediction module, an adjustable rate calculation module, and a cluster analysis module; characterized in that: The user historical load data collection module selects m days of historical adjustable load data, groups them according to different days at a unified time, and constructs a longitudinal load data matrix. The adjustable load forecasting module normalizes historical adjustable load data, divides it into equal probability intervals for different output levels, and transforms the constructed longitudinal load data matrix into a state variable matrix. The adjustability calculation module calculates the state transition matrix of the load under different days at each moment, calculates the load forecast value at each moment of the target day based on the initial state and the transition matrix, obtains the predicted output curve of different loads, and calculates the adjustability curve of the corresponding load based on the predicted load and the historical load average. The clustering analysis module uses a clustering algorithm to cluster the adjustability rate curves of different adjustable loads obtained based on longitudinal trend prediction, and outputs the clustering results of typical users under different adjustability levels.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-7.

Images