Building public power equipment positioning early warning system

Abstract

The application provides a kind of building public power equipment positioning early warning system, it is related to power equipment positioning early warning technical field, including data acquisition module for obtaining line power characteristics;Realize the transformation of line power characteristics-each equipment power characteristics-line power characteristics under normal state and calculate the reconstruction error of encoding-decoding model;When reconstruction error is less than threshold value, the feature adjustment module based on particle swarm algorithm and simulated annealing algorithm obtains the optimal adjustment value of equipment power characteristics;For analyzing the difference degree of equipment power characteristics and optimal adjustment value, to determine the positioning early warning module of abnormal public power equipment, when the existence of abnormality is judged by reconstruction error, the closest equipment power characteristics optimal adjustment value is found based on particle swarm algorithm and simulated annealing algorithm, and then the difference degree between it and equipment power characteristics is analyzed to determine which specific public power equipment is abnormal, to realize the positioning early warning of abnormal public power equipment.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment location and early warning technology, specifically to a building public electrical equipment location and early warning system. Background Technology

[0002] With economic development, high-rise buildings are springing up like mushrooms after rain, leading to a surge in the number of public electrical appliances within these buildings. Monitoring the malfunctions and anomalies of these appliances is crucial for ensuring safety in production and daily life. However, traditional methods for monitoring public electrical appliances require the installation of independent monitoring devices on each appliance. This not only increases initial investment costs but also introduces maintenance complexity, particularly in high-rise systems where the feasibility of this approach is severely challenged. Therefore, a more cost-effective and adaptable monitoring and early warning solution is urgently needed to effectively monitor the public electrical appliances within high-rise buildings.

[0003] In the prior art, a non-intrusive electrical equipment detection method based on big data and supervised learning algorithms (main classification number G06F) with publication number "CN120162674A" includes data acquisition and preprocessing, which collects total load data of the power system through non-intrusive load monitoring equipment and performs outlier removal, data normalization, and sliding window segmentation; feature extraction and importance analysis, which uses supervised learning algorithms to extract key features from the preprocessed data and evaluate the importance of features; and model training and optimization, which constructs a multi-task learning model and improves the accuracy of load decomposition by iteratively optimizing model parameters. By combining big data processing capabilities and high-performance machine learning models, the accuracy and efficiency of load monitoring are significantly improved.

[0004] However, existing technologies still have significant drawbacks. For example, public electrical equipment malfunctions are relatively rare, making it difficult to extract enough data for model training. Furthermore, during malfunctions, power signals are affected by interference and noise, resulting in inaccurate measurements of current and voltage data. Extracting precise equipment power characteristics from these measurements is challenging. More importantly, multiple devices on the same public power line may malfunction simultaneously, and their interactions significantly increase the complexity of power characteristics. Based on the above, training a model that can accurately decompose line power characteristics under abnormal conditions to obtain the power characteristics of each device is quite difficult. Therefore, existing technologies struggle to accurately extract the power characteristics of each device under abnormal conditions, leading to poor accuracy in subsequent fault diagnosis.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a detection system for public electrical equipment in intelligent buildings, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A positioning and early warning system for building public electrical equipment, comprising:

[0009] The data acquisition module is used to obtain the line power characteristics of each public power line in the building during the current monitoring period;

[0010] The encoding-decoding model is used to decompose the line power characteristics of a public power line under the current monitoring period into the equipment power characteristics of each public power device on the public power line, and then restore the line power characteristics of each equipment to the line power characteristics and calculate the reconstruction error. The encoding-decoding model is trained based on the line power characteristics and equipment power characteristics under normal conditions.

[0011] The feature adjustment module is used to adjust the power characteristics of each device output by the encoder-decoder model by combining the particle swarm algorithm and the simulated annealing algorithm when the reconstruction error is greater than the reconstruction error threshold, so as to obtain the optimal adjustment value of the power characteristics of the device that satisfies the reconstruction error not being greater than the reconstruction error threshold.

[0012] The location and early warning module is used to analyze the degree of difference between the equipment power characteristics output by the encoding-decoding model and the optimal adjustment value of the equipment power characteristics, in order to determine the difference index that reflects the degree of abnormality of public power equipment, and to locate and issue early warnings for public power equipment with faults based on the difference index.

[0013] Furthermore, the public power line is a line used to supply power to public electrical equipment within the building, and the data acquisition module includes a current transformer and a voltage probe installed at the public power line.

[0014] Furthermore, the line power characteristics include, but are not limited to, DC component of current, peak current, RMS current, current fluctuation, DC component of voltage, peak voltage, RMS voltage, voltage fluctuation, active power, peak instantaneous power, apparent power, power factor, and phase angle, specifically obtained by analyzing the current value sequence and voltage value sequence obtained by the current transformer and voltage probe.

[0015] Furthermore, the encoding-decoding model consists of an encoder and a decoder. The decoder is used to decompose the line power characteristics into the equipment power characteristics of each public electrical device on the corresponding public power line, and the encoder is used to convert the equipment power characteristics of each public electrical device on the public power line back into the line power characteristics.

[0016] The decoder includes an input layer, a hidden layer, and an output layer, as detailed below:

[0017] The input layer consists of M input nodes, which are used to receive line power characteristics, where M is the number of types of line power characteristics.

[0018] The hidden layers include a first hidden layer and a second hidden layer, as detailed below:

[0019] The first fully connected layer has 32 nodes and uses the ReLU activation function to extract nonlinear combinations of line power characteristics;

[0020] The second fully connected layer has 64 nodes and uses the ReLU activation function to further extract and transform features;

[0021] The output layer adopts a fully connected layer structure with Q×k nodes. It is used to output the equipment power characteristics of each public power device on the public power line, where Q is the number of public power devices on the public power line and k is the number of equipment power characteristic types of the public power devices.

[0022] The encoder also includes an input layer, a hidden layer, and an output layer, as detailed below:

[0023] The input layer adopts a fully connected layer structure, including Q×k input nodes, which are used to receive the equipment power characteristics of each public power equipment on the public power line;

[0024] The hidden layers include a first hidden layer and a second hidden layer, as detailed below:

[0025] The first fully connected layer has 64 nodes and uses the ReLU activation function to extract nonlinear combinations of the device's electrical characteristics.

[0026] The second fully connected layer has 32 nodes and uses the ReLU activation function to further process and transform the device's electrical characteristics.

[0027] The output layer adopts a fully connected layer structure with M nodes and is used to output the power characteristics of the line.

[0028] Furthermore, the electrical characteristics of the device include one or more of the following: active power, phase angle, power factor, effective current value, current fluctuation value, effective voltage value, and voltage fluctuation value.

[0029] Furthermore, the logic for obtaining the reconstruction error threshold is as follows: multiple sets of line power features under normal conditions are obtained as sample line power features, and they are input into the trained encoder-decoder model one by one to obtain the reconstruction error corresponding to each sample line power feature and extract the mean and standard deviation. The standard deviation is multiplied by a margin factor of not less than 1 to form a warning increment, and the sum of the warning increment and the mean is used as the reconstruction error threshold.

[0030] Furthermore, the logic for obtaining the optimal adjustment value of the equipment's power characteristics is as follows:

[0031] 1) Based on the encoder-decoder model, the decoder analyzes the line power characteristics of the public power line during the current monitoring period, uses the obtained power characteristics of each device as the initial particles, and randomly perturbs the initial particles to generate multiple new particles to form a particle swarm.

[0032] The initial particles are represented by a two-dimensional matrix of Q×k. The representation is as follows:

[0033]

[0034] In the formula, Let be the element value in the x-th row and y-th column of the two-dimensional matrix Z, representing the y-th electrical characteristic of the x-th public utility device on the public utility power line, where x is the index of the public utility device on the public utility power line. y is the index of the equipment's electrical characteristic type, and ;

[0035] The logic for randomly perturbing the initial particle to form a particle swarm is as follows: the initial particle is copied multiple times to form multiple copy particles. For each copy particle, a perturbation range is defined for each element value. Within the perturbation range, the element value is randomly perturbed. The perturbed element value is then used as the final element value to update the copy particle, forming multiple new particles to construct the particle swarm. The upper limit of the perturbation range is the product of the element value and the perturbation factor, and the lower limit of the perturbation range is the negative number of the product of the element value and the perturbation factor. The perturbation factor is a decimal greater than 0.

[0036] The specific mathematical expression for perturbing the element values ​​is as follows:

[0037]

[0038] In the formula, As a disturbance factor, Indicates in arrive Take a random value between them. Indicates the after of the disturbance ;

[0039] 2) Input each particle in the particle swarm into the encoder of the encoder-decoder model to generate the line power feature corresponding to the particle. Then input the line power feature into the encoder-decoder model and calculate the reconstruction error corresponding to the particle. Use the difference between the reconstruction error threshold and the reconstruction error as the fitness value of the particle. Record the historical best position and global best position of each particle. Iteratively update the velocity and position of each particle in this way.

[0040] 3) After each iteration, calculate the fitness value of each particle and determine whether the termination condition is met. If it is met, select the best particle as the initial solution to proceed to step 4). If it is not met, record the historical best position and global best position of each particle after this iteration based on the fitness value, and continue to iterate and update the velocity and position of each particle until the termination condition is met.

[0041] 4) Set the initial annealing temperature, the termination temperature, and the cooling rate. Use the initial solution as the current solution in the first round of annealing optimization. In each round of annealing optimization, the current solution is perturbed based on the annealing temperature to generate a neighborhood solution with a fitness value not less than 0. Calculate the mutation index of the neighborhood solution, analyze the mutation index of the current solution and the neighborhood solution to calculate the acceptance probability, and use the probability to accept the neighborhood solution as the current solution. Update the annealing temperature after each round of annealing optimization until the annealing temperature is lower than the termination temperature. Output the final current solution as the optimal adjustment value of the equipment's power characteristics.

[0042] Furthermore, the termination conditions include condition one and condition two, as detailed below:

[0043] Condition 1: At least one particle has a fitness value of not less than 0 after the end of this round of iteration;

[0044] Condition 2: After this round of iterations, at least one particle, in addition to satisfying Condition 1, also has a mutation index that is less than the mutation index threshold. The specific expression for the particle's mutation index is as follows:

[0045]

[0046] In the formula, The variation index, The power characteristics of public power lines during the current monitoring period are represented by a 1×M matrix. To input the electrical characteristics of each device represented by particles satisfying condition one into the encoder of the encoder-decoder model, the resulting line electrical characteristics are also represented as a 1×M matrix. express and Euclidean distance, express The Frobenius norm;

[0047] In the formula, The electrical characteristics of each device, represented by the initial particle, are expressed as a matrix of form Q×k. To satisfy condition one, the electrical characteristics of each device represented by the particles are also represented by a matrix of the form Q×k. express and Euclidean distance, express The Frobenius norm;

[0048] In the formula, To input the reference line power characteristics into the decoder of the encoder-decoder model, the resulting power characteristics of each device are also represented as a matrix of form Q×k. The logic for obtaining the reference line power characteristics is as follows: the most recent historical monitoring time period when the public power line is in a normal state and is closest to the current monitoring time period is taken as the reference time period, and the line power characteristics of the public power line during the reference time period are taken as the reference line power characteristics. express and Euclidean distance, express The Frobenius norm;

[0049] In the formula, , and All are preset weighting coefficients, used to represent , and The weight it occupies in the calculation of the variability index satisfies as well as ;

[0050] Among them, the optimal particle is the particle with the minimum mutation index among the particles that satisfy condition one and whose mutation index is less than the mutation index threshold.

[0051] Furthermore, the logic for perturbing the current solution based on the annealing temperature to generate a neighborhood solution with a fitness value not less than 0 is as follows: In each round of annealing optimization, for each element value of the current solution, a corresponding small perturbation interval is defined. A random value is randomly generated from the small perturbation interval, and its product with the annealing temperature in this round of annealing optimization is calculated as the perturbation value. The perturbation value is summed with the corresponding element value of the current solution as the element value of the neighborhood solution, thereby generating a neighborhood solution. Then, the fitness value of the neighborhood solution is determined. If the fitness value of the neighborhood solution is not less than 0, the next step of calculating the mutation index is performed; otherwise, a random value is randomly generated again to regenerate the neighborhood solution, until the fitness value of the generated neighborhood solution is not less than 0. The upper limit of the small perturbation interval is half the width of the small perturbation interval, and the lower limit is the product of half the width of the small perturbation interval and -1. The half width of the small perturbation interval is less than the perturbation factor.

[0052] Furthermore, the specific mathematical expression for the difference index is as follows:

[0053]

[0054] In the formula, Let y be the optimal adjustment value for the electrical characteristic of the x-th public utility device on the public utility power line. The difference index of the x-th public electrical equipment on the public power line;

[0055] The logic for locating public electrical equipment with faults based on the difference index is as follows: if the difference index of a public electrical equipment is higher than the difference index threshold, then this public electrical equipment is defined as having a fault.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] The building public electrical equipment location and early warning system of the present invention learns the mapping relationship between line power characteristics and equipment power characteristics under normal conditions through an encoding-decoding model. Then, based on the reconstruction error after the current line power characteristics are processed by the encoding-decoding model, it determines whether the public electrical equipment on the public power line is abnormal. Then, based on the particle swarm optimization algorithm and simulated annealing algorithm, it adjusts the equipment power characteristics output by the encoding-decoding model to find the normal equipment power characteristics that are closest to the actual situation as the optimal adjustment value of the equipment power characteristics. Finally, by analyzing the degree of difference between the equipment power characteristics and the optimal adjustment value of the equipment power characteristics, it determines which specific public electrical equipment is abnormal. In this way, it realizes the location and early warning of abnormal public electrical equipment. While avoiding the need to monitor each public electrical equipment individually, it avoids the problem of difficulty in accurately extracting the equipment power characteristics of each electrical equipment under abnormal conditions, and significantly improves the accuracy of abnormal judgment of each public electrical equipment. Attached Figure Description

[0058] Figure 1 This is a modular unit diagram of the overall system of the present invention;

[0059] Figure 2 This is a diagram showing the probability distribution of failures within each difference index interval in this scheme. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0061] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0062] Example 1:

[0063] Please see Figures 1-2 This invention provides a positioning and early warning system for building public electrical equipment, comprising:

[0064] The data acquisition module is used to obtain the line power characteristics of each public power line in the building during the current monitoring period;

[0065] Among them, public power lines are lines used to supply power to public electrical equipment in the building. The specific details can be determined through the building's electrical floor plan or on-site investigation, which can identify the public power lines in the building and the public electrical equipment connected to each public power line. Public electrical equipment includes, but is not limited to, elevators, public lighting, public air conditioning, and fire protection systems, etc.

[0066] The data acquisition module includes a non-intrusive current transformer and a voltage probe, which enables the acquisition of current and voltage values ​​of the line through electromagnetic induction without direct contact with the line. It has the advantages of high safety, simple installation and minimal impact on equipment operation, and is suitable for monitoring the power parameters of various lines in the building in this application.

[0067] As one implementation method, the line power characteristics include, but are not limited to: DC component of current, peak current, RMS current, current fluctuation, DC component of voltage, peak voltage, RMS voltage, voltage fluctuation, active power, peak instantaneous power, apparent power, power factor, and phase angle. By setting the above-mentioned line power characteristics, the power operation status of the public utility lines can be fully extracted, providing a basis for subsequent decomposition to obtain the equipment power characteristics of each public utility device on the public utility line. The logic for obtaining the line power characteristics of each public utility line in the building during the current monitoring period is as follows:

[0068] For each public power line, the current and voltage values ​​are sampled multiple times during the current monitoring period using current transformers and voltage probes to form current and voltage value sequences. After preprocessing, the line's electrical characteristics are extracted. The relevant mathematical expressions are as follows:

[0069]

[0070] In the formula, Indicates the number of times within the current monitoring period. The current value collected during the next sampling. This is the index of the number of samples taken within the current monitoring time period, and , This represents the total number of samples taken during the current monitoring period. This indicates taking the absolute value. This indicates that the maximum value is taken. The current monitoring time period is a period of time that is back a certain length of time from the current moment to the end. The length of the current monitoring time period can generally be 1 second, 2 seconds, 5 seconds, etc. While being close to the current moment, enough current and voltage values ​​are obtained so that the extracted line power characteristics can truly reflect the operating status of each public electrical equipment on the line, avoiding the problem of random errors. When sampling current and voltage values ​​within the current monitoring time period, in order to avoid aliasing, the sampling frequency is generally set to more than twice the AC signal frequency. For example, the frequency of AC signals in my country is generally 50Hz, so the sampling frequency can be set between 100Hz and 1000Hz, that is, the sampling interval is set between 1 millisecond and 10 milliseconds. Of course, it can also be set according to the specific model of public electrical equipment on the public power line, and there is no restriction here.

[0071] In the formula, This refers to the DC component of the current in the public power lines during the current monitoring period. This refers to the peak current of the public power line during the current monitoring period. This refers to the effective current value of the public power line during the current monitoring period. The current fluctuation value of the public power line during the current monitoring period is characterized by the current standard deviation. The larger the value, the greater the degree of current fluctuation of the public power line during the current monitoring period.

[0072]

[0073] In the formula, Indicates the number of times within the current monitoring period. The voltage value collected during the next sampling. This refers to the DC component of the voltage of public power lines during the current monitoring period. The peak voltage of the public power line during the current monitoring period. This refers to the effective voltage value of public power lines during the current monitoring period. The voltage fluctuation value of the public power line during the current monitoring period is characterized by the voltage standard deviation. The larger the value, the greater the voltage fluctuation of the public power line during the current monitoring period.

[0074]

[0075] In the formula, For public power lines during the current monitoring period Instantaneous power at the next sampling point The active power of public power lines during the current monitoring period. This refers to the peak instantaneous power of public power lines during the current monitoring period. The apparent power of public power lines during the current monitoring period. The power factor of public power lines during the current monitoring period. The phase angle of the public power line during the current monitoring period. It is the inverse cosine function;

[0076] It should be noted that all public electrical devices located on the same public utility line share the same power source. That is, the current and voltage signals of all public electrical devices on the same public utility line are superimposed. This means that the line power characteristics of the public utility line can be regarded as a combination of the device power characteristics of each public electrical device. When the public electrical devices are operating normally, their load characteristics are relatively stable, that is, their device power characteristics are relatively stable. Therefore, under normal conditions, by monitoring the line power characteristics, the device power characteristics of each public electrical device can be inferred. Thus, the line power characteristics of the public utility line are obtained here so that they can be decomposed to obtain the device power characteristics of each public electrical device on the public utility line, without having to obtain the device power characteristics of each public electrical device individually, which greatly reduces the data collection cost.

[0077] As one implementation method, preprocessing includes, but is not limited to, denoising and normalization. Specifically, denoising is performed on the current and voltage value sequences using a low-pass filter to remove high-frequency noise. Then, the denoised current and voltage value sequences are normalized using a minimax normalization method to make the element values ​​dimensionless and scaled to between 0 and 1, facilitating subsequent feature extraction and use as model input. During denoising, an RC low-pass filter can be selected, and the cutoff frequency can be set to less than half the sampling frequency, specifically between one-quarter and two-fifths of the sampling frequency, to remove high-frequency noise. During normalization, the maximum and minimum current values ​​of the public power line in the past month can be selected for minimax normalization of the current value sequence, and the maximum and minimum voltage values ​​of the public power line in the past month can be selected for minimax normalization of the voltage value sequence.

[0078] The encoding-decoding model is used to decompose the line power characteristics of a public power line under the current monitoring period into the equipment power characteristics of each public power device on the public power line, and then restore the line power characteristics of each equipment to the line power characteristics and calculate the reconstruction error. The encoding-decoding model is trained based on the line power characteristics and equipment power characteristics under normal conditions.

[0079] The encoder-decoder model consists of an encoder and a decoder. The decoder decomposes the line power characteristics into the equipment power characteristics of each public electrical device on the corresponding public utility line. It includes an input layer, a hidden layer, and an output layer, as detailed below:

[0080] The input layer includes M input nodes, which are used to receive line power characteristics. M is the number of types of line power characteristics. For example, if there are 13 types of line power characteristics, such as DC component of current, peak current, RMS value of current, current fluctuation value, DC component of voltage, peak voltage, RMS value of voltage, voltage fluctuation value, active power, peak instantaneous power, apparent power, power factor and phase angle, then M is 13.

[0081] The hidden layers include a first hidden layer and a second hidden layer, as detailed below:

[0082] The first fully connected layer has 32 nodes and uses the ReLU activation function to extract nonlinear combinations of line power features, enhancing the decoder's expressive power and enabling the decoder to learn more complex feature relationships.

[0083] The second fully connected layer has 64 nodes and uses the ReLU activation function to further extract and transform features, enhance the depth of the decoder, and help the decoder capture higher-dimensional feature information.

[0084] The output layer adopts a fully connected layer structure with Q×k nodes. It is used to output the power characteristics of each public electrical device on the public power line. Q is the number of public electrical devices on the public power line, and k is the number of types of power characteristics of the public electrical devices. It is set according to the specific number of public electrical devices on the public power line.

[0085] The encoder is used to convert the electrical characteristics of various public electrical devices on the public utility line back into the line electrical characteristics. It also includes an input layer, a hidden layer, and an output layer, as detailed below:

[0086] The input layer adopts a fully connected layer structure, including Q×k input nodes, which are used to receive the equipment power characteristics of each public power equipment on the public power line;

[0087] The hidden layers include a first hidden layer and a second hidden layer, as detailed below:

[0088] The first fully connected layer has 64 nodes and uses the ReLU activation function to extract the nonlinear combination of the equipment's electrical characteristics, thereby enhancing the encoder's expressive power and enabling the encoder to effectively capture the nonlinear relationship between the equipment's electrical characteristics and the line's electrical characteristics.

[0089] The second fully connected layer has 32 nodes and uses the ReLU activation function to further process and transform the electrical characteristics of the device, helping the encoder learn more complex feature mappings, thereby improving the accuracy of reconstructing the electrical characteristics of the line.

[0090] The output layer adopts a fully connected layer structure with M nodes and is used to output the power characteristics of the line.

[0091] As one implementation method, the equipment power characteristics may include one or more of the following: active power, phase angle, power factor, RMS current, current fluctuation, RMS voltage, and voltage fluctuation. The specific acquisition method is the same as the acquisition method of the line power characteristics described above, and will not be repeated here. The setting of equipment power characteristics provides an evaluation benchmark for subsequent accurate identification of whether each public power equipment has a fault. For example, if active power, reactive power, power factor, RMS current, current fluctuation, RMS voltage, and voltage fluctuation are all set as the equipment power characteristics in this technical solution, then k=7.

[0092] The process of training the encoder-decoder model is as follows:

[0093] For each public power line, the time-series data of current and voltage values ​​for that public power line, as well as the time-series data of current and voltage values ​​of each public electrical device on that public power line, are simultaneously acquired over a past period. The past period can be selected as the past year, the past six months, or the past three months, etc., and the specific settings are determined by the staff according to the actual situation, which will not be elaborated here. Based on maintenance reports, expert evaluations, and other methods, the time period in which all public electrical devices on that public power line were in a normal state over a past period is determined as the normal time period. Then, the time-series data of current and voltage values ​​of that public power line, as well as the time-series data of current and voltage values ​​of each public electrical device on that public power line, are extracted during the normal time period. Then, the above time-series data are divided into multiple sample time periods according to the length of the current monitoring time period, that is, the length of each sample time period is the same as the current monitoring time period. The power characteristics of the line and the power characteristics of each device under each sample time period are calculated to construct a sample set, and the sample set is divided into a training set, a test set, and a validation set in a ratio of 70:15:15.

[0094] Mean squared error is chosen as the loss function for both the encoder and decoder. The Adam optimizer and stochastic gradient descent algorithm are selected to update the momentum parameters of the encoder and decoder to minimize the loss function. Hyperparameters such as learning rate, batch size, and training epochs are predefined. For example, the learning rate can be set between 0.001 and 0.01 to avoid instability during training due to an excessively large learning rate or slow convergence due to an excessively small learning rate. The batch size can be set to 32 or 64 to ensure appropriate memory usage and training speed. The number of training epochs can be set between 50 and 200 to ensure that the encoder and decoder fully learn the mapping relationship between the line power characteristics and the power characteristics of each device under normal conditions.

[0095] The line power features in the training set are used as input, and the power features of each device are used as output labels to train the decoder. The power features of each device in the training set are used as input, and the line power features are used as output labels to train the encoder. After completing one round of training, the validation set is input into the encoder / decoder for validation. If the mean square error is less than a preset threshold, the encoder / decoder is considered to have completed training. Otherwise, training continues until the predetermined number of training rounds is reached. The encoder / decoder is tested using a test set. If the mean square error of the encoder / decoder is less than a preset threshold, the encoder / decoder is considered to have completed training. Otherwise, the hyperparameters are adjusted and the encoder / decoder is retrained until the mean square error is less than the preset threshold. The preset threshold can be set between 0.01 and 0.1. The specific setting is determined by the staff according to the actual situation and will not be elaborated here.

[0096] It should be noted that public electrical equipment malfunctions are relatively rare, making it difficult to extract enough data for model training. Furthermore, when a malfunction occurs, the power signal is affected by interference and noise, resulting in inaccurate measurements of current and voltage data. Extracting accurate equipment power characteristics from these measurements is challenging. More importantly, multiple devices on the same public power line may malfunction simultaneously, and the interactions between them significantly increase the complexity of the power characteristics. Based on the above, it is evident that training a model capable of accurately decomposing the line power characteristics under abnormal conditions to obtain the power characteristics of each device is quite difficult.

[0097] However, the load characteristics of public electrical equipment are relatively stable during normal operation, that is, the power characteristics of the equipment are relatively stable. It is relatively easy and accurate to train a model that can decompose the power characteristics of the line to obtain the power characteristics of each device under normal conditions. Therefore, this scheme uses data under normal conditions to train the encoder-decoder model. This model only learns the mapping relationship between the power characteristics of the line and the power characteristics of each device under normal conditions. Therefore, the power characteristics of the line under normal conditions are input into the decoder to obtain the power characteristics of each device. Then, the power characteristics of each device are input into the encoder to reconstruct the power characteristics of the line. The reconstructed power characteristics are highly similar to the original values, and the reconstruction error is small. However, when reconstructing the power characteristics of the line under normal conditions, the reconstruction error will increase because the encoder and decoder have failed to learn the mapping relationship under fault conditions.

[0098] The logic for obtaining the reconstruction error is as follows: The line power characteristics of the public power line during the current monitoring period are input into the decoder to obtain the decoder output value, i.e., the equipment power characteristics of each public power device. Then, the decoder output value is input into the encoder to obtain the encoder output value, i.e., the reconstructed line power characteristics. The difference between the line power characteristics of the public power line during the current monitoring period and the encoder output value is compared and analyzed to obtain the reconstruction error. The specific mathematical expression is as follows:

[0099]

[0100] In the formula, For reconstruction error, For public power lines during the current monitoring period, the [number]th [period] The power characteristics of this type of line For the reconstructed first The power characteristics of this type of line This is an index for the types of power characteristics of the line, and Since the power characteristics of the power line are diverse and multidimensional, Euclidean distance is used to quantify the reconstruction error. The larger the reconstruction error, the greater the difference between the power characteristics of the power line under the current monitoring period and the reconstructed power characteristics, which indicates that the public power equipment on the power line is more likely to be faulty or abnormal.

[0101] The feature adjustment module is used to adjust the power characteristics of each device output by the encoder-decoder model by combining the particle swarm algorithm and the simulated annealing algorithm when the reconstruction error is greater than the reconstruction error threshold, so as to obtain the optimal adjustment value of the power characteristics of the device that satisfies the reconstruction error not being greater than the reconstruction error threshold.

[0102] The logic for obtaining the reconstruction error threshold is as follows: multiple sets of line power characteristics under normal conditions are obtained as sample line power characteristics, and then input into the trained encoder-decoder model one by one to obtain the reconstruction error corresponding to each sample line power characteristic and extract the mean and standard deviation. The standard deviation is multiplied by a margin factor of not less than 1 to form a warning increment. The sum of the warning increment and the mean is used as the reconstruction error threshold. The smaller the margin factor value, the stricter the fault judgment of public electrical equipment on public power lines. Conversely, the larger the margin factor value, the more lenient the fault judgment of public electrical equipment on public power lines. The specific value can be set between 1 and 3, and the specific setting is determined by the staff according to the actual situation. For example, the margin factor can be set to 3 to cover the vast majority of normal reconstruction errors and provide a more suitable judgment benchmark.

[0103] It should be noted that if the reconstruction error is not greater than the reconstruction error threshold, it means that the power characteristics of the public power line under the current monitoring period are the power characteristics of the line under normal conditions. That is, it means that all public power equipment on the public power line is in normal condition and there is no need to carry out fault location and early warning.

[0104] The logic for obtaining the optimal adjustment value of the equipment's power characteristics is as follows:

[0105] 1) Based on the encoder-decoder model, the decoder analyzes the line power characteristics of the public power line during the current monitoring period, uses the obtained power characteristics of each device as the initial particles, and randomly perturbs the initial particles to generate multiple new particles to form a particle swarm.

[0106] The initial particles are high-dimensional feature combinations formed by the electrical characteristics of each device, represented by a Q×k two-dimensional matrix. The representation is as follows:

[0107]

[0108] In the formula, Let be the element value in the x-th row and y-th column of the two-dimensional matrix Z, representing the y-th electrical characteristic of the x-th public utility device on the public utility power line, where x is the index of the public utility device on the public utility power line. y is the index of the equipment's electrical characteristic type, and ;

[0109] The logic for randomly perturbing the initial particle to form a particle swarm is as follows: the initial particle is replicated multiple times to form multiple copy particles. For each copy particle, a perturbation range is defined for each element value. Within the perturbation range, the element value is randomly perturbed. The perturbed element value is then used as the final element value to update the copy particle, forming multiple new particles to construct the particle swarm. The upper limit of the perturbation range is the product of the element value and the perturbation factor, and the lower limit of the perturbation range is the negative number of the product of the element value and the perturbation factor. The perturbation factor is a decimal greater than 0, and its specific value is set by the staff according to the actual situation. Generally, it is set between 20% and 40% to avoid the problem that the perturbation factor is too small, which would make the similarity of the particles in the particle swarm too high, which would be not conducive to exploring new solutions, and the perturbation factor is too large, which would seriously deviate from the equipment power characteristics output by the encoding-decoding model, resulting in a decrease in the quality of the final optimal adjustment value of the equipment power characteristics.

[0110] It should be noted that the size of the particle swarm is determined based on the number of features contained in the particles, that is, based on the product of Q and k. If the product of Q and k is less than 10, the size of the particle swarm can be set between 10 and 30, that is, the initial particles are randomly perturbed to generate 10 to 30 new particles to build the particle swarm. If the product of Q and k is greater than 10, the size of the particle swarm can be set between 30 and 100, so as to balance fully exploring the solution space and reducing computational complexity.

[0111] The specific mathematical expression for perturbing the element values ​​is as follows:

[0112]

[0113] In the formula, As a disturbance factor, Indicates in arrive Take a random value between them. Indicates the after of the disturbance ;

[0114] 2) Each particle in the particle swarm is input into the encoder of the encoder-decoder model to generate the line power feature corresponding to that particle. Then, the line power feature is input into the encoder-decoder model and the reconstruction error corresponding to that particle is calculated. The difference between the reconstruction error threshold and the reconstruction error is used as the fitness value of the particle. The historical best position and global best position of each particle are recorded. The velocity and position of each particle are updated iteratively in this way. The specific update mathematical expression is as follows:

[0115]

[0116]

[0117] In the formula, For inertial weights, Let be the velocity of the o-th particle in the t-th iteration. This represents the best historical position of the o-th particle up to the t-th iteration. Let be the position of the o-th particle in the t-th iteration. This represents the globally optimal position of the entire particle swarm up to the t-th iteration. Let be the position of the o-th particle in the (t+1)-th iteration. and All are learning factors. and All are random numbers, where o is the index of a particle in the particle swarm, and t is the index of the iteration number;

[0118] It should be noted that if the power characteristics of each device represented by a particle after the t-th iteration are the power characteristics of the device under normal conditions, then the encoder-decoder model can accurately handle the mapping relationship between the power characteristics of each device represented by the particle and the power characteristics of the line. The power characteristics of each device represented by the particle after the t-th iteration are input into the encoder of the encoder-decoder model to obtain the power characteristics of the line, and then input into the encoder-decoder model again. The corresponding reconstruction error is small. Therefore, the difference between the reconstruction error threshold and the reconstruction error is used as the fitness value. The smaller the reconstruction error, the larger the fitness value, indicating that the power characteristics of each device represented by the particle after the t-th iteration are closer to the normal state. Based on this, after each iteration, the historical best position and global best position of each particle can be recorded based on the fitness value, and then the velocity and position of each particle can be updated so that the particle can be optimized towards the power characteristics of each device under normal conditions.

[0119] As one implementation method, since the electrical characteristics of the device have already been normalized, when updating the velocity and position of particles using the particle swarm optimization algorithm, the initial velocity of the particles can be set between -0.1 and 0.1, i.e. To balance exploring new solutions and maintaining the stability of the exploration optimization, the inertia weight is used to control the particle's tendency, affecting the global and local exploration capabilities of the search. It can generally be set between 0.4 and 0.9. The learning factor... and This is used to control the degree to which a particle learns from its own historical best solution and the global best solution. It can generally be set between 1.5 and 2.5, using random numbers. and Used to adjust the influence of the particle's own historical best solution and global best solution on the particle, in order to increase the randomness of the search, it is generally set to a decimal between 0 and 1;

[0120] 3) After each iteration, calculate the fitness value of each particle and determine whether the termination condition is met. If it is met, select the best particle as the initial solution to proceed to step 4). If it is not met, record the historical best position and global best position of each particle after this iteration based on the fitness value, and continue to iterate and update the velocity and position of each particle until the termination condition is met.

[0121] The termination conditions include condition one and condition two. Only when both condition one and condition two are met is the termination condition considered satisfied. Condition one and condition two are as follows:

[0122] Condition 1: After the current iteration, at least one particle has a fitness value of not less than 0. This setting is used to find particles whose reconstruction error is not greater than the reconstruction error threshold after the current iteration, so as to ensure that the power characteristics of the equipment used as the initial solution for subsequent optimization by the simulated annealing algorithm are in a normal state. The logic is as follows: Since the encoder-decoder model is learned and trained based on the line power characteristics and the power characteristics of each device in a normal state, if the fitness of a particle is not less than 0, it means that the power characteristics of each device represented by the particle are the power characteristics of each device in a normal state, and the corresponding line power characteristics are also the line power characteristics in a normal state.

[0123] Condition 2: After this round of iterations, at least one particle, in addition to satisfying Condition 1, also has a mutation index that is less than the mutation index threshold. The specific expression for the particle's mutation index is as follows:

[0124]

[0125] In the formula, The variation index, The power characteristics of public power lines during the current monitoring period are represented by a 1×M matrix. To input the electrical characteristics of each device represented by particles satisfying condition one into the encoder of the encoder-decoder model, the resulting line electrical characteristics are also represented as a 1×M matrix. express and Euclidean distance, express The Frobenius norm;

[0126] In the formula, The electrical characteristics of each device, represented by the initial particle, are expressed as a matrix of form Q×k. To satisfy condition one, the electrical characteristics of each device represented by the particles are also represented by a matrix of the form Q×k. express and Euclidean distance, express The Frobenius norm;

[0127] In the formula, To input the reference line power characteristics into the decoder of the encoder-decoder model, the resulting power characteristics of each device are also represented as a matrix of form Q×k. The logic for obtaining the reference line power characteristics is as follows: the most recent historical monitoring time period when the public power line is in a normal state and is closest to the current monitoring time period is taken as the reference time period, and the line power characteristics of the public power line during the reference time period are taken as the reference line power characteristics. express and Euclidean distance, express The Frobenius norm;

[0128] It should be noted that when a public utility device on a public utility line malfunctions, its electrical characteristics will change significantly. Other public utility devices connected to the same line that are functioning normally will only experience minor changes or remain unchanged. In other words, when the decoder decomposes the power characteristics of the line under abnormal conditions, the electrical characteristics of the malfunctioning public utility device will show severe changes. This is a significant factor contributing to the reconstruction error exceeding the reconstruction error threshold. Meanwhile, the electrical characteristics of other public utility devices functioning normally will show only minor changes or remain unchanged compared to the true value, having a smaller impact on the reconstruction error exceeding the threshold. Based on this, if a set of electrical characteristics can be found... The optimal adjustment value satisfies the following: the power characteristics of other public electrical equipment in normal state are consistent with the power characteristics of the corresponding equipment in this group of power characteristics, and the power characteristics of the public electrical equipment in this group of power characteristics are the power characteristics of the public electrical equipment in normal state under the fault abnormal state. By comparing this group of power characteristics with each power characteristic output by the decoder, we can find the public electrical equipment corresponding to the power characteristics of the equipment with significantly different power characteristics. The public electrical equipment found is the public electrical equipment in the abnormal state. Therefore, based on this principle, we set a fitness of not less than 0 and a mutation index to find the initial solution of the optimal adjustment value of the power characteristics of the equipment, so as to accelerate the convergence speed of finding the optimal adjustment value of the power characteristics of the equipment through the simulated annealing algorithm in the following text.

[0129] It should be noted that, based on the above, the initial solution needs to be as close as possible to the initial particle while satisfying condition one. Therefore, the mathematical expression for the mutation exponent is set accordingly. This value is used to measure the difference between the particle satisfying condition one and the initial particle. The larger the value, the further the particle is from the optimal adjustment value of the equipment's power characteristics, and the less suitable it is as the initial solution for the optimal adjustment value of the equipment's power characteristics. This measure assesses the difference between the line power characteristics corresponding to particles satisfying condition one and the line power characteristics during the current monitoring period. Essentially, it indirectly reflects the difference between particles satisfying condition one and the initial particles based on the degree of difference in line power characteristics. This is used to measure the difference between the equipment power characteristics corresponding to particles that meet condition one and the equipment power characteristics within a reference time period. Since the reference time period is close to the current monitoring time period, and public electrical equipment under normal conditions generally does not change significantly between two close time periods, the equipment power characteristics within the reference time period can be considered close to the optimal adjustment value of the equipment power characteristics. Therefore, this is done by... This indirectly reflects the distance between the particle and the optimal adjustment value of the equipment's power characteristics at the time level, and is thus used to determine whether the particle is suitable as the initial solution for the optimal adjustment value of the equipment's power characteristics. Therefore, it can be concluded that... , and The larger the value, the farther the particle is from the optimal adjustment value of the equipment's power characteristics, and the less suitable it is as the initial solution for the optimal adjustment value of the equipment's power characteristics. (Variation exponent) The larger the value, the less suitable the particle is as the initial solution for the optimal adjustment value of the equipment's power characteristics.

[0130] In the formula, , and All are preset weighting coefficients, used to represent , and The greater the weight of the particle in the calculation of the variability index, the more important it is to the calculation, that is, the more important it is to the initial solution for measuring whether the particle is suitable as the optimal adjustment value of the equipment's power characteristics. The particle's suitability as an initial solution for the optimal adjustment value of the equipment's power characteristics is directly reflected at the equipment's power characteristic level; therefore, a maximum weight coefficient is assigned to it. , Although both indirectly reflect whether a particle is suitable as the initial solution for the optimal adjustment value of the equipment's power characteristics, there are cases where the equipment's power characteristics differ significantly between two adjacent time periods (such as the switching phase of an elevator from standby to operation). There is a certain degree of random error in assessing whether a particle is suitable as the optimal adjustment value for the electrical characteristics of a device, therefore, it is assigned... A medium weighting coefficient is assigned to A minimum weighting coefficient, that is, in letting Based on , , and The specific value can be set by the staff themselves based on the above relationship, such as... , , No restrictions are imposed here;

[0131] In this context, the optimal particle is the particle with the minimum mutation index among those that satisfy condition one and whose mutation index is less than the mutation index threshold. The mutation index threshold is generally set between 10% and 30% to fully utilize the fast convergence characteristic of the particle swarm optimization algorithm and find an initial solution of acceptable quality. If the mutation index threshold is too large, a poor initial solution may be selected too early, reducing the convergence speed of subsequent simulated annealing optimization. If the mutation index threshold is too small, there is a problem of wasting the computational resources of the particle swarm optimization algorithm. Since condition one was not imposed on the particles when using the particle swarm optimization algorithm in the previous text to improve the convergence speed, a large number of useless particles (i.e., particles that do not satisfy condition one) will be generated in the particle swarm optimization process. Furthermore, the particle swarm optimization algorithm is prone to getting trapped in local optima during the optimization process and cannot perform a global search to find the global optimum like the simulated annealing algorithm. Therefore, it is unnecessary to waste a lot of computational resources and time in the particle swarm optimization process to find the optimal solution. In other words, the mutation index threshold is generally set between 10% and 30%.

[0132] 4) Set the initial annealing temperature, the termination temperature, and the cooling rate. Use the initial solution as the current solution in the first round of annealing optimization. In each round of annealing optimization, the current solution is perturbed based on the annealing temperature to generate a neighborhood solution with a fitness value not less than 0. Calculate the mutation index of the neighborhood solution, analyze the mutation index of the current solution and the neighborhood solution to calculate the acceptance probability, and use the probability to accept the neighborhood solution as the current solution. Update the annealing temperature after each round of annealing optimization until the annealing temperature is lower than the termination temperature. Output the final current solution as the optimal adjustment value of the equipment's power characteristics.

[0133] As an implementation method, since a relatively good initial solution has already been found based on the particle swarm optimization algorithm, the initial annealing temperature does not need to be set too high. It can generally be set between 10-30°C. The termination temperature can be set to a small value such as 1, 0.1 or 0.01. The cooling rate can be set between 0.8-0.99. Here, it can be set to a higher value of 0.95 to allow for a thorough exploration of the global space in order to find the global optimal solution.

[0134] The logic for perturbing the current solution based on the annealing temperature to generate neighborhood solutions with fitness values ​​not less than 0 is as follows: In each round of annealing optimization, for each element value of the current solution, a corresponding small perturbation interval is defined. A random value is randomly generated from the small perturbation interval, and its product with the annealing temperature in this round of annealing optimization is calculated as the perturbation value. The perturbation value is summed with the corresponding element value of the current solution to obtain the element value of the neighborhood solution. This process generates neighborhood solutions. Then, the fitness value of the neighborhood solution is determined using the same method. If the fitness value of the neighborhood solution is not less than 0, the next step is performed. The mutation index is calculated in one step; otherwise, random values ​​are regenerated to regenerate neighborhood solutions until the fitness value of the generated neighborhood solutions is not less than 0. The upper limit of the small perturbation interval is half the width of the small perturbation interval, and the lower limit is the product of half the width of the small perturbation interval and -1. The half width of the small perturbation interval is less than the perturbation factor, so as to explore the global optimal solution in detail in the simulated annealing algorithm optimization. The half width of the small perturbation interval can be set between 0.01 and 0.1 to ensure that the global optimal solution is explored in detail in the simulated annealing algorithm optimization.

[0135] The mathematical expression for the probability of acceptance is as follows:

[0136]

[0137] In the formula, This represents the mutation index of the neighborhood solution during the q-th round of annealing optimization. This represents the mutation index of the current solution during the q-th round of annealing optimization. This represents the annealing temperature during the q-th round of annealing optimization. Let represent the probability of accepting a neighborhood solution in the q-th round of annealing optimization, where q is the index of the annealing optimization round. If the neighboring solution is superior to the current solution, then let... Conversely, when If the current solution is superior to the neighboring solutions, then... Calculate the probability of acceptance , The smaller the value, the worse the neighborhood solution is compared to the current solution, and the lower the acceptance probability. The smaller it is, and the smaller it is. Based on a negative number, the annealing temperature The smaller, The smaller the overall size, the lower the probability of acceptance. This further reduces the perturbation range, reflecting that during the simulated annealing optimization process, when the annealing temperature is high, the algorithm allows for larger random perturbations to explore the solution space. However, as the iteration progresses, the annealing temperature gradually decreases, and the perturbation range becomes smaller, thus focusing more on local search.

[0138] The mathematical expression for updating the annealing temperature after each round of annealing optimization is as follows:

[0139]

[0140] In the formula, This represents the annealing temperature during the (q+1)th round of annealing optimization. Indicates the cooling rate.

[0141] The location and early warning module is used to analyze the degree of difference between the equipment power characteristics output by the encoding-decoding model and the optimal adjustment value of the equipment power characteristics, in order to determine the difference index that reflects the degree of abnormality of public power equipment, and to locate and issue early warnings for public power equipment with faults based on the difference index;

[0142] The specific mathematical expression for the difference index is as follows:

[0143]

[0144] In the formula, Let y be the optimal adjustment value for the electrical characteristic of the x-th public utility device on the public utility power line. Let x be the difference index of the xth public electrical equipment on the public power line. Here, the average relative error of the power characteristics of each device on the public electrical equipment is used to quantify the difference index of the public electrical equipment. The larger the difference index, the higher the probability of the public electrical equipment malfunctioning or abnormal. The specific reasons have been described above and will not be repeated here. Of course, the difference index can also be quantified by Euclidean distance or the maximum absolute error. There is no restriction here.

[0145] The logic for locating faulty public electrical equipment based on the difference index is as follows: if the difference index of a public electrical equipment is higher than the difference index threshold, then this public electrical equipment is defined as having a fault. If the difference index of all public electrical equipment on the public power line is not higher than the difference index threshold, then the public electrical equipment with the highest difference index is designated as a suspected faulty public electrical equipment, and a suspected fault signal is issued so that staff can promptly check the public electrical equipment for abnormalities to avoid potential faults and achieve the technical effect of early detection and early maintenance, thereby reducing subsequent risks and maintenance costs. Specifically, the warning is issued in the form of an early warning signal, which includes at least the equipment ID of the public electrical equipment defined as having a fault.

[0146] It should be noted that the difference index threshold is specifically obtained by analyzing historical data. For example, in this technical solution, 471 sample data points were analyzed for various public electrical devices on the same public power line over the past three months. Of these, 156 samples had a difference index between 0% and 5%, with 3 of these corresponding to equipment malfunctions (a probability of 1.92%). 118 samples had a difference index between 5% and 10%, with 6 of these corresponding to equipment malfunctions (a probability of 5.08%). 95 samples had a difference index between 10% and 15%. Of the 47 data points, 5 correspond to public electrical equipment malfunctions, with a probability of 5.26%. Of the 47 data points with a difference index between 15% and 20%, 4 correspond to public electrical equipment malfunctions, with a probability of 8.51%. Of the 23 data points with a difference index between 20% and 30%, 19 correspond to public electrical equipment malfunctions, with a probability of 82.61%. Of the 32 data points with a difference index exceeding 30%, 29 correspond to public electrical equipment malfunctions, with a probability of 90.63%. See Table 1 and the accompanying diagram in the instruction manual for details. Figure 2 As shown;

[0147] Table 1 Fault Probability Distribution Table

[0148]

[0149] The table above clearly shows that when the difference index exceeds 20%, the probability of abnormalities in public electrical equipment increases significantly. Therefore, the difference index threshold can be set to 20%. Of course, this is just one specific implementation method. The specific value of the difference index needs to be selected and determined based on the actual situation, and no restrictions are imposed here.

[0150] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0151] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0152] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0153] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A positioning and early warning system for public electrical equipment in buildings, characterized in that, include: The data acquisition module is used to obtain the line power characteristics of each public power line in the building during the current monitoring period; The encoding-decoding model is used to decompose the line power characteristics of a public power line under the current monitoring period into the equipment power characteristics of each public power device on the public power line, and then restore the line power characteristics of each equipment to the line power characteristics and calculate the reconstruction error. The encoding-decoding model is trained based on the line power characteristics and equipment power characteristics under normal conditions. The feature adjustment module is used to adjust the power characteristics of each device output by the encoder-decoder model by combining the particle swarm algorithm and the simulated annealing algorithm when the reconstruction error is greater than the reconstruction error threshold, so as to obtain the optimal adjustment value of the power characteristics of the device that satisfies the reconstruction error not being greater than the reconstruction error threshold. The location and early warning module is used to analyze the degree of difference between the equipment power characteristics output by the encoding-decoding model and the optimal adjustment value of the equipment power characteristics, in order to determine the difference index that reflects the degree of abnormality of public power equipment, and to locate and issue early warnings for public power equipment with faults based on the difference index.

2. The positioning and early warning system for building public electrical equipment according to claim 1, characterized in that: The public power line is a line used to supply power to public electrical equipment within the building, and the data acquisition module includes a current transformer and a voltage probe installed at the public power line.

3. The positioning and early warning system for building public electrical equipment according to claim 2, characterized in that: The line power characteristics include, but are not limited to, DC component of current, peak current, RMS current, current fluctuation, DC component of voltage, peak voltage, RMS voltage, voltage fluctuation, active power, peak instantaneous power, apparent power, power factor, and phase angle, which are specifically obtained by analyzing the current value sequence and voltage value sequence obtained by current transformers and voltage probes.

4. The building public electrical equipment positioning and early warning system according to claim 1, characterized in that: The encoding-decoding model consists of an encoder and a decoder. The decoder is used to decompose the line power characteristics into the equipment power characteristics of each public electrical device on the corresponding public power line, and the encoder is used to convert the equipment power characteristics of each public electrical device on the public power line back into the line power characteristics. The decoder includes an input layer, a hidden layer, and an output layer, as detailed below: The input layer consists of M input nodes, which are used to receive line power characteristics, where M is the number of types of line power characteristics. The hidden layers include a first hidden layer and a second hidden layer, as detailed below: The first fully connected layer has 32 nodes and uses the ReLU activation function to extract nonlinear combinations of line power characteristics; The second fully connected layer has 64 nodes and uses the ReLU activation function to further extract and transform features; The output layer adopts a fully connected layer structure with Q×k nodes. It is used to output the equipment power characteristics of each public power device on the public power line, where Q is the number of public power devices on the public power line and k is the number of equipment power characteristic types of the public power devices. The encoder also includes an input layer, a hidden layer, and an output layer, as detailed below: The input layer adopts a fully connected layer structure, including Q×k input nodes, which are used to receive the equipment power characteristics of each public power equipment on the public power line; The hidden layers include a first hidden layer and a second hidden layer, as detailed below: The first fully connected layer has 64 nodes and uses the ReLU activation function to extract nonlinear combinations of the device's electrical characteristics. The second fully connected layer has 32 nodes and uses the ReLU activation function to further process and transform the device's electrical characteristics. The output layer adopts a fully connected layer structure with M nodes and is used to output the power characteristics of the line.

5. A positioning and early warning system for building public electrical equipment according to claim 1, characterized in that: The electrical characteristics of the device include one or more of the following: active power, phase angle, power factor, effective current value, current fluctuation value, effective voltage value, and voltage fluctuation value.

6. The positioning and early warning system for building public electrical equipment according to claim 1, characterized in that: The logic for obtaining the reconstruction error threshold is as follows: multiple sets of line power features under normal conditions are obtained as sample line power features, and they are input into the trained encoder-decoder model one by one to obtain the reconstruction error corresponding to each sample line power feature and extract the mean and standard deviation. The standard deviation is multiplied by a margin factor of not less than 1 to form a warning increment, and the sum of the warning increment and the mean is used as the reconstruction error threshold.

7. A positioning and early warning system for public electrical equipment in buildings according to claim 4, characterized in that: The logic for obtaining the optimal adjustment value of the equipment's power characteristics is as follows: 1) Based on the encoder-decoder model, the decoder analyzes the line power characteristics of the public power line during the current monitoring period, uses the obtained power characteristics of each device as the initial particles, and randomly perturbs the initial particles to generate multiple new particles to form a particle swarm. The initial particles are represented by a two-dimensional matrix of Q×k. The representation is as follows: In the formula, Let be the element value in the x-th row and y-th column of the two-dimensional matrix Z, representing the y-th electrical characteristic of the x-th public utility device on the public utility power line, where x is the index of the public utility device on the public utility power line. y is the index of the equipment's electrical characteristic type, and ; The logic for randomly perturbing the initial particle to form a particle swarm is as follows: the initial particle is copied multiple times to form multiple copy particles. For each copy particle, a perturbation range is defined for each element value. Within the perturbation range, the element value is randomly perturbed. The perturbed element value is then used as the final element value to update the copy particle, forming multiple new particles to construct the particle swarm. The upper limit of the perturbation range is the product of the element value and the perturbation factor, and the lower limit of the perturbation range is the negative number of the product of the element value and the perturbation factor. The perturbation factor is a decimal greater than 0. The specific mathematical expression for perturbing the element values ​​is as follows: In the formula, As a disturbance factor, Indicates in arrive Take a random value between them. Indicates the after of the disturbance ; 2) Input each particle in the particle swarm into the encoder of the encoder-decoder model to generate the line power feature corresponding to the particle. Then input the line power feature into the encoder-decoder model and calculate the reconstruction error corresponding to the particle. Use the difference between the reconstruction error threshold and the reconstruction error as the fitness value of the particle. Record the historical best position and global best position of each particle. Iteratively update the velocity and position of each particle in this way. 3) After each iteration, calculate the fitness value of each particle and determine whether the termination condition is met. If it is met, select the best particle as the initial solution to proceed to step 4). If it is not met, record the historical best position and global best position of each particle after this iteration based on the fitness value, and continue to iterate and update the velocity and position of each particle until the termination condition is met. 4) Set the initial annealing temperature, the termination temperature, and the cooling rate. Use the initial solution as the current solution in the first round of annealing optimization. In each round of annealing optimization, the current solution is perturbed based on the annealing temperature to generate a neighborhood solution with a fitness value not less than 0. Calculate the mutation index of the neighborhood solution, analyze the mutation index of the current solution and the neighborhood solution to calculate the acceptance probability, and use the probability to accept the neighborhood solution as the current solution. Update the annealing temperature after each round of annealing optimization until the annealing temperature is lower than the termination temperature. Output the final current solution as the optimal adjustment value of the equipment's power characteristics.

8. A positioning and early warning system for building public electrical equipment according to claim 7, characterized in that: The termination conditions include condition one and condition two, as detailed below: Condition 1: At least one particle has a fitness value of not less than 0 after the end of this round of iteration; Condition 2: After this round of iterations, at least one particle, in addition to satisfying Condition 1, also has a mutation index that is less than the mutation index threshold. The specific expression for the particle's mutation index is as follows: In the formula, The variation index, The power characteristics of public power lines during the current monitoring period are represented by a 1×M matrix. To input the electrical characteristics of each device represented by particles satisfying condition one into the encoder of the encoder-decoder model, the resulting line electrical characteristics are also represented as a 1×M matrix. express and Euclidean distance, express The Frobenius norm; In the formula, The electrical characteristics of each device, represented by the initial particle, are expressed as a matrix of form Q×k. To satisfy condition one, the electrical characteristics of each device represented by the particles are also represented by a matrix of the form Q×k. express and Euclidean distance, express The Frobenius norm; In the formula, To input the reference line power characteristics into the decoder of the encoder-decoder model, the resulting power characteristics of each device are also represented as a matrix of form Q×k. The logic for obtaining the reference line power characteristics is as follows: the most recent historical monitoring time period when the public power line is in a normal state and is closest to the current monitoring time period is taken as the reference time period, and the line power characteristics of the public power line during the reference time period are taken as the reference line power characteristics. express and Euclidean distance, express The Frobenius norm; In the formula, , and All are preset weighting coefficients, used to represent , and The weight it occupies in the calculation of the variability index satisfies as well as ; Among them, the optimal particle is the particle with the minimum mutation index among the particles that satisfy condition one and whose mutation index is less than the mutation index threshold.

9. A positioning and early warning system for public electrical equipment in buildings according to claim 7, characterized in that, The logic for perturbing the current solution based on the annealing temperature to generate a neighborhood solution with a fitness value not less than 0 is as follows: In each round of annealing optimization, for each element value of the current solution, a corresponding small perturbation interval is defined. A random value is randomly generated from the small perturbation interval, and its product with the annealing temperature in this round of annealing optimization is calculated as the perturbation value. The perturbation value is summed with the corresponding element value of the current solution to obtain the element value of the neighborhood solution. This process generates a neighborhood solution, and then the fitness value of the neighborhood solution is determined. If the fitness value of the neighborhood solution is not less than 0, the mutation index is calculated in the next step. Otherwise, a random value is randomly generated again to regenerate the neighborhood solution until the fitness value of the generated neighborhood solution is not less than 0. The upper limit of the small perturbation interval is half the width of the small perturbation interval, and the lower limit is the product of half the width of the small perturbation interval and -1. The half width of the small perturbation interval is less than the perturbation factor.

10. A positioning and early warning system for public electrical equipment in buildings according to claim 7, characterized in that, The specific mathematical expression for the difference index is as follows: In the formula, Let y be the optimal adjustment value for the electrical characteristic of the x-th public utility device on the public utility power line. The difference index of the x-th public electrical equipment on the public power line; The logic for locating public electrical equipment with faults based on the difference index is as follows: if the difference index of a public electrical equipment is higher than the difference index threshold, then this public electrical equipment is defined as having a fault.

Images