Household electricity consumption analysis method based on non-intrusive load monitoring and related equipment

Abstract

The invention belongs to the field of non-intrusive load monitoring, and discloses a household electricity consumption analysis method based on non-intrusive load monitoring and related equipment, firstly, non-intrusive monitoring is realized by using household total electricity meter data instead of a single equipment electricity meter, and the complexity and cost of household electricity consumption data collection are reduced; secondly, an unsupervised learning method is adopted to cluster the power load state data, data does not need to be marked in advance, and different load events are automatically identified and classified, so that the problems of load type diversity and high identification difficulty are effectively solved, and the accuracy and efficiency of load event detection are improved; in combination with a depth-first search algorithm and load event constraint conditions, fine search is carried out on load event labels, an undetermined event sequence set is generated, the logicality and continuity of event sequences are enhanced, electricity utilization behavior analysis of the electric equipment is deeper and finer, home users can better understand and optimize own electricity utilization modes, and the user experience is improved. And energy conservation, emission reduction and efficient utilization of energy are promoted.

Description

Technical Field

[0001] The present invention belongs to the technical field of non-invasive load monitoring, and in particular relates to a household electricity consumption analysis method based on non-invasive load monitoring and related equipment. Background Art

[0002] With the improvement of social living standards, household electricity consumption has become an important part of the peak load and peak load of the power grid. The rapid increase in household electricity consumption has brought severe challenges to the safe operation of the power grid. The popularization of advanced metering infrastructure and smart home appliances has created conditions for residents to participate in the load regulation of the power grid. Through the detailed analysis of household electricity consumption data, its demand response potential can be fully utilized. However, the diversification of load types and the personalization of electricity demand have brought more difficulties to the load regulation in the power system.

[0003] As an important way to extract user electricity information and analyze user electricity behavior, non-intrusive load monitoring (NILM) is not only low in complexity, but also has a high accuracy in load identification. By analyzing the relevant data around the event, it can accurately capture the change of the switch state of the equipment, and then accurately identify the power load state. However, the current research on NILM mainly stays on the exploration of load identification algorithms, and there is a lack of relevant research on using NILM to analyze the electricity behavior of household users. At present, most methods for implementing NILM are based on unsupervised event detection, identifying the occurrence of load events from the total electricity consumption signal, and supervised classifiers match load events with electrical equipment to estimate the power trajectory of a single device; the widely used classifiers mainly include support vector machines, neural networks, decision trees and hybrid classification methods; however, the above methods are difficult to classify multi-state devices, resulting in poor accuracy of the classification results for electrical equipment, and it is impossible to accurately analyze the electricity behavior of electrical equipment based on the classification results, which affects the visualization of the carbon footprint of electrical equipment.

[0004] It can be seen that the existing non-intrusive load monitoring methods have poor accuracy in classifying electrical equipment due to the difficulty in classifying multi-state devices, which in turn makes it impossible to accurately analyze the power consumption behavior of electrical equipment based on the classification results. Summary of the invention

[0005] The present invention provides a household electricity consumption analysis method based on non-invasive load monitoring and related equipment to solve the technical problem that the existing non-invasive load monitoring method is difficult to classify multi-state devices, the classification result of the electrical equipment is poor in accuracy, and thus it is impossible to accurately analyze the electricity consumption behavior of the electrical equipment according to the classification result.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: A household electricity consumption analysis method based on non-intrusive load monitoring, comprising: According to the total meter data of household electricity consumption, the load event detection time is obtained; Using unsupervised learning method, multiple groups of power load status data corresponding to electrical equipment within the same load event detection time are clustered to obtain multiple groups of load event labels; Based on the depth-first search algorithm combined with load event constraints, multiple groups of load event labels are searched and a set of pending event sequences is output to realize the power consumption behavior analysis of power-consuming equipment according to the set of pending event sequences.

[0007] Furthermore, the specific steps of obtaining the load event detection time according to the total electric meter data of household electricity consumption include: For the total meter data of household electricity consumption, the sliding window accumulation method and cumulative sum algorithm are used to detect load events and obtain the load event detection time.

[0008] Furthermore, for the total meter data of household electricity consumption, the specific steps of using the sliding window accumulation method and the cumulative sum algorithm to detect load events include: According to the set dynamic change threshold, the detection window gradually slides forward from the starting point; the detection window is divided into a mean value calculation window and a transient detection window; The positive cumulative sum of the load and the negative cumulative sum of the load are calculated respectively according to the power mean of the mean calculation window and the power mean of the transient detection window; According to the comparison result of the positive cumulative sum of the load, the negative cumulative sum of the load and the preset load event threshold, the occurrence time of the load event is obtained, and then the load event detection time is obtained; Among them, k is the starting time corresponding to the detection window; Mean calculation window W1 window power mean E1:

[0009] Transient detection window W2 window power average E2:

[0010] The total length of the detection window is m+n; W1=[P1,…,P m ]; W2=[P m+1 ,…,P m+n ]; Load forward accumulation :

[0011] Negative load accumulation :

[0012] Where: β represents the steady-state power fluctuation value. When the power change is greater than β, it is determined that a load event occurs.

[0013] Furthermore, the specific steps of clustering multiple groups of power load status data corresponding to power consumption equipment within the same load event detection time by using an unsupervised learning method include: Within the same load event detection time, the power load status data corresponding to the power-consuming equipment is obtained; wherein the power load status data at least includes active power, reactive power, harmonic current, start and stop time of the power-consuming equipment, and operating time; The power load state data is used to construct a feature vector as a data set, and the data set is divided into a training set and a test set; The pre-built SOM network model is trained based on the training set, and the trained SOM network model is tested based on the test set, and the load event labels are output to achieve clustering of load event samples of load state transition of power equipment.

[0014] Furthermore, the feature vector is normalized, and the specific formula is as follows:

[0015] In the formula, x t is the total load or raw power of the target device; x mean 、x std Represents the mean and standard deviation of the total load or target equipment power, respectively; is the normalized value of the total load or target equipment power at time t; In the process of constructing the SOM network model, the sparrow search algorithm is used to optimize the initial weights of the SOM network model.

[0016] Furthermore, the specific steps of searching for multiple groups of load event labels based on the depth-first search algorithm combined with the load event constraint conditions include: According to the load event detection time, the event occurrence time and power change amount are obtained; Divide multiple groups of load event labels into increase events and decrease events according to the power change amount; Using the depth-first search algorithm, taking the increase event as the starting point and the event that occurs after the increase event as the next node, searching from top to bottom according to the event occurrence time, outputting the pending sequence set that meets the load event constraint conditions; During the search process, the load event constraints include: first, a non-negative power constraint: the sum of the powers corresponding to all event labels of the output pending event sequence set is greater than zero; second, a zero-loop constraint: the sum of the power changes corresponding to all event labels of the output pending event sequence set is equal to zero.

[0017] Furthermore, the step of searching from top to bottom according to the event occurrence time and outputting a pending sequence set that satisfies the load event constraint condition includes: First, determine whether the event sequence satisfies the power non-negative constraint; if not, stop the path search; if satisfied, continue to determine whether it satisfies the zero-loop and constraint; if satisfied, add the load event label on the current search path to the pending event sequence set; if not satisfied, continue to search downward until the zero-loop and constraint are satisfied; if the zero-loop and constraint are still not satisfied until the last event in the time period, delete this path.

[0018] A household electricity consumption analysis system based on non-intrusive load monitoring, comprising: A load event detection module, used to obtain the load event detection time according to the total electric meter data of household electricity consumption; A load event identification module is used to cluster multiple groups of power load status data corresponding to electrical equipment within the same load event detection time using an unsupervised learning method to obtain multiple groups of load event labels; The power consumption behavior analysis module is used to search multiple groups of load event tags based on the depth-first search algorithm combined with load event constraints, and output a set of pending event sequences to implement power consumption behavior analysis of power-consuming equipment based on the set of pending event sequences.

[0019] The load event detection module is specifically used for: The specific steps of obtaining the load event detection time according to the total electric meter data of household electricity consumption include: For the total meter data of household electricity consumption, the sliding window accumulation method and cumulative sum algorithm are used to detect load events and obtain the load event detection time.

[0020] Furthermore, for the total meter data of household electricity consumption, the specific steps of using the sliding window accumulation method and the cumulative sum algorithm to detect load events include: According to the set dynamic change threshold, the detection window gradually slides forward from the starting point; the detection window is divided into a mean value calculation window and a transient detection window; The positive cumulative sum of the load and the negative cumulative sum of the load are calculated respectively according to the power mean of the mean calculation window and the power mean of the transient detection window; According to the comparison result of the positive cumulative sum of the load, the negative cumulative sum of the load and the preset load event threshold, the occurrence time of the load event is obtained, and then the load event detection time is obtained; Among them, k is the starting time corresponding to the detection window; Mean calculation window W1 window power mean E1:

[0021] Transient detection window W2 window power average E2:

[0022] The total length of the detection window is m+n; W1=[P1,…,P m ]; W2=[P m+1 ,…,P m+n ]; Load forward accumulation :

[0023] Negative load accumulation :

[0024] Where: β represents the steady-state power fluctuation value. When the power change is greater than β, it is determined that a load event occurs.

[0025] The load event identification module is specifically used for: The specific steps of clustering multiple groups of power load status data corresponding to power-consuming equipment within the same load event detection time using an unsupervised learning method include: Within the same load event detection time, the power load status data corresponding to the power-consuming equipment is obtained; wherein the power load status data at least includes active power, reactive power, harmonic current, start and stop time of the power-consuming equipment, and operating time; The power load state data is used to construct a feature vector as a data set, and the data set is divided into a training set and a test set; The pre-built SOM network model is trained based on the training set, and the trained SOM network model is tested based on the test set, and the load event labels are output to achieve clustering of load event samples of load state transition of power equipment.

[0026] Furthermore, the feature vector is normalized, and the specific formula is as follows:

[0027] In the formula, x t is the total load or raw power of the target device; x mean 、x std Represents the mean and standard deviation of the total load or target equipment power, respectively; is the normalized value of the total load or target equipment power at time t; In the process of constructing the SOM network model, the sparrow search algorithm is used to optimize the initial weights of the SOM network model.

[0028] The power consumption behavior analysis module is specifically used for: The specific steps of searching multiple groups of load event labels based on the depth-first search algorithm combined with load event constraints include: According to the load event detection time, the event occurrence time and power change amount are obtained; Divide multiple groups of load event labels into increase events and decrease events according to the power change amount; Using the depth-first search algorithm, taking the increase event as the starting point and the event that occurs after the increase event as the next node, searching from top to bottom according to the event occurrence time, outputting the pending sequence set that meets the load event constraint conditions; During the search process, the load event constraints include: first, a non-negative power constraint: the sum of the powers corresponding to all event labels of the output pending event sequence set is greater than zero; second, a zero-loop constraint: the sum of the power changes corresponding to all event labels of the output pending event sequence set is equal to zero.

[0029] Furthermore, the step of searching from top to bottom according to the event occurrence time and outputting a pending sequence set that satisfies the load event constraint condition includes: First, determine whether the event sequence satisfies the power non-negative constraint; if not, stop the path search; if satisfied, continue to determine whether it satisfies the zero-loop and constraint; if satisfied, add the load event label on the current search path to the pending event sequence set; if not satisfied, continue to search downward until the zero-loop and constraint are satisfied; if the zero-loop and constraint are still not satisfied until the last event in the time period, delete this path.

[0030] Compared with the prior art, the present invention has the following beneficial effects: The present invention also provides a household electricity analysis method based on non-intrusive load monitoring. The method firstly realizes non-intrusive monitoring by directly using the household total meter data instead of the single device meter, which greatly reduces the complexity and cost of household electricity data collection and improves the user experience; secondly, the unsupervised learning method is used to cluster the power load state data, without pre-labeling the data, and automatically identifies and classifies different load events, which effectively solves the problems of load type diversity and difficulty in identification, and improves the accuracy and efficiency of load event detection; furthermore, combined with the depth-first search algorithm and the load event constraint conditions, the load event label is finely searched to generate a pending event sequence set, which not only enhances the logic and continuity of the event sequence, but also makes the power consumption behavior analysis of the electrical equipment more in-depth and meticulous, which helps household users to better understand and optimize their own power consumption patterns, and promotes energy conservation, emission reduction and efficient use of energy. The use of this method has shown significant advantages in improving the convenience, accuracy and depth of household electricity monitoring and power consumption behavior analysis.

[0031] Preferably, in the present invention, the sliding window accumulation method and the cumulative sum algorithm are used for load event detection, which can accurately capture the transient changes of the power load, effectively distinguish different load events, and improve the sensitivity and accuracy of load event detection. The setting of the dynamic change threshold makes the detection method more flexible and can adapt to the changes in different household power consumption patterns.

[0032] Preferably, in the present invention, the specific steps of load event detection include division of detection windows, mean calculation, calculation of load cumulative sum and comparison with preset threshold values. The above steps together constitute an efficient and accurate load event detection process. Through sophisticated algorithm design, the time of occurrence of load events can be more accurately identified, providing a reliable data basis for subsequent analysis.

[0033] Preferably, in the present invention, the power load state data is constructed as a feature vector and clustered through a SOM (self-organizing map) network model, thereby achieving effective classification of load event samples of load state transitions of electrical equipment; this unsupervised learning method not only simplifies the complexity of data processing, but also improves the accuracy and stability of clustering, and the normalization of the feature vector further improves the generalization ability of the model.

[0034] Further preferably, in the present invention, the feature vector is normalized so that feature data of different dimensions and magnitudes can be compared and analyzed at the same scale, thereby improving the training efficiency and prediction accuracy of the model. In addition, the sparrow search algorithm is used to optimize the initial weights of the SOM network model, which helps the model to quickly converge to the global optimal solution, further improving the performance of the model and ensuring the clustering effect.

[0035] Preferably, in the present invention, the load event tags are searched based on the depth-first search algorithm combined with the load event constraints, and a set of pending event sequences that meet specific conditions can be generated. This search strategy not only considers the temporal order relationship between events, but also considers the constraints of the power change amount, so that the generated event sequence is more consistent with the actual power consumption situation, providing more accurate data support for power consumption behavior analysis.

[0036] Preferably, in the present invention, the specific steps of generating a set of pending sequences that meet the load event constraints based on the depth-first search algorithm include judgment conditions for power non-negative constraints and zero-cycle and constraints; the setting of these constraints helps to eliminate search paths that do not conform to actual conditions, thereby improving search efficiency and accuracy of results. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flowchart of a household electricity consumption analysis method based on non-intrusive load monitoring provided by an embodiment of the present invention; Figure 2 A functional architecture diagram of a household electricity consumption analysis method based on non-intrusive load monitoring provided by an embodiment of the present invention; Figure 3 The load event label provided by the present invention and the corresponding load event occurrence time and power change diagram; Figure 4 A power consumption behavior result diagram obtained by decomposing the load event labels provided by the present invention; Figure 5 A load event matching result diagram provided by the present invention; Figure 6 A flowchart of a household electricity consumption analysis method based on non-intrusive load monitoring provided by the present invention; Figure 7 A schematic diagram of the structure of a household electricity analysis system based on non-intrusive load monitoring provided by the present invention. DETAILED DESCRIPTION

[0038] Example 1 The present invention provides a household electricity consumption analysis method based on non-intrusive load monitoring, such as Figure 6 As shown, the following steps are included: According to the total meter data of household electricity consumption, the load event detection time is obtained; For the total meter data of household electricity consumption, the sliding window accumulation method and cumulative sum algorithm are used to detect load events and obtain the load event detection time.

[0039] Here, the specific steps include: According to the set dynamic change threshold, the detection window gradually slides forward from the starting point; the detection window is divided into a mean value calculation window and a transient detection window; The positive cumulative sum of the load and the negative cumulative sum of the load are calculated respectively according to the power mean of the mean calculation window and the power mean of the transient detection window; According to the comparison result of the positive cumulative sum of the load, the negative cumulative sum of the load and the preset load event threshold, the occurrence time of the load event is obtained, and then the load event detection time is obtained; Among them, k is the starting time corresponding to the detection window; Mean calculation window W1 window power mean E1:

[0040] Transient detection window W2 window power average E2:

[0041] The total length of the detection window is m+n; W1=[P1,…,P m ]; W2=[P m+1 ,…,P m+n ]; Load forward accumulation :

[0042] Negative load accumulation :

[0043] Where: β represents the steady-state power fluctuation value. When the power change is greater than β, it is determined that a load event occurs.

[0044] Using unsupervised learning method, multiple groups of power load status data corresponding to electrical equipment within the same load event detection time are clustered to obtain multiple groups of load event labels; Here, the specific steps include: Within the same load event detection time, the power load status data corresponding to the power-consuming equipment is obtained; wherein the power load status data at least includes active power, reactive power, harmonic current, start and stop time of the power-consuming equipment, and operating time; The power load state data is used to construct a feature vector as a data set, and the data set is divided into a training set and a test set; The pre-built SOM network model is trained based on the training set, and the trained SOM network model is tested based on the test set, and the load event labels are output to achieve clustering of load event samples of load state transition of power equipment.

[0045] The eigenvector is normalized, and the specific formula is as follows:

[0046] In the formula, x t is the total load or raw power of the target device; x mean 、x std Represents the mean and standard deviation of the total load or target equipment power, respectively; is the normalized value of the total load or target equipment power at time t; In the process of constructing the SOM network model, the sparrow search algorithm is used to optimize the initial weights of the SOM network model.

[0047] Based on the depth-first search algorithm combined with load event constraints, multiple groups of load event labels are searched and a set of pending event sequences is output to realize the power consumption behavior analysis of power-consuming equipment according to the set of pending event sequences.

[0048] In this embodiment, the specific steps include: According to the load event detection time, the event occurrence time and power change amount are obtained; Divide multiple groups of load event labels into increase events and decrease events according to the power change amount; Using the depth-first search algorithm, taking the increase event as the starting point and the event that occurs after the increase event as the next node, searching from top to bottom according to the event occurrence time, outputting the pending sequence set that meets the load event constraint conditions; During the search process, the load event constraints include: first, a non-negative power constraint: the sum of the powers corresponding to all event labels of the output pending event sequence set is greater than zero; second, a zero-loop constraint: the sum of the power changes corresponding to all event labels of the output pending event sequence set is equal to zero.

[0049] Here, the steps of searching from top to bottom according to the event occurrence time and outputting a pending sequence set that meets the load event constraint conditions include: First, determine whether the event sequence satisfies the power non-negative constraint; if not, stop the path search; if satisfied, continue to determine whether it satisfies the zero-loop and constraint; if satisfied, include the load event label on the current search path into the pending event sequence set; if not satisfied, continue to search downward until the zero-loop and constraint are satisfied; if the zero-loop and constraint are still not satisfied until the last event in the time period, delete the current search path.

[0050] like Figure 7As shown, the present invention also provides a household electricity analysis system based on non-intrusive load monitoring, including: a load event detection module, which is used to obtain the load event detection time according to the total meter data of household electricity consumption; a load event identification module, which is used to cluster multiple groups of power load status data corresponding to electrical equipment within the same load event detection time using an unsupervised learning method to obtain multiple groups of load event labels; a power consumption behavior analysis module, which is used to search multiple groups of load event labels based on a depth-first search algorithm combined with load event constraints, and output a pending event sequence set, so as to realize the power consumption behavior analysis of the electrical equipment according to the pending event sequence set.

[0051] The present invention also provides a device, comprising: a memory for storing a computer program; a processor for implementing the steps of the household electricity analysis method based on non-intrusive load monitoring when executing the computer program.

[0052] When the processor executes the computer program, the steps of the household electricity consumption analysis based on non-intrusive load monitoring are implemented, for example: according to the total electric meter data of household electricity consumption, the load event detection time is obtained; using an unsupervised learning method, multiple groups of power load status data corresponding to electrical equipment within the same load event detection time are clustered to obtain multiple groups of load event labels; based on a depth-first search algorithm combined with load event constraints, multiple groups of load event labels are searched, and a pending event sequence set is output, so as to implement the electricity consumption behavior analysis of the electrical equipment according to the pending event sequence set.

[0053] Alternatively, when the processor executes the computer program, it implements the functions of each module in the above-mentioned system, for example: a load event detection module, which is used to obtain the load event detection time based on the total electricity meter data of household electricity consumption; a load event identification module, which is used to cluster multiple groups of power load status data corresponding to electrical equipment within the same load event detection time using an unsupervised learning method to obtain multiple groups of load event labels; a power consumption behavior analysis module, which is used to search for multiple groups of load event labels based on a depth-first search algorithm combined with load event constraints, and output a set of pending event sequences, so as to implement power consumption behavior analysis of electrical equipment based on the set of pending event sequences.

[0054] Exemplarily, the computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments that can complete preset functions, and the instruction segments are used to describe the execution process of the computer program in the household electricity analysis device based on non-intrusive load monitoring. For example, the computer program can be divided into a load event detection module, a load event identification module, and a power consumption behavior analysis module; the specific functions of each module are as follows: a load event detection module, which is used to obtain the load event detection time according to the total meter data of household electricity consumption; a load event identification module, which is used to cluster multiple groups of power load state data corresponding to electrical equipment within the same load event detection time using an unsupervised learning method to obtain multiple groups of load event labels; a power consumption behavior analysis module, which is used to search for multiple groups of load event labels based on a depth-first search algorithm combined with load event constraints, and output a pending event sequence set to implement power consumption behavior analysis of electrical equipment according to the pending event sequence set.

[0055] The household electricity analysis device based on non-invasive load monitoring can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The household electricity analysis device based on non-invasive load monitoring may include, but is not limited to, a processor and a memory. Those skilled in the art will understand that the above are examples of household electricity analysis devices based on non-invasive load monitoring, and do not constitute a limitation on household electricity analysis devices based on non-invasive load monitoring, and may include more components than the above, or a combination of certain components, or different components. For example, the household electricity analysis device based on non-invasive load monitoring may also include input and output devices, network access devices, buses, etc.

[0056] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The processor is the control center of the household electricity analysis based on non-intrusive load monitoring, and uses various interfaces and lines to connect various parts of the household electricity analysis device based on non-intrusive load monitoring.

[0057] The memory can be used to store the computer program and / or module, and the processor implements various functions of the household electricity analysis device based on non-intrusive load monitoring by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0058] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0059] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the household electricity analysis method based on non-invasive load monitoring are implemented.

[0060] If the module / unit integrated in the household electricity analysis system based on non-intrusive load monitoring is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0061] Based on such understanding, the present invention implements all or part of the processes in the above-mentioned household electricity analysis method based on non-intrusive load monitoring, and can also be completed by instructing related hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned household electricity analysis method based on non-intrusive load monitoring can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or preset intermediate form, etc.

[0062] The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0063] It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0064] The present invention will be further described below in conjunction with the embodiments and drawings: Example 2 As described in the background technology, the current research on NILM mainly stays on the exploration of load identification algorithms, and there is a lack of relevant research on using NILM to analyze the electricity consumption behavior of household users. At present, most methods for implementing NILM are based on unsupervised event detection, identifying the occurrence of load events from the total electricity consumption signal, and supervised classifiers match load events with electrical equipment to estimate the power trajectory of a single device; the widely used classifiers currently include support vector machines, neural networks, decision trees, and hybrid classification methods; however, the above methods are difficult to classify multi-state devices, resulting in poor accuracy of the classification results for electrical equipment, and it is impossible to accurately analyze the electricity consumption behavior of electrical equipment based on the classification results, which affects the visualization of the carbon footprint of electrical equipment.

[0065] In order to solve the above problems, the present invention provides a household electricity consumption analysis method based on non-intrusive load monitoring. The method constructs a household electricity consumption perception implementation architecture based on NILM, fully explores the potential of demand response, and fills the current research gap in the application of non-intrusive technology for household electricity facilities; and proposes an improved SOM neural network based on the Sparrow Search Algorithm (SSA), optimizes the initial weights of the network to improve the recognition accuracy of the SOM network; uses the DFS search algorithm to match known categories of load events with corresponding devices, solving the problems of high memory utilization, slow solution speed, and shallow traversal path in the traditional BFS search algorithm.

[0066] like Figure 1 As shown, this embodiment provides a household electricity consumption analysis method based on non-intrusive load monitoring, and the steps are as follows: Step 1: Load event detection: take the total electric meter data as input and return the detected load event time, that is, obtain the load event detection time; the purpose is to determine the time of commissioning and exiting of household loads; Step 2: Load event identification: extract the power, current, harmonic characteristics of multiple groups of loads within the same detection time to form feature vectors, and use unsupervised learning methods to map samples with similar characteristics to adjacent output nodes through competitive learning between neurons, thereby realizing the clustering of load state transition event samples and obtaining load event labels; Step 3: Load event matching: Based on the load event labels obtained by SOM clustering and the data at the time when the load event was detected, the depth-first search algorithm and similarity measurement are used to complete the matching of load events and corresponding equipment.

[0067] The above step 1 specifically includes the following steps: S101: Use the cumulative sum control chart CUSUM algorithm based on sliding windows to detect load events. According to the set dynamic change threshold, the detection window gradually slides forward from the starting point, with a length of m+n, including the mean calculation window W1=[P1,…,P m ] and transient detection window W2=[P m+1 ,…,P m+n ]; Considering that the equipment will cause power fluctuations in the network, the window W 1 and W The power average of the sampling points within 2 is used to calculate the positive and negative cumulative sum of the load, and based on this, it is determined whether there is a load event in the window. The calculation of relevant parameters is shown in the following formula: W1 window power average:

[0068] W2 window power mean:

[0069] In the formula, k The corresponding starting time of the window.

[0070] Load forward cumulative sum:

[0071] Negative cumulative sum of load:

[0072] Where: β represents the steady-state power fluctuation value. When the power change is greater than β, it is determined that a load event occurs.

[0073] In this embodiment, the data needs to be preprocessed. Considering that various interferences may be encountered during the data collection process, the initially acquired data needs to be processed to reduce the interference of noise on the load identification result.

[0074] First, the data is resampled to ensure that the current and voltage signals are aligned. Second, the data is normalized to standardize it for subsequent analysis and to compensate for related issues that affect power quality. Finally, a pre-filtering operation is performed to extract specific harmonic features from the data for subsequent analysis.

[0075] S102: When the load is in a stable state, , When the load is turned on or the gear is shifted up, the power increases, and if the offset power mean is greater than the load fluctuation value, β ,Right now E 2>( E 1+ β ), Start to accumulate positively and gradually increase. < H , indicating that a load event may have occurred, but the state change has not yet been completed. In this case, a delay factor is introduced. d =0, continue to slide forward to detect, let d = d +1 until the total Exceeding the set threshold H When , it is considered that a load event occurs in this detection window. The reverse method is used to determine the time when the event occurs. t = k + m + n - d Similarly, if the load is downshifted or shut down, causing the power mean deviation to be greater than the steady-state fluctuation value, Gradually increase, the detection method is similar to the load opening, determine the time when the load event occurs t = k + m + n - d .

[0076] The above step 2 specifically includes the following steps: S201. According to the load event detection time obtained in step 1, the active power, reactive power, harmonic current, equipment start and stop time and operating time data of the electrical equipment within the same load event detection time are obtained, and the above power load state data is constructed as a feature vector. At the same time, the principal component analysis method is used to convert the m-dimensional related feature vectors into a set of linearly unrelated variables through orthogonal transformation to establish training set and test set samples. Here, in the load feature extraction, the relevant features that can reflect the load characteristics and status are extracted from the load operation data. Different load types are divided into steady-state, transient characteristics and non-electrical quantity characteristics according to the characteristics of household electricity consumption.

[0077] S202, in order to reduce the amount of calculation in the subsequent process and make the load characteristics have the same measurement for easy comparison, this embodiment uses Z-score normalization to process the data and normalize all the initial data to [0,1];

[0078] Where: x t is the total load or raw power of the target device; x mean 、x std Represents the mean and standard deviation of the total load or target equipment power, respectively; It is the normalized value of the total load or target equipment power at time t.

[0079] S203, given the number of neurons in the input and output layers of the SOM network k , v ; Maximum number of learning times , initial winning neighborhood width ; Initial learning rate and the required learning rate ;Number of sparrows n , maximum number of optimization iterations , the ratio of explorer and follower sparrows; In this embodiment, considering that the selection of the initial weight of the network has a great influence on the clustering effect of the model, the following step S204 is performed; S204, executing the Sparrow Search Algorithm (SSA) algorithm to optimize the initial weights of the SOM network; S205: Determine the winning neighborhood and update the weights, update the learning rate and the winning neighborhood width until the learning rate meets the requirements, the training ends, and the load event label is output LE s , realize the clustering of load state transition event samples; that is, determine the winning neighborhood and update the weights, learning rate and winning neighborhood width.

[0080] S206. By continuously training the weight vector and rolling updating the data set, unfamiliar events can be integrated into the existing SOM network clustering results as new categories, thereby continuously improving the completeness and accuracy of the SOM model recognition clustering results.

[0081] The above step three specifically includes the following steps: S301. When the CUSUM algorithm detects that a load event has occurred, the time t at which the event occurred and the power change are recorded. , the SOM network model outputs load event labels LEs according to the feature vector. According to the power increase or decrease caused by the load state change, the load events are divided into increase events and decrease events. Figure 3 As shown, the state change of the device can be described by the above three parameters, assuming that only one device changes at the same time; S302, the load event labels obtained above are divided into increase events and decrease events according to power fluctuations, and the load event labels are grouped into several event sequences using the DFS algorithm. The power consumption behavior information of each device in the household total power curve is further inferred based on the pending event sequence set, such as Figure 4 As shown in the figure, the 12 load event labels can be decomposed to obtain the power consumption behavior results of 5 types of equipment, thereby realizing the power consumption behavior analysis of power-consuming equipment.

[0082] Specifically, the increase events in the matching time window are used as the starting point, and the event occurring after the increase event is used as the next node. The search is performed from top to bottom according to the time of event occurrence. During the search, first determine whether the event sequence satisfies the non-negative power constraint (during the search, the sum of the power of the events added to the pending sequence set is greater than zero, that is, the power of the equipment is always positive during operation). If not, stop the path search; if satisfied, determine whether the zero-loop and constraint are satisfied (the generated load event sequence set must satisfy that the sum of the power changes corresponding to all event sequences is zero, that is, the equipment will not affect the total power of the network after a start-stop working cycle). If satisfied, the events on this search path are included in the pending event sequence set; if not satisfied, continue to search downward until the constraint is satisfied. If the constraint is still not satisfied until the last event in the time period, delete this path; like Figure 3 As shown in the figure, the DFS search tree of the device power consumption behavior in the time period; Figure 5 As shown, LE 1, LE 2 is the increase event in this period, and the search is carried out downwards with it as the starting point. The cross in the figure indicates that this path does not meet the above constraints, and the search is stopped. The set of pending load event sequences that meet the constraints is , , .

[0083] Considering that the working states of the equipment usually do not exceed 3, this method is used to determine whether to retain the sequence containing more events: if a pending load event sequence is the union of other sequences, it can be removed. Figure 5 Take the search results shown as an example. can be , Composition, therefore, will Remove, the final set of pending load event sequences is , The purpose is to disassemble the total power and ensure that the event sequence after disassembly cannot be disassembled again, so that one event sequence corresponds to one action of the electrical equipment.

[0084] S303: In the pending event sequence set obtained through DFS search, if each event appears only once, it indicates that the power fluctuations of different power-consuming devices are different. At this time, the pending event sequence set is the final load event matching result; each sequence change can correspond to a power-consuming device with the same power consumption behavior characteristics. Here, in the pending event sequence set, each disassembled sequence corresponds to an power-consuming device, such as , which can correspond to the air conditioner starting up and shutting down.

[0085] It can be seen that the load event matching process mainly consists of the following three sub-steps: Step 31: Use the DFS algorithm to perform preliminary power matching and generate a pending load event sequence set. The search process must meet the following constraints: Zero-loop and constraint: The generated load event sequence set must satisfy that the sum of the power changes corresponding to all event sequences is 0, that is, the device will not affect the total power of the network after a start-stop working cycle; During the search process, the sum of the event power added to the pending sequence set is greater than zero, that is, the power of the device is always positive during operation; Step 32: Match the incremental events in the time window as the starting point, and the event that occurs after the incremental event as the next node. Search from top to bottom according to the event occurrence time. During the search process, first determine whether the event sequence satisfies the power non-negative constraint. If not, stop the path search; if satisfied, determine whether the zero loop and constraint are satisfied. If satisfied, include the events on this search path in the pending event sequence set; if not satisfied, continue to search downward until the constraint is satisfied. If the constraint is still not satisfied until the last event in the time period, delete this path; Step 33: The set of pending load event sequences obtained by DFS search is selected from which the best matching event sequence is selected, and the final matching of the events is completed using a similarity measurement method.

[0086] In summary, the present invention provides a household electricity consumption analysis method based on non-intrusive load monitoring, which has the following advantages over existing sensing measures: First, this method uses the sliding window accumulation method and cumulative sum algorithm of the total meter data to achieve accurate detection of load events, effectively capture the subtle changes in household electricity consumption, and provide a reliable time benchmark for subsequent electricity consumption analysis. Secondly, the unsupervised learning method is used to cluster the power load status data, which not only avoids the tediousness of manual labeling, but also improves the degree of automation of data processing, and can accurately identify the load status of different electrical equipment, laying the foundation for subsequent behavior analysis. Furthermore, combined with the depth-first search algorithm and load event constraints, multiple groups of load event labels are efficiently searched, and a set of pending event sequences that conform to the actual electricity consumption situation is generated, providing strong support for the analysis of the electricity consumption behavior of electrical equipment. Finally, by normalizing the feature vector and optimizing the initial weights of the SOM network model, the accuracy and robustness of the algorithm are further improved, making this method have broad application prospects and practical value in household electricity management.

[0087] The above embodiment is only one of the implementation methods that can realize the technical solution of the present invention. The scope of protection claimed by the present invention is not limited only to this embodiment, but also includes changes, replacements and other implementation methods that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention.

Claims

1. A household electricity consumption analysis method based on non-intrusive load monitoring, characterized in that: include: According to the total meter data of household electricity consumption, the load event detection time is obtained; Using unsupervised learning method, multiple groups of power load status data corresponding to electrical equipment within the same load event detection time are clustered to obtain multiple groups of load event labels; Based on the depth-first search algorithm combined with load event constraints, multiple groups of load event labels are searched and a set of pending event sequences is output to realize the power consumption behavior analysis of power-consuming equipment according to the set of pending event sequences.

2. The household electricity consumption analysis method based on non-intrusive load monitoring according to claim 1 is characterized in that: The specific steps of obtaining the load event detection time according to the total electric meter data of household electricity consumption include: For the total meter data of household electricity consumption, the sliding window accumulation method and cumulative sum algorithm are used to detect load events and obtain the load event detection time.

3. The household electricity consumption analysis method based on non-intrusive load monitoring according to claim 2 is characterized in that: For the total meter data of household electricity consumption, the specific steps of using the sliding window accumulation method and cumulative sum algorithm to detect load events include: According to the set dynamic change threshold, the detection window gradually slides forward from the starting point; the detection window is divided into a mean value calculation window and a transient detection window; The positive cumulative sum of the load and the negative cumulative sum of the load are calculated respectively according to the power mean of the mean calculation window and the power mean of the transient detection window; According to the comparison result of the positive cumulative sum of the load, the negative cumulative sum of the load and the preset load event threshold, the occurrence time of the load event is obtained, and then the load event detection time is obtained; Among them, k is the starting time corresponding to the detection window; Mean calculation window W1 window power mean E1: Transient detection window W2 window power average E2: The total length of the detection window is m+n; W1=[P1,…,P m ]; W2=[P m+1 ,…,P m+n ]; Load forward accumulation : Negative load accumulation : Where: β represents the steady-state power fluctuation value. When the power change is greater than β, it is determined that a load event occurs.

4. The household electricity consumption analysis method based on non-intrusive load monitoring according to claim 1 is characterized in that: The specific steps of clustering multiple groups of power load status data corresponding to power-consuming equipment within the same load event detection time using an unsupervised learning method include: Within the same load event detection time, the power load status data corresponding to the power-consuming equipment is obtained; wherein the power load status data at least includes active power, reactive power, harmonic current, start and stop time of the power-consuming equipment, and operating time; The power load state data is used to construct a feature vector as a data set, and the data set is divided into a training set and a test set; The pre-built SOM network model is trained based on the training set, and the trained SOM network model is tested based on the test set, and the load event labels are output to achieve clustering of load event samples of load state transition of power equipment.

5. The household electricity consumption analysis method based on non-intrusive load monitoring according to claim 4 is characterized in that: The eigenvector is normalized, and the specific formula is as follows: In the formula, x t is the total load or raw power of the target device; x mean 、x std Represents the mean and standard deviation of the total load or target equipment power, respectively; is the normalized value of the total load or target equipment power at time t; In the process of constructing the SOM network model, the sparrow search algorithm is used to optimize the initial weights of the SOM network model.

6. The household electricity consumption analysis method based on non-intrusive load monitoring according to claim 1 is characterized in that: The specific steps of searching multiple groups of load event labels based on the depth-first search algorithm combined with load event constraints include: According to the load event detection time, the event occurrence time and power change amount are obtained; Divide multiple groups of load event labels into increase events and decrease events according to the power change amount; Using the depth-first search algorithm, taking the increase event as the starting point and the event that occurs after the increase event as the next node, searching from top to bottom according to the event occurrence time, outputting the pending sequence set that meets the load event constraint conditions; During the search process, the load event constraints include: first, a non-negative power constraint: the sum of the powers corresponding to all event labels of the output pending event sequence set is greater than zero; second, a zero-loop constraint: the sum of the power changes corresponding to all event labels of the output pending event sequence set is equal to zero.

7. The household electricity consumption analysis method based on non-intrusive load monitoring according to claim 6 is characterized in that: The steps of searching from top to bottom according to the event occurrence time and outputting the pending sequence set that meets the load event constraint conditions include: First, determine whether the event sequence satisfies the power non-negative constraint; if not, stop the path search; if satisfied, continue to determine whether it satisfies the zero-loop and constraint; if satisfied, add the load event label on the current search path to the pending event sequence set; if not satisfied, continue to search downward until the zero-loop and constraint are satisfied; if the zero-loop and constraint are still not satisfied until the last event in the time period, delete this path.

8. A household electricity analysis system based on non-intrusive load monitoring, characterized in that: include: A load event detection module, used to obtain the load event detection time according to the total electric meter data of household electricity consumption; A load event identification module is used to cluster multiple groups of power load status data corresponding to electrical equipment within the same load event detection time using an unsupervised learning method to obtain multiple groups of load event labels; The power consumption behavior analysis module is used to search multiple groups of load event tags based on the depth-first search algorithm combined with load event constraints, and output a set of pending event sequences to implement power consumption behavior analysis of power-consuming equipment based on the set of pending event sequences.

9. A device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the household electricity consumption analysis method based on non-intrusive load monitoring as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it is used to implement the steps of the household electricity consumption analysis method based on non-intrusive load monitoring as described in any one of claims 1-7.

Images