A method and system for early warning of electrical fire in TOD buildings

Abstract

The present invention relates to the field of fire warning technology, and specifically to an electrical fire warning method and system for a TOD building, including real-time acquisition of current data of multiple power units; predicting the current data corresponding to each power unit based on a preset fire prediction neural network to obtain a prediction result; judging the initial fire assessment coefficient corresponding to the suspected fire unit according to the arc state and confidence change in the prediction result corresponding to each power unit; real-time acquisition of the corresponding monitoring image of the suspected fire unit; correcting the initial fire assessment coefficient based on the smoke state in the monitoring image to obtain a fire assessment coefficient; judging whether to send fire warning information to the target terminal based on the fire assessment coefficient and a preset fire threshold. The present invention not only relies on the fire prediction neural network results, but also considers the complexity of the current time series data, and makes a comprehensive judgment in combination with the confidence of the network to deal with the problem of not being able to cover all situations.

Description

Technical Field

[0001] The present invention relates to the technical field of fire warning, and in particular to an electrical fire warning method and system for a TOD building. Background Art

[0002] TOD (Transit-Oriented Development) is an urban development concept that focuses on high-density, high-mixed-use development around public transportation hubs to reduce cities’ dependence on private cars and promote sustainable development and community convenience.

[0003] The main characteristics of TOD include high-density development: that is, high-density development around public transportation stations. By concentrating commercial and office facilities around public transportation hubs, high-density development can improve land use efficiency and at the same time attract more people to live and work in these areas, thereby improving land use efficiency.

[0004] However, at the same time, there are many types of electrical equipment in TOD buildings, the power load is large, and the electrical circuits are complex. The probability of fire in electrical equipment is high. In addition, the population density in TOD buildings is high. When there is a false alarm in the fire warning, it will cause unnecessary panic. If there is a missed alarm, it will cause serious fire hazards. Therefore, a more accurate warning method for fire risks that meets the needs of TOD buildings is now needed. Summary of the invention

[0005] In order to solve the above technical problems, the purpose of the present invention is to provide an electrical fire warning method and system for a TOD building. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present application provides an electrical fire early warning method for a TOD building, the method comprising:

[0007] Acquire current data of multiple power consumption units in real time, and convert it into current time series data corresponding to each power consumption unit at the current moment;

[0008] Based on a preset fire prediction neural network, arc prediction is performed on the current time series data corresponding to each of the power consumption units at the current moment to obtain a prediction result at the current moment, wherein the prediction result includes an arc state and a first confidence level;

[0009] The prediction results corresponding to the current moment and the historical moment are converted to obtain the time series prediction data corresponding to each of the power consumption units;

[0010] Determine the fire initial assessment coefficient corresponding to the suspected fire unit according to the arc state and confidence change in the time series prediction data corresponding to each power consumption unit;

[0011] Acquire the corresponding monitoring image of the suspected fire unit in real time;

[0012] Correcting the initial fire assessment coefficient based on the smoke state in the monitoring image to obtain a fire assessment coefficient;

[0013] Based on the fire assessment coefficient and the preset fire threshold, it is determined whether to send fire warning information to the target terminal.

[0014] In combination with the first aspect, in a possible implementation, arc prediction is performed on the current time series data corresponding to each of the power consumption units at the current moment based on a preset fire prediction neural network to obtain a prediction result at the current moment, including:

[0015] The current time series data are decomposed differently to obtain two sets of current components;

[0016] Perform characteristic calculations based on the two current component sets to obtain two current characteristic vectors;

[0017] Using a feature extraction network model to extract features from each current feature vector to obtain a feature value vector;

[0018] Based on three preset prediction models, state prediction is performed on the eigenvalue vector and the two current eigenvectors respectively to obtain state prediction results, wherein the state prediction results include three predicted states and a second confidence level corresponding to each predicted state;

[0019] A prediction result is obtained based on the state prediction result judgment.

[0020] In combination with the first aspect, in a possible implementation manner, characteristic calculations are performed based on the two current component sets respectively to obtain two current characteristic vectors, including:

[0021] Performing sample entropy calculation on the components in the current component set to obtain a sample entropy vector;

[0022] Classifying the components in the current component set to obtain a high-frequency component set, a medium-frequency component set and a low-frequency component set;

[0023] Performing component energy calculation on the high-frequency component set and the low-frequency component set to obtain total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively;

[0024] An energy ratio is obtained by performing ratio calculation based on the total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively;

[0025] The sample entropy vector and the energy ratio are concatenated and normalized to obtain a current feature vector.

[0026] In combination with the first aspect, in a possible implementation manner, obtaining a prediction result based on the state prediction result judgment includes:

[0027] confirming the frequencies of obtaining different prediction states according to the three prediction states;

[0028] Confirming the arc state based on the frequency of all the predicted states;

[0029] A first confidence is calculated based on the frequencies of all the predicted states and all the second confidences.

[0030] In combination with the first aspect, in a possible implementation, determining the fire initial assessment coefficient corresponding to the suspected fire unit according to the arc state and confidence change in the time series prediction data corresponding to each of the power consumption units includes:

[0031] When the arc state in the time series prediction data corresponding to the power consumption unit changes, the initial fire assessment coefficient is a preset value;

[0032] When the arc state in the time series prediction data corresponding to the power consumption unit has not changed, a straight line fitting is performed on the first confidence level in the time series prediction data, and the fire initial assessment coefficient is calculated according to the slope of the straight line fitting;

[0033] It is determined whether the power unit is a suspected fire unit according to the initial fire assessment coefficient corresponding to the power unit and a preset initial threshold value, and the initial fire assessment coefficient of the suspected fire unit is greater than the preset initial threshold value.

[0034] In a second aspect, the present application also provides an electrical fire warning system for a TOD building, comprising:

[0035] A current data acquisition module is used to acquire the current data of multiple power consumption units in real time and convert it into the current time series data corresponding to each of the power consumption units at the current moment;

[0036] A prediction module, configured to perform arc prediction on the current time series data corresponding to each of the power consumption units at the current moment based on a preset fire prediction neural network to obtain a prediction result at the current moment, wherein the prediction result includes an arc state and a first confidence level;

[0037] A data integration module, used to convert the prediction results corresponding to the current moment and the historical moment into the time series prediction data corresponding to each of the power consumption units;

[0038] A logic module, used to determine the initial fire assessment coefficient corresponding to the suspected fire unit according to the arc state and confidence change in the time series prediction data corresponding to each of the power consumption units;

[0039] An image data module, used for acquiring corresponding monitoring images of the suspected fire unit in real time;

[0040] A correction module, used for correcting the initial fire assessment coefficient based on the smoke state in the monitoring image to obtain a fire assessment coefficient;

[0041] The early warning module is used to determine whether to send fire early warning information to the target terminal based on the fire assessment coefficient and the preset fire threshold.

[0042] In conjunction with the second aspect, in a possible implementation manner, the prediction module includes:

[0043] A data decomposition module is used to decompose the current time series data in different ways to obtain two sets of current components;

[0044] A feature calculation module, used for performing feature calculations based on the two current component sets respectively to obtain two current feature vectors;

[0045] A feature extraction module, used for extracting features from each current feature vector using a feature extraction network model to obtain a feature value vector;

[0046] A state prediction module, used for performing state prediction on the eigenvalue vector and the two current eigenvectors respectively based on three preset prediction models to obtain state prediction results, wherein the state prediction results include three predicted states and a second confidence level corresponding to each predicted state;

[0047] A result merging module is used to determine and obtain a prediction result based on the state prediction result.

[0048] In conjunction with the second aspect, in a possible implementation manner, the feature calculation module includes:

[0049] A sample entropy calculation module, used to perform sample entropy calculation on the components in the current component set to obtain a sample entropy vector;

[0050] A component classification module, used for classifying the components in the current component set to obtain a high-frequency component set, a medium-frequency component set and a low-frequency component set;

[0051] An energy calculation module, used to perform component energy calculation on the high-frequency component set and the low-frequency component set to obtain total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively;

[0052] A ratio module, configured to calculate an energy ratio based on the total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively;

[0053] The splicing module is used to splice the sample entropy vector and the energy ratio and normalize them to obtain a current feature vector.

[0054] In conjunction with the second aspect, in a possible implementation manner, the result merging module includes:

[0055] A frequency statistics module, used for confirming the frequencies of obtaining different prediction states according to the three prediction states;

[0056] A state confirmation module, used to confirm the arc state according to the frequencies of all the predicted states;

[0057] A confidence calculation module is used to calculate a first confidence based on the frequencies of all the predicted states and all the second confidences.

[0058] In conjunction with the second aspect, in a possible implementation manner, the logic module includes:

[0059] A first judgment module, configured to determine that when the arc state in the time series prediction data corresponding to the power consumption unit changes, the fire initial assessment coefficient is a preset value;

[0060] A second judgment module is used for performing a straight line fitting on the first confidence level in the time series prediction data when the arc state in the time series prediction data corresponding to the power consumption unit has not changed, and calculating the fire initial assessment coefficient according to the slope of the straight line fitting;

[0061] The third judgment module is used to judge whether the power unit is a suspected fire unit according to the initial fire assessment coefficient corresponding to the power unit and a preset initial threshold, and the initial fire assessment coefficient of the suspected fire unit is greater than the preset initial threshold.

[0062] The present invention has the following beneficial effects:

[0063] The present invention adopts a dual-dimensional combined judgment method, that is, only when the arc that causes the electrical fire exists and is accompanied by smoke, can it be judged that an electrical fire has occurred. In addition, the present invention does not only rely on the prediction results of the fire prediction neural network to make judgments. In order to overcome the complexity of the current time series data brought by various electrical equipment contained in the power unit, such as capacitors, resistors and other electrical components, the confidence of the fire prediction neural network is also combined for comprehensive judgment. In this way, it is possible to effectively deal with the problem that the preset fire prediction neural network may not be able to fully cover all situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0065] Figure 1 A flow chart of an electrical fire warning method for a TOD building provided in Embodiment 1 of the present invention;

[0066] Figure 2 A schematic diagram of the flow of step S2 provided in Embodiment 2 of the present invention;

[0067] Figure 3 A schematic diagram of the flow of step S6 provided in Embodiment 3 of the present invention;

[0068] Figure 4 This is a schematic diagram of the structure of the electrical fire warning system for a TOD building described in Example 4 of the present invention. DETAILED DESCRIPTION

[0069] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the electrical fire warning method and system for a TOD building proposed by the present invention, its specific implementation method, structure, features and effects are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0070] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0071] Embodiment 1:

[0072] The specific scheme of the electrical fire early warning method for TOD buildings provided by the present invention is described in detail below with reference to the accompanying drawings.

[0073] See also Figure 1 , which shows a flow chart of an electrical fire warning method for a TOD building provided by an embodiment of the present invention, wherein in this embodiment, steps S1 to S7 are included.

[0074] S1. Acquire current data of multiple power consumption units in real time and convert it into current time series data corresponding to each power consumption unit at the current moment;

[0075] In this embodiment, the power consumption unit refers to the area where a merchant is located or the area where the office area is located in the TOD building. In addition, since the application object involved in this embodiment is a TOD building, the concept of the power consumption unit can also cover facilities such as lighting groups in the underground area. As for the current data mentioned in this embodiment, it can be data directly obtained through an electric meter, or it can be data collected through a current transformer in the Internet of Things technology. These current data acquisition methods all belong to the scope of the prior art, so this embodiment does not limit these specific data acquisition methods, and those skilled in the art can flexibly select appropriate current data acquisition methods according to actual conditions.

[0076] Furthermore, in the above steps, the acquisition frequency of the current data is 12.8 kHz. Meanwhile, the duration of the current time series data is 2 s.

[0077] S2. Based on a preset fire prediction neural network, arc prediction is performed on the current time series data corresponding to each of the power consumption units at the current moment to obtain a prediction result at the current moment, wherein the prediction result includes an arc state and a first confidence level;

[0078] In this embodiment, the implementation of the fire prediction neural network can adopt a variety of types, including but not limited to back propagation (BP) neural network or recurrent neural network (RNN). The specific selection of which network type can be determined by those skilled in the art according to the needs, and this embodiment does not impose specific restrictions on this. At the same time, regarding the use of current time series data for model construction, model training and arc prediction process, it is common knowledge in the field and will not be elaborated in detail in this embodiment. As for the arc state, it generally includes two situations: presence and absence, and can be expressed in numerical form. Specifically, the value 0 represents the absence of an arc, and the value 1 represents the presence of an arc.

[0079] S3, converting the prediction results corresponding to the current moment and the historical moment to obtain the time series prediction data corresponding to each of the power consumption units;

[0080] In this embodiment, the historical moments referred to refer to multiple moments before the current moment. The time interval between the current moment and the historical moment is set to 1 second, and accordingly, the time interval between multiple historical moments is also 1 second. In this embodiment, the duration of the time series prediction data is determined to be 10 seconds. However, for those skilled in the art, the duration of the time series prediction data and the time interval between the current moment and the historical moment can be appropriately adjusted according to actual conditions, and no specific restrictions are made in this embodiment.

[0081] At the same time, it should be pointed out that in step S4 of this embodiment, the change rate of the first confidence level is used to further confirm the initial fire assessment coefficient. Therefore, it is recommended that the data on the first confidence level in the time series prediction data should not be too small, and at least 5 time series points should be included to ensure the accuracy and reliability of the data. At the same time, the accuracy of fire assessment can also be improved.

[0082] S4, determining the fire initial assessment coefficient corresponding to the suspected fire unit according to the arc state and confidence change in the time series prediction data corresponding to each of the power consumption units;

[0083] In order to facilitate understanding by those skilled in the art, this embodiment uses the time series prediction data of one power consumption unit as an example to illustrate how to determine the suspected fire unit. Specifically, this step includes steps S41 to S43.

[0084] S41, when the arc state in the time series prediction data corresponding to the power consumption unit changes, the fire initial assessment coefficient is a preset value;

[0085] In this embodiment, real-time current monitoring is mainly implemented for the power-consuming units that use electricity normally to ensure that the power use is safe. Therefore, in the current monitoring process, this embodiment is based on the information about the change of the arc state in the time series prediction data to judge the fire warning. If the time series prediction data all show that the arc state does not exist, that is, the value of the arc state is 0, it means that the power-consuming unit does not generate an arc during the power use process, and the probability of causing a fire is extremely low. However, once the arc state at the current moment in the time series prediction data changes, that is, the value originally 0 changes to 1, this means that the power-consuming unit generates an arc during the power use process. In view of the potential danger of the arc, it may ignite the surrounding combustible materials and then cause a fire. In order to ensure the accuracy of the fire risk assessment, the initial fire assessment coefficient needs to be adjusted in time. In this embodiment, it is preferred to set the initial fire assessment coefficient to a preset value of 1. Once the change of the arc state is detected, other fire prevention and control measures can be quickly started for monitoring, so that the source of the fire can be confirmed when the flame first appears, and it can be extinguished in the early stage of the fire.

[0086] S42: When the arc state in the time series prediction data corresponding to the power consumption unit has not changed, a straight line fitting is performed on the first confidence level in the time series prediction data, and the initial fire assessment coefficient is calculated based on the slope of the straight line fitting.

[0087] It should be noted that in this step, the calculation function of the fire initial assessment coefficient is as follows:

[0088] ;

[0089] in, represents the fire initial assessment coefficient; represents the slope of the straight line fit; Represents the adjustment coefficient, which is a real number greater than 1 and is used to adjust The amplitude approaches 1. The preferred value is 1.5. Those skilled in the art may adjust the value according to their needs, and no specific limitation is made in this embodiment.

[0090] This embodiment believes that the phenomenon that the first confidence level gradually decreases in the time dimension indicates that the credibility of the preset fire prediction neural network is gradually decreasing when predicting based on the current time series data. Although the arc state has not changed at the current moment, certain measures should be taken for such a state at this time in view of the phenomenon of reduced credibility. Therefore, in this embodiment, the fire initial assessment coefficient is determined by analyzing the degree of change in the slope of the straight line fitting. Specifically, when the slope decreases, it indicates that the speed at which the confidence level decreases is accelerating, which means that the current fire initial assessment coefficient should be changed, but it does not need to be equal to the preset value. Therefore, the above curve function can effectively describe this change trend. For professionals and technicians in this field, different slope threshold ranges can also be set to correspond to different fire initial assessment coefficients. In this embodiment, no specific restrictions are set for this, so that professionals and technicians can flexibly adjust according to actual conditions.

[0091] S43. Determine whether the power unit is a suspected fire unit based on the initial fire assessment coefficient corresponding to the power unit and a preset initial threshold, and the initial fire assessment coefficient of the suspected fire unit is greater than the preset initial threshold.

[0092] Specifically, in this embodiment, the preset initial threshold value can be set to 0.7, and the power consumption unit with the fire initial assessment coefficient greater than 0.7 is identified as a suspected fire unit, thereby reducing the possibility of misjudgment.

[0093] S5, acquiring the corresponding monitoring image of the suspected fire unit in real time;

[0094] It should be noted that in this embodiment, the monitoring image can be obtained by directly retrieving the monitoring video according to the area where the suspected fire unit is located. At the same time, if there are multiple monitorings in a suspected fire unit, the monitoring image in each monitoring video can be obtained separately. This is a prior art and is not limited in this embodiment.

[0095] S6, correcting the initial fire assessment coefficient based on the smoke state in the monitoring image to obtain a fire assessment coefficient;

[0096] Specifically, in this embodiment, if there is smoke in the monitoring image, there may be a small fire and smoke. Then the fire initial assessment coefficient should be adjusted to be greater than the preset fire threshold, so as to trigger an alarm and extinguish the fire as soon as possible to reduce the possibility of flame spread.

[0097] S7. Determine whether to send fire warning information to the target terminal based on the fire assessment coefficient and the preset fire threshold.

[0098] In this embodiment, the target terminal device may be a mobile phone terminal or a computer terminal equipped with an alarm program. In addition, the target terminal device may also be a fire alarm system in a TOD building, which can realize the linkage function with the alarm program. The specific type of terminal device should be determined by a technician with relevant professional knowledge according to the actual situation, and this embodiment does not impose any specific restrictions on this.

[0099] At the same time, the preset fire threshold mentioned in this step is preferably set to 0.85. This means that when the fire assessment coefficient exceeds 0.85, the method considers that there is a risk of electrical fire and immediately triggers a fire warning signal. At the same time, those skilled in the art can also set other values, and through the preset fire threshold, the method can issue a warning in time at the early stage of the fire, thereby buying valuable time for the evacuation of personnel and the implementation of fire response measures.

[0100] In this embodiment, the above steps are first used to identify the state of the arc that may cause a fire. Specifically, it includes identification in the dimension of current change to detect whether there is an arc that may cause a fire. Secondly, considering that in an actual power consumption environment, although an arc may be generated, some electrical materials have a certain fire resistance and can prevent the arc from causing a fire. Therefore, in this embodiment, in addition to identifying the arc state, a comprehensive judgment is made based on whether smoke appears in the arc generation area. In other words, this embodiment adopts a two-dimensional combined judgment method, that is, only when an arc that causes an electrical fire exists and is accompanied by smoke, it is judged that an electrical fire has been constituted. In this case, the power unit should be immediately extinguished to prevent the further spread of the fire.

[0101] In addition, in this embodiment, the judgment is not solely based on the prediction results of the fire prediction neural network. In order to overcome the complexity of the current timing data caused by the various electrical devices contained in the power unit, such as capacitors, resistors and other electrical components, this embodiment also combines the confidence of the fire prediction neural network for comprehensive judgment. Doing so can effectively deal with the problem that the preset fire prediction neural network may not be able to fully cover all situations. Even in some cases, there is a possibility of misjudgment of the fire prediction neural network, this embodiment further combines the performance of the corresponding monitoring image to further correct whether a fire warning signal should be sent. This judgment method that combines multiple aspects of information has a high fault tolerance and can more accurately identify and respond to fire risks.

[0102] Embodiment 2:

[0103] The difference between this embodiment and embodiment 1 is that a feature fusion prediction method is provided for step S2 to improve the prediction accuracy of the fire prediction neural network. Figure 2 The figure takes the current time series data of an electric unit as an example to illustrate how to perform arc prediction, see steps S21 to S25 for details.

[0104] S21, performing different decompositions on the current time series data to obtain two current component sets;

[0105] It should be noted that the decomposition mentioned in this step is a collective empirical mode decomposition model and a local mean decomposition model. That is, in this step, the current time series data is decomposed by the collective empirical mode decomposition model and the local mean decomposition model respectively to obtain multiple IMF (Intrinsic Mode Function) components and multiple PF (Product Function) components, and all IMF (Intrinsic Mode Function) components and PF (Product Function) components are respectively regarded as a current component set, so that the IMF component set and the PF component set can be obtained. In this embodiment, the local mean decomposition model is used to overcome the endpoint effect phenomenon existing in the decomposition process of the collective empirical mode decomposition model. At the same time, the collective empirical mode decomposition model is also used to overcome the problem of over-smoothing in the decomposition process of the local mean decomposition model. In this way, the two current components are obtained and then mixed prediction is performed to effectively improve the prediction accuracy.

[0106] In this step, this embodiment adopts two decomposition methods, namely, the ensemble empirical mode decomposition model and the local mean decomposition model, to achieve the purpose of decomposing the current time series data. Multiple IMF components can be obtained through the ensemble empirical mode decomposition model, which can reveal the intrinsic frequency characteristics of the current signal. At the same time, multiple PF components can be obtained using the local mean decomposition model, which can also reflect the frequency characteristics of the current signal, but at different angles.

[0107] Then, all IMF components and PF components are respectively formed into current component sets, and the IMF component set and the PF component set can be obtained. In this embodiment, the endpoint effect phenomenon that may occur in the decomposition process of the ensemble empirical mode decomposition model is considered, and the local mean decomposition model is adopted to effectively alleviate this problem. In addition, the local mean decomposition model may cause the problem of over-smoothing during the decomposition process, so the advantages of the ensemble empirical mode decomposition model are combined to retain signal details.

[0108] By mixing the current components obtained by the two decomposition methods, their respective advantages can be combined to effectively improve the sophistication of subsequent feature extraction of current time series data. This combined decomposition method can not only more accurately capture the intrinsic characteristics of the current signal, but also improve the reliability of the prediction results, providing more solid data support for subsequent power system analysis and decision-making.

[0109] S22, performing characteristic calculations based on the two current component sets respectively to obtain two current characteristic vectors;

[0110] Specifically, in this embodiment, a characteristic calculation process of a current component combination is taken as an example, see steps S221 to S225 for details.

[0111] S221, performing sample entropy calculation on the components in the current component set to obtain a sample entropy vector;

[0112] Among them, in this embodiment, the sample entropy (Sample Entropy, referred to as SampEn) is used as a quantitative indicator to evaluate the complexity of the time series data. This indicator reflects the complexity of the time series by measuring the probability of the generation of new patterns in the signal. Specifically, an arc is an instantaneous high-energy discharge phenomenon, usually accompanied by violent fluctuations in current. By observing the changes in sample entropy, abnormal fluctuations in the current time series data can be discovered in time, so as to determine whether an arc has occurred. When the sample entropy value suddenly increases, it indicates that a new pattern has appeared in the current time series data, which may be a signal of the occurrence of an arc.

[0113] S222, classifying the components in the current component set to obtain a high-frequency component set, a medium-frequency component set and a low-frequency component set;

[0114] At the same time, in the process of classifying components, those skilled in the art can adopt a variety of methods to achieve this goal. Specifically, the components can be divided according to a fixed boundary value, which is simple and easy to implement. For example, a threshold can be set to divide the components into three categories according to their numerical values: one category is less than the threshold, one category is equal to the threshold, and another category is greater than the threshold. In addition, other calculation methods can be used to divide the components, such as using statistical methods, machine learning algorithms or other mathematical models to determine the classification criteria of the components. These methods can be selected and adjusted according to the actual application scenarios and needs to achieve the best classification effect. Since these methods belong to the prior art, the specific process will not be repeated.

[0115] S223, performing component energy calculation on the high-frequency component set and the low-frequency component set to obtain total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively;

[0116] The component energy calculation mentioned in this step refers to the calculation of the energy of each individual component. Specifically, technicians in this field can use two different methods to perform this calculation. The first method is a direct method, which involves calculating the sum of the squares of the amplitudes of all sampling points to obtain the energy value of the component. The second method is an indirect method, which first uses Hilbert transform to extract the envelope signal of the IMF (empirical mode decomposition) component, and then calculates the energy of this envelope signal to obtain the energy value of the component. Since the above method belongs to the prior art, the specific process will not be repeated.

[0117] S224, performing ratio calculation based on the total energy values ​​corresponding to the high-frequency component set and the low-frequency component set to obtain an energy ratio;

[0118] In this step, the calculated energy ratio actually reflects the difference between the high-frequency component and the low-frequency component. This ratio can effectively reveal the relative strength of different frequency components in the signal. At the same time, in the process of residential electricity use, when the arc phenomenon occurs, although the frequency characteristics of the arc generation are quite complex and difficult to measure accurately, it can be determined that the frequency components generated by the arc will not cover both the high-frequency and low-frequency areas. Therefore, in this embodiment, the energy ratio is selected as one of the important features to distinguish and identify the arc phenomenon.

[0119] S225 , concatenating the sample entropy vector and the energy ratio and normalizing them to obtain a current feature vector.

[0120] It should be noted that the concatenation operation is implemented through the contact function. The contact function is a commonly used function for concatenating multiple data or arrays into a continuous whole. Specifically, this function will merge the various parts of the input according to a certain order and rule to form a new, longer data sequence or array. In addition, the present embodiment also involves normalization. Normalization is a common data preprocessing method, and its purpose is to adjust the numerical range of the data to a unified standard interval for subsequent processing and analysis. There are many specific methods for normalization, such as minimum-maximum normalization, z-score standardization, etc. Through normalization, the dimensional differences between different data sources can be eliminated, making the data comparable, thereby improving the accuracy and efficiency of data processing.

[0121] Furthermore, for those skilled in the art, feature calculations in other dimensions may also be performed, and this embodiment does not impose any specific limitation on this.

[0122] S23, extracting features from each current feature vector using a feature extraction network model to obtain a feature value vector;

[0123] In this step, the feature extraction network model mentioned is Convolutional Neural Networks (CNN). The feature extraction network model consists of multiple layers. Specifically, it starts with an input layer, followed by a convolutional layer, a pooling layer, a rectified linear unit layer, a convolutional layer, a pooling layer, a rectified linear unit layer, a batch normalization layer, and an output layer.

[0124] At the same time, in this embodiment, two current feature vectors are first input into the input layer. These two current feature vectors respectively represent the features of the current data under different decomposition methods. And in order to more effectively extract and fuse these features, this embodiment adopts a recurrent neural network model for feature fusion. In this way, the features of the current time series data under two different decomposition methods can be effectively fused, thereby extracting more comprehensive and in-depth feature information. This feature extraction method not only improves the accuracy of the model, but also enhances the sensitivity and adaptability of the model to changes in current data.

[0125] S24, performing state prediction on the eigenvalue vector and the two current eigenvectors respectively based on three preset prediction models to obtain state prediction results, wherein the state prediction results include three predicted states and a second confidence level corresponding to each predicted state;

[0126] In this embodiment, a trained long short-term memory network (LSTM) is used as a prediction model. LSTM is a special recurrent neural network (RNN) structure, which is designed to solve the gradient vanishing problem faced by traditional RNN when dealing with long-term dependencies. This gradient vanishing problem causes the model to gradually lose its memory of early input information during the learning process, thereby affecting its performance on long-term sequence data.

[0127] The LSTM effectively solves this problem by introducing a gating mechanism. It contains three main gates: forget gate, input gate, and output gate. The forget gate is responsible for deciding which information should be discarded from the cell state, the input gate controls the inflow of new information, and the output gate determines what information to output at each time step. The combination of these gates enables LSTM to maintain and transmit important information in long sequences, thereby better capturing long-term dependencies.

[0128] In this embodiment, the training data set of the long short-term memory network is also obtained by the processing method of steps S21 to S23. As for the training process of the long short-term memory network, it involves initializing network parameters, forward propagation calculation loss, back propagation update weights and iterative optimization until convergence. In the forward propagation stage, the input data is transmitted layer by layer through the network to finally obtain the prediction result. The loss function is used to measure the difference between the prediction result and the true value. In the back propagation stage, the gradient of the loss function with respect to the network parameters is calculated and used to update the network weights. Through multiple iterations, the network parameters are gradually adjusted, and the prediction performance of the model is gradually improved until the desired accuracy is reached. The above-mentioned specific training processes are all prior art and will not be repeated in this embodiment.

[0129] S25. Determine and obtain a prediction result based on the state prediction result.

[0130] It can be seen from the above step S24 that in this embodiment, a total of three state prediction results output by the trained long short-term memory network are obtained. Therefore, it is necessary to summarize the above three state prediction results, see steps S251 to S253 for details.

[0131] S251, confirming the frequencies of obtaining different prediction states according to the three prediction states;

[0132] S252, confirming that the predicted state with the highest frequency is an arc state according to the frequencies of all the predicted states;

[0133] That is, in reality, if two predicted states are 0 and the other predicted state is 1, then the confirmed arc state is 0, that is, the arc does not exist. Vice versa.

[0134] S253. Calculate a first confidence based on the frequencies of all the predicted states and all the second confidences.

[0135] Specifically, in this embodiment, the first confidence calculation function is as follows:

[0136] ;

[0137] in, represents the first confidence level; , , Represent three prediction states respectively; , , Respectively represent the second confidences corresponding to the three predicted states; represents a counting function; It indicates the number of occurrences of 1 in the statistical prediction state, and also indicates the number of occurrences of arc in the statistical prediction state; It indicates the number of times 0 occurs in the statistical prediction state, and also indicates the number of times the arc does not exist in the statistical prediction state.

[0138] Embodiment 3:

[0139] The difference between this embodiment and embodiment 1 is that a smoke recognition method is provided for step S6 to improve the accuracy of smoke recognition. Figure 3 The figure takes the current time series data of an electric unit as an example to illustrate how to perform arc prediction, see steps S61 to S64 for details.

[0140] S61, obtaining a monitoring dark channel image based on performing dark channel processing on the monitoring image;

[0141] In this embodiment, the process of how to calculate a channel-based image by analyzing an image containing three color channels of RGB belongs to the prior art, so it will not be repeated here. Based on the dark channel prior theory, the size of the dark channel value can reflect the degree to which each pixel is affected by external factors such as light and smoke. Therefore, a monitoring dark channel image can be obtained by calculating the dark channel value of each pixel in the monitoring image. However, since each power unit in a building is usually equipped with a variety of supplementary lighting equipment, such as spotlights, downlights, etc., it is impossible to directly use the dark channel value to approximate the degree to which the pixel is affected by smoke. In order to accurately extract the smoke area, it is necessary to effectively distinguish the smoke area from the light area. The specific extraction process can refer to the following steps: step S62 and step S63.

[0142] S62, performing edge detection on the monitored dark channel image using a Sobel operator edge detection model to obtain at least one suspected smoke region and a gradient of each pixel point in each suspected smoke region;

[0143] The Sobel operator edge detection model mentioned in this step is usually referred to as the Sobel operator edge detection model. The main function of this model is to effectively extract the edge information in the image by calculating the gradient of the grayscale value of each pixel in the image. Specifically, the Sobel operator edge detection model utilizes the local gradient information of the image and determines the position and direction of the edge by weighted summation of the grayscale changes in the horizontal and vertical directions. This method has a wide range of applications in image processing and is commonly used in fields such as image segmentation, feature extraction, and target recognition. In this way, the Sobel operator edge detection model in this step can effectively enhance the edge features of the image, making subsequent image analysis and processing easier and more accurate.

[0144] S63, calculating the smoke characteristic value of each of the suspected smoke areas based on the gradient change of each pixel point in each of the suspected smoke areas;

[0145] In this embodiment, the smoke generated by the fire has a significant fluidity characteristic. As the smoke concentration increases, the difference in pixels in the area where the smoke is located gradually decreases, which is manifested as a decrease in the gradient amplitude in the gradient calculation of image processing. In addition, the gradient direction of the smoke usually shows complexity and irregularity. Given that the smoke is generated by an open flame igniting organic matter, one end of the smoke exhibits the characteristic of minimizing the gradient due to the presence of the flame. Therefore, this embodiment uses one of the suspected smoke areas as an example to illustrate the confirmation of the smoke feature value, and the details can be found in steps S631-S633.

[0146] S631, performing amplitude mean calculation according to the gradient of each pixel point in the suspected smoke area to obtain the gradient amplitude mean of the suspected smoke area;

[0147] It is well known to those skilled in the art that the gradient is a vector that includes direction and value. Therefore, in this step, the mean value of the gradient amplitude is calculated based on the amplitude corresponding to each pixel point in the suspected smoke area, which can represent the uniformity of the dark channel value inside the suspected smoke area, that is, the brightness of the suspected smoke area.

[0148] S632, calculating the diffusion direction based on the gradient of each pixel point in the suspected smoke area to obtain a smoke diffusion evaluation coefficient of the suspected smoke area;

[0149] Specifically, first, the flame point O is screened in the suspected smoke area. The flame point is the point with the smallest gradient amplitude. If there are multiple points with the smallest gradient amplitude, the point at the center is selected. The method of selecting the point at the center is the existing technology and will not be repeated in this embodiment. Further, the direction of each pixel point pointing to the flame point in the suspected smoke area is recorded respectively, and the angle θ with the gradient direction of the pixel point is calculated; finally, the cosine value of the angle θ corresponding to each pixel point is calculated one by one and the average is calculated to obtain the smoke diffusion assessment coefficient.

[0150] S633. Calculate the smoke characteristic value based on the edge coefficient corresponding to the suspected smoke area, the smoke diffusion assessment coefficient and the gradient amplitude mean value.

[0151] Specifically, the calculation function of the smoke characteristic value is as follows:

[0152] ;

[0153] in, Indicates the smoke characteristic value of the suspected smoke area; Indicates the smoke diffusion assessment coefficient of the suspected smoke area; Represents the mean value of the gradient amplitude in the suspected smoke area.

[0154] In the above function, It can indicate the uniformity of the dark channel value inside the suspected smoke area, that is, the brightness of the suspected smoke area. The larger the value, the more likely it is a light spot formed by a lamp. It represents the mean value of the cosine value of the angle between the gradient direction of the pixel point and the flame point. If the spot formed by the lamp is symmetrical, it is better reflected in The value is close to 1, but due to the randomness of smoke diffusion, the contour is uncontrollable and the symmetry is poor. is much smaller than 1. Therefore, in this embodiment, the above functional formula better reflects the characteristics of smoke.

[0155] In summary, The larger the value is, the more likely the suspected smoke area is to be a smoke-generating area.

[0156] S64. A smoke area is obtained by screening based on the smoke characteristic value, and the initial fire assessment coefficient is corrected based on the smoke characteristic value of the smoke area to obtain a fire assessment coefficient, and the smoke characteristic value corresponding to the smoke area is the maximum value.

[0157] Specifically, the fire assessment coefficient calculation function is:

[0158] ;

[0159] in, represents the fire assessment factor; represents the normalization function; represents the fire initial assessment coefficient; Indicates the smoke characteristic value of the suspected smoke area.

[0160] In this embodiment, the smoke recognition rate can be effectively improved by mining smoke features in the dark channel image.

[0161] Embodiment 4:

[0162] like Figure 4 As shown, this embodiment provides an electrical fire alarm system for a TOD building, the system comprising:

[0163] A current data acquisition module is used to acquire the current data of multiple power consumption units in real time and convert it into the current time series data corresponding to each of the power consumption units at the current moment;

[0164] A prediction module, configured to perform arc prediction on the current time series data corresponding to each of the power consumption units at the current moment based on a preset fire prediction neural network to obtain a prediction result at the current moment, wherein the prediction result includes an arc state and a first confidence level;

[0165] A data integration module, used to convert the prediction results corresponding to the current moment and the historical moment into the time series prediction data corresponding to each of the power consumption units;

[0166] A logic module, used to determine the initial fire assessment coefficient corresponding to the suspected fire unit according to the arc state and confidence change in the time series prediction data corresponding to each of the power consumption units;

[0167] An image data module, used for acquiring corresponding monitoring images of the suspected fire unit in real time;

[0168] A correction module, used for correcting the initial fire assessment coefficient based on the smoke state in the monitoring image to obtain a fire assessment coefficient;

[0169] The early warning module is used to determine whether to send fire early warning information to the target terminal based on the fire assessment coefficient and the preset fire threshold.

[0170] In some specific embodiments, the prediction module includes:

[0171] A data decomposition module is used to decompose the current time series data in different ways to obtain two sets of current components;

[0172] A feature calculation module, used for performing feature calculations based on the two current component sets respectively to obtain two current feature vectors;

[0173] A feature extraction module, used for extracting features from each current feature vector using a feature extraction network model to obtain a feature value vector;

[0174] A state prediction module, used for performing state prediction on the eigenvalue vector and the two current eigenvectors respectively based on three preset prediction models to obtain state prediction results, wherein the state prediction results include three predicted states and a second confidence level corresponding to each predicted state;

[0175] A result merging module is used to determine and obtain a prediction result based on the state prediction result.

[0176] In some specific embodiments, the feature calculation module includes:

[0177] A sample entropy calculation module, used to perform sample entropy calculation on the components in the current component set to obtain a sample entropy vector;

[0178] A component classification module, used for classifying the components in the current component set to obtain a high-frequency component set, a medium-frequency component set and a low-frequency component set;

[0179] An energy calculation module, used to perform component energy calculation on the high-frequency component set and the low-frequency component set to obtain total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively;

[0180] A ratio module, configured to calculate an energy ratio based on the total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively;

[0181] The splicing module is used to splice the sample entropy vector and the energy ratio and normalize them to obtain a current feature vector.

[0182] In some specific embodiments, the result merging module includes:

[0183] A frequency statistics module, used for confirming the frequencies of obtaining different prediction states according to the three prediction states;

[0184] A state confirmation module, used to confirm the arc state according to the frequencies of all the predicted states;

[0185] A confidence calculation module is used to calculate a first confidence based on the frequencies of all the predicted states and all the second confidences.

[0186] In some specific embodiments, the logic module includes:

[0187] A first judgment module, configured to determine that when the arc state in the time series prediction data corresponding to the power consumption unit changes, the fire initial assessment coefficient is a preset value;

[0188] A second judgment module is used for performing a straight line fitting on the first confidence level in the time series prediction data when the arc state in the time series prediction data corresponding to the power consumption unit has not changed, and calculating the fire initial assessment coefficient according to the slope of the straight line fitting;

[0189] The third judgment module is used to judge whether the power unit is a suspected fire unit according to the initial fire assessment coefficient corresponding to the power unit and a preset initial threshold, and the initial fire assessment coefficient of the suspected fire unit is greater than the preset initial threshold.

[0190] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0191] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0192] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. An electrical fire early warning method for a TOD building, characterized in that: The method comprises: Acquire current data of multiple power consumption units in real time, and convert it into current time series data corresponding to each power consumption unit at the current moment; Based on a preset fire prediction neural network, arc prediction is performed on the current time series data corresponding to each of the power consumption units at the current moment to obtain a prediction result at the current moment, wherein the prediction result includes an arc state and a first confidence level; The prediction results corresponding to the current moment and the historical moment are converted to obtain the time series prediction data corresponding to each of the power consumption units; Determine the fire initial assessment coefficient corresponding to the suspected fire unit according to the arc state and confidence change in the time series prediction data corresponding to each power consumption unit; Acquire the corresponding monitoring image of the suspected fire unit in real time; Correcting the initial fire assessment coefficient based on the smoke state in the monitoring image to obtain a fire assessment coefficient; Determining whether to send fire warning information to the target terminal based on the fire assessment coefficient and the preset fire threshold; Among them, based on the preset fire prediction neural network, arc prediction is performed on the current time series data corresponding to each of the power consumption units at the current moment to obtain the prediction result at the current moment, including: The current time series data are decomposed differently to obtain two sets of current components; decomposition into a set empirical mode decomposition model and a local mean decomposition model; Perform characteristic calculations based on the two current component sets to obtain two current characteristic vectors; Using a feature extraction network model to extract features from each current feature vector to obtain a feature value vector; Based on three preset prediction models, state prediction is performed on the eigenvalue vector and the two current eigenvectors respectively to obtain state prediction results, wherein the state prediction results include three predicted states and a second confidence level corresponding to each predicted state; The prediction result is obtained based on the state prediction result judgment; specifically, the frequencies of different prediction states are confirmed according to the three prediction states; the prediction state with the highest frequency is confirmed to be the arc state according to the frequencies of all the prediction states; the first confidence is calculated based on the frequencies of all the prediction states and all the second confidences; the first confidence calculation function is as follows: Among them, CL represents the first confidence; P1, P2, and P3 represent three prediction states respectively; CL1, CL2, and CL3 represent the second confidences corresponding to the three prediction states respectively; count() represents the counting function; count('1') represents the number of occurrences of 1 in the statistical prediction state, and also represents the number of occurrences of the arc in the statistical prediction state; count('0') represents the number of occurrences of 0 in the statistical prediction state, and also represents the number of occurrences of the arc not existing in the statistical prediction state.

2. The electrical fire early warning method for TOD buildings according to claim 1, characterized in that: Based on the two current component sets, characteristic calculations are performed respectively to obtain two current characteristic vectors, including: Performing sample entropy calculation on the components in the current component set to obtain a sample entropy vector; Classifying the components in the current component set to obtain a high-frequency component set, a medium-frequency component set and a low-frequency component set; Performing component energy calculation on the high-frequency component set and the low-frequency component set to obtain total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively; An energy ratio is obtained by performing ratio calculation based on the total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively; The sample entropy vector and the energy ratio are concatenated and normalized to obtain a current feature vector.

3. The electrical fire early warning method for TOD buildings according to claim 1, characterized in that: The fire initial assessment coefficient corresponding to the suspected fire unit is obtained according to the arc state and confidence change in the time series prediction data corresponding to each power consumption unit, including: When the arc state in the time series prediction data corresponding to the power consumption unit changes, the initial fire assessment coefficient is a preset value; When the arc state in the time series prediction data corresponding to the power consumption unit has not changed, a straight line fitting is performed on the first confidence level in the time series prediction data, and the fire initial assessment coefficient is calculated according to the slope of the straight line fitting; It is determined whether the power unit is a suspected fire unit according to the initial fire assessment coefficient corresponding to the power unit and a preset initial threshold value, and the initial fire assessment coefficient of the suspected fire unit is greater than the preset initial threshold value.

4. An electrical fire warning system for a TOD building, characterized in that: include: A current data acquisition module is used to acquire the current data of multiple power consumption units in real time and convert it into the current time series data corresponding to each of the power consumption units at the current moment; A prediction module, configured to perform arc prediction on the current time series data corresponding to each of the power consumption units at the current moment based on a preset fire prediction neural network to obtain a prediction result at the current moment, wherein the prediction result includes an arc state and a first confidence level; A data integration module, used to convert the prediction results corresponding to the current moment and the historical moment into the time series prediction data corresponding to each of the power consumption units; A logic module, used to determine the initial fire assessment coefficient corresponding to the suspected fire unit according to the arc state and confidence change in the time series prediction data corresponding to each of the power consumption units; An image data module, used for acquiring corresponding monitoring images of the suspected fire unit in real time; A correction module, used for correcting the initial fire assessment coefficient based on the smoke state in the monitoring image to obtain a fire assessment coefficient; An early warning module, used to determine whether to send fire early warning information to a target terminal based on the fire assessment coefficient and a preset fire threshold; Among them, the prediction module includes: A data decomposition module is used to decompose the current time series data in different ways to obtain two sets of current components; A feature calculation module, used for performing feature calculations based on the two current component sets respectively to obtain two current feature vectors; A feature extraction module, used for extracting features from each current feature vector using a feature extraction network model to obtain a feature value vector; A state prediction module, used for performing state prediction on the eigenvalue vector and the two current eigenvectors respectively based on three preset prediction models to obtain state prediction results, wherein the state prediction results include three predicted states and a second confidence level corresponding to each predicted state; A result merging module, used to determine and obtain a prediction result based on the state prediction result; Wherein, the result merging module comprises: A frequency statistics module, used for confirming the frequencies of obtaining different prediction states according to the three prediction states; A state confirmation module, used to confirm the arc state according to the frequencies of all the predicted states; A confidence calculation module is used to calculate a first confidence based on the frequencies of all the predicted states and all the second confidences, wherein the first confidence calculation function is as follows: Among them, CL represents the first confidence; P1, P2, and P3 represent three prediction states respectively; CL1, CL2, and CL3 represent the second confidences corresponding to the three prediction states respectively; count() represents the counting function; count('1') represents the number of occurrences of 1 in the statistical prediction state, and also represents the number of occurrences of the arc in the statistical prediction state; count('0') represents the number of occurrences of 0 in the statistical prediction state, and also represents the number of occurrences of the arc not existing in the statistical prediction state.

5. The electrical fire warning system for TOD buildings according to claim 4, characterized in that: The feature calculation module comprises: A sample entropy calculation module, used to perform sample entropy calculation on the components in the current component set to obtain a sample entropy vector; A component classification module, used for classifying the components in the current component set to obtain a high-frequency component set, a medium-frequency component set and a low-frequency component set; An energy calculation module, used to perform component energy calculation on the high-frequency component set and the low-frequency component set to obtain total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively; A ratio module, configured to calculate an energy ratio based on the total energy values ​​corresponding to the high-frequency component set and the low-frequency component set respectively; The splicing module is used to splice the sample entropy vector and the energy ratio and normalize them to obtain a current feature vector.

6. The electrical fire warning system for TOD buildings according to claim 4, characterized in that: The logic module includes: A first judgment module, configured to determine that when the arc state in the time series prediction data corresponding to the power consumption unit changes, the fire initial assessment coefficient is a preset value; A second judgment module is used for performing a straight line fitting on the first confidence level in the time series prediction data when the arc state in the time series prediction data corresponding to the power consumption unit has not changed, and calculating the fire initial assessment coefficient according to the slope of the straight line fitting; The third judgment module is used to judge whether the power unit is a suspected fire unit according to the initial fire assessment coefficient corresponding to the power unit and a preset initial threshold, and the initial fire assessment coefficient of the suspected fire unit is greater than the preset initial threshold.

Images