Adaptive load event detection method based on signal normalization and two-level judgment

Abstract

The present invention discloses an adaptive load event detection method of signal normalization combined with two-level judgment, which includes the following steps: 1. Collect the signal of the power supply line as the input signal, filter the input signal, and obtain the filtered signal Y; 2. Normalize the filtered signal Y obtained in step 1 to obtain a normalized pre-processed signal; 3. Use a two-level judgment detection algorithm to perform event detection on the normalized pre-processed signal, that is, use the amplitude threshold THR1 to perform load event detection, and use the judgment threshold THR2 to perform effective load event judgment, wherein the judgment threshold THR2 is greater than the amplitude threshold THR1. A set of algorithms can be implemented to uniformly adapt to users of different energy levels and devices with different power change trends for event detection, improve the detection integrity and accuracy of slowly changing, segmented, and oscillating load events, and improve the anti-noise capability.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grid and smart electricity consumption, and relates to an adaptive load event detection method combining signal normalization with two-level determination. Background Art

[0002] As smart grids continue to develop rapidly, the need to optimize power management and improve energy efficiency is becoming more and more prominent. Industrial and commercial loads consume huge amounts of energy, so it is particularly important to monitor and manage the power consumption of industrial and commercial loads. Currently, the research and application of non-intrusive load identification are mainly focused on residential user scenarios, and there is little research on industrial and commercial users.

[0003] Usually, a complete load identification process requires device type identification after event detection. The present invention only includes two parts: signal acquisition processing and load event detection, and does not perform subsequent device type identification.

[0004] Generally, in different industrial and commercial scenarios, the energy consumption levels of equipment vary significantly. It is a common phenomenon that industrial and commercial loads start and stop in stages or slowly change. It is also a common phenomenon for industrial and commercial equipment to have variable capacity or continuous frequent oscillations during operation. These will cause great interference to event detection. However, most of the current research on industrial and commercial event detection is based on the prerequisite that the power of industrial load units is large, the operating state is relatively stable, and the switching is not frequent. Event detection is achieved by setting appropriate active cycle increment difference thresholds and observation cycle numbers. Different industrial and commercial scenarios require separate increment thresholds, and the scenario compatibility is low. This method is not applicable to scenarios where equipment starts and stops in stages or slowly changes, or the operating state is accompanied by frequent oscillations. Obviously, this method is not suitable for complex industrial and commercial power scenarios.

[0005] Therefore, an adaptive load event detection method is needed to solve the above problems. Summary of the invention

[0006] The object of the present invention is to provide an adaptive load event detection method combining signal normalization with two-level determination to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a signal normalization combined with two-level determination adaptive load event detection method, comprising the following steps:

[0008] 1. Collect the signal of the power supply line as the input signal, filter the input signal to obtain the filtered signal Y;

[0009] 2. Normalize the filtered signal Y obtained in step 1 to obtain a normalized preprocessed signal;

[0010] 3. Use a two-level determination detection algorithm to perform event detection on the normalized preprocessed signal, that is, use the amplitude threshold THR1 to perform load event detection, and use the determination threshold THR2 to perform effective load event determination, wherein the determination threshold THR2 is greater than the amplitude threshold THR1.

[0011] Furthermore, the signal of the power supply line in step 1 includes an active signal and / or a reactive signal.

[0012] Furthermore, the filtering process of the input signal in step 1 is specifically as follows: using the filtering interval as the filtering sliding window length, sliding filtering, sorting the signals in the filtering sliding window according to the amplitude, and taking the median as the value of the filtering sliding window after downsampling.

[0013] Furthermore, the filtering sliding window length is 2-5 seconds.

[0014] Furthermore, in step 2, normalizing the filtered signal obtained in step 1 includes the following steps:

[0015] 2.1. Set a normalized sliding window, slide and traverse the filtered signal Y, and take and record a reference value for each normalized sliding window. The reference value is the maximum value of the absolute value of the signal amplitude in the normalized sliding window;

[0016] 2.2. Divide the signal amplitude in each normalized sliding window by the reference value of the normalized sliding window to obtain a normalized preprocessed signal.

[0017] The reference value used for signal normalization is dynamically taken in the time dimension to avoid missing low-power events in high-power scenarios. That is, a normalized sliding window with a fixed window length of W1 is set, and the signal after sliding traversal filtering is traversed. The reference value of each normalized sliding window is taken and recorded. The normalized reference value of the normalized sliding window is the maximum absolute value of the amplitude of the signal in this sliding window. The duration of the start-stop process of industrial and commercial loads and their own operation time are usually long, so the value of W1 needs to be larger.

[0018] The normalized sliding window length in the present invention ranges from 1 hour to 2 hours. Since the switching of electrical equipment in industrial and commercial scenarios is not frequent and the equipment overlap is not high, the normalized sliding window does not need to be too small to reduce unnecessary calculations of the algorithm. The normalized sliding window should not be set too large to ensure the algorithm's adaptability to changes in signal amplitude. After testing, the normalized sliding window value of 1-2 hours is more preferred.

[0019] Furthermore, in step 3, event detection on the normalized preprocessed signal using a two-level determination detection algorithm includes the following steps:

[0020] 3.1. Set the event detection sliding window, traverse the normalized preprocessed signal, and use the amplitude threshold THR1 to perform load event detection every time an event detection sliding window is slid;

[0021] 3.2. If there are events that meet the splicing conditions, event splicing is performed. If there are ended load events, the judgment threshold THR2 is used to perform effective load event judgment.

[0022] In the present invention, the amplitude threshold THR1 ranges from 0.03 to 0.05, and the determination threshold THR2 ranges from 0.2 to 0.3. The values ​​of the amplitude threshold THR1 and the determination threshold THR2 have no units and are the optimal ranges obtained by statistically analyzing the data of multiple experimental results.

[0023] A lower amplitude threshold THR1 is set to detect as many tiny load events as possible with high sensitivity: that is, tiny load events with relatively small amplitude changes and close time are considered to be components of real load events. The amplitude change value is taken as the difference between the medians of two adjacent sliding window signals. A lower amplitude threshold is set as the high-sensitivity detection threshold THR1, and load events with an absolute value of signal amplitude change greater than this threshold are detected. The purpose is to detect as many events as possible to avoid missed detections.

[0024] A higher threshold is set as the judgment threshold THR2 for high-accuracy screening of valid events: a higher amplitude threshold is set as the high-accuracy judgment threshold THR2. For the merged event, a valid event is considered when the change in signal amplitude before and after the event is greater than the threshold THR2. This filters out invalid events caused by interference such as load oscillation, and achieves high-accuracy screening of valid load events.

[0025] Furthermore, the length of the event detection window is 10-30 times the length of the filter window.

[0026] The lower limit of the detection window length is set to 10 signal time points after filtering. If it is lower, the amount of information contained in the detection sliding window will be too small, and unnecessary calculation will be increased to affect the algorithm performance. The upper limit of the detection window length is set to 30 signal time points after filtering to ensure the detection sensitivity. If the detection window length is too large, it will not be possible to distinguish two independent events when the events occur close to each other.

[0027] Furthermore, the load event detection performed in step 3.1 includes the following steps:

[0028] 3.1.1. When traversing to the jth event detection sliding window, calculate the median of the jth event detection sliding window , the median of the j-th event detection sliding window Subtract the median of the j-1th event detection sliding window , get the signal change value ;

[0029] 3.1.2 When When the absolute value of is greater than the amplitude threshold THR1, it is determined that a load event has occurred. If it is positive, it is recorded as a load-on event. A negative value indicates a load-off event.

[0030] Furthermore, if there are events that meet the splicing conditions in step 3.2, event splicing is performed as follows: extract the last load event that occurred before the current event detection sliding window, determine whether the load event detected by the current event detection sliding window meets the splicing conditions, and merge the load events that meet the splicing conditions.

[0031] Furthermore, the splicing condition in step 3.2 is to simultaneously meet the following conditions:

[0032] (1) The current load event and the last load event that occurred before must both be rising or falling;

[0033] (2) The time difference between the current load event and the last load event that occurred before Not greater than the minimum limit LIMIT1; where, in, is the start time of the current load event, is the end time of the last load event that occurred previously;

[0034] (3) Length of load event after splicing The maximum limit LIMIT2 must not be exceeded. This limit is intended to prevent the algorithm from malfunctioning due to excessive event splicing caused by continuous and violent load fluctuations.

[0035] Due to the relatively long and irregular variable capacity interval of variable capacity devices, in order to ensure the completeness of variable capacity device event detection, the algorithm will send information about whether there is a variable capacity device in a certain scene. If there is a variable capacity device, a larger time difference limit is set, and if not, a smaller limit is set, so as to realize adaptive adjustment of the time difference limit to adapt to different scenes.

[0036] Furthermore, in step 3.2, the load events that meet the splicing conditions are merged as follows:

[0037] Assume that the load event detected by the current event detection sliding window is , the load event that meets the splicing conditions is ,Will Merge into , that is, use The end time T_end, the median value after the event occurs and post-event baseline values replace end time, median and baseline values.

[0038] Furthermore, in step 3.2, determining whether there is a completed load event is specifically as follows: calculating the time difference between the end time of the ongoing load event and the starting time of the current event detection sliding window. If the time difference is greater than LIMIT1, the load event is considered to be completed.

[0039] Furthermore, the effective load event determination in step 3.2 is specifically as follows: setting the determination threshold THR2, for a load event that has ended, if the change in signal amplitude before and after the load event occurs is greater than the determination threshold THR2, then the load event is a valid load event, otherwise it is an invalid load event.

[0040] Furthermore, if the current load event is a drop event and the last load event before this load event is a rise event, a supplementary determination process is also included, specifically: if the difference between the start time of the drop event and the end time of the rise event is not greater than the minimum limit LIMIT1, and the difference in signal amplitude after the end of the drop event and before the start of the rise event is greater than the determination threshold THR2, then the two load events are merged into one rise event. In practice, there may be factors such as data anomalies, rapid misoperation, and arcing impact current that lead to false detection, so the above-mentioned supplementary determination logic needs to be added.

[0041] Furthermore, the amplitude threshold THR1 ranges from 0.03 to 0.05, and the determination threshold THR2 ranges from 0.2 to 0.3, which are the optimal ranges obtained by statistically analyzing the data of multiple experimental results.

[0042] Principle of the invention: The present invention adopts a signal normalization combined with a two-level judgment method to detect events. Signal normalization can solve the problem of large differences in energy levels in different industrial and commercial scenarios, fix the signal amplitude between 0 and 1, and thus fix the amplitude threshold; the two-level judgment detection method, that is, using high-sensitivity, high-accuracy two-level amplitude thresholds, using duration and time difference two-level time limits, can completely identify the slowly changing or segmented event process, solve the problem of conventional detection methods missing or incompletely detecting such events, effectively avoid misoperation, equipment load self-capacitance or violent oscillation caused by false detection, and effectively avoid excessive event splicing, improve the ability of the algorithm to handle loads accompanied by violent oscillations.

[0043] Beneficial effect: The adaptive load event detection method of signal normalization combined with two-level judgment of the present invention adds signal filtering and dynamic normalization to realize adaptive adjustment of the signal in the amplitude dimension to adapt to different industrial and commercial scenarios. Combined with the two-level judgment method, a set of algorithms can be implemented to uniformly adapt to users of different energy levels and equipment with different power change trends to perform event detection, effectively improving the detection integrity and accuracy of slowly changing, segmented, and oscillating load events, and improving noise resistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of the present invention;

[0045] Figure 2 : Effect diagram of variable capacity start-up detection of laminating machine in ceramic industry scene;

[0046] Figure 3 : Effect diagram of variable capacitance shutdown detection of laminating machine in ceramic industry scene;

[0047] Figure 4 : Construction site scene pile driver vibration start detection effect diagram;

[0048] Figure 5 : Construction site scene pile driver vibration shutdown detection effect diagram;

[0049] Figure 6 : Electric vehicle charging slow change accompanied by noise shutdown detection effect diagram. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] Step 1: Signal acquisition and filtering

[0052] Use a non-intrusive intelligent terminal to collect the active or reactive signal of the power supply line at high frequency as the input signal, and perform median downsampling filtering on the input signal, that is, set the filter interval, use the filter interval as the sliding window length, perform sliding filtering, sort the signals in the filter sliding window according to the amplitude, and then take the median as the value after this sliding window downsampling. The filter interval here is 3 seconds. This interval range can achieve a relatively ideal filtering effect for different industrial scenarios, and can also retain important feature information.

[0053] Step 2: Signal Dynamic Normalization

[0054] After filtering, the signal is normalized. After signal normalization, the signal amplitude is limited to between 0 and 1. In this way, adaptive adjustment of the signal amplitude is achieved to adapt to different industrial and commercial scenarios, and the compatibility of the same set of amplitude thresholds with scenarios of different energy levels is achieved, without the need to set thresholds for different scenarios.

[0055] The reference value used for signal normalization is dynamically taken in the time dimension. That is, a normalized sliding window with a fixed window length W1 is set as the reference value sliding window, and the signal Y after sliding traversal filtering is recorded. The reference value of each sliding window is taken and recorded. The normalized reference value of the sliding window is the maximum absolute value of the amplitude of the signal Y in this sliding window. The reference value sliding window length is 2 hours. The signal in the sliding window is normalized, that is, the signal amplitude in the sliding window is divided by the sliding window reference value. Assuming that the current normalized sliding window is the i-th sliding window, the signal Yi after normalization of the current sliding window is obtained.

[0056] Step 3: Use a two-level decision method for load event detection

[0057] Set the fixed window length W2 to 1 minute event detection sliding window as the event detection sliding window, and the traversed signal is the normalized signal Yi in the current normalized sliding window. Every time a short window is slid in the time dimension, a small load event detection is performed, and then the event results detected in the previous window are observed. If there is a small event that meets the splicing conditions, event splicing is performed. If there is a load event process that has ended, the effective load event is determined and stored.

[0058] First, high sensitivity detection of small load events

[0059] Set a lower amplitude threshold THR1 to detect as many small load events as possible with high sensitivity. Traverse to the jth event detection sliding window and calculate the median of the jth sliding window , the median value in the jth window Subtract the median of the j-1th window , get the signal change value ,determination Is the absolute value greater than THR1? If so, it is considered that a minor event has occurred. If it is positive, it is recorded as an on event, and if it is negative, it is a off event. Record the event start time t_begin and end time t_end. The start time and end time are taken from the start time of the jth window. Record As the median value after a minor event , As the median value before the occurrence , record the normalized reference values ​​of the j-1th window and the jth window as the reference values ​​before and after the event , , the reference value is multiplied by the normalized signal amplitude to restore the signal.

[0060] Filter out payload events with high accuracy

[0061] (1) Extract the last event that occurred before the current window, determine whether it meets the splicing conditions with the small load event detected in the current window, and merge the small events that meet the splicing conditions. The splicing limit conditions are as follows:

[0062] a) Small load events need to be both rising or both falling;

[0063] b) Time difference of small load events Not greater than the minimum limit LIMIT1.

[0064] in The start time of the current event process. The end time of the process of the event closest to the current event.

[0065] c) The length of the event after splicing The maximum limit LIMIT2 must not be exceeded. Assuming that the sequence number of the event completed by splicing is k, then , is the process end time of the kth event, is the start time of the event process. This restriction is intended to prevent the algorithm from malfunctioning due to excessive splicing of events caused by continuous and violent oscillations in the load.

[0066] Due to the relatively long and irregular capacity-changing intervals of variable-capacity devices, in order to ensure the completeness of the detection of variable-capacity device events, the algorithm will send information about whether there is a variable-capacity device in a certain scene. If there is a variable-capacity device, a larger time limit is set; if not, a smaller limit is set, so as to achieve adaptive adjustment of the time limit to adapt to different scenes.

[0067] The specific steps of event merging are as follows: Assume that the event that meets the splicing conditions is , the tiny events detected in the current window are , then Merge into , that is, use The end time t_end, the median value after the event occurs , Baseline value after the event replace The corresponding value of .

[0068] (2) Determine whether the ongoing event has ended. Calculate the time difference between the end time of the ongoing event and the start time of the current window. If the difference is greater than LIMIT1, the event is considered to have been completed and enter the high-accuracy event screening process.

[0069] (3) High-accuracy event screening process

[0070] A higher threshold is set as the high-accuracy judgment threshold THR2. For an event that has ended, it is considered a valid event only when the change in signal amplitude before and after the event is greater than the threshold THR2, otherwise it is considered invalid.

[0071] (4) If the current event is a falling event, and the event that occurs before it and is most adjacent to it is a rising event, then enter the supplementary judgment process: if the difference between the start time of the current falling event and the end time of the adjacent rising event is not greater than the minimum limit LIMIT1, and the difference in signal amplitude after the end of this falling event and before the start of the rising event is greater than THR2, then these two events are merged into one rising event.

[0072] Figures 2 to 6 This is a detection effect diagram of the opening and closing events of electrical appliances with different power change characteristics in three industrial and commercial scenarios.

[0073] The red legend represents the input signal amplitude curve, and the signal curve segment covered by the blue dots represents the entire event process detected by the algorithm. Among them, the three scenarios of THR1 and THR2 take the same value, that is, THR1 takes 0.05 and THR2 takes 0.3; the laminating machine in the ceramic industry scenario is a variable capacitance appliance, LIMIT1 takes 8 minutes and LIMIT2 takes 20 minutes; the construction site scenario of the pile driver and electric vehicle charging pile, LIMIT1 takes 3 minutes and LIMIT2 takes 10 minutes.

[0074] Depend on Figure 2 arrive Figure 6 It can be seen that the method of the present invention can detect the complete event occurrence process of oscillation, variable capacitance and slow-changing events, and has the ability to resist noise interference.

[0075] The experimental results verify that this algorithm can uniformly adapt to users with different energy levels and devices with different power change trends to perform event detection, improve the detection completeness and accuracy of slowly changing, segmented, and oscillating load events, and enhance noise resistance.

Claims

1. A signal normalization combined with two-level determination adaptive load event detection method, characterized in that: The following steps are involved:

1. Collect the signal of the power supply line as the input signal, filter the input signal to obtain the filtered signal Y; The filtering process of the input signal in step 1 is specifically as follows: using the filtering interval as the filtering sliding window length, sliding filtering, sorting the signals in the filtering sliding window according to the amplitude, and taking the median as the value of the filtering sliding window after downsampling; The filtering sliding window length is 2-5 seconds; 2. Normalize the filtered signal Y obtained in step 1 to obtain a normalized preprocessed signal; In step 2, normalizing the filtered signal obtained in step 1 includes the following steps: 2.

1. Set a normalized sliding window, slide and traverse the filtered signal Y, and take and record a reference value for each normalized sliding window. The reference value is the maximum value of the absolute value of the signal amplitude in the normalized sliding window; 2.

2. Divide the signal amplitude in each normalized sliding window by the reference value of the normalized sliding window to obtain a normalized preprocessed signal; 3. Use a two-level determination detection algorithm to perform event detection on the normalized preprocessed signal, that is, use the amplitude threshold THR1 to perform load event detection, and use the determination threshold THR2 to perform effective load event determination, wherein the determination threshold THR2 is greater than the amplitude threshold THR1; In step 3, event detection of the normalized preprocessed signal using a two-level decision detection algorithm includes the following steps: 3.

1. Set the event detection sliding window, traverse the normalized preprocessed signal, and use the amplitude threshold THR1 to perform load event detection every time an event detection sliding window is slid; The load event detection performed in step 3.1 includes the following steps: 3.1.

1. When traversing to the j-th event detection sliding window, calculate the median of the j-th event detection sliding window , the median of the j-th event detection sliding window Subtract the median of the j-1th event detection sliding window , get the signal change value ; 3.1.2 When When the absolute value of is greater than the amplitude threshold THR1, it is determined that a load event has occurred. If it is positive, it is recorded as a load-on event. If it is negative, it is a load shutdown event. 3.

2. If there are events that meet the splicing conditions, event splicing is performed. If there are ended load events, the judgment threshold THR2 is used to judge the effective load event; If there are events that meet the splicing conditions in step 3.2, event splicing is performed as follows: extract the last load event that occurred before the current event detection sliding window, determine whether the load event detected by the current event detection sliding window meets the splicing conditions, and merge the load events that meet the splicing conditions; The splicing conditions in step 3.2 are to meet the following conditions at the same time: (1) The current load event and the last load event that occurred before must both be rising or falling; (2) The time difference between the current load event and the last load event that occurred before Not greater than the minimum limit LIMIT1; where, in is the start time of the current load event, is the end time of the last load event that occurred previously; (3) Length of load event after splicing The maximum limit LIMIT2 must not be exceeded.

2. The adaptive load event detection method of signal normalization combined with two-level determination according to claim 1 is characterized by: The signal of the power supply line in step 1 includes an active signal and / or a reactive signal.

3. The adaptive load event detection method of signal normalization combined with two-level determination according to claim 1 is characterized in that: The length of the event detection sliding window is 10-30 times the length of the filter window.

4. The adaptive load event detection method of signal normalization combined with two-level determination according to claim 1 is characterized in that: In step 3.2, the load events that meet the splicing conditions are merged as follows: Assume that the load event detected by the current event detection sliding window is , the load event that meets the splicing conditions is ,Will Merge into , that is, use The end time T_end, the median value after the event occurs and post-event baseline values replace end time, median and baseline values.

5. The adaptive load event detection method of signal normalization combined with two-level determination according to claim 1 is characterized in that: In step 3.2, determining whether there is a completed load event is specifically as follows: calculating the time difference between the end time of the ongoing load event and the starting time of the current event detection sliding window. If the time difference is greater than LIMIT1, the load event is considered to be completed.

6. The adaptive load event detection method of signal normalization combined with two-level determination according to claim 1 is characterized by: The effective load event determination in step 3.2 is specifically as follows: setting the determination threshold THR2. For a load event that has ended, if the absolute value of the signal amplitude change before and after the load event occurs is greater than the determination threshold THR2, then the load event is a valid load event, otherwise it is an invalid load event.

7. The adaptive load event detection method of signal normalization combined with two-level determination according to claim 1 is characterized in that: If the current load event is a decreasing event and the last load event that occurred before this load event was an increasing event, a supplementary judgment process is also included, specifically: if the difference between the start time of the decreasing event and the end time of the increasing event is not greater than the minimum limit LIMIT1, and the difference in signal amplitude after the end of the decreasing event and before the start of the increasing event is greater than the judgment threshold THR2, then the two load events are merged into one increasing event.

8. The adaptive load event detection method of signal normalization combined with two-level determination according to claim 1 is characterized by: The amplitude threshold THR1 ranges from 0.03 to 0.05, and the determination threshold THR2 ranges from 0.2 to 0.3.

Images