A method and system for monitoring abnormal behavior of a heat user
By constructing a variation prediction learning machine model by screening climate and time factors, and optimizing the input layer weights and hidden layer thresholds, the problem of inefficient thermal user anomaly detection in traditional methods is solved, and high-precision monitoring results are achieved.
Patent Information
- Application Number
- CN202211053172.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-30
- Publication Date
- 2026-02-27
- Estimated Expiration
- 2042-08-30
AI Technical Summary
Traditional methods for detecting anomalies in heating users rely on manual inspection, which is inefficient. Furthermore, monitoring methods based on neural networks suffer from poor monitoring results because the randomness of the input layer weights and hidden layer thresholds affects the output accuracy.
By acquiring climatic and temporal factors that affect the heat consumption of heating users, filtering feature vectors, constructing a variation prediction learning machine model, optimizing the input layer weights and hidden layer thresholds, and using the optimal parameters for prediction, abnormal operating states can be determined.
This improved the prediction accuracy and monitoring precision of abnormal behavior monitoring of heating users, effectively reducing the occurrence of false alarms and missed alarms.
Smart Images
Figure CN115423015B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of heat user abnormal behavior monitoring, and particularly relates to a heat user abnormal behavior monitoring method and system. BACKGROUND
[0002] With the rapid development of economy, the heat consumption of heat users is increasing. If the heat user behavior is abnormal, it will not only cause safety hazards to the heat user, but also greatly damage the interests of the heat company and cause losses to the heat system. The traditional heat user abnormality detection method has a large dependence on people, and usually relies on periodic inspection, heat user reporting and other means, which has low efficiency.
[0003] At present, the heat user abnormal behavior is usually monitored based on a neural network. The extreme learning machine (ELM) is more suitable for application in the monitoring field due to its fast training speed, simplicity and other advantages. However, since the input layer weight and the hidden layer threshold value are randomly input, it may affect the accuracy of the output and thus affect the monitoring effect. SUMMARY
[0004] The purpose of the present application is to provide a heat user abnormal behavior monitoring method and system to solve the problem of poor monitoring effect.
[0005] To achieve the above purpose, the present application provides the following solutions.
[0006] A heat user abnormal behavior monitoring method comprises the following steps:
[0007] An influencing factor affecting the heat consumption of a heat user is obtained, and the influencing factor is screened to determine a feature vector; the influencing factor includes a climate influencing factor and a time influencing factor; the climate influencing factor includes the daily maximum temperature, the daily minimum temperature, the current temperature, the daily average wind speed, the relative humidity and the daily sunshine duration; the time influencing factor includes the month and the date type;
[0008] The feature vector is preprocessed to generate a preprocessed feature vector;
[0009] The preprocessed feature vector is divided into a training set, a test set and a prediction set;
[0010] A variation prediction learning machine prediction model is constructed using the training set;
[0011] The test set is input into the variation prediction learning machine prediction model to output the optimal input layer weight and the optimal hidden layer threshold value;
[0012] The prediction set, the optimal input layer weights, and the optimal hidden layer thresholds are input into the mutation prediction learning machine prediction model to determine the predicted value of the heat consumption of the heat user.
[0013] The actual heat consumption of the heat user is obtained, and the current operating status of the heat user is determined based on the deviation between the predicted heat consumption and the actual heat consumption; the current operating status includes normal operating status and abnormal operating status.
[0014] Optionally, the step of acquiring the influencing factors affecting the heat consumption of heat users, filtering the influencing factors, and determining the feature vector specifically includes:
[0015] Using formula Factors with relevance higher than a set threshold are selected and defined as feature vectors; where P is the relevance; M... a This is a sample that affects users' heat consumption; The average value of the sample that affects user heat consumption; M b A sample of influencing factors; is the average value of the influencing factors in the sample; k is the sample size; i is the sample number.
[0016] Optionally, the step of preprocessing the feature vector to generate a preprocessed feature vector specifically includes:
[0017] Determine whether the feature vector is abnormal; if it is abnormal, remove the feature vector.
[0018] If no anomalies are found, fill in the feature vector and determine the filled feature vector;
[0019] The filled feature vector is normalized to generate a preprocessed feature vector.
[0020] Optionally, the mutation prediction learning machine prediction model is:
[0021] u = 1, 2, ..., C, v = 1, 2, ..., M; where f(x) v ) represents the mutation prediction learning machine prediction model; C represents the number of hidden layer nodes; M represents the number of output layer nodes; x v For the sample feature set; w u For input weights; b u The threshold of the hidden layer neurons; β u The connection weights between the u-th hidden layer neuron and the output neuron are: g(·) is the activation function; t v is the output of the mutation prediction learning machine prediction model; v is the output layer node.
[0022] Optionally, the test set is input into the mutation prediction learning machine prediction model to output optimal input layer weights and optimal hidden layer thresholds, and the method specifically comprises the following steps:
[0023] The test set is input into the mutation prediction learning machine prediction model, and the region of the individual input layer weights and the hidden layer thresholds is taken as the best region;
[0024] The first position update is performed in the best region, and the current position is locked;
[0025] Continuous searching is performed around the locked current position, the second position update is performed, and the position after the second update is locked;
[0026] The position update is continuously performed by moving towards the position after the second update until the target position is reached;
[0027] When the number of iterations reaches the set number of iterations or the fitness reaches the set fitness, the optimal input layer weights and the optimal hidden layer thresholds are output.
[0028] Optionally, the first position update is performed in the best region, and the current position is locked, and the method specifically comprises the following steps:
[0029] The formula is used The first position update is performed in the best region, and the current position is locked; wherein Z n,i is the latest position after the current individual update; η is a constant in [1.5, 2]; P l is a chaos operator; Z b is the region of the previous best position; Z m is the average position of the current population; Z i is the position before the current individual update, and t is time.
[0030] Optionally, continuous searching is performed around the locked current position, the second position update is performed, and the position after the second update is locked, and the method specifically comprises the following steps:
[0031] Continuous searching is performed around the locked current position, and the formula is used The second position update is performed, and the position after the second update is locked; wherein e(i) and f(i) are two positions of the individual in polar coordinates; ω(t) is a weight factor; Z i+1 is the second position.
[0032] Optionally, the position update is continuously performed by moving towards the position after the second update until the target position is reached, and the method specifically comprises the following steps:
[0033] The formula is used Continuously moving to the position updated after the second time of updating the lock until moving to the target position; wherein, e k (i) and f k (i) are the other two positions of the individual in polar coordinates; c1=c2=1.
[0034] A heat user abnormal behavior monitoring system, comprising:
[0035] A screening module is configured to acquire influence factors affecting heat consumption of a heat user, screen the influence factors, and determine a feature vector; the influence factors include climate influence factors and time influence factors; the climate influence factors include daily maximum temperature, daily minimum temperature, current temperature, daily average wind speed, relative humidity, and daily sunshine duration; the time influence factors include month and date type;
[0036] A preprocessing module is configured to preprocess the feature vector to generate a preprocessed feature vector;
[0037] A division module is configured to divide the preprocessed feature vector into a training set, a test set, and a prediction set;
[0038] A variation prediction learning machine prediction model construction module is configured to construct a variation prediction learning machine prediction model by using the training set;
[0039] An optimal input layer weight and optimal hidden layer threshold output module is configured to input the test set into the variation prediction learning machine prediction model to output an optimal input layer weight and an optimal hidden layer threshold;
[0040] A predicted value of heat consumption of a heat user determination module is configured to input the prediction set, the optimal input layer weight, and the optimal hidden layer threshold into the variation prediction learning machine prediction model to determine a predicted value of heat consumption of a heat user;
[0041] A current running state determination module is configured to acquire an actual value of heat consumption of a heat user, and determine a current running state of the heat user according to a deviation between the predicted value of heat consumption of the heat user and the actual value of heat consumption of the heat user; the current running state includes a normal running state and an abnormal running state.
[0042] Optionally, the screening module specifically includes:
[0043] A screening unit is configured to use a formula to screen out influence factors with a correlation degree higher than a set threshold to determine a feature vector; wherein, P is the correlation degree; M a is a sample affecting heat consumption of a user; M a is an average value of the sample affecting heat consumption of a user; M bThe sample of the influencing factor; M b The average value of the sample of the influencing factor; k is the sample number; i is the sample serial number.
[0044] According to the specific embodiments provided by the present application, the following technical effects are disclosed: the present application provides a heat user abnormal behavior monitoring method and system, the preprocessed feature vectors are divided into a training set, a test set and a prediction set, and a variation prediction learning machine prediction model is constructed and optimized by using the training set and the test set, and the optimal input layer weight and the optimal hidden layer threshold value are output, finally, based on the optimal input layer weight and the optimal hidden layer threshold value, the heat consumption of the heat user is predicted by using the variation prediction learning machine prediction model, which can effectively improve the prediction accuracy and further improve the monitoring accuracy. BRIEF DESCRIPTION OF DRAWINGS
[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below, and obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.
[0046] Figure 1 The flow chart of the heat user abnormal behavior monitoring method provided by the present application;
[0047] Figure 2 The structure diagram of the heat user abnormal behavior monitoring system provided by the present application;
[0048] Figure 3 The comparison diagram of the predicted daily heat consumption of a user and the measured result provided by the present application. DETAILED DESCRIPTION
[0049] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application, and obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0050] The purpose of the present application is to provide a heat user abnormal behavior monitoring method and system, which can effectively improve the prediction accuracy and further improve the monitoring accuracy.
[0051] In order to make the above-mentioned purposes, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.
[0052] Figure 1As shown in a flow chart of a heat user abnormal behavior monitoring method provided by the present application, Figure 1 a heat user abnormal behavior monitoring method comprises the following steps:
[0053] Step 101: acquiring influence factors affecting heat consumption of a heat user, screening the influence factors, and determining a feature vector; the influence factors include climate influence factors and time influence factors; the climate influence factors include daily maximum temperature, daily minimum temperature, current temperature, daily average wind speed, relative humidity, and daily sunshine duration; the time influence factors include month and date type.
[0054] The correlation degree of the above influence factors and user heat consumption is calculated by using formula (1), and the factors with strong correlation degree are extracted as feature parameters and then input into a prediction model.
[0055]
[0056] wherein, M a , M b is a data sample of influence factors affecting user heat consumption; k is the number of data samples; P is in (-1, 1), P is in (-1, 0) indicating negative correlation, P is in (0, 1) indicating positive correlation, and the greater the absolute value of P is, the stronger the correlation degree is.
[0057] Step 102: preprocessing the feature vector to generate a preprocessed feature vector.
[0058] The data is preprocessed. First, formula (2) is used to check whether the data is abnormal, if it is abnormal, it is excluded, and formula (3) is used for filling. In addition, since the input data has a large difference in dimension, formula (4) is used for normalization processing.
[0059]
[0060] wherein, d k is a heat consumption change rate; d a and d b are adjacent two heat consumption data; τ is a change rate threshold.
[0061]
[0062] wherein, d r is the average heat consumption of the missing period; d e and d f are the readings of the front heat meter and the rear heat meter before the missing period, respectively; d l is the missing period.
[0063]
[0064] where t' is the normalized data; t is the original data; t min , t max are the minimum value and the maximum value of the input data, respectively.
[0065] Step 103: dividing the preprocessed feature vector into a training set, a test set, and a prediction set.
[0066] Step 104: constructing a mutation prediction learning machine prediction model using the training set.
[0067] The mutation extreme learning machine prediction model is constructed to predict the user's heat consumption. The steps are as follows:
[0068] 1) dividing the preprocessed sample set into a training set, a test set, and a prediction set.
[0069] 2) setting the number of populations, the number of iterations, and setting the position of each individual as Z = (z 1 ,z 2 ,…,z t ). Input the training set sample (x j ,y j ), where x j = (x j1 ,x j2 ,…,x jn ) T , y j = (y j1 ,y j2 ,…,y jm ) T , determine the number of hidden nodes, input the input weight and threshold value corresponding to each individual, select the sigmoid function as the activation function, and construct the network structure of the extreme learning machine according to the above, construct the prediction model, and train it.
[0070] The expression of the extreme learning machine is shown in formula (5).
[0071]
[0072] where f(x v ) is the mutation prediction learning machine prediction model; C is the number of hidden layer nodes; M is the number of output layer nodes; x v is the sample feature set; w u is the input weight; b u is the hidden layer neuron threshold value; β u is the connection weight value between the u-th hidden layer neuron and the output neuron; g(·) is the activation function; t v is the output result of the mutation prediction learning machine prediction model; and v is the output layer node.
[0073] The matrix expression of the above formula is shown as formula (6).
[0074] Hβ=T (6)
[0075] Wherein, H is the output matrix of the hidden layer; β is the output weight matrix; T is the expected value matrix.
[0076] Based on the least square criterion, the output connection weight Can be obtained by formula (7).
[0077]
[0078] Wherein, H + Is the Moore-Penrose generalized inverse matrix of H.
[0079] Step 105: input the test set into the mutation prediction learning machine prediction model, output the optimal input layer weight and the optimal hidden layer threshold value.
[0080] The step 105 specifically comprises:
[0081] The test set is input into the mutation prediction learning machine prediction model, and the region where the individual input layer weight and the hidden layer threshold value are located is regarded as the best region.
[0082] The first position update is performed in the best region, and the current position is locked.
[0083] Constant search is performed around the locked current position, the second position update is performed, and the position after the second update locking is locked.
[0084] The position update is continuously performed by moving to the position after the second update locking until the target position is moved to.
[0085] When the number of iterations reaches the iteration number setting value or the fitness reaches the fitness setting, the optimal input layer weight and the optimal hidden layer threshold value are output.
[0086] Wherein, the first position update is performed in the best region, and the current position is locked, specifically comprising:
[0087] The test set is input into the above model, the mean square error between the predicted value and the actual value is obtained as the fitness, the region where the initial input weight and the threshold value of the individual are located is regarded as the best region, the first position update is performed according to formula (8), and the current fitness is calculated.
[0088]
[0089] Wherein, Z n,iis the latest position of the current individual after updating; η is a constant in [1.5, 2]; P l = 4P l × (1 - sinP l ), P l is a chaotic operator; Z b is the region where the previous best position is located; Z m is the average position of the current population; Z i is the position of the current individual before updating, and t is time.
[0090] Constantly search around the current position locked, perform the second position updating, lock the position after the second updating, and specifically include:
[0091] Constantly search around the new position obtained, lock the second position according to formula (9), and calculate the fitness.
[0092]
[0093]
[0094]
[0095]
[0096] wherein σ is a constant in [5, 10]; R is a constant in [0.5, 2]; r is a random number uniformly distributed in [0, 1]; e(i) and y(i) are two positions of the individual in the polar coordinate, and the values are usually (-1, 1); ω(t) is a weight factor, s is the current iteration number, S is the maximum iteration number; θ(i) is the polar angle in the polar coordinate; r(i) is the polar radius in the polar coordinate; er(i) is the value of the y-axis in the rectangular coordinate system converted from the polar coordinate; fr(i) is the value of the x-axis in the rectangular coordinate system converted from the polar coordinate.
[0097] Constantly move to the position after the second updating and lock to update the position until the target position is reached and the iteration number reaches the iteration number setting value or the fitness reaches the fitness setting value, and output the optimal input layer weight and the optimal hidden layer threshold value, and specifically include:
[0098] Subsequently, constantly move to the position after the second updating and lock, until the target position is reached. Update the position using formula (13). If the iteration number is reached or the fitness setting value is reached, the updating is ended, and the optimal input weight and threshold value are output. Otherwise, constantly repeat the updating and locking step until the condition is met.
[0099]
[0100]
[0101]
[0102]
[0103] where e k (i) and f k (i) are the other two positions of the individual in polar coordinates; c1, c2 are constants, usually taking 1.
[0104] Step 106: inputting the prediction set, the optimal input layer weight and the optimal hidden layer threshold value into the mutation prediction learning machine prediction model to determine the predicted value of heat consumption of the heat user.
[0105] Step 107: obtaining the actual value of heat consumption of the heat user, and determining the current running state of the heat user according to the deviation between the predicted value of heat consumption of the heat user and the actual value of heat consumption of the heat user; the current running state includes a normal running state and an abnormal running state.
[0106] It is judged whether the heat user is abnormal. The deviation between the predicted value and the actual value of the user heat consumption is calculated by using formula (17). If the deviation exceeds the threshold range, it is considered that the heat user is abnormal, and an alarm is given, otherwise it is considered that the user is normally running.
[0107] δ t = |M t -N t | (17)
[0108] where δ t is the deviation value at time t; M t is the predicted heat consumption of the user at time t; N t is the actual heat consumption of the user at time t.
[0109] Figure 2 The structure diagram of the heat user abnormal behavior monitoring system provided by the application is shown in FIG. 1, which is a heat user abnormal behavior monitoring system, comprising: Figure 2
[0110] The screening module 201 is used for obtaining influence factors affecting heat consumption of the heat user, screening the influence factors, and determining a feature vector; the influence factors include climate influence factors and time influence factors; the climate influence factors include daily maximum temperature, daily minimum temperature, current temperature, daily average wind speed, relative humidity and daily sunshine duration; the time influence factors include month and date type.
[0111] The screening module 201 specifically comprises a screening unit, which is used for screening the influence factors by using formula Screening the impact factors with a correlation degree higher than a set threshold value to determine a feature vector; wherein P is the correlation degree; M a is a sample of influencing user heat consumption; is an average value of the sample of influencing user heat consumption; M b is a sample of influencing factors; is an average value of the sample of influencing factors; k is the number of samples; and i is the sample serial number.
[0112] The preprocessing module 202 is configured to preprocess the feature vector to generate a preprocessed feature vector.
[0113] The division module 203 is configured to divide the preprocessed feature vector into a training set, a test set, and a prediction set.
[0114] The mutation prediction learning machine prediction model construction module 204 is configured to construct a mutation prediction learning machine prediction model using the training set.
[0115] The optimal input layer weight and optimal hidden layer threshold output module 205 is configured to input the test set into the mutation prediction learning machine prediction model to output an optimal input layer weight and an optimal hidden layer threshold.
[0116] The predicted value of heat consumption of a heat user determination module 206 is configured to input the prediction set, the optimal input layer weight, and the optimal hidden layer threshold into the mutation prediction learning machine prediction model to determine a predicted value of heat consumption of a heat user.
[0117] The current running state determination module 207 is configured to obtain an actual value of heat consumption of a heat user and determine a current running state of the heat user according to a deviation between the predicted value of heat consumption of the heat user and the actual value of heat consumption of the heat user; the current running state includes a normal running state and an abnormal running state.
[0118] The heating period of a certain user from November 2020 to March 31, 2021 is selected, and 3744 hours of heat consumption are collected as samples at an interval of 1 hour to establish a heat consumption prediction model, and the data from March 28, 2021 to March 31, 2021 are selected to detect the monitoring effect of the application.
[0119] First, the features are screened, and the correlation degrees of the calculated influencing factors and user heat consumption are shown in Table 1.
[0120] Table 1 Correlation degrees of influencing factors and user heat consumption
[0121]
[0122] After screening, the daily maximum temperature, the daily minimum temperature, the current temperature, the relative humidity and the daily sunshine time are selected as the input feature vectors.
[0123] Then, the data is processed, the abnormal data is eliminated, and the missing values are filled in. Since the input data set includes temperature, humidity, time and other different characteristics, the dimensions differ greatly, and normalization processing is also required.
[0124] Then, the processed data is input into the user heat consumption prediction model of the variation extreme learning machine, and then the difference between the predicted value and the actual value is compared, and the result is as shown in Figure 3
[0125] Figure 3 It can be concluded that the predicted value of the user heat consumption deviates greatly from the actual value from 18:00 to 20:00 on March 30, and it is considered that there is an abnormal situation. After investigation, it is found that the user indeed has an abnormal situation in this period. It is proved that the present application can well monitor the abnormal behavior of the user, and the accuracy is high.
[0126] The present application alarms the user abnormal behavior by calculating the deviation between the predicted value and the actual value of the user heat consumption, so the accuracy of the predicted value of the heat consumption is required. Although the traditional extreme learning machine has the advantages of few training parameters, fast learning speed and strong generalization ability, the input layer weight and the hidden layer threshold are randomly input, and the training effect may be affected to a certain extent. The present application proposes a variation learning machine model, which continuously optimizes and updates to obtain the optimal input weight and threshold, and then uses the model for prediction, which can effectively improve the prediction accuracy and further improve the monitoring accuracy.
[0127] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.
[0128] In the present application, specific examples are used to illustrate the principles and implementation modes of the present application. The above embodiment is only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In view of the above, the content of the specification should not be understood as a limitation of the present application.
Claims
1. A method for monitoring abnormal behavior of heating users, characterized in that, include: The factors influencing the heat consumption of heating users are identified, and these factors are filtered to determine feature vectors. These factors include climate factors and time factors. The climate factors include the highest temperature of the day, the lowest temperature of the day, the current temperature, the average wind speed of the day, relative humidity, and the duration of sunshine of the day. The time factors include the month and date type. The feature vector is preprocessed to generate a preprocessed feature vector; The preprocessed feature vectors are divided into a training set, a test set, and a prediction set. A mutation prediction learning machine prediction model is constructed using the training set; The test set is input into the mutation prediction learning machine prediction model, and the optimal input layer weights and optimal hidden layer thresholds are output, specifically including: The test set is input into the mutation prediction learning machine prediction model, and the region where the individual input layer weights and hidden layer thresholds are located is taken as the optimal region. The first position update within the optimal area, and the current position locked, specifically includes: Using formula The position is updated for the first time within the optimal region, and the current position is locked; where Z n,i The current individual's latest position after the update; η is a constant within [1.5, 2]; P l Z is a chaotic operator. b This refers to the area where the previous optimal position was located; Z m Z represents the average position of the current population. i The current position of the individual before the update, where t is the time; The system continuously searches around the current locked position, performs a second position update, and locks the position after the second update. The position is continuously updated by moving towards the position locked after the second update until the target position is reached. When the number of iterations reaches the set value or the fitness reaches the set value, the optimal input layer weight and the optimal hidden layer threshold are output. The prediction set, the optimal input layer weights, and the optimal hidden layer thresholds are input into the mutation prediction learning machine prediction model to determine the predicted value of the heat consumption of the heat user. The actual heat consumption of the heat user is obtained, and the current operating status of the heat user is determined based on the deviation between the predicted heat consumption and the actual heat consumption; the current operating status includes normal operating status and abnormal operating status.
2. The method for monitoring abnormal behavior of heating users according to claim 1, characterized in that, The process of acquiring the influencing factors affecting the heat consumption of heating users, filtering the influencing factors, and determining the feature vector specifically includes: Using formula Factors with relevance higher than a set threshold are selected and identified as feature vectors; where P is the relevance; M... a This is a sample that affects users' heat consumption; The average value of the sample that affects user heat consumption; M b A sample of influencing factors; is the average value of the influencing factors in the sample; k is the sample size; i is the sample number.
3. The method for monitoring abnormal behavior of heating users according to claim 2, characterized in that, The step of preprocessing the feature vector to generate a preprocessed feature vector specifically includes: Determine whether the feature vector is abnormal; if it is abnormal, remove the feature vector. If no anomalies are found, fill in the feature vector and determine the filled feature vector; The filled feature vector is normalized to generate a preprocessed feature vector.
4. The method for monitoring abnormal behavior of heating users according to claim 3, characterized in that, The mutation prediction learning machine prediction model is as follows: Where, f(x) v ) represents the mutation prediction learning machine prediction model; C represents the number of hidden layer nodes; M represents the number of output layer nodes; x v For the sample feature set; w u For input weights; b u The threshold of the hidden layer neurons; β u The connection weights between the u-th hidden layer neuron and the output neuron are: g(·) is the activation function; t v is the output of the mutation prediction learning machine prediction model; v is the output layer node.
5. The method for monitoring abnormal behavior of heating users according to claim 1, characterized in that, The process of continuously searching around the locked current position, performing a second position update, and locking the position after the second update specifically includes: Continuously search around the locked current position, using the formula Perform a second position update and lock the position after the second update; where e(i) and f(i) are the two positions of the individual in polar coordinates; ω(t) is the weight factor; Z i+1 This is the second position.
6. The method for monitoring abnormal behavior of heating users according to claim 5, characterized in that, The step of continuously moving to the locked position after the second update to update the position until the target position is reached specifically includes: Using formula The position is continuously updated by moving towards the position locked after the second update until the target position is reached; where e k (i) and f k (i) represents the other two positions of the individual in polar coordinates; c1 = c2 = 1.
7. A system for monitoring abnormal behavior of heating users, characterized in that, The abnormal behavior monitoring system for heat users executes the abnormal behavior monitoring method for heat users according to any one of claims 1-6, wherein the abnormal behavior monitoring system for heat users comprises: The filtering module is used to acquire factors affecting the heat consumption of heating users, filter the factors, and determine feature vectors. The factors include climate factors and time factors. The climate factors include the highest temperature of the day, the lowest temperature of the day, the current temperature, the average wind speed of the day, the relative humidity, and the sunshine duration of the day. The time factors include the month and date type. The preprocessing module is used to preprocess the feature vector to generate a preprocessed feature vector; The partitioning module is used to divide the preprocessed feature vectors into a training set, a test set, and a prediction set. A mutation prediction learning machine prediction model building module is used to build a mutation prediction learning machine prediction model using the training set. The optimal input layer weights and optimal hidden layer thresholds output module is used to input the test set into the mutation prediction learning machine prediction model and output the optimal input layer weights and optimal hidden layer thresholds. The module for determining the predicted value of heat consumption of heat users is used to input the prediction set, the optimal input layer weights and the optimal hidden layer thresholds into the mutation prediction learning machine prediction model to determine the predicted value of heat consumption of heat users. The current operating status determination module is used to obtain the actual value of the heat consumption of the heat user, and determine the current operating status of the heat user based on the deviation between the predicted value of the heat consumption and the actual value of the heat consumption; the current operating status includes normal operating status and abnormal operating status.
8. The abnormal behavior monitoring system for heat users according to claim 7, characterized in that, The filtering module specifically includes: Filtering unit, used to utilize formulas Factors with relevance higher than a set threshold are selected and identified as feature vectors; where P is the relevance; M... a This is a sample that affects users' heat consumption; The average value of the sample that affects user heat consumption; M b A sample of influencing factors; is the average value of the influencing factors in the sample; k is the sample size; i is the sample number.
Citation Information
Patent Citations
Intelligent electric power searching method based on load prediction
CN114707422A