Intelligent Power Consumption Risk Identification System and Method Based on Multi-Sensor Fusion

Abstract

The present invention relates to the field of multi-sensor fusion technology, and specifically to a smart power consumption risk identification system and method based on multi-sensor fusion, including step S100: configuring sensors, obtaining an initial alarm threshold, generating a sensor data sequence group, and preprocessing the data; step S200: recording risk events, marking and extracting abnormal data during the duration of the risk event, adding labels to the abnormal data and recording the time stamp, and then training through a long short-term memory network model to define the type and probability of the risk event; step S300: establishing a linear regression equation to obtain the influence relationship between the abnormal data and the initial alarm threshold, and obtaining the influence score of the abnormal data through weighted fusion; step S400: calculating the real-time alarm threshold, adjusting the acquisition frequency according to the influence score, warning the user of the possible type and probability of the risk event, and generating corresponding solutions and specific steps by using a matching algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of multi - sensor fusion, and specifically to a smart power consumption risk identification system and method based on multi - sensor fusion. Background Technique

[0002] The identification of power consumption risks in modern buildings is crucial for ensuring building safety and operation efficiency. Especially in the complex mechanical and electrical installation environment of high - rise buildings, if power consumption risks cannot be detected and handled in a timely manner, it may lead to serious accidents and even casualties.

[0003] In addition, the deficiencies of existing power consumption risk identification systems are as follows: their adaptability to complex and dynamic environments is poor; due to the large differences in data between different types of sensors, such as temperature sensors, current sensors, and pressure sensors, each having different data characteristics and processing requirements, existing systems often have difficulty effectively fusing and processing and analyzing these heterogeneous data, resulting in untimely or inaccurate risk assessment; existing power consumption risk identification systems have limitations in wide - range application, as they are often designed for specific types of electrical systems or equipment and lack sufficient applicability; existing power consumption risk identification systems cannot give possible situations, occurrence probabilities, and corresponding countermeasures when identifying risks, wasting valuable risk - handling time. Summary of the Invention

[0004] The purpose of the present invention is to provide a smart power consumption risk identification system and method based on multi - sensor fusion to solve the problems raised in the above - mentioned background technique.

[0005] To solve the above - mentioned technical problems, the present invention provides the following technical solutions: A smart power consumption risk identification system and method based on multi - sensor fusion, including:

[0006] Step S100: Set types of sensors, set the importance score of each sensor, obtain the initial alarm threshold of each sensor, collect data at the set acquisition frequency and generate a sensor data sequence group, and pre - process the data in the sensor data sequence group.

[0007] Step S200: Record a primary alarm issued by the electrical system as a primary risk event, extract the sensor data sequence group during the duration of the risk event, mark the abnormal sensor data sequence group, add type labels to the abnormal data and record the time stamp, and train through a long short - term memory network model to define the type and probability of the risk event.

[0008] Step S300: Extract the duration, data peak value, and importance score from the abnormal data, establish a linear regression equation, obtain the influence relationship between the abnormal data and the initial alarm threshold, and obtain the abnormal data influence score through weighted fusion.

[0009] Step S400: Calculate the real-time alarm threshold according to the influence relationship, monitor the operation of the electrical system with the real-time alarm threshold, adjust the sensor acquisition frequency according to the influence score of abnormal data, update the type and probability of possible risk events in real time and give early warnings to users, establish a database, and generate corresponding solutions and specific steps using a matching algorithm.

[0010] Further, step S100 includes:

[0011] Step S101: Obtain the safety intervals of the data detected by types of sensors respectively, and use the upper limit and the lower limit of the safety interval as the initial alarm threshold;

[0012] The user sets the importance score of each sensor according to the actual scenario ;

[0013] Among them, is the number of sensor types, represents the importance score of the th sensor, and the value range of

[0014] is 0 to 100;

[0015] In the actual usage scenario, different types of sensors may have different sensitivities to the same risk event. Setting the importance score according to the actual usage scenario can enable the system to respond faster to signals with high risk recognition capabilities, and perform weighting according to the specific functions of each sensor and its sensitivity to data changes, improving the accuracy of early warnings;

[0015] Step S102: Set the initial acquisition frequency of the sensors, and collect the data in the electrical system at the initial acquisition frequency to obtain a sensor data sequence group, where the sensor data sequence group includes sensor data sequences collected at the initial acquisition frequency ;

[0016] Step S103: Preprocess the data in the sensor data sequence group, and use the linear scaling method to make all data points in the sensor data sequence group fall within the range of 0 to 100;

[0017] Normalizing the data to the range of 0 to 100 through the linear scaling method can significantly enhance the consistency and efficiency of data processing, and at the same time help reduce the influence of outliers.

[0018] Further, step S200 includes:

[0019] Step S201: Extract the sensor data sequence group during the risk duration, perform anomaly detection on the extracted sensor data sequence group, and mark the sensor data sequences exceeding the initial alarm threshold as abnormal data sequences; sensor data sequences are marked as abnormal data sequences;

[0020] Step S202: According to the sensor type corresponding to the abnormal data sequence, add type labels to the groups of abnormal data, record the timestamps of the abnormal data, and input the abnormal data, the type labels of the abnormal data, and the abnormal data timestamps into the long short-term memory network model;

[0021] Step S203: The long short-term memory network model performs learning and training by receiving and processing the abnormal data, the type labels of the abnormal data, and the abnormal data timestamps;

[0022] The long short-term memory network model uses the abnormal data as input features, differentiates and marks different anomalies using the type labels of the abnormal data, and constructs the time correlation between data based on the time series information contained in the timestamps of the abnormal data, predicting the type and occurrence probability of the risk event;

[0023] Using the long short-term memory network model to process the abnormal data, utilizing the type labels of the abnormal data and tracking the time development trajectory of the abnormal events, different types of risk events can be accurately classified and responded to, providing higher accuracy and foresight for risk identification.

[0024] Further, step S300 includes:

[0025] Step S301: Extract the duration , peak data , and the importance score of the sensor corresponding to each type of data for each group of abnormal data in the risk event respectively, where the subscript is the index of the th type of abnormal data in the groups of abnormal data; groups of abnormal data;

[0026] Step S302: Establish a real-time alarm threshold upper and lower limit adjustment formula according to the linear regression equation;

[0027] Real-time alarm threshold upper limit adjustment formula:

[0028] ;

[0029] where, is the initial alarm threshold upper limit of the th type of abnormal data, and the subscript is the th type of abnormal data in the Index in the group of abnormal data; Is the intercept term, indicating the basic adjustment amount of abnormal data to the upper limit of the alarm threshold in the absence of any significant input features; Is the coefficient of the linear term, indicating the direct impact of each data on the upper limit of the real-time alarm threshold; is the coefficient of the quadratic term, indicating the non-linear relationship in the data; Is the coefficient of the interaction term, indicating the additional impact on the upper limit under the combined action of two variables; All are calculated from the sensor data by statistical methods;

[0030] The formula for adjusting the lower limit of the real-time alarm threshold is:

[0031] ;

[0032] Among them, Is the initial lower limit of the alarm threshold for the th type of abnormal data, and the subscript Is the th type of abnormal data at the index in the group of abnormal data Is the intercept term, indicating the basic adjustment amount of abnormal data to the lower limit of the alarm threshold in the absence of any significant input features; Is the coefficient of the linear term, indicating the direct impact of each data on the lower limit of the real-time alarm threshold; Is the coefficient of the quadratic term, indicating the non-linear relationship in the data; Is the coefficient of the interaction term, indicating the additional impact on the lower limit under the combined action of two variables; All are calculated from the sensor data by statistical methods;

[0033] Obtain the upper limit of the real-time alarm threshold for the th type of abnormal data and the lower limit of the real-time alarm threshold;

[0034] By calculating the real-time alarm threshold for each type of abnormal data, the system can dynamically adjust the sensitivity to potential risks, thereby achieving more accurate monitoring and early warning;

[0035] Step S303: Establish a weighted fusion formula:

[0036] ;

[0037] Among them, Is the duration of the th type of abnormal data, Is the peak value of the th type of abnormal data, Is the Importance score of the sensor corresponding to a certain type of abnormal data; is the weight of the duration of the th type of abnormal data, is the weight of the peak value of the th type of abnormal data, The weight of the importance score of the sensor corresponding to the ~ th type of abnormal data is given based on actual requirements and data characteristics;

[0038] Obtain the influence score of the th type of abnormal data , and the range of the influence score is 0 to 100;

[0039] By calculating the influence score between 0 and 100 for each type of abnormal data, the potential impact of each abnormal data on safety can be quantified, reflecting the high-risk state of the data, so as to dynamically adjust the acquisition frequency in the warning state according to the severity of the abnormal data;

[0040] Furthermore, step S400 includes:

[0041] Step S401: Monitor the operation of the electrical system with the real-time alarm threshold. When the real-time data in the sensor data sequence group exceeds the upper and lower limits of the real-time alarm threshold, obtain the types of abnormal data in the real-time data;

[0042] According to the influence score of the th type of abnormal data, obtain the acquisition frequency formula in the warning state:

[0043] ;

[0044] ;

[0045] Among them, is the initial acquisition frequency set by the user, is the adjustment step size, is the real-time acquisition frequency of each sensor when the real-time data exceeds the upper and lower limits of the real-time alarm threshold, and the maximum value among all real-time acquisition frequencies is selected as the acquisition frequency in the warning state; Using a non-linear model can achieve a larger adjustment amplitude after reaching a certain threshold to better respond to possible risk events;

[0046] Obtain the acquisition frequency in the warning state of the electrical system in the warning state ;

[0047] ​By calculating and setting the acquisition frequency of the electrical system in the early warning state, the intelligent power consumption risk identification system can achieve faster response and more detailed monitoring of potential risks. This adaptive acquisition frequency adjustment enables the system to collect more key data at critical moments, enhancing the real-time monitoring and analysis capabilities for abnormal events;

[0048] Step S402: According to the abnormal data impact score Change the acquisition frequency to obtain the acquisition frequency in the early warning state , at the acquisition frequency in the early warning state Monitor the electrical system data. Based on the training results of the type tags and timestamps corresponding to the real-time data in the long short-term memory network model, warn the user and update the type and probability of possible risk events in real time;

[0049] Step S403: Establish a database to store the solutions and specific steps of risk events. When warning the user, use a matching algorithm to obtain the solutions corresponding to the types of possible risk events and feedback them to the user in real time;

[0050] By establishing a database to store the solutions and specific steps of risk events and using a matching algorithm to provide corresponding solutions to the user during early warning, the intelligent power consumption risk identification system can achieve efficient risk response and rapid problem-solving. This solution matching and feedback mechanism ensures that when potential risks are discovered, the user can obtain targeted guidance and suggestions in a shorter response time, improving the ability to handle emergencies and thus optimizing the overall operation safety of the system.

[0051] Furthermore, the system includes a sensor configuration and data acquisition module, a risk event identification and data marking module, a data analysis and scoring module, and a real-time monitoring and warning module;

[0052] The sensor configuration and data acquisition module includes different types of sensors. The sensor configuration and data acquisition module is used to obtain the initial alarm threshold of each sensor and set the importance score, determine the acquisition frequency of the sensors and collect the operation data of the electrical system at the acquisition frequency, generating a sensor data sequence group;

[0053] The risk event identification and data marking module is used to record risk events when the electrical system issues an alarm, extract the sensor data sequences in the risk events and mark the data sequences exceeding the initial alarm threshold as abnormal data sequences, add labels and timestamps to the abnormal data and input them into the long short-term memory network model, and the long short-term memory network model learns to define and identify the types and occurrence probabilities of risk events;

[0054] The data analysis and scoring module is used to extract the duration, data peak value in the abnormal data, and the importance score of the sensors corresponding to the abnormal data types, establish a linear regression model, obtain the influence relationship between the abnormal data and the initial alarm threshold, and calculate the influence score of the abnormal data through weighted fusion;

[0055] The real-time monitoring and early warning module is used to calculate and adjust the alarm threshold in real time according to the influence relationship, adjust the acquisition frequency according to the influence score of the abnormal data, so as to monitor the operation status of the electrical system. The real-time monitoring and early warning module warns the user and updates the possible types and probabilities of risk events in real time, establishes a database and uses a matching algorithm to generate solutions and specific steps for the corresponding risk events;

[0056] The risk event identification and data marking module is connected to the output end of the sensor configuration and data acquisition module; the data analysis and scoring module is connected to the output end of the risk event identification and data marking module, and the real-time monitoring and early warning module is connected to the output end of the data analysis and scoring module.

[0057] Furthermore, the sensor configuration and data acquisition module includes a sensor configuration unit, a data acquisition unit, and a data preprocessing unit;

[0058] The sensor configuration unit includes several different types of sensors, each sensor has a safety interval for the corresponding data, and the sensor configuration unit sets the acquisition frequency of the sensors and the main importance score of the sensors;

[0059] The data acquisition unit is used to collect the operation data of the electrical system regularly according to the acquisition frequency and generate a sensor data sequence group;

[0060] The data preprocessing unit preprocesses the data in the sensor data sequence group.

[0061] Furthermore, the risk event identification and data marking module includes an event monitoring unit, an anomaly detection unit, and a data learning unit;

[0062] The event monitoring unit is used to monitor the alarms issued by the electrical system under the standard of the initial alarm threshold and record the occurrence of risk events;

[0063] The anomaly detection unit is used to extract the sensor data sequence group during the duration of the risk event, mark the abnormal data exceeding the initial alarm threshold, add type labels and record timestamps;

[0064] The data learning unit is used to input the abnormal data into the long short-term memory network model for training, define and identify the types and occurrence probabilities of risk events.

[0065] Furthermore, the data analysis and scoring module includes a feature extraction unit, a regression analysis unit, and a scoring calculation unit;

[0066] The feature extraction unit is used to extract the duration, peak data from the abnormal data, and extract the importance score according to the corresponding type;

[0067] The regression analysis unit analyzes the influence relationship between the abnormal data and the initial alarm threshold by establishing a linear regression model;

[0068] The scoring calculation unit calculates the influence score of the abnormal data through a weighted fusion formula.

[0069] Furthermore, the real-time monitoring and early warning module includes a monitoring condition adjustment unit, a monitoring and early warning unit, and a solution generation unit;

[0070] The monitoring condition adjustment unit calculates and adjusts the real-time alarm threshold according to the influence relationship, and updates the sensor acquisition frequency according to the influence score of the abnormal data;

[0071] The monitoring and early warning unit monitors the operation of the electrical system using the updated real-time alarm threshold and acquisition frequency, extracts the type label and timestamp of the abnormal data, and warns the user of the possible risk event type and probability;

[0072] The solution generation unit is used to match the corresponding solutions and specific steps according to the possible risk event type and probability, and provide feedback to the user.

[0073] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: The intelligent power consumption risk identification system and method based on multi-sensor fusion of the present invention configures various types of sensors and performs data fusion on them, which can comprehensively and timely detect abnormal situations in the electrical system, thereby improving the efficiency and accuracy of risk identification; Secondly, using the long short-term memory network model to train and analyze abnormal data can effectively predict the type and occurrence probability of risk events, enhancing the early warning ability of the system; In addition, by establishing a linear regression model and a weighted fusion algorithm, the alarm threshold and acquisition frequency are adjusted in real time, making the system have better adaptability in a dynamic environment, ensuring that risk events can be timely warned and corresponding solutions are provided; The present invention not only improves the intelligent level of power consumption risk identification, but also can warn of the occurrence of risk events and respond quickly, reducing the possibility of accidents and ensuring the safe operation of the electrical system. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0075] Figure 1It is a schematic structural diagram of the intelligent power consumption risk identification system and method based on multi-sensor fusion according to the present invention;

[0076] Figure 2 It is a schematic flowchart of the method of the intelligent power consumption risk identification system and method based on multi-sensor fusion according to the present invention. Specific embodiments

[0077] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0078] Please refer to Figure 1 - Figure 2 , the present invention provides a technical solution:

[0079] According to an embodiment of the present invention, the intelligent power consumption risk identification system and method based on multi-sensor fusion include:

[0080] Step S100: Set 5 types of sensors, set the importance score of each sensor, obtain the initial alarm threshold of each sensor, collect data at the set collection frequency and generate a sensor data sequence group, and preprocess the data in the sensor data sequence group;

[0081] Among them, step S100 includes:

[0082] Step S101: Respectively obtain the safety intervals of the data detected by the 5 sensors, and use the upper and lower limits of the safety intervals as the initial alarm thresholds. The initial alarm thresholds include the upper limit of the safety interval and the lower limit of the safety interval ;

[0083] The user sets the importance scores of each sensor to be respectively according to the actual scenario;

[0084] Step S102: Set the initial collection frequency of the sensors , and collect the data in the electrical system at the initial collection frequency of to obtain a sensor data sequence group. The sensor data sequence group includes all sensor data sequences collected at the initial collection frequency of ;

[0085] Step S103: Preprocess the data in the sensor data sequence group, and make all data points in the sensor data sequence group fall within the range of 0 to 100 through the linear scaling method.

[0086] Step S200: Record the first alarm issued by the electrical system as a risk event, extract the sensor data sequence group during the duration of the risk event, mark the abnormal sensor data sequence group, add type tags to the abnormal data and record the timestamps, and train through a long short-term memory network model to define the type and probability of the risk event;

[0087] Among them, step S200 includes:

[0088] Step S201: Extract the sensor data sequence group during the risk duration, perform anomaly detection on the extracted sensor data sequence group, and mark the sensor data sequences exceeding the initial alarm threshold as abnormal data sequences;

[0089] Step S202: According to the sensor type corresponding to the abnormal data sequence, add type tags to the group of abnormal data, record the timestamps of the abnormal data, and input the abnormal data, the type tags of the abnormal data, and the abnormal data timestamps into the long short-term memory network model;

[0090] Step S203: The long short-term memory network model performs learning and training by receiving and processing the abnormal data, the type tags of the abnormal data, and the abnormal data timestamps;

[0091] The long short-term memory network model takes the abnormal data as input features, uses the type tags of the abnormal data to distinguish and mark different anomalies, and at the same time constructs the time association between data based on the time series information contained in the timestamps of the abnormal data to predict the type and occurrence probability of the risk event.

[0092] Step S300: Extract the duration, data peak, and importance score in the abnormal data, establish a linear regression equation, obtain the influence relationship between the abnormal data and the initial alarm threshold, and obtain the abnormal data influence score through weighted fusion;

[0093] Among them, step S300 includes:

[0094] Step S301: Extract the groups of abnormal data in the risk event respectively; among them, the duration of abnormal data 1 is 90 seconds, the peak data is 40 units, and the sensor importance score is 70; the duration of abnormal data 2 is 120 seconds, the peak data is 60 units, and the sensor importance score is 85;

[0095] Step S302: Establish an adjustment formula for the upper and lower limits of the real-time alarm threshold according to the linear regression equation;

[0096] Real-time alarm threshold upper limit adjustment formula:

[0097] ;

[0098] Among them, is the upper limit of the initial alarm threshold for the th type of abnormal data, and the subscript is the index of the th type of abnormal data in the group of abnormal data; is the intercept term, indicating the basic adjustment amount of the abnormal data to the upper limit of the alarm threshold in the absence of any significant input features; is the coefficient of the linear term, indicating the direct impact of each data on the upper limit of the real-time alarm threshold; is the coefficient of the quadratic term, indicating the non-linear relationship in the data; is the coefficient of the interaction term, indicating the additional impact on the upper limit under the combined action of two variables; All are calculated from the sensor data by statistical methods;

[0099] The formula for adjusting the lower limit of the real-time alarm threshold is:

[0100] ;

[0101] Among them, is the lower limit of the initial alarm threshold for the th type of abnormal data, and the subscript is the index of the th type of abnormal data in the group of abnormal data is the intercept term, indicating the basic adjustment amount of the abnormal data to the lower limit of the alarm threshold in the absence of any significant input features; is the coefficient of the linear term, indicating the direct impact of each data on the lower limit of the real-time alarm threshold; is the coefficient of the quadratic term, indicating the non-linear relationship in the data; is the coefficient of the interaction term, indicating the additional impact on the lower limit under the combined action of two variables; All are calculated from the sensor data by statistical methods;

[0102] Obtain the upper limit of the real-time alarm threshold and the lower limit of the real-time alarm threshold for the 1st type of abnormal data, and obtain the upper limit of the real-time alarm threshold and the lower limit of the real-time alarm threshold for the 2nd type of abnormal data;

[0103] Step S303: Establish a weighted fusion formula:

[0104] ;

[0105] Among them, is the The duration of the i-th abnormal data is the peak value of the i-th abnormal data is the importance score of the sensor corresponding to the i-th abnormal data; is the weight of the duration of the i-th abnormal data is the weight of the peak value of the i-th abnormal data, the weight of the importance score of the sensor corresponding to the i-th abnormal data, ~ are given based on actual requirements and data characteristics;

[0106] Obtain the influence score of the first type of abnormal data , obtain the influence score of the first type of abnormal data 82.5;

[0107] Step S400: Calculate the real-time alarm threshold according to the influence relationship, monitor the operation of the electrical system with the real-time alarm threshold, adjust the sensor acquisition frequency according to the influence score of the abnormal data, update the types and probabilities of possible risk events in real time and give an early warning to the user, establish a database, and generate corresponding solutions and specific steps using a matching algorithm;

[0108] Among them, step S400 includes:

[0109] Step S401: Monitor the operation of the electrical system with the real-time alarm threshold. When the real-time data in the sensor data sequence group exceeds the upper and lower limits of the real-time alarm threshold, obtain the types of abnormal data in the real-time data;

[0110] According to the influence score of the abnormal data Obtain the formula for the acquisition frequency in the warning state:

[0111] ;

[0112] ;

[0113] Among them, is the initial acquisition frequency set by the user, is the adjustment step size, is the real-time acquisition frequency of each sensor when the real-time data exceeds the upper and lower limits of the real-time alarm threshold, and the maximum value among all real-time acquisition frequencies is selected as the acquisition frequency in the warning state; A non-linear model can be used to increase the adjustment amplitude after reaching a certain threshold to better respond to possible risk events; ​

[0114] Obtain the real-time acquisition frequency of the first type of abnormal data , obtain the real-time acquisition frequency of the second type of abnormal data ,

[0115] Select the maximum item among all real-time acquisition frequencies to obtain the warning state acquisition frequency of the electrical system in the warning state ;

[0116] Step S402: According to the impact score of abnormal data Change the acquisition frequency to obtain the warning state acquisition frequency , and monitor the electrical system data at the warning state acquisition frequency Based on the training results of the type tags and timestamps corresponding to the real-time data in the long short-term memory network model, warn the user and update the types and probabilities of possible risk events in real time;

[0117] Step S403: Establish a database to store the solutions and specific steps of risk events. When warning the user, use a matching algorithm to obtain the solutions corresponding to the types of possible risk events and feedback them to the user in real time.

[0118] The system includes a sensor configuration and data acquisition module, a risk event identification and data marking module, a data analysis and scoring module, and a real-time monitoring and warning module;

[0119] The sensor configuration and data acquisition module includes different types of sensors. The sensor configuration and data acquisition module is used to obtain the initial alarm threshold of each sensor and set the importance score, determine the acquisition frequency of the sensor, collect the operation data of the electrical system at the acquisition frequency, and generate a sensor data sequence group;

[0120] The risk event identification and data marking module is used to record risk events when the electrical system issues an alarm, extract the sensor data sequences in the risk events, mark the data sequences exceeding the initial alarm threshold as abnormal data sequences, add labels and timestamps to the abnormal data and input them into the long short-term memory network model, and the long short-term memory network model learns to define and identify the types and occurrence probabilities of risk events;

[0121] The data analysis and scoring module is used to extract the duration, data peak in the abnormal data, and the importance score of the type of sensor corresponding to the abnormal data, establish a linear regression model, obtain the impact relationship between the abnormal data and the initial alarm threshold, and calculate the impact score of the abnormal data through weighted fusion;

[0122] The real-time monitoring and early warning module is used to calculate and adjust the alarm threshold in real time according to the influence relationship, adjust the acquisition frequency according to the influence score of abnormal data, so as to monitor the operation status of the electrical system. The real-time monitoring and early warning module warns the user and updates the possible types and probabilities of risk events in real time, establishes a database and uses a matching algorithm to generate solutions and specific steps for corresponding risk events;

[0123] The risk event identification and data marking module is connected to the output end of the sensor configuration and data acquisition module; the data analysis and scoring module is connected to the output end of the risk event identification and data marking module, and the real-time monitoring and early warning module is connected to the output end of the data analysis and scoring module.

[0124] The sensor configuration and data acquisition module includes a sensor configuration unit, a data acquisition unit and a data preprocessing unit; the sensor configuration unit includes several different types of sensors. Each sensor has a safe range for corresponding data. The sensor configuration unit sets the acquisition frequency of the sensors and the main importance score of the sensors; the data acquisition unit is used to collect the operation data of the electrical system regularly according to the acquisition frequency and generate a sensor data sequence group; the data preprocessing unit preprocesses the data in the sensor data sequence group.

[0125] The risk event identification and data marking module includes an event monitoring unit, an anomaly detection unit and a data learning unit; the event monitoring unit is used to monitor the alarms issued by the electrical system under the standard of the initial alarm threshold and record the occurrence of risk events; the anomaly detection unit is used to extract the sensor data sequence group during the duration of the risk event, mark the abnormal data exceeding the initial alarm threshold, add type tags and record timestamps; the data learning unit is used to input the abnormal data into the long short-term memory network model for training to define and identify the types and occurrence probabilities of risk events.

[0126] The data analysis and scoring module includes a feature extraction unit, a regression analysis unit and a scoring calculation unit; the feature extraction unit is used to extract the duration, peak data from the abnormal data and extract the importance score according to the corresponding type; the regression analysis unit analyzes the influence relationship between the abnormal data and the initial alarm threshold by establishing a linear regression model; the scoring calculation unit calculates the influence score of the abnormal data through a weighted fusion formula.

[0127] The real-time monitoring and early warning module includes a monitoring condition adjustment unit, a monitoring and early warning unit, and a solution generation unit. The monitoring condition adjustment unit calculates and adjusts the real-time alarm threshold according to the influence relationship, and updates the sensor acquisition frequency according to the influence score of abnormal data. The monitoring and early warning unit monitors the operation of the electrical system using the updated real-time alarm threshold and acquisition frequency, extracts the type tags and timestamps of abnormal data, and warns the user of the possible types and probabilities of risk events. The solution generation unit is used to match the corresponding solutions and specific steps according to the possible types and probabilities of risk events, and provide feedback to the user.

[0128] It should be noted that in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variation thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or elements inherent to such process, method, article or device.

[0129] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for identifying intelligent power consumption risks based on multi-sensor fusion, characterized in that, The method includes: Step S100: Set types of sensors, set the importance score for each sensor, obtain the initial alarm threshold for each sensor, collect data at the set acquisition frequency and generate a sensor data sequence group, and preprocess the data in the sensor data sequence group; Step S200: Extract the sensor data sequence group during the duration of the risk event, mark the abnormal sensor data sequence group, add type tags to the abnormal data and record the timestamps, train through a long short-term memory network model, and define the type and probability of the risk event; Step S300: Extract the duration and data peak in the abnormal data, as well as the importance score of the corresponding sensor, establish a linear regression equation, obtain the influence relationship between the abnormal data and the initial alarm threshold, and obtain the influence score of the abnormal data through weighted fusion; The said Step S300 includes: Step S301: Extract the duration of continuous abnormal data in each group , peak data and the importance score of the sensor corresponding to each type of data , where the subscript is the index of the th type of abnormal data in the group of abnormal data; Step S302: Establish an adjustment formula for the upper and lower limits of the real-time alarm threshold according to the linear regression equation; Real-time alarm threshold upper limit adjustment formula: ； Among them, is the upper limit of the initial alarm threshold for the th type of abnormal data, and the subscript is the index of the th type of abnormal data in the group of abnormal data; is the intercept term, indicating the basic adjustment amount of the abnormal data to the upper limit of the alarm threshold in the absence of any significant input features; is the coefficient of the linear term, indicating the direct impact of each data on the upper limit of the real-time alarm threshold; is the coefficient of the quadratic term, indicating the non-linear relationship in the data; is the coefficient of the interaction term, indicating the additional impact on the upper limit under the combined action of two variables; the are all calculated from the sensor data by statistical methods; Real-time alarm threshold lower limit adjustment formula: ； Among them, is the lower limit of the initial alarm threshold for the th type of abnormal data, and the subscript is the index of the th type of abnormal data in the th group of abnormal data; is the intercept term, indicating the basic adjustment amount of the abnormal data to the lower limit of the alarm threshold in the absence of any significant input features; is the coefficient of the linear term, indicating the direct impact of each data on the lower limit of the real-time alarm threshold; is the coefficient of the quadratic term, indicating the non-linear relationship in the data; is the coefficient of the interaction term, indicating the additional impact on the lower limit under the combined action of two variables; the are all calculated from the sensor data by statistical methods; Obtain the upper limit of the real-time alarm threshold for the abnormal data and the lower limit of the real-time alarm threshold ; Step S303: Establish a weighted fusion formula: ； Among them, is the duration of the th abnormal data, is the peak value of the th abnormal data, is the importance score of the sensor corresponding to the th abnormal data; is the weight of the duration of the th abnormal data, is the weight of the peak value of the th abnormal data, is the weight of the importance score of the sensor corresponding to the th abnormal data, and the ~ are given based on actual requirements and data characteristics; Obtain the influence score of the abnormal data , and the range of the influence score is 0 to 100; Step S400: Calculate the real-time alarm threshold according to the influence relationship, monitor the operation of the electrical system with the real-time alarm threshold, adjust the sensor acquisition frequency according to the influence score of the abnormal data, update the type and probability of the possible risk events in real time and give a warning to the user, establish a database, and generate corresponding solutions and specific steps by using a matching algorithm.

2. The intelligent power consumption risk identification method based on multi-sensor fusion according to claim 1, wherein The said Step S100 includes: Step S101: Obtain respectively the safety intervals of the data detected by the sensors, and use the upper limit and the lower limit of the safety interval as the initial alarm thresholds; The user sets the importance score of each sensor according to the actual scenario ; Among them, is the number of sensor types, represents the importance score of the th sensor, ranging from 0 to 100; Step S102: Set the initial acquisition frequency of the sensor , at the initial acquisition frequency collect data within the electrical system to obtain a sensor data sequence group, where the sensor data sequence group includes collected at sensor data sequences; Step S103: Preprocess the data in the sensor data sequence group, and make all data points in the sensor data sequence group fall within the range of 0 to 100 through the linear scaling method.

3. A method for identifying intelligent power consumption risks based on multi-sensor fusion according to claim 2, characterized in that The said Step S200 includes: Step S201: Extract the sensor data sequence group within the risk duration, perform anomaly detection on the extracted sensor data sequence group, and mark the sensor data sequences that exceed the initial alarm threshold as abnormal data sequences; sensor data sequences as abnormal data sequences; Step S202: According to the sensor type corresponding to the abnormal data sequence, for types of abnormal data, add type labels, record the timestamps of the abnormal data, and input the abnormal data, the type labels of the abnormal data, and the timestamps of the abnormal data into the long short-term memory network model; Step S203: The long short-term memory network model learns and trains by receiving and processing the abnormal data, the type tags of the abnormal data, and the timestamps of the abnormal data; The long short-term memory network model takes the abnormal data as the input feature, uses the type tags of the abnormal data to distinguish and mark different abnormalities, and at the same time constructs the time association between the data based on the time series information contained in the timestamps of the abnormal data, and predicts the type and occurrence probability of the risk event.

4. A method for identifying intelligent power consumption risks based on multi-sensor fusion according to claim 1, characterized in that, The said Step S400 includes: Step S401: Monitor the operation of the electrical system with the real-time alarm threshold. When it is detected that the real-time data in the sensor data sequence group exceeds the upper and lower limits of the real-time alarm threshold, obtain the types of abnormal data in the real-time data; According to the influence score of the first type of abnormal data, the formula for collecting the warning state frequency is obtained: ； ； Among them, is the initial acquisition frequency set for the user, is the adjustment step, is the real-time acquisition frequency of each sensor when the real-time data exceeds the upper and lower limits of the real-time alarm threshold, and the maximum item among all real-time acquisition frequencies is selected as the acquisition frequency in the early warning state; Obtain the warning status acquisition frequency of the electrical system in the warning state ; Step S402: Affect the score according to the abnormal data Change the acquisition frequency to obtain the acquisition frequency in the warning state , and use the acquisition frequency in the warning state Monitor the electrical system data, and based on the training results of the type tags and timestamps corresponding to the real-time data in the long short-term memory network model, warn the user and update the possible risk event types and probabilities in real time; Step S403: Establish a database to store the solutions and specific steps of the risk events. When giving a warning to the user, use a matching algorithm to obtain the solutions corresponding to the possible risk event types and feedback them to the user in real time.

5. A multi-sensor fusion-based intelligent power consumption risk identification system applied to the multi-sensor fusion-based intelligent power consumption risk identification method according to any one of claims 1-4, characterized in that, The system includes a sensor configuration and data acquisition module, a risk event identification and data marking module, a data analysis and scoring module, and a real-time monitoring and warning module; The sensor configuration and data acquisition module includes different types of sensors. The sensor configuration and data acquisition module is used to obtain the initial alarm threshold of each sensor and set the importance score, determine the acquisition frequency of the sensor and collect the operation data of the electrical system at the acquisition frequency, and generate a sensor data sequence group; The risk event identification and data marking module is used to record risk events when the electrical system issues an alarm, extract the sensor data sequence in the risk event, mark the data sequence exceeding the initial alarm threshold as an abnormal data sequence, add tags and timestamps to the abnormal data, and input it into the long short-term memory network model, which is learned by the long short-term memory network model to define and identify the type and occurrence probability of the risk event; The data analysis and scoring module is used to extract the duration, data peak in the abnormal data, and the importance score of the sensors corresponding to the abnormal data type, establish a linear regression model, obtain the influence relationship between the abnormal data and the initial alarm threshold, and calculate the influence score of the abnormal data through weighted fusion; The real-time monitoring and early warning module is used to calculate and adjust the alarm threshold in real time according to the influence relationship, adjust the acquisition frequency according to the influence score of the abnormal data, so as to monitor the operation status of the electrical system. The real-time monitoring and early warning module warns the user and updates the possible type and probability of the risk event in real time, establishes a database and uses a matching algorithm to generate solutions and specific steps for the corresponding risk event; The risk event identification and data marking module is connected to the output end of the sensor configuration and data acquisition module; the data analysis and scoring module is connected to the output end of the risk event identification and data marking module, and the real-time monitoring and early warning module is connected to the output end of the data analysis and scoring module.

6. The intelligent power consumption risk identification system based on multi-sensor fusion according to claim 5, wherein The sensor configuration and data acquisition module includes a sensor configuration unit, a data acquisition unit, and a data preprocessing unit; The sensor configuration unit includes different types of sensors, each of which has a safety interval for corresponding data. The sensor configuration unit sets the acquisition frequency of the sensors and the main importance score of the sensors; The data acquisition unit is used to regularly collect the operation data of the electrical system according to the acquisition frequency and generate a group of sensor data sequences; The data preprocessing unit preprocesses the data in the group of sensor data sequences.

7. A smart power consumption risk identification system based on multi-sensor fusion according to claim 5, characterized in that The risk event identification and data marking module includes an event monitoring unit, an anomaly detection unit, and a data learning unit; The event monitoring unit is used to monitor the alarm issued by the electrical system under the standard of the initial alarm threshold and record the occurrence of the risk event; The anomaly detection unit is used to extract the group of sensor data sequences during the duration of the risk event, mark the abnormal data exceeding the initial alarm threshold, add type tags and record timestamps; The data learning unit is used to input the abnormal data into the long short-term memory network model for training to define and identify the type and occurrence probability of the risk event.

8. The intelligent power consumption risk identification system based on multi-sensor fusion according to claim 5, characterized in that, The data analysis and scoring module includes a feature extraction unit, a regression analysis unit, and a scoring calculation unit; The feature extraction unit is used to extract the duration, peak data from the abnormal data, and extract the importance score according to the corresponding type; The regression analysis unit analyzes the influence relationship between the abnormal data and the initial alarm threshold by establishing a linear regression model; The scoring calculation unit calculates the influence score of the abnormal data through a weighted fusion formula.

9. The intelligent power consumption risk identification system based on multi-sensor fusion according to claim 5, characterized in that, The real-time monitoring and early warning module includes a monitoring condition adjustment unit, a monitoring and early warning unit, and a solution generation unit; The monitoring condition adjustment unit calculates and adjusts the real-time alarm threshold according to the influence relationship, and updates the sensor acquisition frequency according to the influence score of the abnormal data; The monitoring and warning unit monitors the operation of the electrical system using the updated real-time alarm threshold and acquisition frequency, extracts the type tags and timestamps of abnormal data, and warns the user of the possible types and probabilities of risk events. The solution generation unit is used to match the corresponding solutions and specific steps according to the possible types and probabilities of risk events and provide feedback to the user.

Images