A station energy consumption abnormality detection method, system, device and medium

Abstract

The present invention relates to a method, system, device and medium for detecting abnormal energy consumption of a station, and the method comprises: automatically collecting data of a station, and loading it into a pre-deployed data processing environment and handing it over to at least one processing node for processing to form a data set; obtaining the characteristics of at least one dimension of the energy consumption data based on the data set, combining the distribution law and correlation of each influencing factor with the energy consumption data, and outputting the data analysis result; constructing a feature set reflecting the energy consumption influencing factors according to the data analysis result; inputting the feature set reflecting the energy consumption influencing factors into a pre-built and trained energy consumption abnormality detection model, so that the energy consumption abnormality detection model performs single or batch abnormality detection on the energy consumption data according to the input feature set, and outputs the abnormality detection result. The present invention realizes comprehensive monitoring and intelligent management of station energy consumption through a series of steps such as data integration, in-depth analysis, feature engineering, model training and abnormality detection.

Description

Technical Field

[0001] The present invention relates to the technical field of energy consumption monitoring, and in particular to a method, system, equipment and medium for detecting abnormal energy consumption at a station. Background Art

[0002] With the acceleration of industrialization and urbanization, global energy demand is showing a trend of continuous growth. In this context, the rational use and conservation of energy has become increasingly prominent, triggering widespread attention from all walks of life to energy management efficiency. Energy consumption anomaly detection technology has therefore emerged and has rapidly developed into an important means to improve energy management and optimize energy efficiency.

[0003] The core of energy consumption anomaly detection is to accurately capture and identify energy consumption changes that exceed the normal fluctuation range through real-time monitoring and in-depth analysis of energy consumption data. The timeliness and accuracy of this technology enable energy managers to respond quickly, adjust and optimize energy use strategies, thereby improving energy efficiency, reducing operating costs, ensuring safe operation of equipment, and promoting the realization of sustainable development goals.

[0004] However, in the current energy consumption anomaly identification technology, statistical methods, machine learning algorithms or deep learning technologies are mainly used. Although these technologies have performed well in some aspects, each of them has limitations that cannot be ignored. Specifically, when processing energy consumption data, traditional statistical methods can often only detect those obvious abnormal values ​​that deviate from the normal range, but they are powerless for more complex and hidden abnormal patterns. On the other hand, although machine learning algorithms show strong flexibility in adapting to data changes, they need to update the model regularly to adapt to the ever-changing working environment. At the same time, in a dynamically changing working environment, machine learning algorithms also require frequent manual adjustment of parameters, which not only increases the cost of manual intervention, but also affects the real-time and accuracy of detection to a certain extent. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a station energy consumption anomaly detection method, system, device and medium, which solves the technical problems that the existing energy consumption anomaly detection technology cannot fully identify complex abnormal patterns and requires high-frequency manual adjustment of parameters.

[0007] (II) Technical solution

[0008] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0009] In a first aspect, an embodiment of the present invention provides a method for detecting abnormal energy consumption in a station, comprising:

[0010] Automatically collect at least one type of data of the station and load it into a pre-deployed data processing environment for processing by at least one processing node to form a data set;

[0011] Based on the data set, the characteristics of at least one dimension of energy consumption data are obtained, and the distribution law and correlation between various influencing factors and energy consumption data obtained by statistical methods are combined to output data analysis results;

[0012] According to the data analysis results, feature engineering is performed on different types of data in the data set to construct a feature set that reflects the factors affecting energy consumption;

[0013] The feature set reflecting the factors affecting energy consumption is input into a pre-built and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set and outputs the anomaly detection results.

[0014] Optionally, at least one type of data of the station is automatically collected and loaded into a pre-deployed data processing environment for processing by at least one processing node to form a data set including:

[0015] Selecting and collecting at least one type of data of a station in response to a user request or automatically;

[0016] Integrate all types of collected data into a unified data platform to obtain an integrated data set;

[0017] Dividing the integrated data set into at least one subset and loading it onto at least one independent processing node in the pre-deployed data processing environment according to a subset-to-node mapping rule;

[0018] On each independent processing node, the following processing is performed:

[0019] Perform logical checks on subsets of the integrated data to identify and flag missing values, duplicate values, and malformed data;

[0020] For missing values ​​in the integrated data subset, at least one imputation method is used based on the distribution characteristics of the data, the proportion of missing values, or the analysis requirements, and the marked missing values ​​are imputed according to the selected imputation method;

[0021] By checking the timestamp or date field in the integrated data subset, the time series data and the granularity and frequency of the statistical time series are identified and analyzed, and the integrated data subsets are aligned according to the granularity and frequency of the statistical time series;

[0022] According to the node-to-subset mapping rules and / or analysis requirements, the integrated data subsets processed by the independent processing nodes are reconstructed into the data set to be analyzed;

[0023] in,

[0024] The data types include: station basic information, equipment basic information, environmental data, equipment operation data, energy consumption data, operation data, passenger flow data and number of tags;

[0025] The mapping rules from subsets to nodes are:

[0026] ;

[0027] In the formula, f is a mapping function, D i For the i subsets, n is the number of subsets to be divided, N j For the j processing nodes, argmax k It means to find the independent variable that makes the function reach the maximum value. α is the data volume weight coefficient, β is the data feature weight coefficient, | D i ∣ is a data subset D i The amount of data, is the sum of the data volumes of all data subsets, For data subset D i With processing nodes N k The similarity of characteristics between k Is the index variable used to traverse all processing nodes N k , and find the processing node index that maximizes the expression value j ;

[0028] The mapping rules from nodes to subsets are:

[0029] ;

[0030] In the formula, g is the remapping function, After remapping i subsets, N j For the j processing nodes, γ is the node processing capacity weight coefficient, δ is the weight coefficient of matching degree between data subset and node characteristics, For processing nodes Nk The capacity size, is the sum of the capacities of all processing nodes, is the remapped data subset With processing nodes N k The matching degree between the characteristics.

[0031] Optionally, based on the data set, characteristics of at least one dimension of the energy consumption data are obtained, and combined with the distribution law and correlation of each influencing factor and the energy consumption data obtained by a statistical method, the output data analysis results include:

[0032] Extract features of at least one dimension from the energy consumption data in the data set, and construct initial feature data including trend features, seasonal features, periodic features, energy consumption pattern features, nonlinear features, and complex correlation features of the energy consumption data;

[0033] A recursive feature elimination algorithm is used to iteratively train the preset base model and gradually reduce the number of features in the initial feature set according to the importance score or coefficient of the base model until the preset conditions are met, thereby obtaining the feature data after the initial simplification.

[0034] According to the characteristics of energy consumption data and prediction targets, a feature selection model is selected, and the remaining features in the feature data after the initial simplification are trained and screened using the feature selection model to select the features that contribute most to energy consumption prediction or analysis, thereby obtaining the feature data after further simplification;

[0035] Through box plot analysis, identify and compare the energy consumption levels corresponding to different types of data in the data set;

[0036] According to the distribution characteristics of the box plot, determine the distribution of energy consumption data under the influence of different types of data;

[0037] Use correlation analysis or regression analysis to quantify the strength and direction of the monotonic relationship between different types of data and energy consumption data;

[0038] The data analysis results are formed based on the further simplified characteristic data, the distribution of energy consumption data under the influence of different types of data, and the strength and direction of the monotonic relationship between different types of data and energy consumption data.

[0039] Optionally, according to the data analysis results, feature engineering processing is performed on different types of data in the data set to construct a feature set reflecting the factors affecting energy consumption, including:

[0040] The non-numeric data in the dataset, including station type, station layout, whether there is a property area, season type, weather type, whether there is a special event, and energy consumption label, are transformed into numerical results that can be used for model input by labeling.

[0041] The continuous numerical data in the data set, including station building area, weather temperature, weather humidity, and passenger volume, are standardized by calculating the mean value and standard deviation of the data set to obtain the standardized result;

[0042] The specific continuous numerical variables in the data set, including the operation time and various energy consumption data, are normalized and transformed. The data is scaled to a specific range by determining the minimum and maximum values ​​in the data set to obtain the normalized results.

[0043] Using the results of numeration, standardization and normalization to train multiple base learners, and obtain outputs of multiple base learners;

[0044] A soft voting mechanism is introduced to convert the output of each base learner into a probability distribution, and weighted average fusion is performed through the weights assigned according to the performance and confidence of each base learner to form a feature set that reflects the factors affecting energy consumption.

[0045] Optionally, a feature set reflecting the energy consumption influencing factors is input into a pre-built and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set. Before outputting the anomaly detection result, the following is also included:

[0046] Determine the input items including station type, station layout method, station area, number of station floors, whether there is a property area, season type, weather type, weather temperature, weather humidity, average operating power of each device, average operating time of each device, failure data of each device, power consumption of each system, operating time, whether there are special events, number of train departures and passenger flow data as the input of the model;

[0047] Determine that the output of the model is energy consumption label data;

[0048] The acquired historical data is integrated with a fixed time interval as the time granularity to form a training data structure table containing time, input data and corresponding energy consumption labels, and the training data structure table is divided into a training set and a validation set;

[0049] According to the determined model input and output, select the composite local anomaly factor to build the initial energy consumption anomaly detection model, configure the model parameters, and ensure that the model can receive the corresponding input items and output energy consumption label data;

[0050] Use the data in the training set to train the initial energy consumption anomaly detection model. By adjusting the model parameters and optimizing the algorithm, the model can learn the mapping relationship between the input data and the energy consumption label. During the training process, the validation set is used to evaluate the model regularly.

[0051] After at least one iterative training and optimization until the preset standard is reached, an energy consumption anomaly detection model is obtained.

[0052] Optionally, according to the determined model input and output, a local anomaly factor is selected to construct an initial energy consumption anomaly detection model, and model parameters are configured to ensure that the model can receive corresponding input items and output energy consumption label data including:

[0053] Construct a basic model of local abnormal factors;

[0054] According to the basic model of local anomaly factors, the kernel method is introduced to obtain the local anomaly factor model with the kernel method;

[0055] By adopting the strategy of ensemble learning, multiple local anomaly factor models with kernel methods are trained, and the output results are fused to obtain a composite local anomaly factor.

[0056] According to the anomaly detection goal, the parameters of the composite local anomaly factor are dynamically adjusted to ensure that the model can receive the corresponding input items and output energy consumption label data;

[0057] in,

[0058] The basic local anomaly factor model is:

[0059] ;

[0060] In the formula, l ( x ) represents the sample point x The local anomaly factor is used to measure the sample point x The relative density to the neighbors, N k ( x ) represents the sample point x of k nearest neighbors, o yes N k ( x ) is an element in the set, represented by x A nearest neighbor of k represents the number of nearest neighbors, lrd ( x ) represents the sample point x The local reachable density of

[0061] The local anomaly factor model introduced by the kernel method is:

[0062] ;

[0063] In the formula, K ( x , y ) represents the Gaussian kernel function, which is used to map the original data into a high-dimensional space. x , y are the two data points being compared, σ represents the bandwidth parameter of the Gaussian kernel, l k ( x ) represents the sample points calculated in high-dimensional space x The local anomaly factor, lrd k ( x ) represents the sample points calculated in high-dimensional space x The local reachable density of express x All k Nearest Neighbor o and x The kernel function value between K ( x , o ) and the local reachable density in high-dimensional space The sum of the products of

[0064] The local anomaly factor model of ensemble learning is:

[0065] ;

[0066] in, l e ( x ) represents the sample point after ensemble learning x The local anomaly factor, M Indicates the number of models trained ，l km ( x ) indicates the m Sample points calculated by the local anomaly factor model using the kernel method x The local anomaly factor.

[0067] Optionally, a feature set reflecting the factors affecting energy consumption is input into a pre-built and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set, and the output anomaly detection results include:

[0068] Input the feature set reflecting the factors affecting energy consumption into the trained energy consumption anomaly detection model to perform anomaly detection, and obtain the normal or abnormal category judgment results of the current station energy consumption for a single or batch;

[0069] At each test, the model input data and the corresponding energy consumption label are stored in the training data structure table;

[0070] At every set period, the detection accuracy of the current model is calculated. If the accuracy is lower than the set lower limit, the retraining process is started, the model is retrained using all the accumulated data, and the model parameters are updated;

[0071] Using visualization tools, the energy consumption data, anomaly detection results, and cause analysis conclusions are presented to the user, and when an energy consumption anomaly is detected, the user is notified through at least one communication channel;

[0072] Prepare and regularly release to users energy consumption analysis reports that include regular summaries of abnormal situations, cause analysis results, and improvement measures taken and their effects.

[0073] In a second aspect, an embodiment of the present invention provides a station energy consumption abnormality detection system, comprising:

[0074] The collection and processing module is used to automatically collect at least one type of data of the station and load it into a pre-deployed data processing environment for processing by at least one processing node to form a data set.

[0075] The data analysis module is used to obtain the characteristics of at least one dimension of energy consumption data based on the data set, combine the distribution law and correlation of various influencing factors and energy consumption data obtained by statistical methods, and output the data analysis results.

[0076] The feature set output module is used to perform feature engineering processing on different types of data in the data set according to the data analysis results, and construct a feature set that reflects the factors affecting energy consumption.

[0077] The anomaly detection module is used to input the feature set reflecting the factors affecting energy consumption into a pre-built and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set and outputs the anomaly detection results.

[0078] In a third aspect, an embodiment of the present invention provides a station energy consumption abnormality detection device, including:

[0079] At least one database; and a memory communicatively connected to the at least one database; wherein the memory stores instructions executable by the at least one database, and the instructions are executed by the at least one database so that the at least one database can execute the station energy consumption abnormality detection method as described above.

[0080] In a fourth aspect, an embodiment of the present invention provides a computer-readable medium having computer executable instructions stored thereon, and when the executable instructions are executed by a processor, the station energy consumption anomaly detection method as described above is implemented.

[0081] (III) Beneficial effects

[0082] The beneficial effects of the present invention are:

[0083] First, the present invention collects and processes various types of data from stations to form a comprehensive and rich data set. The benefit of this step is that it can integrate data from different sources, break down information silos, and provide a solid data foundation for subsequent energy consumption analysis.

[0084] Secondly, based on the constructed data set, this method deeply analyzes the multi-dimensional characteristics of energy consumption data, and combines statistical methods to reveal the distribution patterns and correlations between various influencing factors and energy consumption data. This in-depth analysis not only helps to understand the complex behavior of energy consumption, but also accurately identifies the key factors affecting energy consumption, providing a scientific basis for energy management.

[0085] Furthermore, based on the data analysis results, feature engineering is performed on the data to construct a feature set that reflects the factors affecting energy consumption. The cleverness of this step is that it can transform the original data into more representative and discriminative features, thereby improving the performance and accuracy of the energy consumption anomaly detection model.

[0086] In addition, by building an energy consumption anomaly detection model and training it with historical data and corresponding energy consumption category data, this method enables the model to learn the inherent laws and abnormal patterns of energy consumption behavior. This self-learning ability means that the model can continuously adapt to changes in the working environment and maintain sensitivity and accuracy to energy consumption anomalies.

[0087] Finally, the processed feature set is input into the trained energy consumption anomaly detection model to perform anomaly detection and visualize the detection results. The practicality of this step lies in that it can not only detect and locate energy consumption anomalies in a timely manner, but also present the detection results in an intuitive way, which is convenient for managers to quickly understand and respond, thereby greatly improving the intelligence level and response speed of energy management.

[0088] Therefore, the present invention realizes comprehensive monitoring and intelligent management of station energy consumption through a series of steps such as data integration, in-depth analysis, feature engineering, model training and visual display. This can not only improve energy utilization efficiency and achieve energy conservation and emission reduction goals, but also reduce operating costs and improve the overall benefits of station operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 A schematic diagram of a flow chart of a method for detecting abnormal energy consumption in a station provided by an embodiment of the present invention;

[0090] Figure 2 A schematic diagram of a specific flow chart of step S1 of the method for detecting abnormal energy consumption at a station provided by an embodiment of the present invention;

[0091] Figure 3 A schematic diagram of a specific flow chart of step S2 of the method for detecting abnormal energy consumption in a station provided by an embodiment of the present invention;

[0092] Figure 4 A box plot of the station energy consumption anomaly detection method provided by an embodiment of the present invention;

[0093] Figure 5 A schematic diagram of a specific flow chart of step S3 of the method for detecting abnormal energy consumption in a station provided by an embodiment of the present invention;

[0094] Figure 6 A schematic diagram of a specific process before step S4 of the station energy consumption abnormality detection method provided by an embodiment of the present invention;

[0095] Figure 7 A schematic diagram of a neural network model for detecting abnormal energy consumption in a station according to an embodiment of the present invention;

[0096] Figure 8 A schematic diagram of a specific flow chart of step S4 of the method for detecting abnormal energy consumption in a station provided by an embodiment of the present invention;

[0097] Fig. 9 A schematic diagram of the model self-optimization process of step S4 of the station energy consumption anomaly detection method provided by an embodiment of the present invention;

[0098] Fig.10 A schematic diagram of the overall process of the station energy consumption anomaly detection method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0099] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0100] like Figure 1As shown, a station energy consumption anomaly detection method proposed in an embodiment of the present invention includes: automatically collecting at least one type of data of the station, and loading it into a pre-deployed data processing environment for processing by at least one processing node to form a data set; obtaining features of at least one dimension of the energy consumption data based on the data set, combining the distribution law and correlation of each influencing factor and the energy consumption data obtained by a statistical method, and outputting a data analysis result; according to the data analysis result, performing feature engineering processing on different types of data in the data set to construct a feature set reflecting energy consumption influencing factors; inputting the feature set reflecting energy consumption influencing factors into a pre-constructed and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set, and outputs the anomaly detection result.

[0101] First, the present invention collects and processes various types of data from stations to form a comprehensive and rich data set. The benefit of this step is that it can integrate data from different sources, break down information silos, and provide a solid data foundation for subsequent energy consumption analysis.

[0102] Secondly, based on the constructed data set, this method deeply analyzes the multi-dimensional characteristics of energy consumption data, and combines statistical methods to reveal the distribution patterns and correlations between various influencing factors and energy consumption data. This in-depth analysis not only helps to understand the complex behavior of energy consumption, but also accurately identifies the key factors affecting energy consumption, providing a scientific basis for energy management.

[0103] Furthermore, based on the data analysis results, feature engineering is performed on the data to construct a feature set that reflects the factors affecting energy consumption. The cleverness of this step is that it can transform the original data into more representative and discriminative features, thereby improving the performance and accuracy of the energy consumption anomaly detection model.

[0104] In addition, by building an energy consumption anomaly detection model and training it with historical data and corresponding energy consumption category data, this method enables the model to learn the inherent laws and abnormal patterns of energy consumption behavior. This self-learning ability means that the model can continuously adapt to changes in the working environment and maintain sensitivity and accuracy to energy consumption anomalies.

[0105] Finally, the processed feature set is input into the trained energy consumption anomaly detection model to perform anomaly detection and visualize the detection results. The practicality of this step lies in that it can not only detect and locate energy consumption anomalies in a timely manner, but also present the detection results in an intuitive way, which is convenient for managers to quickly understand and respond, thereby greatly improving the intelligence level and response speed of energy management.

[0106] Therefore, the present invention realizes comprehensive monitoring and intelligent management of station energy consumption through a series of steps such as data integration, in-depth analysis, feature engineering, model training and visual display. This can not only improve energy utilization efficiency and achieve energy conservation and emission reduction goals, but also reduce operating costs and improve the overall benefits of station operations.

[0107] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0108] Specifically, an embodiment of the present invention provides a method for detecting abnormal energy consumption of a station, which includes:

[0109] S1. Automatically collect at least one type of data of the station and load it into a pre-deployed data processing environment for processing by at least one processing node to form a data set.

[0110] Furthermore, if Figure 2 As shown, step S1 includes:

[0111] S11. In response to user demand or automatically selecting and collecting at least one type of data of the station, integrating the collected various types of data into a unified data platform to obtain an integrated data set.

[0112] The data that needs to be collected in the embodiment of the present invention include: station basic information, equipment basic information, environmental data, equipment operation data, energy consumption data, operational data, passenger flow data, and label data. The corresponding information can be collected in the following ways: Use the configuration tool to obtain station basic information and equipment basic information data through a one-time configuration. Deploy various sensors (such as environmental monitoring equipment, equipment status monitoring equipment, power monitoring equipment, etc.) on key facilities (such as traction substations, stations, depots, trains) to collect environmental data, equipment operation data, energy consumption data, etc. in real time. In addition, obtain operational data and passenger flow data from the data platform, and classify the total power consumption data of the station.

[0113] Specifically, the basic information of the station includes: station type, station layout, station area, number of station floors, and whether there is a property area. Among them, the station type is classified data, which is divided into general station, transfer station, and hub station; the station layout is classified data, which is divided into underground station, ground station, and elevated station; the station area is numerical data, and the unit is m 2 ; The number of floors of the station is numerical data and dimensionless; whether there is a property area is categorical data, which is divided into yes and no.

[0114] The basic equipment information includes: the rated power and rated frequency of each equipment. The rated power is a numerical data in kw; the rated frequency is a numerical data in Hz.

[0115] Environmental data includes: season type, weather type, weather temperature, and weather humidity. Among them, season type is categorical data, which is divided into ventilation season and air conditioning season. The ventilation season is the season when the water system is not turned on, and the air conditioning season is the season when the water system is turned on; weather type is categorical data, which is divided into sunny day, cloudy day, rainy day, snowy day (can also be divided into more detailed); weather temperature is numerical data, the unit is ℃; weather humidity is numerical data, the unit is %.

[0116] Equipment operation data includes: real-time power, real-time frequency, fault status, and operation status of each device (at least one of the real-time power and real-time frequency is sufficient). Real-time power is numerical data in kw; real-time frequency is numerical data in Hz; fault status is classified data, divided into normal and fault; operation status is classified data, divided into running and stopped.

[0117] Energy consumption data include: escalator system power consumption, lighting system power consumption, large system power consumption, small system power consumption, water system power consumption, VRV system power consumption, other system power consumption, and total station power consumption. The eight energy consumption data are numerical data in kw·h.

[0118] Operation data include: operation time, whether there are special events, and number of train departure pairs. Among them, operation time is numerical data, the unit is h; whether there are special events is categorical data, divided into yes and no; number of train departure pairs is numerical data, the unit is pair.

[0119] Passenger flow data includes: passenger volume. Among them, passenger volume is numerical data, the unit is person-time.

[0120] Energy consumption label data includes: station power consumption label. Labels are classified data, which are divided into normal, escalator system power consumption is too high, escalator equipment problems lead to high power consumption, lighting system power consumption is too high, lighting equipment problems lead to high power consumption, large system power consumption is too high, combined air conditioning equipment problems lead to high power consumption, small system power consumption is too high, small system fan equipment problems lead to high power consumption, water system power consumption is too high, chiller equipment problems lead to high power consumption, VRV system power consumption is too high, VRV equipment problems lead to high power consumption, weather temperature is too high lead to high power consumption, weather humidity is too high lead to high power consumption, weather temperature is too low lead to high power consumption, passenger flow is too large lead to high power consumption, operation time is too long lead to high power consumption, large passenger flow events lead to high power consumption, and additional train operation logs lead to high power consumption. Label classification can be divided into more detailed categories according to actual conditions.

[0121] S12: Divide the integrated data set into at least one subset, and load it into at least one independent processing node in the pre-deployed data processing environment according to a subset-to-node mapping rule.

[0122] S13. On each independent processing node, perform the following preprocessing: perform a logical check on the integrated data subset to identify and mark missing values, duplicate values, and data with incorrect formats; for the missing values ​​of the integrated data subset, use at least one filling method based on the distribution characteristics of the data, the proportion of missing values, or analysis requirements, and fill in the marked missing values ​​based on the selected filling method; identify and analyze the time series data and the granularity and frequency of the statistical time series by checking the timestamp or date field in the integrated data subset, and align the various integrated data subsets based on the granularity and frequency of the statistical time series.

[0123] Preferably, the following filling methods can be used: Mean filling: For data with approximately normal distribution, the controller calculates the mean of non-missing values ​​and fills the missing values ​​with the mean. Median filling: For data with extreme values ​​or skewed distribution, the controller uses the median for filling. Mode filling: For categorical data or data with obvious central tendency, the controller uses the mode for filling. Machine learning-based prediction filling: For complex data sets, the controller trains machine learning models (such as regression models, decision trees, etc.) to predict missing values ​​and uses the prediction results for filling.

[0124] S14. According to the mapping rule from node to subset and / or analysis requirements, the integrated data subset processed by the independent processing node is reconstructed into a data set to be analyzed.

[0125] In a specific embodiment, the mapping rule from subset to node is:

[0126] ;

[0127] In the formula, f is a mapping function, D i For the i subsets, n is the number of subsets to be divided, N j For the j processing nodes, argmax k It means to find the independent variable that makes the function reach the maximum value. α is the data volume weight coefficient, β is the data feature weight coefficient, | D i ∣ is a data subset D i The amount of data, is the sum of the data volumes of all data subsets, For data subset D i With processing nodes N k The similarity of characteristics between k Is the index variable used to traverse all processing nodes N k , and find the processing node index that maximizes the expression value j。 in other words, k It does not directly represent a specific parameter here, but serves as an intermediate variable to implement the function of the argmax function, that is, to find the processing node index that maximizes the value of a given expression. j .

[0128] A clearer explanation is: j = argmax(...) is actually performing a selection operation to find the expression that maximizes the expression. j value (i.e. the index of the processing node). Once this selection operation is completed, a specific j value, which can be used to identify N j (i.e., specific processing nodes), similarly, in the mapping rules from nodes to subsets i It also plays the same role.

[0129] The mapping rules from nodes to subsets are:

[0130] ;

[0131] In the formula, g is the remapping function, After remapping i subsets, N j For the j processing nodes, γ is the node processing capacity weight coefficient, δ is the weight coefficient of matching degree between data subset and node characteristics, For processing nodes N k The capacity size, is the sum of the capacities of all processing nodes, is the remapped data subset With processing nodes N k The matching degree between the characteristics.

[0132] S2. Based on the data set, the characteristics of at least one dimension of the energy consumption data are obtained, and the distribution law and correlation between each influencing factor and the energy consumption data obtained by statistical methods are combined to output the data analysis results.

[0133] Furthermore, if Figure 3 As shown, step S2 includes:

[0134] S21. Extract at least one dimension of features from the energy consumption data in the data set, and construct initial feature data including trend features, seasonal features, periodic features, energy consumption pattern features, nonlinear features, and complex correlation features of the energy consumption data. The following is a detailed explanation of each feature:

[0135] Trend characteristics: Use linear regression or time series decomposition technology to extract long-term trends and short-term fluctuations in energy consumption data to form trend characteristics.

[0136] Seasonal characteristics: Through seasonal decomposition or Fourier transform, seasonal changes in energy consumption data, such as annual, quarterly or monthly periodic changes, are identified and seasonal characteristics are constructed.

[0137] Periodic features: Use autocorrelation function or periodogram method to detect periodic components in energy consumption data, such as daily or weekly repetitive patterns, and generate periodic features.

[0138] Energy consumption pattern characteristics: Based on the device type, usage scenario and time period, the energy consumption data is clustered and analyzed to identify different energy consumption patterns and form corresponding characteristics.

[0139] Nonlinear features: Use nonlinear models such as kernel methods, neural networks or support vector machines to explore nonlinear relationships in energy consumption data and construct nonlinear features.

[0140] Complex correlation features: Through association rule mining, causal relationship analysis or graph theory methods, the complex correlation between energy consumption data and other variables is revealed to form complex correlation features.

[0141] S22. A recursive feature elimination algorithm is used to iteratively train a preset base model and gradually reduce the number of features in the initial feature set according to the importance score or coefficient of the base model until a preset condition is met, thereby obtaining feature data after initial simplification.

[0142] Specifically, first, a base model is selected for feature importance assessment. Common base models include linear regression, logistic regression, decision tree, etc. The selection of the base model should be based on the characteristics of the energy consumption data and the prediction target. Next, the base model is trained using the initial feature set, and the model's importance score or coefficient is obtained. These scores or coefficients reflect the contribution of each feature to the model's prediction results. Furthermore, the features are sorted according to the model's importance score or coefficient. Then, the features with the lowest score or the smallest coefficient, that is, the least important features, are removed. Finally, the feature set is updated, the removed features are removed, and the base model is retrained using the new feature set. Repeat this process, removing one of the least important features in each iteration, until the preset conditions are met.

[0143] It should be clear that the preset condition can be a lower limit on the number of features (such as retaining a certain number of features), a threshold of model performance (such as reaching a certain level of accuracy, R² score, etc.), or an upper limit on the number of iterations. When any of the conditions is met, the recursive feature elimination process stops.

[0144] S23. According to the characteristics of the energy consumption data and the prediction target, a feature selection model is selected, and the feature selection model is used to train and screen the remaining features in the feature data after the initial simplification, so as to select the features that contribute most to the energy consumption prediction or analysis, and obtain the feature data after further simplification.

[0145] Specifically, according to the characteristics of the energy consumption data and the prediction target, select a suitable feature selection model. These include linear models (such as linear regression, logistic regression), tree models (such as decision trees, random forests), support vector machines, etc. These models can provide feature importance scores or weights to measure the contribution of features to the prediction results. Next, use the streamlined feature set as input to train the selected feature selection model. During the training process, the model learns the relationship between the features and the target variables, and gives the feature importance scores or weights based on this relationship. Evaluate the contribution of each feature to energy consumption prediction or analysis based on the importance scores or weights provided by the feature selection model. Generally, the higher the score or weight, the greater the contribution of the feature to the prediction result. Based on the evaluation results, select the features that contribute most to energy consumption prediction or analysis. This can be done by setting a threshold or selecting a front. N The selected features will constitute the final feature set.

[0146] S24. Through box plot analysis, identify and compare the energy consumption levels corresponding to different types of data in the data set.

[0147] S25. According to the distribution characteristics of the box plot, determine the distribution of energy consumption data under the influence of different types of data.

[0148] S26. Use correlation analysis or regression analysis to quantitatively analyze the strength and direction of the monotonic relationship between different types of data and energy consumption data.

[0149] S27. According to the further simplified characteristic data, the distribution of the energy consumption data under the influence of different types of data, and the strength and direction of the monotonic relationship between different types of data and the energy consumption data, a data analysis result is formed.

[0150] In a specific embodiment, statistical methods (such as box plots, correlation coefficients, etc.) are used to analyze the trend changes of various influencing factors and energy consumption, and preliminarily determine the impact of various factors on energy consumption.

[0151] like Figure 4 As shown in the figure, the box plot can analyze the distribution of station power consumption under the influence of different characteristic factors. For example, the box plot can be used to analyze the impact of different station types on the total power consumption of the station. Six days were randomly selected to analyze the distribution of power consumption under different station types, and the box plot was used to represent it. From the figure, it can be seen that in general, the power consumption of hub stations is greater than that of transfer stations, and greater than that of general stations.

[0152] The correlation coefficient can be used to calculate the strength and direction of the monotonic relationship between two variables. For example, the correlation coefficient is used to analyze the strength and direction of the correlation between the total power consumption of the station and the weather temperature. Referring to Table 1, the Spearman correlation coefficient is used to calculate the correlation between the station power consumption and the temperature of each station every day in 2023, and the average correlation coefficient is 0.644652. The correlation results of 10 stations are shown in the following table. From the correlation coefficient results, it can be seen that the temperature has a strong positive correlation with the station power consumption.

[0153] Table 1 Correlation between station power consumption and temperature

[0154]

[0155] Spearman's Rank Correlation Coefficient, calculated as:

[0156] ;

[0157] in, ρ is the Spearman correlation coefficient; d i Represents the rank difference of each pair of observations in the two variables; k is the number of samples. The coefficient range is [-1, 1]. A value less than 0 indicates a negative correlation, a value greater than 0 indicates a positive correlation, and a value equal to 0 indicates that there is no correlation. The closer the absolute value of the correlation coefficient is to 0, the weaker the correlation; the closer the absolute value is to 1, the stronger the correlation.

[0158] S3. Based on the data analysis results, feature engineering is performed on different types of data in the data set to construct a feature set that reflects the factors affecting energy consumption. Specifically, feature engineering can be performed on different data in a numerical, standardized, and normalized manner.

[0159] Furthermore, if Figure 5 As shown, step S3 includes:

[0160] S31. The non-numeric data in the data set, including station type, station layout method, whether there is a property area, season type, weather type, whether there is a special event, and energy consumption label, are transformed into numerical results that can be used for model input using labeling.

[0161] Non-numeric data (such as categorical variables) need to be converted into numerical form for model input. Station type, station layout, whether there is a property area, season type, weather type, whether there is a special event, and energy consumption label are categorical data and need to be converted into numerical values ​​that can be used for model input. For example, in the station type data, general stations are converted into numerical values ​​1, transfer stations are converted into numerical values ​​2, and hub stations are converted into numerical values ​​3.

[0162] S32. The continuous numerical data including station building area, weather temperature, weather humidity, and passenger volume in the data set are standardized by calculating the mean value and standard deviation of the data set to obtain a standardized result.

[0163] By standardizing, the data is converted into a standard normal distribution, that is, a distribution with a mean of 0 and a standard deviation of 1, to reduce the impact of different features on the model. For example, station building area, weather temperature, weather humidity, and passenger volume can be standardized.

[0164] ;

[0165] in, x is the original data value; μ is the mean value of the data set; σ is the standard deviation of the data set. In the present invention, the μ , σ It is recorded as a parameter and used for data conversion in the subsequent model training and model detection stages.

[0166] S33. Perform normalization transformation on specific continuous numerical variables including operation time and various energy consumption data in the data set, and scale the data to a specific range by determining the minimum and maximum values ​​in the data set to obtain a normalized result.

[0167] By normalizing the data, we can scale it to the range of [0, 1] to reduce the impact of different features on the model. For example, the operation time and energy consumption data can be normalized.

[0168] ;

[0169] in, x is the original data value, min ( x ) is the minimum value in the data set, max ( x ) is the maximum value in the data set. In the present invention, the value of each feature is min ( x ), max ( x ) is recorded as a parameter for data conversion in the subsequent model training and model detection stages.

[0170] S34. Train multiple base learners using the results of digitization, standardization, and normalization to obtain outputs of the multiple base learners.

[0171] S35. Introduce a soft voting mechanism to convert the output of each base learner into a probability distribution, and perform weighted average fusion based on the weights assigned according to the performance and confidence of each base learner to form a feature set that reflects the factors affecting energy consumption.

[0172] Soft voting is a method that converts the outputs of multiple classifiers into probability distributions and fuses them by weighted averaging. Unlike hard voting, soft voting takes into account the confidence of each classifier and is therefore more flexible and accurate.

[0173] For the output of each base learner, convert it to a probability distribution. If the base learner itself outputs a probability distribution (such as logistic regression, naive Bayes, etc.), use it directly; if the output is a category label (such as decision tree, support vector machine, etc.), convert it to a probability distribution through certain methods (such as Platt scaling, one-to-many conversion, etc.).

[0174] According to the performance and confidence of each base learner, a weight is assigned to it. Then, the probability outputs of all base learners are weighted averaged according to the weights to obtain the final fusion result. This fusion result can be regarded as a feature set of energy consumption influencing factors that combines the advantages of multiple base learners.

[0175] S4. Input the feature set reflecting the factors affecting energy consumption into a pre-built and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set, and outputs the anomaly detection result.

[0176] Before step S4, the method further includes: constructing an energy consumption anomaly detection model, and training the energy consumption anomaly detection model using the acquired historical data and the corresponding energy consumption category data. Figure 6 As shown, including:

[0177] F41. Determine the input items including station type, station layout, station area, number of station floors, whether there is a property area, season type, weather type, weather temperature, weather humidity, average operating power of each device, average operating time of each device, failure data of each device, power consumption of each system, operating time, whether there are special events, number of train departures and passenger flow data as the input of the model.

[0178] refer to Figure 7 , it is determined that the model has 39 inputs and 1 output.

[0179] The inputs include: station type, station layout method, station area, number of station floors, whether there is a property area, season type, weather type, weather temperature, weather humidity, average operating power of escalators, average operating time of escalators, escalator failures, average operating power of lighting equipment, average operating time of lighting equipment, lighting equipment failures, average operating power of combined air conditioners, average operating time of combined air conditioners, combined air conditioner failures, average operating power of small system fans, average operating time of small system fans, small system fan failures, average operating power of chillers, average operating time of chillers, chiller failures, average operating power of VRV equipment, average operating time of VRV equipment, VRV equipment failures, escalator system power consumption, lighting system power consumption, large system power consumption, small system power consumption, water system power consumption, VRV system power consumption, other system power consumption, total station power consumption, operating hours, whether there are special events, number of train departures, and passenger flow data. Among the input data, the average operating power, average operating time and fault data of each device are obtained through secondary calculation, and the remaining data can be directly obtained through data collection.

[0180] At the same time, the output is determined to be energy consumption label data.

[0181] Take escalators as an example. There are multiple escalators in a station. Through data collection, the rated power, rated frequency, real-time power in seconds, real-time frequency in seconds, real-time fault status in seconds, and real-time operation status of each escalator are obtained. The time period of these data should be unified. For example:

[0182] (1) There are two ways to calculate the average operating power of the equipment:

[0183] The first is to directly obtain real-time power conditions:

[0184] ;

[0185] in, is the average operating power of the escalator in the station during the time period [0,T], n is the number of escalators in the station, This is the station i Taipower escalator t Real-time operating power at all times.

[0186] The second method is to only obtain the real-time frequency situation, which can be obtained by calculating the following formula: t The real-time power at the moment, and then the average operating power of the escalator is obtained through formula (1).

[0187] ;

[0188] in, This is the station t Moment i The real-time operating frequency of Taipower escalators. This is the first i The rated frequency of Taipower escalator, This is the first i Rated power of Taipower escalator.

[0189] (2) Average equipment operating time method:

[0190] ;

[0191] in, is the average running time of the escalator at the station in the time period [0,T], n is the number of escalators in the station, This is the station i Taipower escalator t The real-time running status at the moment, 1 means running, 0 means stopped.

[0192] (3) Equipment failure data:

[0193] ;

[0194] in, is the escalator failure data of the station in the time period [0,T], n is the number of escalators in the station; This is the station i Taipower escalator t Fault status at the moment, 1 means fault, 0 means normal.

[0195] F42. Determine that the output of the model is energy consumption label data.

[0196] F43. Integrate the acquired historical data with a fixed time interval as the time granularity to form a training data structure table containing time, input data and corresponding energy consumption labels, and divide the training data structure table into a training set and a validation set.

[0197] F44. According to the determined model input and output, select the composite local anomaly factor to build the initial energy consumption anomaly detection model, configure the model parameters, and ensure that the model can receive the corresponding input items and output the energy consumption label data.

[0198] In a specific embodiment, step F44 includes: constructing a local abnormal factor basic model;

[0199] According to the basic model of local anomaly factors, the kernel method is introduced to obtain the local anomaly factor model of the kernel method; by adopting the strategy of ensemble learning, multiple local anomaly factor models of the kernel method are trained, and the output results are fused to obtain a composite local anomaly factor; according to the goal of anomaly detection, the parameters of the composite local anomaly factor are dynamically adjusted to ensure that the model can receive the corresponding input items and output energy consumption label data.

[0200] in,

[0201] The basic local anomaly factor model is:

[0202] ;

[0203] In the formula, l ( x ) represents the sample point x The local anomaly factor is used to measure the sample point x The relative density to the neighbors, N k ( x ) represents the sample point x of k nearest neighbors, o yes N k ( x ) is an element in the set, represented by x A nearest neighbor of k represents the number of nearest neighbors, lrd ( x ) represents the sample point x The local reachable density of .

[0204] The local anomaly factor model introduced by the kernel method is:

[0205] ;

[0206] In the formula, K ( x ,y ) represents the Gaussian kernel function, which is used to map the original data into a high-dimensional space and then calculate the local anomaly factor of each sample point in the high-dimensional space; x , y are the two data points being compared, σ represents the bandwidth parameter of the Gaussian kernel, l k ( x ) represents the sample points calculated in high-dimensional space x The local anomaly factor, lrd k ( x ) represents the sample points calculated in high-dimensional space x The local reachable density of express x All k Nearest Neighbor o and x The kernel function value between K ( x , o ) and the local reachable density in high-dimensional space The sum of the products of .

[0207] The local anomaly factor model of ensemble learning is:

[0208] ;

[0209] in, l e ( x ) represents the sample point after ensemble learning x The local anomaly factor, M Indicates the number of models trained ，l km ( x ) indicates the m Sample points calculated by the local anomaly factor model using the kernel method x The local anomaly factor.

[0210] F45. Use the data in the training set to train the initial energy consumption anomaly detection model. By adjusting the model parameters and optimizing the algorithm, the model can learn the mapping relationship between the input data and the energy consumption label. At the same time, during the training process, the validation set is used regularly to evaluate the model.

[0211] F46. After at least one iteration of training and optimization until the preset standard is reached, an energy consumption anomaly detection model is obtained.

[0212] In another specific embodiment, the historical data is integrated with a fixed time interval (eg, 24 hours) as the time granularity and organized as shown in Table 2.

[0213] Table 2 Training data structure table

[0214]

[0215] The energy consumption label is used as the model output, and the remaining data is used as the model input. The energy consumption anomaly detection model is trained using historical data to obtain the target model.

[0216] Furthermore, if Figure 8 As shown, step S4 includes:

[0217] S41. Input the feature set reflecting the factors affecting energy consumption into the trained energy consumption anomaly detection model to perform anomaly detection, and obtain the normal or abnormal category judgment results of the current station energy consumption in single or batch.

[0218] S42, at each test, the model input data and the corresponding energy consumption label are stored in the training data structure table. The energy consumption anomaly detection personnel configure the power consumption category of the observed data through the page, and store the category as the energy consumption label in the training data structure table.

[0219] S43. Calculate the detection accuracy of the current model at every set time period. If the accuracy is lower than the set lower limit, start the retraining process, use all the accumulated data to retrain the model, and update the model parameters.

[0220] Next, obtain the basic information data, equipment basic information data, environmental data, equipment operation data, energy consumption data, operation data, and passenger flow data of the current station, convert the data into a format acceptable to the model according to the method described in the model construction phase, and input it into the trained neural network model to obtain the normal or abnormal category of the current station's power consumption. During each detection, the model input data is stored in the training data structure table, and the energy consumption anomaly detection personnel configure the power consumption category of this observation data through the page, and store it in the training data structure table as an energy consumption label.

[0221] Also refer to Fig. 9 This method includes an automatic training function, which calculates the detection accuracy at regular intervals (such as every Monday at 0:00). If the accuracy is lower than the set lower limit, the model is retrained using all accumulated data and the model parameters are replaced. The accuracy calculation formula is as follows:

[0222] ;

[0223] in, n is the number of samples in the observation period, is the category judgment result, is the actual result of the category, It is used to determine whether the two are the same, if they are the same, it is 1, otherwise it is 0.

[0224] S44. Using visualization tools, the energy consumption data, abnormality detection results, and cause analysis conclusions are presented to the user, and when energy consumption abnormalities are detected, the user is notified through at least one communication channel.

[0225] S45. Prepare and regularly release to users energy consumption analysis reports that include regular summaries of abnormal conditions, cause analysis results, and improvement measures taken and their effects.

[0226] Afterwards, data dashboards, GIS maps, heat maps and other tools are used to present energy consumption data, abnormal detection results, and cause analysis conclusions in an intuitive and easy-to-understand way, so that decision makers can quickly understand them. A real-time alarm system is set up to notify relevant personnel through emails, text messages, APP push, etc. when energy consumption anomalies are detected to ensure timely response. An energy consumption analysis report is compiled to summarize abnormal situations, cause analysis results, improvement measures and effects, providing a basis for continuous improvement.

[0227] In addition, an embodiment of the present invention provides a method for detecting abnormal energy consumption in a station, comprising:

[0228] The collection and processing module is used to automatically collect at least one type of data of the station and load it into a pre-deployed data processing environment for processing by at least one processing node to form a data set.

[0229] The data analysis module is used to obtain the characteristics of at least one dimension of energy consumption data based on the data set, combine the distribution law and correlation of various influencing factors and energy consumption data obtained by statistical methods, and output the data analysis results.

[0230] The feature set output module is used to perform feature engineering processing on different types of data in the data set according to the data analysis results, and construct a feature set that reflects the factors affecting energy consumption.

[0231] The anomaly detection module is used to input the feature set reflecting the factors affecting energy consumption into a pre-built and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set and outputs the anomaly detection results.

[0232] At the same time, an embodiment of the present invention provides an energy consumption anomaly detection device, including: at least one database; and a memory communicatively connected to the at least one database; wherein the memory stores instructions that can be executed by at least one database, the instructions are executed by at least one database, and at least one database is capable of executing the station energy consumption anomaly detection method as described above.

[0233] The device contains one or more databases, which constitute the core data storage and processing unit of the device. The database is not only used to store massive energy consumption data, but also responsible for performing various data query, analysis and processing operations. Through these operations, the database can extract key information from the energy consumption data and provide data support for subsequent anomaly detection.

[0234] The memory connected to the database is another important part of the device. The memory stores a set of instructions that can be executed by the database. These instructions are specially designed for the task of energy consumption anomaly detection. When the database needs to perform energy consumption anomaly detection, it reads the corresponding instructions from the memory and operates according to the instructions.

[0235] Specifically, after the instructions in the memory are executed by the database, they can guide the database to execute a series of preset station energy consumption anomaly detection methods. These methods may include but are not limited to data preprocessing, feature extraction, anomaly identification and other steps. Through the execution of these steps, the database can accurately detect anomalies in energy consumption data and report them to users in a timely manner.

[0236] Furthermore, an embodiment of the present invention provides a computer-readable medium having computer-executable instructions stored thereon, and when the executable instructions are executed by a processor, the station energy consumption abnormality detection method as described above is implemented.

[0237] The embodiment of the present invention further provides a computer-readable medium. This medium has a special storage function, on which computer executable instructions are stored. When these instructions are read and executed by the corresponding processor, they will guide the processor to accurately implement the station energy consumption anomaly detection method described in detail above according to the preset process. In this way, the present invention not only provides an efficient software solution, but also greatly facilitates the practical application and deployment of the station energy consumption anomaly detection method, thereby helping to more widely promote and optimize energy efficiency.

[0238] In summary, the embodiments of the present invention provide a method, system, device and medium for detecting abnormal energy consumption in a station, such as Fig.10 As shown, first, the present invention provides a solid foundation for subsequent analysis by collecting diversified data, including historical energy consumption data, environmental data, operational data, equipment operation data, and passenger flow data. Then, the data is cleaned and preprocessed, such as processing missing values ​​and unifying timestamps, to ensure the integrity and consistency of the data set.

[0239] In the data analysis stage, the present invention deeply explores the inherent characteristics of energy consumption data, such as trends, periodicity, seasonality, etc., and explores the distribution patterns and correlations between various influencing factors and energy consumption data. Based on these analyses, a feature set reflecting the factors affecting energy consumption, such as time characteristics, weather characteristics, operating status characteristics, etc., is constructed, laying a solid foundation for feature engineering.

[0240] Subsequently, the present invention uses an advanced convolutional neural network model, takes various factors that affect energy consumption as model input, and outputs classification results of normal or abnormal energy consumption. The model is trained by collecting historical data and energy consumption category data to obtain an energy consumption abnormality target detection model with high accuracy and high efficiency.

[0241] In practical applications, the model can detect anomalies in energy consumption data and output clear classification results. In addition, the present invention also visualizes the anomaly detection results to facilitate users to intuitively understand the energy consumption status.

[0242] It is worth mentioning that the energy consumption anomaly detection model provided by the present invention is universal and can be widely used in energy consumption monitoring of various places, and can accurately give the cause of the anomaly. This not only provides a scientific basis for energy management in places such as stations, but also helps to achieve the goals of energy conservation and emission reduction and improve energy utilization efficiency.

[0243] Compared with the prior art, the present invention shows significant advantages in many aspects. First, it can quickly adapt to changes in the working environment, efficiently and accurately identify the links and causes of energy consumption anomalies in the operating system, and provide strong support for decision-making optimization and energy management. Secondly, by deeply exploring the relationship between various influencing factors and energy consumption values, the present invention solves the limitations of traditional statistical methods in dealing with complex situations. Finally, the energy consumption anomaly detection system constructed by the present invention has the functions of automated data collection and model updating, which reduces the cost of manual intervention, improves operational efficiency and saves costs.

[0244] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0245] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.

[0246] It should be noted that in the claims, any reference numerals placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by the same hardware. The use of the words first, second, third, etc., is for convenience of expression only and does not indicate any order. These words may be understood as part of the component name.

[0247] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0248] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0249] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.

Claims

1. A method for detecting abnormal energy consumption at a station, characterized in that: include: Automatically collect at least one type of data of the station and load it into a pre-deployed data processing environment for processing by at least one processing node to form a data set; Based on the data set, the characteristics of at least one dimension of energy consumption data are obtained, and the distribution law and correlation between various influencing factors and energy consumption data obtained by statistical methods are combined to output data analysis results; According to the data analysis results, feature engineering is performed on different types of data in the data set to construct a feature set that reflects the factors affecting energy consumption; Input a feature set reflecting the factors affecting energy consumption into a pre-built and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set and outputs anomaly detection results; Among them, automatically collecting at least one type of data of the station and loading it into a pre-deployed data processing environment for processing by at least one processing node to form a data set includes: Selecting and collecting at least one type of data of a station in response to a user request or automatically; Integrate all types of collected data into a unified data platform to obtain an integrated data set; Dividing the integrated data set into at least one subset and loading it onto at least one independent processing node in the pre-deployed data processing environment according to a subset-to-node mapping rule; On each independent processing node, the following processing is performed: Perform logical checks on subsets of the integrated data to identify and flag missing values, duplicate values, and malformed data; For missing values ​​in the integrated data subset, at least one imputation method is used based on the distribution characteristics of the data, the proportion of missing values, or the analysis requirements, and the marked missing values ​​are imputed according to the selected imputation method; By checking the timestamp or date field in the integrated data subset, the time series data and the granularity and frequency of the statistical time series are identified and analyzed, and the integrated data subsets are aligned according to the granularity and frequency of the statistical time series; According to the node-to-subset mapping rules and / or analysis requirements, the integrated data subsets processed by the independent processing nodes are reconstructed into the data set to be analyzed; in, The data types include: station basic information, equipment basic information, environmental data, equipment operation data, energy consumption data, operation data, passenger flow data and number of tags; The mapping rules from subsets to nodes are: ； In the formula, f is a mapping function, D i For the i subsets, n is the number of subsets to be divided, N j For the j processing nodes, argmax k It means to find the independent variable that makes the function reach the maximum value. α is the data volume weight coefficient, β is the data feature weight coefficient, | D i ∣ is a data subset D i The amount of data, is the sum of the data volumes of all data subsets, For data subset D i With processing nodes N k The similarity of characteristics between k Is the index variable used to traverse all processing nodes N k , and find the processing node index that maximizes the expression value j ; The mapping rules from nodes to subsets are: ； In the formula, g is the remapping function, After remapping i subsets, N j For the j processing nodes, γ is the node processing capacity weight coefficient, δ is the weight coefficient of matching degree between data subset and node characteristics, For processing nodes N k The capacity size, is the sum of the capacities of all processing nodes, is the remapped data subset With processing nodes N k The matching degree between the characteristics.

2. The station energy consumption abnormality detection method according to claim 1, characterized in that: Based on the data set, the characteristics of at least one dimension of energy consumption data are obtained. Combined with the distribution law and correlation of various influencing factors and energy consumption data obtained by statistical methods, the output data analysis results include: Extract features of at least one dimension from the energy consumption data in the data set, and construct initial feature data including trend features, seasonal features, periodic features, energy consumption pattern features, nonlinear features, and complex correlation features of the energy consumption data; A recursive feature elimination algorithm is used to iteratively train the preset base model and gradually reduce the number of features in the initial feature set according to the importance score or coefficient of the base model until the preset conditions are met, thereby obtaining the feature data after the initial simplification. According to the characteristics of energy consumption data and prediction targets, a feature selection model is selected, and the remaining features in the feature data after the initial simplification are trained and screened using the feature selection model to select the features that contribute most to energy consumption prediction or analysis, thereby obtaining the feature data after further simplification; Through box plot analysis, identify and compare the energy consumption levels corresponding to different types of data in the data set; According to the distribution characteristics of the box plot, determine the distribution of energy consumption data under the influence of different types of data; Use correlation analysis or regression analysis to quantify the strength and direction of the monotonic relationship between different types of data and energy consumption data; The data analysis results are formed based on the further simplified characteristic data, the distribution of energy consumption data under the influence of different types of data, and the strength and direction of the monotonic relationship between different types of data and energy consumption data.

3. The station energy consumption abnormality detection method according to claim 1, characterized in that: According to the data analysis results, feature engineering is performed on different types of data in the data set to construct feature sets that reflect the factors affecting energy consumption, including: The non-numeric data in the dataset, including station type, station layout, whether there is a property area, season type, weather type, whether there is a special event, and energy consumption label, are transformed into numerical results that can be used for model input by labeling. The continuous numerical data in the data set, including station building area, weather temperature, weather humidity, and passenger volume, are standardized by calculating the mean value and standard deviation of the data set to obtain the standardized result; The specific continuous numerical variables in the data set, including the operation time and various energy consumption data, are normalized and transformed. The data is scaled to a specific range by determining the minimum and maximum values ​​in the data set to obtain the normalized results. Using the results of numeration, standardization and normalization to train multiple base learners, and obtain outputs of multiple base learners; A soft voting mechanism is introduced to convert the output of each base learner into a probability distribution, and weighted average fusion is performed through the weights assigned according to the performance and confidence of each base learner to form a feature set that reflects the factors affecting energy consumption.

4. The station energy consumption abnormality detection method according to claim 1, characterized in that: The feature set reflecting the factors affecting energy consumption is input into the pre-built and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set. Before outputting the anomaly detection result, it also includes: Determine the input items including station type, station layout method, station area, number of station floors, whether there is a property area, season type, weather type, weather temperature, weather humidity, average operating power of each device, average operating time of each device, failure data of each device, power consumption of each system, operating time, whether there are special events, number of train departures and passenger flow data as the input of the model; Determine that the output of the model is energy consumption label data; The acquired historical data is integrated with a fixed time interval as the time granularity to form a training data structure table containing time, input data and corresponding energy consumption labels, and the training data structure table is divided into a training set and a validation set; According to the determined model input and output, select the composite local anomaly factor to build the initial energy consumption anomaly detection model, configure the model parameters, and ensure that the model can receive the corresponding input items and output energy consumption label data; Use the data in the training set to train the initial energy consumption anomaly detection model. By adjusting the model parameters and optimizing the algorithm, the model can learn the mapping relationship between the input data and the energy consumption label. During the training process, the validation set is used to evaluate the model regularly. After at least one iterative training and optimization until the preset standard is reached, an energy consumption anomaly detection model is obtained.

5. The station energy consumption abnormality detection method according to claim 4, characterized in that: According to the determined model input and output, select local anomaly factors to build the initial energy consumption anomaly detection model, configure model parameters, and ensure that the model can receive the corresponding input items and output energy consumption label data including: Construct a basic model of local abnormal factors; According to the basic model of local anomaly factors, the kernel method is introduced to obtain the local anomaly factor model with the kernel method; By adopting the strategy of ensemble learning, multiple local anomaly factor models with kernel methods are trained, and the output results are fused to obtain a composite local anomaly factor. According to the anomaly detection goal, the parameters of the composite local anomaly factor are dynamically adjusted to ensure that the model can receive the corresponding input items and output energy consumption label data; in, The basic local anomaly factor model is: ； In the formula, l ( x ) represents the sample point x The local anomaly factor is used to measure the sample point x The relative density to the neighbors, N k ( x ) represents the sample point x of k nearest neighbors, o yes N k ( x ) is an element in the set, represented by x A nearest neighbor of k represents the number of nearest neighbors, lrd ( x ) represents the sample point x The local reachable density of The local anomaly factor model introduced by the kernel method is: ； In the formula, K ( x , y ) represents the Gaussian kernel function, which is used to map the original data into a high-dimensional space. x , y are the two data points being compared, σ represents the bandwidth parameter of the Gaussian kernel, l k ( x ) represents the sample points calculated in high-dimensional space x The local anomaly factor, lrd k ( x ) represents the sample points calculated in high-dimensional space x The local reachable density of express x All k Nearest Neighbor o and x The kernel function value between K ( x , o ) and the local reachable density in high-dimensional space The sum of the products of The local anomaly factor model of ensemble learning is: ； in, l e ( x ) represents the sample point after ensemble learning x The local anomaly factor, M Indicates the number of models trained ，l km ( x ) indicates the m Sample points calculated by the local anomaly factor model using the kernel method x The local anomaly factor.

6. The station energy consumption abnormality detection method according to claim 4 or 5, characterized in that: The feature set reflecting the factors affecting energy consumption is input into the pre-built and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set, and the output anomaly detection results include: Input the feature set reflecting the factors affecting energy consumption into the trained energy consumption anomaly detection model to perform anomaly detection, and obtain the normal or abnormal category judgment results of the current station energy consumption for a single or batch; At each test, the model input data and the corresponding energy consumption label are stored in the training data structure table; At every set period, the detection accuracy of the current model is calculated. If the accuracy is lower than the set lower limit, the retraining process is started, the model is retrained using all the accumulated data, and the model parameters are updated; Using visualization tools, the energy consumption data, anomaly detection results, and cause analysis conclusions are presented to the user, and when an energy consumption anomaly is detected, the user is notified through at least one communication channel; Prepare and regularly release to users energy consumption analysis reports that include regular summaries of abnormal situations, cause analysis results, and improvement measures taken and their effects.

7. A station energy consumption abnormality detection system, characterized in that: include: A collection and processing module, used to automatically collect at least one type of data of the station, and load it into a pre-deployed data processing environment for processing by at least one processing node to form a data set; A data analysis module is used to obtain the characteristics of at least one dimension of energy consumption data based on the data set, combine the distribution law and correlation of various influencing factors and energy consumption data obtained by statistical methods, and output the data analysis results; The feature set output module is used to perform feature engineering processing on different types of data in the data set according to the data analysis results, and construct a feature set that reflects the factors affecting energy consumption; The anomaly detection module is used to input the feature set reflecting the energy consumption influencing factors into the pre-built and trained energy consumption anomaly detection model, so that the energy consumption anomaly detection model performs single or batch anomaly detection on the energy consumption data according to the input feature set and outputs the anomaly detection result; Among them, automatically collecting at least one type of data of the station and loading it into a pre-deployed data processing environment for processing by at least one processing node to form a data set includes: Selecting and collecting at least one type of data of a station in response to a user request or automatically; Integrate all types of collected data into a unified data platform to obtain an integrated data set; Dividing the integrated data set into at least one subset and loading it onto at least one independent processing node in the pre-deployed data processing environment according to a subset-to-node mapping rule; On each independent processing node, the following processing is performed: Perform logical checks on subsets of the integrated data to identify and flag missing values, duplicate values, and malformed data; For missing values ​​in the integrated data subset, at least one imputation method is used based on the distribution characteristics of the data, the proportion of missing values, or the analysis requirements, and the marked missing values ​​are imputed according to the selected imputation method; By checking the timestamp or date field in the integrated data subset, the time series data and the granularity and frequency of the statistical time series are identified and analyzed, and the integrated data subsets are aligned according to the granularity and frequency of the statistical time series; According to the node-to-subset mapping rules and / or analysis requirements, the integrated data subsets processed by the independent processing nodes are reconstructed into the data set to be analyzed; in, The data types include: station basic information, equipment basic information, environmental data, equipment operation data, energy consumption data, operation data, passenger flow data and number of tags; The mapping rules from subsets to nodes are: ； In the formula, f is a mapping function, D i For the i subsets, n is the number of subsets to be divided, N j For the j processing nodes, argmax k It means to find the independent variable that makes the function reach the maximum value. α is the data volume weight coefficient, β is the data feature weight coefficient, | D i ∣ is a data subset D i The amount of data, is the sum of the data volumes of all data subsets, For data subset D i With processing nodes N k The similarity of characteristics between k Is the index variable used to traverse all processing nodes N k , and find the processing node index that maximizes the expression value j ; The mapping rules from nodes to subsets are: ； In the formula, g is the remapping function, After remapping i subsets, N j For the j processing nodes, γ is the node processing capacity weight coefficient, δ is the weight coefficient of matching degree between data subset and node characteristics, For processing nodes N k The capacity size, is the sum of the capacities of all processing nodes, is the remapped data subset With processing nodes N k The matching degree between the characteristics.

8. A station energy consumption abnormality detection device, characterized in that: include: at least one database; And a memory communicatively connected to the at least one database; wherein the memory stores instructions executable by the at least one database, and the instructions are executed by the at least one database so that the at least one database can execute the station energy consumption abnormality detection method as described in any one of claims 1-6.

9. A computer-readable medium having computer-executable instructions stored thereon, characterized in that: When the executable instructions are executed by the processor, the station energy consumption abnormality detection method as described in any one of claims 1-6 is implemented.

Images