Training-free non-intrusive variable frequency air conditioner load monitoring method

Abstract

The present invention discloses a training-free non-invasive variable frequency air conditioner load monitoring method. According to the variable frequency air conditioner power waveform, the load events of the variable frequency air conditioner are divided into three categories: gentle slope type, step type and peak type; the step type and peak type load events are detected and marked by the difference method, and the gentle slope type load events are detected and marked by the curve segmentation linearization method; by serially reorganizing the detected load events, the variable frequency air conditioner gentle slope type load events that are as complete as possible are obtained; according to the specificity of the air conditioner startup power waveform, the variable frequency air conditioner startup event is screened out from the obtained complete gentle slope type load event, and other events matching the variable frequency air conditioner startup event are further searched, and a legal variable frequency air conditioner load event sequence is formed. The method of the present invention is more adaptable to the randomness of the variable frequency air conditioner operating state change, and the overall detection effect is better. At the same time, it takes into account economic practicality and can effectively promote the air conditioning load to actively participate in the power system demand response.

Description

Technical Field

[0001] The invention belongs to the field of air-conditioning load monitoring and identification, and in particular relates to a training-free non-intrusive variable frequency air-conditioning load monitoring method. Background Art

[0002] In terms of energy consumption, it is foreseeable that the impact of air conditioning load on the power system will become more and more obvious. Given that air conditioning load is a typical flexible load, monitoring air conditioning load can effectively promote residents' air conditioning load to actively participate in the power system demand response and ensure the economic, safe and stable operation of the power grid.

[0003] There are two existing monitoring solutions [1-2] : One is to directly install a measuring device on the air conditioner, which is called intrusive monitoring (ILM), and the other is to identify the working status of the air conditioner by analyzing the total load data, which is called non-intrusive load monitoring (NILM). The latter has the advantages of easy installation, low cost, high reliability, good data integrity and easy and rapid promotion.

[0004] For non-intrusive variable frequency air conditioner load monitoring, there are two types of methods: supervised and unsupervised. The former requires the establishment of a recognition model through learning when the power consumption data samples of the air conditioner in a given scenario are known. The latter can directly establish an air conditioner recognition model by analyzing the original total load data without individually marking the air conditioners in advance. Relatively speaking, the monitoring accuracy of supervised methods is better, but its adaptability to unfamiliar scenes is poor and it is not suitable for large-scale promotion. Although unsupervised methods overcome this defect, in order to establish an air conditioner recognition model, most unsupervised methods usually need to accumulate a certain amount of effective data that can be used for model training and learning, and at the same time, they need to manually adjust the key parameters in the algorithm for different scenarios, which will affect the application effect of such methods in actual engineering.

[0005] [References]

[0006] [1] Cui Can. Research on key technologies of non-intrusive residential load monitoring for smart electricity consumption[D]. North China Electric Power University (Beijing), 2017.

[0007] [2] Yu Yixin, Liu Bo, Luan Wenpeng. Non-intrusive residential power load monitoring and decomposition technology [J]. China Southern Power Grid Technology, 2013, 7(04): 1-5.

[0008] [3]GW Hart. Nonintrusive Appliance Load Monitoring[J].Proceedings of the IEEE, 1992, 80(12):1870-1891.

[0009] [4]Berlin E,Laerhoven K V.An on-line piecewise linear approximationtechnique for wireless sensor networks[C] / / The 35th Annual IEEE Conference onLocal Computer Networks, LCN 2010,10-14October 2010,Denver,Colorado,USA,Proceedings.IEEE,2010.

[0010] [5]Zhao B,Stankovic L,Stankovic V,“On a Training-Less Solution forNon-Intrusive Appliance Load Monitoring Using Graph Signal Processing,”IEEEAccess,vol.4,pp:1784-1799, 2016.

[0011] [6]Oliver Parson,Grant Fisher,April Hersey,et al,“Dataport andNILMTK:A Building Data Set Designed for Non-intrusive Load Monitoring,”in 3rdIEEE Global Conference on Signal and Information Processing(GlobalSIP),pp:210-214,2015.

[0012] [7]Zhao B,He K,Stankovic L,et al.Improving Event-based Non-intrusiveLoad Monitoring using Graph Signal Processing[J].IEEE Access,2018:1-1.

[0013] [8]Liu B, Yu Y, Luan W, et al. An unsupervised electrical appliance modeling framework for non-intrusive load monitoring [C] / / 2017IEEE Power&Energy Society General Meeting (PESGM).IEEE, 2017. Summary of the invention

[0014] In view of the above-mentioned prior art, the present invention provides a training-free non-intrusive variable frequency air conditioner load monitoring method, which is more adaptable to the strong randomness of variable frequency air conditioner operation status changes and has better overall detection effect. At the same time, it takes into account economic practicality, and can effectively promote residents' air conditioning load to actively participate in the power system demand response, and ensure the economic, safe and stable operation of the power grid.

[0015] In order to solve the above technical problems, the present invention proposes a training-free non-intrusive variable frequency air conditioner load monitoring method. In this method, the load events of the variable frequency air conditioner are divided into gentle slope load events, step load events and peak load events according to the variable frequency air conditioner power waveform. The method comprises the following steps:

[0016] Step 1: Adaptive load event detection: Use the difference method to detect and mark step load events and peak load events, and record the detection results one by one in the load event list E; each load event record e includes the start time ts, stop time td, active power increment ΔP, reactive power increment ΔQ and active power change rate K of the load event transient process, and the calculation method is shown in formula (1), formula (2) and formula (3):

[0017] ΔP=P(td)-P(ts) (1)

[0018] ΔQ=Q(td)-Q(ts) (2)

[0019]

[0020] In equations (1) to (3), P(td), P(ts), Q(td), and Q(ts) are the total active and reactive power values ​​of the load at the start and stop time of the transient process, respectively;

[0021] According to the load event detection results, the remaining power time series other than the load event power time series detected by the difference method in the total power time series is split and cut into multiple time segments; each of the split time segments is then subjected to curve piecewise linearization processing to further generate multiple time series sub-segments;

[0022] Step 2: Preprocessing of the candidate variable frequency air conditioner load events: For any two gentle slope load events e in the load event list E i and e j , where i and j are the indexes of the load event in the load event list E. If the gentle slope load event e i and gentle ramp load events j The occurrence time interval Δt is less than the threshold Δt 0 , and the active power growth direction is consistent, then the two gentle slope load events are considered to be a coherent change process and are processed as follows:

[0023] 1) If there is no other load event between the two gentle slope load events, they are directly connected in series and merged into one load event; 2) If there is a step load event or a peak load event between the two gentle slope load events, the step load event or the peak load event is removed from the total load power curve, and then the two gentle slope load events are reorganized into one load event; the above 1) or 2) is iterated one by one for the gentle slope load events in the load event list E, and after each operation, the relevant records in the load event list E are adjusted and updated;

[0024] Threshold Δt 0 Adaptive adjustment needs to be made according to the duration of the two gentle slope load events to be connected in series or recombined, and is determined by formula (4):

[0025] Δt 0 =max{5,0.2×(td j -ts j +td i -ts i )} (4)

[0026] In formula (4), ts i and td i They represent the gentle slope load event e i The start and stop time, ts j and td j They represent the gentle slope load event e j start and stop times;

[0027] Step 3: Identification of variable frequency air conditioners based on load event matching, including:

[0028] Step 3-1) Performing a balance area detection according to the load active power characteristic parameters, wherein the balance area refers to a load operation period including a number of complete working cycles of the electrical appliances, the load working state before and after the period is the same, and the total active power at any time in the period is not less than the total active power at the initial time of the period;

[0029] Step 3-2) Identification of load events during the inverter air conditioner startup process: For all the gentle ramp load events in the load event list E, find the inverter air conditioner startup load events that meet the following two conditions: Condition 1: The duration of the transient process is greater than the threshold t 0 ; Condition 2: The event active power increment is greater than the threshold ΔP th1 ;

[0030] Step 3-3) Generate a candidate set of variable frequency air conditioning load events: remove the load events that meet any of the following conditions from the load event list E: Condition 1: non-slope positive event, that is, ΔP>0; Condition 2: the absolute value of ΔP is less than ΔP th2 The non-slope negative event, that is, ΔP<0, thus forming the variable frequency air conditioning load event candidate list E 1 ;

[0031] Step 3-4) Generate variable frequency air conditioning load event sequence: For list E 1 The events between two adjacent variable frequency air conditioner start-up events in a certain balance area are used to generate all candidate load event sequences that meet the loop sum constraint (ZLSC) and power non-negative constraint. The number of candidate sequences is less than 50. For all candidate sequences, the load event sequence with the smallest zero loop sum error is found according to formula (5) as the final matching result. All events in the sequence are marked as variable frequency air conditioners.

[0032]

[0033] In formula (5), ε p is the active power weight, ε q is the reactive power weight, and ε is taken in residential application scenarios p =0.25,ε q =0.75; N is the number of events in the candidate load event sequence.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] With the advancement of technology, the proportion of variable frequency air conditioners is increasing day by day. Based on the in-depth analysis of the operating mechanism of variable frequency air conditioners and a large amount of actual measurement experience, the adaptive load event detection and variable frequency air conditioner identification based on load event matching proposed in the method of the present invention do not need to establish and implement variable frequency air conditioner identification based on a certain finite state machine model, which can better adapt to the strong randomness of the operating state changes of variable frequency air conditioners. Compared with the existing non-invasive variable frequency air conditioner monitoring algorithm, the overall detection effect is better. At the same time, due to the use of non-invasive monitoring technology, the economic practicality of the proposed framework is taken into account, which can effectively promote the active participation of residential air conditioning loads in the power system demand response and ensure the economic, safe and stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is the power waveform of the startup process of variable frequency air conditioner, electric water heater and refrigerator;

[0037] Figure 2 It is a schematic diagram of the training-free non-intrusive variable frequency air conditioner load monitoring framework of the present invention;

[0038] Figure 3 This is a diagram showing the effect of adaptive load event detection in the present invention;

[0039] Figure 4 Schematic diagram of the load event preprocessing process of the alternative variable frequency air conditioner in the present invention, wherein the line segments with different labels represent the transient processes of different load events respectively;

[0040] Figure 5 It is a schematic diagram of the balance point and balance area detected from the total port power time series in the variable frequency air conditioner identification process based on load event matching described in the present invention. DETAILED DESCRIPTION

[0041] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but the following embodiments are by no means intended to limit the present invention in any way.

[0042] In principle, unlike non-inverter air conditioners, inverter air conditioners need to go through a slow frequency increase process of the compressor in order to reduce the surge current during the startup process and improve the operating efficiency. According to a lot of actual test experience, it takes more than 10 seconds or several minutes for an inverter air conditioner to complete the startup process, and the power consumption during this period does not change monotonically, but may experience multiple fluctuations. This makes the power waveform of the inverter air conditioner startup process have obvious specificity compared with other common electrical appliances. Figure 1 Examples of power waveforms during the startup process of variable frequency air conditioners, electric water heaters, and refrigerators are given. Based on experience, the power waveforms of common electrical load events (corresponding to the start-up, shutdown, and other working state transition processes, etc.) can be divided into Figure 1Therefore, if the load event information can be accurately extracted from the total load information, it will be helpful for variable frequency air conditioning monitoring and identification.

[0043] In addition, during operation, non-inverter air conditioners rely on adjusting the duty cycle of their on-off working cycle to maintain (follow the indoor temperature changes) and adjust the temperature, while inverter air conditioners use power electronics and other technologies to achieve this through stepless speed regulation of the compressor. In actual projects, natural changes in external temperature and manual adjustment of set temperatures by users often occur, and accordingly, inverter air conditioners will produce different levels of power output. Therefore, the on-off working cycle of inverter air conditioners usually contains multiple load events of varying numbers, which is different from other electrical appliances (such as electric water heaters and refrigerators).

[0044] Therefore, in order to make full use of the operating characteristics of variable frequency air conditioners and accurately monitor and identify the operating status of various variable frequency air conditioners in unfamiliar scenarios, the present invention proposes a training-free and non-intrusive variable frequency air conditioner load monitoring framework. Figure 2 As shown in the figure, the framework includes four functional modules: load data acquisition module, adaptive load event detection module, alternative variable frequency air conditioner load event preprocessing module, and variable frequency air conditioner identification module based on load event matching.

[0045] The total load data acquisition module is responsible for continuously providing the necessary target raw data for non-intrusive variable frequency air conditioner detection, which is the total load data for a certain period of time (generally one day), including multi-dimensional data such as voltage, current, active power, reactive power, etc.

[0046] The adaptive load event detection module has a built-in differential method load event detection algorithm and a curve segmented linearization load event detection algorithm. The differential method load event detection algorithm is responsible for detecting and marking step-type and peak-type load events, and the curve segmented linearization load event detection algorithm is responsible for detecting and marking gentle slope-type load events.

[0047] The alternative variable frequency air conditioner load event preprocessing module is responsible for serially reorganizing the detected load events so as to obtain the variable frequency air conditioner gentle slope load events as complete as possible. In actual projects, on the one hand, in the face of unfamiliar scenes, the complete startup process of the variable frequency air conditioner may be forcibly divided due to unreasonable preset parameters related to the curve segmentation linearization; on the other hand, the startup process lasts for a long time and is more likely to be cut off by load events generated by other electrical appliances. This module can ensure the accuracy of the relevant load characteristics by serially reorganizing them.

[0048] The variable frequency air conditioner identification module based on load event matching can find the variable frequency air conditioner startup event from the gentle slope load events obtained after event detection and preprocessing according to the specificity of the air conditioner startup power waveform, and further find other events that can match the variable frequency air conditioner startup event to form a legal load event sequence, and identify it as an air conditioner, thereby realizing the detection of the variable frequency air conditioner without the need for a training step.

[0049] Based on the above content, a training-free non-intrusive variable frequency air conditioner load monitoring framework is given, that is, a specific implementation method is given for the main modules in the above framework, as shown below:

[0050] Step 1: Adaptive load event detection:

[0051] In order to detect the gentle ramp load events generated by the variable frequency air conditioner, the commonly used difference method needs to adjust the detection threshold value when performing event detection, and its anti-noise ability will be reduced at this time, and a large number of false events will usually be generated. In this regard, the steps of the present invention propose an adaptive load event detection method that can cope with different load event types, including:

[0052] The first step is to use the difference method [3] Detect and mark step-type and peak-type load events, and record the detection results e one by one in the load event list E. Each load event record e includes the start time ts (in seconds, the rest is the same) of the transient process of the load event, the stop time td, the active power increment ΔP, the reactive power increment ΔQ and the active power change rate K, which are calculated as shown in formulas (1), (2) and (3):

[0053] ΔP=P(td)-P(ts) (1)

[0054] ΔQ=Q(td)-Q(ts) (2)

[0055]

[0056] In equations (1) to (3), P(td), P(ts), Q(td), and Q(ts) are the total active and reactive power values ​​of the load at the start and stop times of the transient process, respectively.

[0057] like Figure 3 As shown, the solid line segment marked ① is the detection result of the difference method. At this time, except for the total power time series corresponding to the detected load event, the remaining total power time series has been split and cut into multiple segments.

[0058] In the second step, each total power time series segment is subjected to a curve segment-by-segment linearization process to further generate multiple time series sub-segments. Classic segment-by-segment linearization methods include bottom-up algorithms, top-down algorithms, sliding window algorithms, etc. The present invention does not limit the specific algorithm and can be selected according to the needs during the implementation process. [4] Medium SWAB algorithm.

[0059] When the active power increment ΔP of a sub-segment is greater than the threshold ΔP th0 And the active power change rate K is greater than the threshold K 0 When ΔP is reached, it is marked as a gentle slope load event and recorded in the load event list E in chronological order. th0 =100W, K 0 =15W / s. The processing results are as follows Figure 3 As shown by the line segments marked ② and ③ (the solid line and double dashed line types are used to distinguish adjacent sub-segments).

[0060] Step 2: Preprocessing of load events of alternative variable frequency air conditioners:

[0061] The proposed event detection method can more effectively capture the gentle slope load events generated by the variable frequency air conditioner, laying the foundation for identifying its working state. Figure 4 As shown in (the complete curve composed of multiple line segments is the original active power curve), on the one hand, facing unfamiliar scenes, the complete startup process of the variable frequency air conditioner may be forcibly divided due to unreasonable preset parameters related to the curve segmentation linearization, see Figure 4 (a) The parts marked ① and ②; on the other hand, the startup process lasts longer and is more easily affected by load events generated by other electrical appliances (such as Figure 4 In this regard, this module designs an alternative variable frequency air conditioner load event preprocessing method, the purpose of which is to obtain the variable frequency air conditioner gentle slope load event as complete as possible by serially reorganizing the detected load events, thereby ensuring the accuracy of the relevant load characteristics. Figure 4 The load event preprocessing process is shown (line segments with different labels represent different load event transient processes).

[0062] For any two gentle slope load events e in the load event list E i and e j (i and j are the indexes of the load event in the load event list E), if the gentle slope load event e i and gentle ramp load events j The occurrence time interval Δt is less than the threshold Δt 0, and the active power growth direction is consistent, then the two gentle slope load events are considered to be a coherent change process. Further, they are divided into two cases and handled one by one:

[0063] First, if there is no other load event between the two gentle slope load events, they are directly connected in series to form one load event, and the relevant records in list E are adjusted and updated. For example, Figure 4 The parts marked ① and ② in (a) are obviously part of the variable frequency air conditioner startup process. Due to the different waveform slopes, they are divided into two sections after the curve segmentation linearization process. After connecting the front and back in series, the following is obtained: Figure 4 (b) shows the result (line segment ① and line segment ② are connected in series to form one line segment).

[0064] Second, if there is a step load event or a peak load event between the two gentle ramp load events, it is generally believed that they are not generated by the air conditioner. Here, it is necessary to remove the step load event or the peak load event from the total load power curve, and then reorganize the first and second gentle ramp load events into one load event. For example, Figure 4 (b) The line segments ①, ② and ③ are separated by the step load event generated by the electric water heater, and the green and blue parts are separated by the peak load event generated by the refrigerator. After reorganization, the results are as follows: Figure 4 (c) The area formed by multiple series connection of ①, ②, ④, ⑥ is shown in the bold dashed line segment area.

[0065] The above series reorganization process needs to iterate the gentle slope load events in the load event list E one by one. After each operation, the relevant records in the load event list E are adjusted and updated.

[0066] It should be noted that, in order to ensure the effect, the threshold Δt 0 Adaptive adjustment needs to be made according to the duration of the two gentle slope load events to be connected in series or recombined, which can be calculated and determined using formula (4):

[0067] Δt 0 =max{5,0.2×(td j -ts j +td i -ts i )} (4)

[0068] In formula (4), ts i and td i They represent the gentle slope load event e i The start and stop time of j and td j They represent the gentle slope load event e jstart and stop times.

[0069] Step 3: Identification of variable frequency air conditioners based on load event matching:

[0070] The basic idea of ​​the variable frequency air conditioner identification based on load event matching proposed in this step is: according to the specificity of the air conditioner startup power waveform, find the startup event of the variable frequency air conditioner from the gentle slope load events obtained after event detection and preprocessing, and further find other events that can match the startup event of the variable frequency air conditioner to form a legal load event sequence, and identify it as an air conditioner. Here, the legal load event sequence corresponds to a complete working cycle of the appliance, and generally needs to meet the constraints such as zero cycle and non-negative power. The specific principle can be found in references [5-6].

[0071] The first step is to detect the balance area. In essence, load event matching is to find a combination of load events that meets certain constraints. Generally, the more load events involved in matching, the longer the matching process will take, and the accuracy of the results will be affected to a certain extent. [7] In order to improve the efficiency and accuracy of air conditioning load event matching, the present invention introduces the concept of balance area [8] .

[0072] The balance area detection is performed according to the load active power characteristic parameters. The balance area refers to a period of load operation that includes several complete working cycles of electrical appliances. Its characteristics are: ① The load working state before and after this period is the same, and ② The total active power at any time in this period is not less than the total active power at the initial time of this period. Figure 5 As shown in Figure 1, there are two balancing areas in this period, which overlap in time. According to the definition of balancing areas, events in area 2 do not need to participate in load matching in area 1, which helps to reduce the number of load events involved in matching, improve efficiency, and ensure the accuracy of matching results.

[0073] For the balance area detection, it can be carried out according to its characteristics and characteristic parameters such as load active power. The specific implementation method can be found in reference [8] and will not be repeated here.

[0074] The second step is to identify the load events during the startup process of the variable frequency air conditioner. For all the gentle ramp load events in the load event list E, find the startup load events of the variable frequency air conditioner that meet the following two conditions: ① The duration of the transient process is greater than the threshold t 0 , usually t 0 =18s;②The active power increment of the event is greater than the threshold ΔP th1 , in the specific implementation process, ΔP th1 =200W.

[0075] The third step is to generate a candidate set of variable frequency air conditioner load events. According to the working characteristics of variable frequency air conditioners, which need to reduce the inrush current during the equipment startup process and improve the operating energy efficiency by slowly increasing the frequency of their high-power compressors, it is considered that load events that meet one of the following two conditions are generated by other electrical appliances. Load events that meet any of the following conditions are removed from the load event list E: ① non-slope positive events, that is, ΔP>0; ② the absolute value of ΔP is less than ΔP th2 A non-slope negative event, that is, ΔP < 0, takes ΔP in the specific implementation process. th2 = 200W. Therefore, in order to improve the efficiency of subsequent load event matching, these events are removed from list E, thereby forming a variable frequency air conditioning load event candidate list E 1 .

[0076] Step 4: Generate variable frequency air conditioning load event sequence. 1 The events between two adjacent variable frequency air conditioner start-up events in a certain balance area are generated to generate all candidate load event sequences that meet the loop sum constraint (ZLSC) and power non-negative constraint. Since a large number of non-variable frequency air conditioner load events have been eliminated and the search area is limited, the number of candidate sequences generated in this step is not large. According to actual experience, it is generally less than 50. For all candidate sequences, the load event sequence with the smallest zero loop sum error is found according to formula (5). As the final matching result, all events in the sequence are marked as variable frequency air conditioners.

[0077]

[0078] Among them, ε p is the active power weight, ε q is the reactive power weight; N is the number of events in the candidate load event sequence. According to actual test experience, variable frequency air conditioner startup events are prone to mismatch with high-power electric heating appliances shutdown events. Therefore, appropriately increasing the reactive power weight helps improve accuracy. This invention recommends taking ε in residential application scenarios. p =0.25,ε q =0.75.

[0079] Although the present invention has been described above in conjunction with the accompanying drawings, the present invention is not limited to the above-mentioned specific embodiments. The above-mentioned specific embodiments are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can make many modifications without departing from the purpose of the present invention, which are all within the protection of the present invention.

Claims

1. A training-free, non-intrusive variable frequency air conditioner load monitoring method, wherein the load events of the variable frequency air conditioner are divided into gentle slope load events, step load events and peak load events according to the variable frequency air conditioner power waveform, characterized in that: The method comprises the following steps: Step 1: Adaptive load event detection: The difference method is used to detect and mark step load events and peak load events, and the detection results are recorded one by one in the load event list E. Each load event record e includes the start time ts, stop time td, active power increment ΔP, reactive power increment ΔQ and active power change rate K of the transient process of the load event. The calculation method is shown in formula (1), formula (2) and formula (3): ΔP=P(td)-P(ts) (1) ΔQ=Q(td)-Q(ts) (2) In equations (1) to (3), P(td), P(ts), Q(td), and Q(ts) are the total active and reactive power values ​​of the load at the start and stop time of the transient process, respectively; According to the load event detection result, the remaining power time series other than the load event power time series detected by the difference method in the total power time series is split and cut into multiple time segments; Then, for each split time segment, curve piecewise linearization processing is performed to further generate multiple time series sub-segments; Step 2: Preprocessing of load events of alternative variable frequency air conditioners: For any two gentle slope load events e in the load event list E i and e j , where i and j are the indexes of the load event in the load event list E. If the gentle slope load event e i and gentle ramp load events j If the occurrence time interval Δt is less than the threshold Δt0 and the active power growth direction is consistent, then the two gentle slope load events are considered to be a coherent change process and are processed as follows: 1) If there is no other load event between the two gentle ramp load events, they are directly connected in series and combined into one load event; 2) If there is a step load event or a peak load event between the two gentle ramp load events, the step load event or the peak load event is removed from the total load power curve, and then the two gentle ramp load events are reorganized into one load event; The above 1) or 2) is iterated one by one for the gentle slope load events in the load event list E. After each operation, the relevant records in the load event list E are adjusted and updated; The threshold Δt0 needs to be adaptively adjusted according to the duration of the two gentle slope load events to be connected in series or recombined, and is calculated and determined using formula (4): Δt0=max{5,0.2×(td j -ts j +td i -ts i )} (4) In formula (4), ts i and td i They represent the gentle slope load event e i The start and stop time, ts j and td j They represent the gentle slope load event e j start and stop times; Step 3: Identification of variable frequency air conditioners based on load event matching, including: Step 3-1) Performing a balance area detection according to the load active power characteristic parameters, wherein the balance area refers to a load operation period including a number of complete working cycles of the electrical appliances, the load working state before and after the period is the same, and the total active power at any time in the period is not less than the total active power at the initial time of the period; Step 3-2) Identification of load events during the start-up process of the variable frequency air conditioner: For all the gentle ramp load events in the load event list E, find the start-up load events of the variable frequency air conditioner that meet the following two conditions: Condition 1: The duration of the transient process is greater than the threshold t0; Condition 2: The active power increment of the event is greater than the threshold ΔP th1 ; Step 3-3) Generate a candidate set of variable frequency air conditioning load events: remove the load events that meet any of the following conditions from the load event list E: Condition 1: non-slope positive event, that is, ΔP>0; Condition 2: the absolute value of ΔP is less than ΔP th2 The non-slope negative event, that is, ΔP<0, thus forming the variable frequency air conditioning load event candidate list E1; Step 3-4) Generate variable frequency air conditioner load event sequence: For the events between two adjacent variable frequency air conditioner start-up events in a certain balance area in list E1, generate all candidate load event sequences that satisfy the loop sum constraint (ZLSC) and the power non-negative constraint; for all candidate sequences, find the load event sequence with the smallest zero loop sum error according to formula (5), as the final matching result, all events in the sequence are marked as variable frequency air conditioners; In formula (5), ε p is the active power weight, ε q is the reactive power weight, and N is the number of events in the candidate load event sequence.

Images