Residential Load Location Method

Abstract

The present invention relates to a method for locating resident loads, the method comprising: defining a resident load network M × N The user side load network matrix, M is the number of air switches in the distribution box, N is the maximum number of electrical devices connected behind the air switch. The elements in the user-side load network matrix are the carrier input impedance or working state of each electrical device. The position of the elements in the user-side load network matrix depends on the relative position of the corresponding electrical device to the electric meter in the residential load network. The carrier input impedance of each electrical device is measured at the electric meter and fitted using the XGBoost algorithm. M , N The value of is then iterated by the XGBoost algorithm to fill the carrier input impedance or working state of each electrical device into the user-side load network matrix, thereby determining the relative position of each electrical device based on the filled user-side load network matrix. The present invention can locate each electrical device in the residential load network and improve the level of residential load identification.

Description

Technical Field

[0001] The present invention belongs to the field of non-intrusive load monitoring and decomposition on the power demand side, serves the expansion and promotion of the comprehensive energy business of power grid companies, and specifically relates to a method for locating residential loads. Background Art

[0002] The construction of new power systems promotes the coupling development of power systems and advanced information communication control technologies, and transforms the configuration mode of traditional energy and electricity from partial perception, one-way control, and planning to high perception, two-way interaction, intelligence and efficiency. Traditional load monitoring adopts an intrusive method, which requires the installation of specific sensors on each monitored electrical equipment. However, in reality, considering that the electrical equipment of residential users belongs to user assets, power supply companies have no right to deal with it, and there is a potential risk of infringing user privacy, non-intrusive load monitoring methods are more easily accepted by all parties. Non-intrusive load monitoring (NILM) only needs to install a load monitoring device at the user's power entrance, perform load identification and power consumption behavior decomposition to obtain accurate power consumption information of various electrical appliances inside the user. Because it can conveniently and quickly obtain detailed household power consumption, monitor the user's power consumption behavior online, and realize real-time two-way interaction between users and the power grid, it has become a hot topic of research at home and abroad. The unique information that reflects the power consumption status reflected by the power equipment during operation is called load signatures (LS), which is the main judgment basis that NILM relies on.

[0003] The existing implementation architecture based on NILM residential electricity consumption behavior analysis is as follows: Figure 1 As shown, it relies on the LS data of the power equipment during operation collected by the smart meter, in which the historical data is used for load feature training and classifier design, and the real-time data is compared with the trained historical data for the decomposition and detection of the current power consumption behavior. The current NILM has not yet found a way to break through this architecture.

[0004] With the continuous deepening of NILM research and the introduction of various training algorithms, NILM's load recognition rate has been significantly improved, but there are still difficult problems that have not been effectively solved. The "positioning" problem of electrical equipment, that is, the existing NILM technology can only answer the question of whether a specific electrical equipment is running and its operating status at a specific time, but cannot answer the "physical location positioning" problem of a specific electrical equipment in the entire household load network, making it difficult to achieve accurate identification of the load location. Summary of the invention

[0005] The object of the present invention is to provide a resident load positioning method capable of identifying the location of resident loads.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A residential load positioning method is used to position each electrical device in a residential load network. An M×N user-side load network matrix is ​​defined for the residential load network, where M is the number of air switches in a distribution box corresponding to the residential load network, and N is the maximum number of electrical devices connected behind the air switches. The elements in the user-side load network matrix are the carrier input impedance or working state of each electrical device, and the position of the elements in the user-side load network matrix depends on the relative position of the corresponding electrical device to an electric meter in the residential load network; the carrier input impedance of each electrical device is measured at the electric meter, and the values ​​of M and N are obtained by fitting using an XGBoost algorithm, and then the carrier input impedance or working state of each electrical device is filled into the user-side load network matrix by iteratively using the XGBoost algorithm, so as to determine the relative position of each electrical device based on the filled user-side load network matrix.

[0008] Actively inject a carrier signal into the electrical devices, thereby obtaining the carrier input impedance of each of the electrical devices.

[0009] The device for measuring the carrier input impedance of each of the electrical equipment includes a carrier signal transceiver unit, a transformer, a measuring resistor, a capacitor, a voltage and current measurement module and a signal processing module; the carrier signal transceiver unit, the measuring resistor and the primary side of the transformer are connected in series to form a first loop, the capacitor, the secondary side of the transformer and the electrical equipment to be detected are connected in series to form a second loop, the voltage and current measurement module is connected to the second loop and is used to measure the voltage across the residential load to be detected and the current in the second loop and output a corresponding measurement signal, and the signal processing module is connected to the voltage and current measurement module and is used to perform data processing to obtain the carrier input impedance of the residential load to be detected.

[0010] The carrier signal transceiver unit is embedded in the electric meter.

[0011] When the XGBoost algorithm is used for iteration, the known constraints and location distribution of the electrical equipment adopt preset values.

[0012] Due to the application of the above technical solution, the present invention has the following advantages compared with the prior art: the present invention can locate each electrical device in the residential load network, thereby improving the identification level of the residential load. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Attached Figure 1 This is the architecture diagram for analyzing residential electricity consumption behavior based on NILM.

[0014] Attached Figure 2 The figure is a schematic diagram of a device for measuring the carrier input impedance of each electrical device.

[0015] Attached Figure 3 This is the user side load network architecture diagram.

[0016] Attached Figure 4 Schematic diagram of a method for measuring impedance of a power line carrier communication channel.

[0017] Attached Figure 5 Schematic diagram of a non-intrusive residential load monitoring device.

[0018] Attached Figure 6 This is the flowchart of XGBoost.

[0019] Attached Figure 7 It is a flow chart of the load positioning method of the present invention.

[0020] Attached Figure 8 Power line carrier impedance spectrum with five lights on.

[0021] Attached Fig. 9 Power line carrier impedance spectrum with ten lights on.

[0022] Attached Fig.10 This is a diagram of the electricity consumption curve of a user's home on a certain day. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with the embodiments shown in the accompanying drawings.

[0024] Embodiment 1: As shown in the attached Figure 7 As shown, a method for locating a residential load for locating each electrical device in a residential load network is specifically as follows: define an M×N user-side load network matrix for the residential load network, M is the number of air switches in the distribution box corresponding to the residential load network, N is the maximum number of electrical devices connected after the air switch, the elements in the user-side load network matrix are the carrier input impedance or working state of each electrical device, and the position of the elements in the user-side load network matrix depends on the relative position of the corresponding electrical device to the electric meter in the residential load network; measure the carrier input impedance of each electrical device at the electric meter, use the XGBoost algorithm to fit the values ​​of M and N, and then use the XGBoost algorithm to iterate and fill the carrier input impedance or working state of each electrical device into the user-side load network matrix, so as to determine the relative position of each electrical device based on the filled user-side load network matrix. When iterating using the XGBoost algorithm, the known constraints and position distribution of the electrical devices use preset values.

[0025] The method for measuring the carrier input impedance of each electrical device is to actively inject carrier signals (of different frequencies) into the electrical device, thereby obtaining the carrier input impedance (spectrum) of each electrical device. Figure 2 As shown, the device for measuring the carrier input impedance of each electrical device includes a carrier signal transceiver unit, a transformer, a measuring resistor, a capacitor, a voltage and current measuring module and a signal processing module; the carrier signal transceiver unit, the measuring resistor and the primary side of the transformer are connected in series to form a first loop, the capacitor, the secondary side of the transformer and the electrical device to be detected are connected in series to form a second loop, the voltage and current measuring module is connected to the second loop and is used to measure the voltage at both ends of the residential load to be detected and the current in the second loop and output a corresponding measurement signal, and the signal processing module is connected to the voltage and current measuring module and is used to perform data processing to obtain the carrier input impedance of the residential load to be detected. The carrier signal transceiver unit is embedded in the electric meter.

[0026] The explanations for the above scheme are as follows:

[0027] (1) User-side load network

[0028] In order to solve the problems existing in the existing NILM method and realize the effective supervision of the electrical equipment that is not in working state and the accurate positioning of the physical location of the electrical equipment, it is necessary to first establish the user-side load network architecture. Figure 3 As shown in the figure, for residential users, the outlet end of the household meter is the asset demarcation point between the power grid company and the user. The user-side load network starts from the household meter and is connected to various electrical devices through multiple groups of air switches in the distribution box. Figure 2 The figure shows all 39 electrical devices in a residential user. The structure of the user-side load network is the basis for realizing the identification and positioning of electrical devices. Compared with the original method, the position relationship of each electrical device in the load network can be determined. When the state of a certain electrical device changes, the entire load network will change.

[0029] Define the M×N user side load network matrix Z l As shown in formula (1):

[0030]

[0031] M is the number of air switches in the distribution box corresponding to the residential load network, N is the maximum number of electrical equipment connected after the air switch, and the element Z in the user side load network matrix is loadij is the carrier input impedance or working state of each electrical device. The position (i, j) of the element in the user-side load network matrix depends on the relative position (distance) of the electrical device from the electric meter in the residential load network. If there is no electrical device, it is represented by "0".

[0032] Affected by the multipath effect and attenuation characteristics of the user-side load network, the various electrical devices in the network are interrelated, and the corresponding N×N correlation coefficient matrix ρ is shown in formula (2):

[0033]

[0034] (2) Power line carrier input impedance and measurement

[0035] Since the transmission line is regarded as a part of the load network structure under distributed parameters, the input impedance of the power load of the same type, same parameters and same working conditions at the meter is different, such as Figure 4 FIG. 1 is a typical method for measuring impedance of a power line carrier communication channel. The frequency-converting carrier signal source is applied to realize spectrum scanning of the input carrier at the measurement point. The input impedance of the power line carrier communication channel at the measurement point is equivalent to: and A high-precision measurement impedance is connected in series Its value is known and low temperature drift. Voltage on and Voltage on It can be accurately measured and regarded as a known quantity. Using formula (3), we can know the power line impedance The specific value of .

[0036]

[0037] Where: U XX is the real part of the voltage shared by the power line input impedance; U XY is the imaginary part of the voltage shared by the power line input impedance, U SX To measure the real part of the voltage shared by the impedance with high precision, U SY It is used to measure the imaginary part of the voltage shared by the impedance with high precision. The schematic diagram of the carrier transceiver unit is as follows: Figure 5 shown.

[0038] In the actual working environment, the carrier transmitting and receiving unit is embedded in the smart meter, and the carrier signal is connected to the capacitor element through a 1:1 transformer to form a coupling circuit. On the one hand, it couples the transmitted signal to the low-voltage power line channel, and on the other hand, it isolates the strong current, avoids the interference of the industrial frequency power signal on the measurement result, and improves the signal-to-noise ratio. Figure 5 The voltage U between points B1 and B2 is B With current I B The ratio of is the input impedance of the channel. When the transformer ratio is 1:1, U C It can be approximately regarded as B Equal, I C It can be approximately regarded asB Equal, and I C =U S / Z S . Therefore, the modulus of the system input impedance is:

[0039]

[0040] (3) XGBoost Classifier Algorithm

[0041] XGBoost algorithm, or extreme gradient boosting algorithm, is a machine learning algorithm based on tree model. The basic principle of XGBoost algorithm is to stack multiple tree models with low accuracy to form a model that can meet actual needs. XGBoost algorithm is integrated by multiple classification and regression trees (CART), and the XGBoost model can be expressed as:

[0042]

[0043] Where: is the predicted value, K is the number of trees, F is the possible CART set, x i is the i-th sample of the input, f k For x i The corresponding value in the F set. The XGBoost algorithm uses a gradient boosting strategy, and only one tree is added each time to make up for the previous fitting result. The objective function of the XGBoost model is defined as:

[0044]

[0045] Where: is the training residual of the model, is the regularized sum of K trees. Performing a second-order Taylor expansion on equation (6) yields:

[0046]

[0047]

[0048]

[0049] Where: g i 、h i Represent the first and second order derivatives respectively. The regularization term is defined as:

[0050]

[0051] Where: jis the value represented by the jth leaf in tree f, T is the total number of leaves in tree f, γ and λ are the penalty coefficients of regularization term L1 and regularization term L2, respectively, which are used to represent the complexity of tree f and the weight of restricted nodes, and the final objective function is:

[0052]

[0053]

[0054] Where: I is the sample set on each leaf, I j ={i|q(x i )=j},I j is the leaf set on the jth leaf, q(x i ) = j is the sample x i The corresponding index number j in the tree. The smaller the value of the objective function, the better the overall structure of the tree. The XGBoost algorithm uses continuous iteration to strengthen weak learners into strong learners. The flowchart of XGBoost is as follows Figure 6 As shown in Figure 1, a large number of weak learners are iteratively stacked into a strong learner. The residual of each iteration is used to correct the previous predictor, which can optimize the specified loss function. A regular phase is added to the loss function to establish the objective function of XGBoost to measure the performance of the model.

[0055] The XGBoost algorithm is mainly used to fit the M and N values ​​of the load network matrix. After the values ​​of M and N are determined, the load network matrix needs to be filled and the position of each household appliance in the load network matrix needs to be determined. The XGBoost algorithm is also needed. i The possible values ​​of M and N before become possible load network matrices. In order to reduce the number of iterative fitting, the common constraints and location distribution of household appliance load networks are preset values. The proposed user-side load network solves the problem that existing methods cannot achieve load location positioning and the state supervision of electrical equipment that is not in working state. The parameter fitting and filling method of the residential load network matrix based on XGBoost, that is, the load location method flow chart is as follows: Figure 7 shown.

[0056] For the load perception and identification on the power demand side, this solution proposes an active perception method of the power system load network based on the power line carrier impedance characteristics on the basis of the existing NILM. In order to deal with the problems that the current NILM has not yet solved, the patent proposes a method of the concept of the user side load network, which solves the problem of locating the location of residential electrical equipment. The advantages of the technical solution applied for compared with the existing technology can be summarized as follows:

[0057] (1) The concept of the power demand side load network is proposed. While taking user privacy into consideration, the power lines are considered as part of the load network, providing network coordinates for the precise positioning of the residential load network.

[0058] (2) It realizes effective supervision of household electrical appliances that are not in working state. In the load network matrix, even household appliances that are not in working state are displayed as an "effective" working condition;

[0059] (3) The XGBoost algorithm is introduced to realize the parameter fitting of the load network matrix, and the specific values ​​of the matrix rows M and columns N are determined through continuous training;

[0060] (4) The XGBoost algorithm is also used to fill in the load network matrix parameters. Through continuous iterative training of the impedance spectrum, various types of household appliances can be filled into the load matrix.

[0061] Test verification

[0062] In order to verify the effectiveness and scientificity of the proposed method, two groups of verification experiments were designed. One group of experiments was a 4×4 lamp matrix experiment in the laboratory, where the electrical load equipment was only a group of 4×4 lighting lamps; the other group of experiments was a residential load under real working conditions.

[0063] ①4×4 lamp matrix test

[0064] Similar to the ceiling lights in ordinary workplaces, the 4×4 light matrix test load is 16 straight-light LEDs, which are connected to the power supply and carrier transmitter terminal through a soft-core 2*1 square power line. The specifications of each lighting lamp are exactly the same, and the power line distance between each lamp and the carrier transmitter terminal is different.

[0065] Define the state when the light is on as "1" and the state when the light is off as "0", then a state matrix of the light matrix is:

[0066]

[0067] When five lights are on, the possible state matrices are shown in equation (11).

[0068]

[0069]

[0070]

[0071] The corresponding power line carrier impedance spectrum is as follows: Figure 8 shown.

[0072] When 10 lights are on, the possible state matrices are shown in equation (12).

[0073]

[0074]

[0075]

[0076] The corresponding power line carrier impedance spectrum is as follows: Fig. 9 shown.

[0077] like Figure 8 and Fig. 9 As shown in the impedance spectrum, the power line carrier impedance fluctuates at different frequencies, mainly due to the peaks and valleys of the input impedance under high-frequency distributed parameters. Considering the characteristic impedance of the power line, whether it is the five-lamp working condition or the ten-lamp working condition, the lamp matrix can be effectively distinguished and identified.

[0078] ② Actual working condition test

[0079] In order to further verify the feasibility of the proposed method in actual situations, an experiment was conducted in a household. The details of the household electrical equipment are shown in Table 1. "√" indicates operation and "×" indicates non-operation. There are three air conditioners of the same type, parameters and working conditions, which are powered by three-phase four-wire system. According to the different electrical properties exhibited by household appliances when working, household appliances can be roughly divided into three categories: resistive load, motor load and power electronic load. Resistive loads such as water heaters, electric heaters and rice cookers almost show completely resistive characteristics in steady state; motor loads such as refrigerators, air conditioners and range hoods are the same loads to be identified in the experiment; power electronic loads such as rectifiers, vacuum cleaners and computers have obvious distortion in the current waveform and contain a large number of harmonic components, which may have a certain impact on the measurement of input impedance.

[0080] Table 1 Electrical equipment used in a certain household

[0081]

[0082] like Fig.10 As shown in the figure, this is the electricity consumption curve of this household on a certain day. Since the sampling period of the electricity meter is 15 minutes, there are 96 sampling points.

[0083] The load network matrix M = 5 and N = 8 are fitted by using the XGBoost algorithm. The load network matrix filling values ​​of the common constraint conditions are brought into the XGBoost algorithm and continuously trained and iterated to adjust the load network matrix of scenario 1 as shown in formula (13):

[0084]

[0085] The load network matrix filling values ​​of common constraints are brought into the XGBoost algorithm and continuously trained and iterated to adjust the load network matrix of scenario 2 as shown in formula (14).

[0086]

[0087] The load network matrix filling values ​​of common constraints are brought into the XGBoost algorithm and continuously trained and iterated to adjust the load network matrix of scenario three as shown in formula (15).

[0088]

[0089] Matrices (13)-(15) represent the positions and corresponding working states of each electrical device in different scenarios, realizing the location of the load. It is worth noting that in actual working conditions, when the load of the same electrical device is small under strong interference of large load, such as when the same load is a low-power lighting lamp under actual working conditions, its location and identification cannot be achieved.

[0090] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for locating residential loads, used to locate each electrical device in a residential load network, characterized in that: For the residential load network definition M × N The user side load network matrix, M is the number of air switches in the distribution box corresponding to the residential load network, N is the maximum number of the electrical devices connected behind the air switch, the elements in the user-side load network matrix are the carrier input impedance or working state of each of the electrical devices, and the positions of the elements in the user-side load network matrix depend on the relative positions of the corresponding electrical devices to the electric meters in the residential load network; The carrier input impedance of each of the electrical devices is measured at the electric meter, and a carrier signal is actively injected into the electrical device to obtain the carrier input impedance of each of the electrical devices. The XGBoost algorithm is used to fit the impedance M , N The value of is then iterated by the XGBoost algorithm to fill the carrier input impedance or working state of each of the electrical devices into the user-side load network matrix, so as to determine the relative position of each of the electrical devices based on the filled user-side load network matrix. When the XGBoost algorithm is iterated, the known constraints and position distribution of the electrical devices adopt preset values.

2. The method for locating resident loads according to claim 1, characterized in that: The device for measuring the carrier input impedance of each of the electrical equipment includes a carrier signal transceiver unit, a transformer, a measuring resistor, a capacitor, a voltage and current measurement module and a signal processing module; the carrier signal transceiver unit, the measuring resistor and the primary side of the transformer are connected in series to form a first loop, the capacitor, the secondary side of the transformer and the electrical equipment to be detected are connected in series to form a second loop, the voltage and current measurement module is connected to the second loop and is used to measure the voltage across the residential load to be detected and the current in the second loop and output a corresponding measurement signal, and the signal processing module is connected to the voltage and current measurement module and is used to perform data processing to obtain the carrier input impedance of the residential load to be detected.

3. The method for locating resident loads according to claim 2, characterized in that: The carrier signal transceiver unit is embedded in the electric meter.

Images