A smart electricity monitoring system
By using a low-cost intelligent electricity monitoring system that combines device-side, edge-side, and cloud platform modules, and utilizing machine learning and time series algorithms for abnormal electricity behavior warnings and equipment health assessments, the system solves the problem of limited functionality in existing systems and achieves multi-dimensional electricity management and enterprise electricity optimization.
Patent Information
- Application Number
- CN202310318398.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-28
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-03-28
AI Technical Summary
Existing electricity monitoring systems lack intelligent decision support, cannot meet the actual business needs of enterprises for electricity management, and have large hardware investments and limited functions, making it difficult to achieve multi-dimensional management of equipment electricity consumption.
A low-cost, lightweight intelligent electricity monitoring system is adopted, including equipment-side modules, edge-side modules, and cloud platform-side modules. It uses industrial intelligence technology to conduct early warning of abnormal electricity consumption behavior, assess equipment health degradation, and comprehensively optimize operational decisions. It also utilizes machine learning and time series algorithms for power prediction and health risk assessment.
It has achieved safe protection, risk warning and operational decision support for equipment power consumption, reduced the probability of power safety accidents, improved the level of enterprise power management, and reduced power consumption and carbon emissions.
Smart Images

Figure CN116365707B_ABST
Abstract
Description
Technical Field
[0001] This application relates to an intelligent electricity consumption monitoring system, applicable to the field of power control technology. Background Technology
[0002] Power management is a crucial way for enterprises to save energy, reduce consumption, and enhance competitiveness. However, conventional power management is asset-heavy, requires significant hardware investment, and primarily relies on output threshold alarms, lacking intelligent decision support. Power management is also highly dependent on human intervention. Therefore, how to improve power management and reduce electricity costs through effective intelligent and systematic methods has become a key factor for enterprise development.
[0003] Currently, intelligent monitoring systems for equipment power consumption focus more on hardware system design. This includes hardware design for data acquisition nodes to collect power data, hardware design for mobile terminals to monitor power consumption, layered wireless network architectures and sampling control units to accurately measure power consumption data, and intelligent power acquisition master and slave stations to monitor various relevant parameters of power users in real time. There are also power consumption monitoring systems that lean towards traditional numerical monitoring, such as judging critical anomalies in power signals, displaying the operating current and voltage of electrical equipment in real time, and issuing alarms when current exceeds limits. Hardware-oriented systems focus more on how to collect equipment power consumption information; traditional numerical monitoring systems offer simple real-time data monitoring, statistics, and threshold alarm functions. Existing equipment power consumption monitoring systems have significant limitations and cannot meet the actual business needs of enterprises for power management.
[0004] CN216697044U discloses a power consumption monitoring and management system, which includes a current sensor, a voltage sensor, a first controller, an IoT gateway, a timing module, and a management platform. The current sensor, voltage sensor, and first controller are all connected to electrical equipment. The current sensor, voltage sensor, and timing module are also connected to the first controller, which is further connected to the IoT gateway. The IoT gateway is connected to the management platform. Staff can view the operating time and power consumption of electrical equipment in real time through the management platform, improving the efficiency of power consumption statistics. However, the management platform in this patent can only view the operating time and power consumption of electrical equipment; its function is limited and it cannot achieve intelligent monitoring of power consumption. Summary of the Invention
[0005] This application provides a low-cost, lightweight intelligent electricity monitoring system based on industrial intelligent technology, which can realize multi-dimensional management of electricity safety protection, equipment risk warning, asset energy efficiency optimization and operational decision support.
[0006] This application relates to an intelligent electricity consumption monitoring system, including a device-side module, an edge-side module, and a cloud platform-side module. The edge-side module collects current and voltage signals from the devices in the device-side module and transmits the collected signals to the cloud platform-side module. The cloud platform-side module includes a data layer unit, a model layer unit, and an application layer unit. The data layer unit accesses the data from the edge-side module, the model layer unit establishes and applies an intelligent model for electricity management, and the application layer unit performs energy safety protection, asset energy efficiency optimization, health risk warning, and operational decision support based on the calculation results of the intelligent model.
[0007] The cloud platform module may include an abnormal electricity consumption behavior early warning model, and the algorithm of the abnormal electricity consumption behavior early warning model includes the following steps:
[0008] (1.1) Obtain current and voltage data related to the energy consumption of the equipment;
[0009] (1.2) The equipment is classified according to its operating conditions, which are divided into high-fluctuation operating conditions and stable operating conditions;
[0010] (1.3) By segmenting the operating conditions, the interference of operating condition fluctuations on the prediction is reduced, and statistics related to equipment energy consumption are extracted;
[0011] (1.4) Establish a power prediction model and use time series or machine learning algorithms to predict future power;
[0012] (1.5) Establish a power upper and lower limit prediction model. Based on the statistical distribution of historical data, train the historical power upper and lower limit prediction model through quantile regression, and predict the future power upper and lower limits at the same time.
[0013] (1.6) By comparing the predicted power value with the predicted upper and lower limits of power, predict whether the future power will exceed the limit; by comparing the difference between the predicted value of the power prediction model and the real-time power measurement value, calculate the distribution deviation of the residual distribution and the training data, and determine whether there is an abnormality in energy consumption.
[0014] (1.7) Output power over-limit and abnormal detection model results, and early warning of over-limit and abnormal states.
[0015] In step (1.3), segmenting the operating conditions may include segmenting the signal by calculating the autocorrelation coefficient of the relevant variables of the operating conditions and finding the periodicity of the signal; segmenting the operating conditions may also include segmenting the operating conditions by the amplitude of the operating condition variables, and using rules to divide the operating condition signal into steady-state operating conditions and transient operating conditions.
[0016] The cloud platform module may further include an asset health risk early warning model, the algorithm of which includes the following steps:
[0017] (2.1) Obtain current and voltage data related to the energy consumption of the equipment;
[0018] (2.2) Determine the similarity of equipment operating conditions;
[0019] (2.3) Identify similar operating conditions of equipment and establish basic conditions for comparing the energy consumption of equipment under the same operating conditions;
[0020] (2.4) Extract the feature vector of equipment degradation characteristics from the variable characteristics of equipment signals;
[0021] (2.5) Establish a healthy baseline model and transform the features into a healthy baseline model that can characterize the training history distribution;
[0022] (2.6) In the prediction process, repeat the above steps (2.1)-(2.3) to obtain the feature vector of the prediction process, calculate the distance between the feature vector and the baseline model to quantify the degree of conformity of the prediction feature relative to the baseline distribution, and calculate the equipment health score.
[0023] (2.7) Accumulate fault samples, label and calibrate health scores according to different time intervals from the fault, and calculate the health risk level of the equipment.
[0024] In step (2.4), the following features can be extracted simultaneously:
[0025] Time-domain statistical characteristics include statistics that can describe the signal morphology of segmented similar operating condition intervals;
[0026] Temporal morphological similarity features include features that can describe the degree of similarity in signal shape during similar operating cycles;
[0027] Frequency domain energy distribution characteristics, including indices that can describe the frequency domain energy distribution of a signal;
[0028] Time-frequency domain energy distribution characteristics include image feature matrices that can describe the changes in energy distribution across different frequency bands within a similar operating cycle.
[0029] In step (2.2), the autocorrelation coefficient can be used to quantify the similarity of equipment operating conditions.
[0030] The cloud platform module may also include an operational decision-making comprehensive optimization model, which can benchmark energy consumption against health status and integrate power prediction values and health assessment values of similar equipment. By comparing the energy consumption prediction values and health values of equipment with the global baseline of similar equipment, the model can derive equipment energy consumption ranking, equipment energy consumption trend compared with the group, equipment health ranking, and equipment health trend compared with the group.
[0031] The integrated optimization model for operational decision-making can be used to quantify equipment performance and support operational optimization. It can customize equipment performance indicators based on the comparison of equipment energy consumption, health assessment values, and groups. The equipment-side module includes high-current wired devices and low-current mobile devices. The intelligent electricity monitoring system also includes a user-side module, through which electricity monitoring information can be obtained.
[0032] This application analyzes the current and voltage information of industrial equipment to provide early warnings of abnormal equipment power consumption behavior, assess the degree of equipment health degradation, and comprehensively optimize operational decisions. The system is easy to implement and use, reduces the probability of serious electrical safety accidents, ensures the reliability of equipment assets, effectively reduces enterprise power consumption and carbon emissions, and improves the level of enterprise power management. Attached Figure Description
[0033] Figure 1 This diagram shows the intelligent electricity monitoring system of this application.
[0034] Figure 2 This paper presents an algorithm flowchart for the abnormal electricity consumption behavior early warning model of this application.
[0035] Figure 3 This paper presents a flowchart illustrating the algorithm of the asset health risk early warning model.
[0036] Figure 4 This paper presents a flowchart illustrating the algorithm of the integrated optimization model for operational decision-making. Detailed Implementation
[0037] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be arbitrarily combined with each other.
[0038] like Figure 1 As shown, the intelligent electricity monitoring system according to this application includes an equipment-side module, an edge-side module, a cloud platform-side module, and a user-side module. This application utilizes industrial intelligent data analysis technology to perform real-time voltage and current data analysis on electrical equipment to provide early warnings of abnormal equipment power consumption behavior, assess the degree of equipment health degradation, and comprehensively optimize operational decisions, thereby improving the enterprise's electricity management level.
[0039] The equipment-side modules mainly include high-current wired devices and low-current mobile devices. High-current wired devices include critical equipment on the enterprise's production line, whose drive mechanisms draw power from the workshop's electrical control cabinet. These devices are characterized by prolonged operation and high power supply current. Low-current mobile devices include auxiliary equipment such as enterprise testing and mobile tools, powered by 220V AC sockets. These devices are characterized by intermittent or on-demand operation and low power supply current. Those skilled in the art will understand that the terms "high current" and "low current" are relative and do not require specifying concrete numerical ranges.
[0040] The edge module is responsible for collecting current and voltage signals from the device-side modules via instruments, and transmitting the collected signals to the cloud platform through the network. This can be categorized into wired and wireless acquisition. For wired acquisition, current is collected by using a current transformer at the power supply line of the electrical control cabinet and then connecting it to the acquisition unit via a cable. Voltage is collected by connecting the voltage signal to the acquisition unit inside the electrical control cabinet via terminal wiring. For wireless acquisition, current is collected by directly plugging the device power plug into the socket, which directly collects the current and voltage signals and transmits the electrical signals to the gateway via a wireless network.
[0041] In the user-side module, users can obtain relevant information on power monitoring through the web, mobile app, and mobile mini-program, improving the ease of use of the system.
[0042] The cloud platform module comprises a data layer unit, a model layer unit, and an application layer unit. The data layer unit is responsible for accessing data from the edge modules and for managing and storing that data. The model layer unit establishes and applies intelligent models for energy management, including anomaly warning models for electricity consumption behavior, asset health risk warning models, comprehensive optimization models for operational decisions, and other statistical analysis models. The application layer unit implements functions such as energy safety protection, asset energy efficiency optimization, health risk warning, and operational decision support. Energy safety protection includes electricity safety and sudden change alarms; asset energy efficiency optimization includes energy consumption monitoring, over-limit alarms, and anomaly alarms; health risk warning includes health scoring and risk classification; and operational decision support includes peer benchmarking and decision support.
[0043] The algorithm flow of the abnormal electricity consumption behavior early warning model is as follows: Figure 2 As shown. Specifically, it includes the following steps:
[0044] (1) Data acquisition: Acquire data related to equipment energy consumption, such as current, voltage, and active power;
[0045] (2) Classify the equipment according to its operating conditions, which are divided into high-fluctuation operating conditions and stable operating conditions. The specific steps include:
[0046] (2.1) Identify relevant variables of the operating conditions, such as speed, current, position, etc.;
[0047] (2.2) Based on industry characteristics and expert experience, distinguish the volatility attributes of the work. For example, equipment in process industries such as chemical engineering is generally in a stable state; while industries that are greatly affected by the environment, such as wind power, or manufacturing processes such as stamping and machining, generally experience greater volatility. At the level of judgment rules, if the relevant variables of the working condition remain constant when the equipment is working, or the fluctuation is less than a certain percentage of the rated working amplitude, such as 10%, it is considered a stable working condition; all others are considered high-volatility working conditions.
[0048] (3) Preprocessing and feature extraction:
[0049] (3.1) Preprocessing: By segmenting the operating conditions, the interference of operating condition fluctuations on prediction is reduced, and the prediction accuracy is improved. The operating condition segmentation method is relatively flexible and requires customized algorithms according to specific business scenarios. The purpose is to distinguish between steady-state data with unchanged time-domain and frequency-domain characteristics and transient data with changing time-domain and frequency-domain characteristics.
[0050] (3.2) By identifying periodic patterns in the signal, for example, the autocorrelation coefficients of the variables related to the aforementioned operating conditions can be calculated. The autocorrelation sequence of a periodic signal has the same periodic characteristics as the original signal. If the signal is shifted along the time domain, the autocorrelation coefficient is maximized when the lag order is an integer multiple of the signal period. Therefore, calculating the autocorrelation coefficient can help verify periodicity and identify repetitive patterns in the signal.
[0051] For signal x t ={x1, x2, ..., x N}, t=1...N, its autocorrelation coefficient is:
[0052]
[0053] Where c k Let k be the autocorrelation coefficient with lag of k. The autocorrelation coefficient is the average value of the signal over a complete time sequence t = 1, ..., N. Based on this calculation method, an autocorrelation sequence composed of autocorrelation coefficients is obtained. Upper and lower limits for the autocorrelation coefficients can be set according to user needs. The lag order corresponding to autocorrelation coefficient values exceeding these limits represents the number of periods in which the potential recurring pattern appears, i.e., the signal period. The definition of these limits can be determined by the user based on experience and the original signal morphology to assess the reasonableness of the calculated periodicity. Generally, an absolute value of the autocorrelation coefficient greater than 0.6 indicates a strong correlation; the higher the value, the stronger the correlation.
[0054] Therefore, the signal can be periodically divided using a sliding window of length k, with each window segment constituting a complete period. Samples within and outside the window can be compared, meaning the data within a window is:
[0055]
[0056] (3.3) Segmentation by amplitude of operating condition variables: For operating condition signals, such as motor speed, steady-state and transient operating conditions are divided using rules, for example:
[0057] For a certain generator speed signal v t ={v1, v2, ..., v N}, t=1...N, its rotational speed at which it reaches rated generating power is V c Its upper limit is V l The working conditions can then be divided according to this rule:
[0058]
[0059] Where 1 represents steady state, 2 represents transient state, and 0 represents abnormal state unsuitable for modeling. Therefore, the operating condition data corresponding to the labeled samples are:
[0060] R i =x i ,where r i =1
[0061] (3.4) Feature extraction: Here, features refer to statistics related to the energy consumption of the equipment, including but not limited to peak values and average values of current or voltage.
[0062] (4) Establish a power prediction model and use time series prediction ARIMA, GARCH, etc., or machine learning and deep learning algorithms such as LSTM and RNN to predict future power.
[0063] (5) Establish a power upper and lower limit prediction model. Based on the statistical distribution of historical data, train the historical power upper and lower limit prediction model through quantile regression, and predict the future power upper and lower limits at the same time. Quantile regression is used to describe the upper and lower limits of historical data. The upper and lower limit quantiles can be set by the user. For example, the upper limit can correspond to τ = 0.95 and the lower limit τ = 0.05.
[0064] (6) Establish a power over-limit and / or anomaly detection model:
[0065] By comparing the predicted power value with the predicted upper and lower power limits, it is possible to predict whether future power will exceed the limits, i.e., "future energy consumption exceeding limits". By comparing the difference between the predicted value of the power prediction model and the real-time power measurement value, the distribution deviation of the residual distribution and the training data is calculated to determine whether there is any abnormality in energy consumption, i.e., "abnormal energy consumption behavior".
[0066] (7) Output power over-limit and abnormal detection model results, and early warning of over-limit and abnormal states.
[0067] Asset health risk early warning model
[0068] Under similar operating conditions and with the same work cycle, equipment energy consumption can typically reflect the degree of equipment health degradation. The algorithm flow of the asset health risk early warning model is as follows: Figure 3 As shown, it includes the following steps:
[0069] (1) Data acquisition: Collect data related to equipment energy consumption, such as current, voltage, and active power.
[0070] (2) Operating Condition Similarity Judgment: Operating condition similarity refers to whether the equipment's operating state is repetitive, that is, whether there is a repetitive pattern in the signal. For example, in mass production processes such as electronics manufacturing, equipment usually follows a fixed operating mode and operates periodically; in wind power, the equipment's operating state depends on the highly uncertain wind speed, so the degree of operating condition similarity is low. For the quantitative judgment of operating condition similarity, the autocorrelation coefficient mentioned above can be used:
[0071]
[0072] Among them, c k Let k be the autocorrelation coefficient with lag of k. This represents the average value of the signal over the complete time sequence t = 1, ..., N. As mentioned above, based on the upper and lower limits of the autocorrelation sequence, the lag order corresponding to autocorrelation coefficient values exceeding these limits represents the number of periods in which the potential recurring pattern occurs. The definition of this limit can be determined by the user based on experience and by judging the reasonableness of the calculated periodicity in conjunction with the original signal morphology. Generally, an absolute value of the autocorrelation coefficient greater than 0.6 indicates a strong correlation; the higher the value, the stronger the correlation.
[0073] (3) Identification of similar operating conditions: Unlike equipment energy consumption prediction, equipment health assessment requires finding similar operating conditions and establishing the basic conditions for comparing equipment energy consumption under the same operating conditions.
[0074] When the similarity of operating conditions is high, this process can be implemented automatically by the model. Optional methods include, but are not limited to: time series pattern recognition methods such as motif recognition, SAX, and change point detection; and time series similarity comparison models such as Dynamic Time Warping (DTW). When the similarity of operating conditions is low, this process can be implemented manually, i.e., by manually defining a certain time window or setting certain conditions. Once these conditions are triggered, the health assessment process begins.
[0075] (4) Equipment Deterioration Feature Extraction: Features related to equipment deterioration are reflected in the variable features from the electrical signals. The feature vector extracted under similar operating conditions is called the "energy fingerprint". The following features are also extracted:
[0076] Time-domain statistical characteristics: statistical quantities that can describe the signal shape of segmented similar operating condition intervals, including but not limited to: mean, root mean square, kurtosis, standard deviation, skewness, peak-to-peak value, maximum value, minimum value, zero-crossing rate, etc.
[0077] Temporal morphological similarity features: features that can describe the degree of similarity in signal shape during similar operating conditions, such as Euclidean distance, cosine distance, DTW dynamic time warping distance, etc.
[0078] Frequency domain energy distribution characteristics: Indicators that can describe the frequency domain energy distribution of a signal, including but not limited to: peak frequency harmonic replication, spectral entropy, etc.
[0079] Time-frequency domain energy distribution characteristics: Image feature matrices that can describe the changes in energy distribution in different frequency bands within a similar operating period, such as the spectrum after fast Fourier transform, the energy proportion of each layer of discrete wavelet transform, etc.
[0080] (5) Establish a healthy baseline model: Use statistical, unsupervised machine learning methods, or representation learning methods, such as principal component analysis, self-organizing map neural networks, or autoencoders, to transform features into a baseline model that can characterize the training history distribution.
[0081] (6) Calculate the equipment health score: In the prediction process, repeat steps (1)-(3) above to obtain the feature vector of the prediction process; calculate the distance between the feature vector and the baseline model mapping, or the "matching degree / fit degree", to quantify the degree of conformity of the predicted features relative to the baseline distribution. For example: using principal component analysis (PCA) modeling, the predicted feature vector The device health score is the feature vector y calculated from this feature vector and the actual measured data. t ={y1, ...,y N The distance between} can be represented by Mahalanobis distance, that is:
[0082]
[0083] Where ∑ is the covariance matrix of the PCA model, and D M This is the Mahalanobis distance.
[0084] (7) Calculate the health risk level of the equipment: When accumulating fault samples, the health score can be marked and calibrated according to different time intervals from the fault, such as 1 month ago, 1 week ago, 1 hour ago, etc., so that the health score can more accurately reflect the urgency of the fault that needs to be dealt with.
[0085] Operational Decision-Making Comprehensive Optimization Model
[0086] The algorithm flow of the comprehensive optimization model for operational decision-making is as follows: Figure 4 As shown.
[0087] (1) Benchmarking Energy Consumption and Health Status: Under the premise of data anonymization, the power prediction values and health assessment values of similar devices are integrated. By comparing the device energy consumption prediction value and health value of a user with the global baseline of similar devices, the device energy consumption ranking, the trend of device energy consumption benchmarking against the group, the device health ranking, and the trend of device health benchmarking against the group are obtained.
[0088] (2) Quantify equipment performance and support operation optimization: Based on the comparison of equipment energy consumption, health assessment value and group, users can customize equipment performance indicators. When the health status of the equipment is lower than the average level or the energy consumption is too high, the performance decline prompt will be given to guide users to switch the operating equipment in time or optimize the operating status of the equipment, extend the service life of the equipment and improve the operating efficiency.
[0089] Example
[0090] The system proposed in this application has been implemented and verified in a certain enterprise. The system reduces the probability of major electrical safety accidents through real-time alarms for abnormal power consumption; it reduces unplanned downtime events and avoids reduced equipment lifespan due to unreasonable production scheduling through equipment health risk alarms; it predicts future power consumption trends to avoid unreasonable production scheduling, effectively reducing the enterprise's electricity consumption and carbon emissions; and it uses multi-dimensional data mining and analysis to identify operational optimization opportunities and assist in achieving the lowest possible electricity cost per unit product. The specific implementation process is as follows:
[0091] 1. Device end
[0092] This verification involved the drive motors of key production equipment and laboratory testing equipment. The drive motors of the key production equipment were powered by electrical cabinets with a supply voltage of 380V and a supply current of less than 100A; the laboratory testing equipment was powered by socket panels with a supply voltage of 220V and a supply current of less than 16A.
[0093] 2. Edge end
[0094] The edge device is responsible for collecting current and voltage signals from the equipment via instruments and transmitting the collected signals to the cloud platform through the operator's network. Current acquisition is divided into wired and wireless methods. Wired current acquisition involves collecting the three-phase current signal of the drive motor through a current transformer at the power supply line of the electrical control cabinet and connecting it to the acquisition unit via a cable. Voltage acquisition involves connecting the three-phase voltage signal of the drive motor to the acquisition unit inside the electrical control cabinet via terminal wiring. Wireless current acquisition involves directly inserting the power socket of the testing equipment into a socket added to the existing socket panel. The socket directly collects the current and voltage signals and transmits them to the gateway via a wireless network.
[0095] 3. Cloud platform
[0096] The cloud platform's data layer unit receives real-time current and voltage data from the edge device and processes and stores the data. The model layer unit establishes and applies abnormal electricity consumption behavior early warning models, asset health risk early warning models, and comprehensive operational decision optimization models, as well as other statistical models.
[0097] Electricity consumption behavior abnormal early warning model:
[0098] a) Data acquisition: Collect data related to equipment energy consumption, such as current, voltage, and active power;
[0099] b) Preprocessing and feature extraction: Based on the equipment's operating patterns, it is divided into two categories: high fluctuation and stable. Highly predictive features are extracted in a targeted and accurate manner. By learning the energy consumption fluctuation patterns and segmenting the operating conditions, the interference of operating condition fluctuations on prediction is reduced.
[0100] c) Establish a power prediction model: Use deep learning algorithms to predict future power;
[0101] d) Establish a power upper and lower limit prediction model. Based on the statistical distribution of historical data, train the historical power upper and lower limit prediction model through quantile regression to predict the future power upper and lower limits.
[0102] e) Establish a power over-limit / abnormality detection model: By comparing the predicted power value with the predicted upper and lower power limits, predict whether the future power will exceed the limit; by comparing the difference between the predicted value of the power prediction model and the real-time power measurement value, calculate the distribution deviation of the residual distribution and the training data, and determine whether there is an abnormality in energy consumption;
[0103] f) Output power over / abnormal detection model results, and early warning of over / abnormal states;
[0104] Asset health risk early warning model:
[0105] a) Data acquisition: Collect data related to equipment energy consumption, such as current, voltage, and active power;
[0106] b) Similar working condition identification: When the similarity of working conditions is not high, a certain time window is manually defined as the comparison period for similar working conditions;
[0107] c) Equipment deterioration feature extraction: Extract "energy fingerprint" health features, including:
[0108] Time-domain characteristics: mean, root mean square, kurtosis, standard deviation, skewness, peak-to-peak value, maximum value, minimum value, zero-crossing rate, etc.
[0109] Frequency domain characteristics: peak frequency harmonics, spectral entropy, energy proportion of each layer of discrete wavelet transform, etc.
[0110] d) Establish a healthy baseline model: Use principal component analysis to transform features into a baseline model that can characterize the training history distribution;
[0111] e) Calculate the equipment health score: In the prediction process, repeat steps a)–c) above to obtain the feature vector of the prediction process; calculate the fitting residual SPE between the feature vector and the baseline model.
[0112] f) Calculate the equipment health risk level: When enough failure samples have been accumulated, health scores are marked according to different time intervals from the failure, such as 1 month ago, 1 week ago, 1 hour ago, etc., and the health level and health value are remapped.
[0113] Comprehensive optimization model for operational decision-making:
[0114] a) Energy consumption and health status benchmarking: After data anonymization, compare a user's predicted device energy consumption and health status with the global baseline of similar devices to obtain device energy consumption ranking, device energy consumption and group benchmarking trend, device health ranking, and device health and group benchmarking trend.
[0115] b) Quantify equipment performance to support operational optimization: Based on user-defined equipment performance indicators, when the equipment health status is below average or energy consumption is too high, a performance decline is indicated;
[0116] The application layer units provide specific system functions, including:
[0117] Electrical safety protection: alarms for overvoltage, undervoltage, overcurrent, and sudden current changes; equipment start / stop status reminders;
[0118] Real-time energy consumption monitoring: Statistics on the current power consumption of electrical equipment, displaying the status values of current, voltage, and power;
[0119] Historical energy consumption statistics: Statistics on the historical power consumption of electrical equipment, showing the trend of current, voltage, and power;
[0120] Power Consumption Exceedance Alarm: An alarm is triggered based on the threshold status to indicate abnormal power consumption exceeding the limit.
[0121] Equipment runtime statistics: Statistics on the runtime of electrical equipment based on different time dimensions such as year, month, and day;
[0122] Equipment utilization statistics: Calculate and display equipment OEE;
[0123] Electricity consumption anomaly warning: Learn electricity consumption habits, predict future electricity consumption trends, and warn of potential electricity consumption anomalies;
[0124] Asset health risk warning: Based on the equipment's "energy consumption fingerprint", assess the equipment's health status, quantify the degree of deterioration, and provide early warning of failure risks;
[0125] Comprehensive optimization of operational decisions: Benchmarking against peers' energy consumption to identify high-energy-consuming processes; integrating time-of-use electricity pricing, electricity consumption forecasting, and equipment health to support operational decision optimization.
[0126] 4. User side
[0127] Users can access electricity monitoring information through three methods: Web, mobile APP, and mobile mini-program, improving the ease of use of the system.
[0128] Although the embodiments disclosed in this application are as described above, the content is merely for the purpose of facilitating understanding of this application and is not intended to limit this application. Any person skilled in the art to which this application pertains may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application; however, the scope of patent protection of this application shall still be determined by the scope defined in the appended claims.
Claims
1. An intelligent electricity consumption monitoring system, comprising a device-side module, an edge-side module, and a cloud platform-side module, characterized in that, The edge module collects the current and voltage signals of the device in the device module and transmits the collected signals to the cloud platform module. The cloud platform module includes a data layer unit, a model layer unit, and an application layer unit. The data layer unit receives the data from the edge module, the model layer unit establishes and applies an intelligent model for power management, and the application layer unit performs energy safety protection, asset energy efficiency optimization, health risk warning, and operational decision support based on the calculation results of the intelligent model. The cloud platform module includes an abnormal electricity consumption behavior early warning model, and the algorithm of the abnormal electricity consumption behavior early warning model includes the following steps: (1.1) Obtain current and voltage data related to the energy consumption of the equipment; (1.2) The equipment is classified according to its operating conditions. The operating conditions are divided into high-fluctuation operating conditions and stable operating conditions. Among them, if the operating condition-related variables remain constant when the equipment is working, or the fluctuation is less than a certain percentage of the rated operating amplitude, it is considered a stable operating condition; all others are considered high-fluctuation operating conditions. (1.3) By segmenting the operating conditions, the interference of operating condition fluctuations on the prediction is reduced, and statistics related to equipment energy consumption are extracted; the segmentation of operating conditions includes segmenting by the amplitude of operating condition variables, and using rules to divide the operating condition signals into steady-state operating conditions and transient operating conditions. (1.4) Establish a power prediction model and use time series or machine learning algorithms to predict future power; (1.5) Establish a power upper and lower limit prediction model. Based on the statistical distribution of historical data, train the historical power upper and lower limit prediction model through quantile regression, and predict the future power upper and lower limits at the same time. (1.6) By comparing the predicted power value with the predicted upper and lower limits of power, predict whether the future power will exceed the limit; by comparing the difference between the predicted value of the power prediction model and the real-time power measurement value, calculate the distribution deviation of the residual distribution and the training data, and determine whether there is an abnormality in energy consumption. (1.7) Output power over-limit and abnormal detection model results, and early warning of over-limit and abnormal states; In step (1.3), segmenting the operating conditions includes calculating the autocorrelation coefficients of relevant variables to find the periodicity of the signal for segmentation; for the signal Its autocorrelation coefficient is: ; in, For lag The autocorrelation coefficient, For the signal in time The average value of the complete sequence; based on this calculation method, the autocorrelation sequence composed of autocorrelation coefficients is obtained; set the upper and lower limits of the autocorrelation coefficients, and the lag order corresponding to the autocorrelation coefficient values exceeding the upper and lower limits is the number of cycles in which the potential repetitive pattern appears, i.e., the signal period; The cloud platform module also includes an asset health risk early warning model, the algorithm of which includes the following steps: (2.1) Obtain current and voltage data related to the energy consumption of the equipment; (2.2) The similarity of equipment operating conditions is quantitatively judged by using the autocorrelation coefficient calculation method described in step (1.3); (2.3) Identify similar operating conditions of equipment and establish basic conditions for comparing the energy consumption of equipment under the same operating conditions and the same working cycle; (2.4) Extract the feature vector of equipment degradation characteristics from the variable characteristics of equipment signals; (2.5) Establish a healthy baseline model and transform the features into a healthy baseline model that can characterize the training history distribution; (2.6) In the prediction process, repeat the above steps (2.1)-(2.3) to obtain the feature vector of the prediction process, calculate the distance between the feature vector and the baseline model to quantify the degree of conformity of the prediction feature relative to the baseline distribution, and calculate the equipment health score; (2.7) Accumulate fault samples, label and calibrate health scores according to different time intervals from the fault, and calculate the health risk level of the equipment.
2. The intelligent electricity consumption monitoring system according to claim 1, characterized in that, In step (2.4), the following features are extracted simultaneously: Time-domain statistical characteristics include statistics that can describe the signal morphology of segmented similar operating condition intervals; Temporal morphological similarity features include features that can describe the degree of similarity in signal shape during similar operating cycles; Frequency domain energy distribution characteristics, including indices that can describe the frequency domain energy distribution of a signal; Time-frequency domain energy distribution characteristics include image feature matrices that can describe the changes in energy distribution across different frequency bands within a similar operating cycle.
3. The intelligent electricity consumption monitoring system according to claim 1, characterized in that, In steps (2.2)-(2.3), when the similarity of working conditions is high, time series pattern recognition method or time series similarity comparison model is used to identify similar working conditions; when the similarity of working conditions is not high, the working condition interval to enter the health assessment process is determined by manually defining the time window or setting trigger conditions.
4. The intelligent electricity consumption monitoring system according to any one of claims 1-3, characterized in that, The cloud platform module also includes an operational decision-making comprehensive optimization model, which can benchmark energy consumption against health status and integrate power prediction values and health assessment values of similar equipment. By comparing the energy consumption prediction values and health values of equipment with the global baseline of similar equipment, the model can derive equipment energy consumption ranking, equipment energy consumption trend compared with the group, equipment health ranking, and equipment health trend compared with the group.
5. The intelligent electricity consumption monitoring system according to claim 4, characterized in that, The comprehensive optimization model for operational decision-making is used to quantify equipment performance and support operational optimization. It can customize equipment performance indicators based on the comparison of equipment energy consumption, health assessment values and the group.
6. The intelligent electricity consumption monitoring system according to any one of claims 1-3 and 5, characterized in that, The device-side module includes high-current wired devices and low-current mobile devices. The intelligent electricity monitoring system also includes a user-side module, through which electricity monitoring information can be obtained.
Citation Information
Patent Citations
Power distribution Internet-of-things data monitoring system and scheduling method based on cloud edge collaboration
CN113836796A
Combined power station operation optimization method and system based on high-precision wind and light output prediction
CN115714420A
