A fire alarm prediction method and device based on a Gaussian process

Abstract

The application provides a fire alarm prediction method and device based on a Gaussian process. The method provided by the application comprises the following steps: determining a first time length based on state data of a fire alarm device in a first region; acquiring a fire alarm frequency of the first region in a first time period, generating alarm time sequence data, determining a third time period which is the same as the first time period based on state data of the first region in a second time period, and taking a time of the first time length which is closest to the current time in the third time period as the first time period; establishing a Gaussian process model based on the alarm time sequence data; and predicting the fire alarm frequency of the first region in the second time period based on the learned Gaussian process model. The fire alarm prediction method and device based on the Gaussian process provided by the application do not require a large amount of data and the construction of a neural network model, and long-term prediction of a fire alarm is realized.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of fire safety management, and in particular to a fire alarm prediction method and device based on a Gaussian process. BACKGROUND

[0002] With the acceleration of urbanization and the increase of building density, fire risk is increasing, and the rational scheduling of fire resources and the accurate implementation of preventive measures are becoming crucial. In the context of high fire incidence, long-term prediction of fire alarms is particularly critical. By conducting long-term prediction on historical alarm data, relevant departments can identify high-risk areas and trends in fire occurrence in advance, thereby optimizing emergency plans and deploying fire forces in advance to minimize casualties and property losses.

[0003] Currently, the industry mainly uses Internet of Things technology and neural network models to achieve long-term prediction of fire alarms. Sensors are connected to the Internet through Internet of Things technology, real-time collection of environmental parameter data (such as temperature, humidity, smoke concentration, etc.), and transmission of data to the data processing system. After receiving the data transmitted by the sensor, the data processing system uses a neural network model to model the data. The neural network model can be of various types, such as multi-layer perceptron (MLP), convolutional neural network (CNN), or recurrent neural network (RNN), and the appropriate model is selected for establishment according to the specific circumstances. The trained and optimized neural network model converts the input data into a fire alarm probability. This probability is based on the model's learning and prediction of data and can be represented as a quantitative indicator of fire potential. The obtained fire alarm probability needs to be further processed to meet actual needs. For example, alarm triggering can be performed according to threshold settings, or more complex analysis and judgment can be performed in combination with other data.

[0004] However, this method relies on a large amount of data, and traditional neural network models require a large amount of known data for training. However, in actual situations, it is difficult to obtain large-scale alarm data, especially for fixed locations with a small number of alarms over a period of time. This leads to overfitting of neural networks on small data sets, which affects their performance on new data. And the adjustment of hyperparameters is cumbersome, neural network models have many hyperparameters that need to be adjusted, including learning rate, number of layers, number of neurons, etc. Selecting the appropriate combination of hyperparameters is a challenge, usually requiring a lot of experiments and experience, which increases the complexity and time cost of parameter adjustment.

[0005] Therefore, there is an urgent need for a method that does not require a large amount of data and the construction of a neural network model to achieve long-term prediction of fire alarms. SUMMARY

[0006] Therefore, the application provides a fire alarm prediction method and device based on a Gaussian process, which realizes long-term prediction of fire alarms without a large amount of data and construction of a neural network model.

[0007] Specifically, the application is realized by the following technical solutions.

[0008] The first aspect of the application provides a fire alarm prediction method based on a Gaussian process, which comprises the following steps.

[0009] determining a first time length based on fire alarm equipment state data in a first region and state data in the first region;

[0010] obtaining the number of fire equipment alarms in the first region in a first time period, and generating alarm time sequence data, wherein the alarm time sequence data is vector data generated after the number of fire equipment alarms is arranged in chronological order, and the time length of the first time period is the first time length;

[0011] wherein a third time period identical to the first region in a second time period is determined based on state data in the first region, the time length of the third time period is greater than the first time length, the time of the first time length closest to the current time in the third time period is taken as the first time period, and the second time period is a time period corresponding to the number of fire equipment alarms predicted based on the Gaussian process;

[0012] establishing a Gaussian process model according to the alarm time sequence data;

[0013] predicting the number of fire equipment alarms in the first region in a second time period based on the learned Gaussian process model, wherein the time length of the second time period is longer than the time length of the first time period.

[0014] The second aspect of the application provides a fire alarm prediction device based on a Gaussian process, which comprises a determination module, a generation module, an establishment module and a prediction module, wherein

[0015] The determination module is configured to determine a first time length based on fire alarm equipment state data in a first region and state data in the first region;

[0016] The generation module is configured to obtain the number of fire equipment alarms in the first region in a first time period, and generate alarm time sequence data, wherein the alarm time sequence data is vector data generated after the number of fire equipment alarms is arranged in chronological order, and the time length of the first time period is the first time length;

[0017] The third time period is determined according to the state data in the first region in the second time period, the third time period has a time length greater than the first time length, a time in the third time period closest to a current time is taken as the first time length, and the second time period is a time period corresponding to the number of fire-fighting equipment alarms predicted based on the Gaussian process.

[0018] The establishing module is configured to establish a Gaussian process model according to the alarm time sequence data.

[0019] The predicting module is configured to predict the number of fire-fighting equipment alarms in the first region in a second time period based on the learned Gaussian process model, and the second time period has a time length greater than the time length of the first time period.

[0020] The fire alarm prediction method and apparatus based on Gaussian processes provided in this application, in three aspects, firstly, determine the number of fire equipment alarms in a first time period based on a first time length, then determine a third time period based on a second time period and the first time period, and select the first time period closest to the current time from the third time period. By determining the first, second, and third time periods and comparing and analyzing them, patterns and trends in alarm time series data can be captured more accurately, improving the accuracy of fire alarm prediction. Secondly, when determining the first time period from the third time period, selecting the first time period closest to the current time ensures that the alarm time series data of the first time period used to predict the number of fire equipment alarms in the second time period can more accurately reflect the latest trends and current patterns, providing more relevant information, helping to reduce prediction errors, improve the accuracy and reliability of prediction, and ensure that the prediction and analysis results are consistent with reality. Thirdly, in alarm time series data, the environment and alarm patterns fluctuate over time; for example, the number of fire equipment alarms may change significantly during specific seasons, holidays, or when specific events occur. By dynamically adjusting the time period for data acquisition, these changes can be adapted, thereby improving the predictive capability of the Gaussian process model. Especially for long-term forecasting, Gaussian process models need to continuously update their historical data to follow changes in data patterns. Furthermore, alarm time series data has unique characteristics: trends are more stable over longer periods. While short-term forecasts are generally more accurate in general time series forecasting, alarm time series data often exhibits potential periodic trends and is significantly influenced by seasonality, population activity, and other factors. Fixed time periods may prevent Gaussian process models from effectively adapting to emerging patterns or long-term trend changes, while dynamically adjusting the time period allows the model to better capture new data characteristics. By extending the time period, Gaussian process models can better eliminate the interference of small fluctuations and grasp the overall trend. This not only improves the robustness of forecasts but also reduces reliance on sudden anomalies or sporadic alarms. By continuously updating and extending the time period, Gaussian process models can eliminate the influence of small fluctuations under different circumstances, thus more accurately predicting the trend of alarm time series data in the long term. Attached Figure Description

[0021] Figure 1 A flowchart of an embodiment of the fire alarm prediction method based on Gaussian processes provided in this application;

[0022] Figure 2 This is a schematic diagram of the structure of a fire alarm prediction device based on Gaussian processes provided in this application. Detailed Implementation

[0023] The exemplary embodiments will be described in detail herein with reference to several drawings. Descriptions of well-known functions and structures are omitted so as not to obscure the concept of the application. The following detailed description is presented in terms of a number of acts that can be implemented as instructions executed by a processor or a controller. Also, the description hereinafter uses the term "couple" or "coupled" to refer to any direct or indirect communication between two or more elements, which can mean the coupling of some elements, but it can also mean the

[0024] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0025] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is to be further understood that the terms "comprise," "comprises," and / or "comprising" when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It is to be understood that the terms "if' and "as if' as used herein can be interpreted either as "when" or "when a" or "in response to determining."

[0026] The following detailed description is presented in terms of a number of acts that can be implemented as instructions executed by a processor or a controller. Also, the description hereinafter uses the term "couple" or "coupled" to refer to any direct or indirect communication between two or more elements, which can mean the coupling of some elements, but it can also mean the

[0027] Figure 1 The flowchart of the embodiment one of the fire alarm prediction method based on Gaussian process provided in the present application is shown in FIG. 1. Please refer to Figure 1 The method provided in the embodiment can include the following steps.

[0028] S101, determining a first time length based on fire alarm equipment state data in a first region and state data in the first region.

[0029] Specifically, the first region is set according to actual needs, and in the embodiment, it is not limited. It should be noted that in the present application, the acquisition of the fire alarm frequency and the prediction of the fire alarm frequency are both directed to the first region. For example, in an embodiment, the first region is XX city. The fire alarm equipment state data represents the working state of the fire alarm equipment, including alarm records, equipment running state, equipment maintenance records, equipment response time, sensor data, etc. The state data represents the regional environmental conditions and external factors of the fire alarm equipment, including environmental conditions (temperature, humidity, air pressure), regional activities (personnel activities, equipment usage), building structure and layout, external events (weather conditions, construction, decoration), etc.

[0030] In specific implementation, determining the first time length based on the status data of fire alarm equipment in the first area and the status data in the first area includes: analyzing the status data of fire alarm equipment in the first area and the status data based on time series analysis methods to determine the status changes of fire alarm equipment and the behavior changes related to fire alarms in the first area, and obtaining a first change trend; determining the change time point based on the first change trend; and determining the first time length based on the longest interval between the first change trend and the change time point.

[0031] Specifically, based on time series analysis methods (moving average, exponential smoothing, seasonal decomposition, etc.), the status data of fire alarm equipment in the first area are analyzed to identify changes in the status of fire alarm equipment (changes in alarm frequency and intensity) and changes in behavior related to fire alarms (increases or decreases in activity within the first area). Based on these status and behavioral changes, a primary trend (long-term trend, seasonal variation, and periodic pattern) is determined. Further, based on this primary trend, change time points are determined, namely, the time points when the alarm frequency of fire alarm equipment suddenly increases or decreases, the time points when environmental conditions within the first area change significantly (e.g., rapid temperature changes), and the time points when personnel activity patterns within the first area change significantly (e.g., peak holiday periods). Based on the intervals between all change time points, the two change time points with the longest intervals are identified, and the interval between these two change time points is defined as the primary time length.

[0032] The method provided in this embodiment, in a first aspect, analyzes the status data of fire alarm equipment in a first area to identify periodicity and long-term trends in the data. This helps ensure that the first time length can adapt to the natural periodic changes in the data, reducing prediction errors caused by mismatches in periodic changes. In a second aspect, when determining the first time length, the longest interval between the changing time points is defined as the first time length. This ensures that the selected first time length can cover the main changing trends in the data and avoids potential problems caused by excessively short or long time periods.

[0033] S102. Obtain the number of fire equipment alarms in the first time period within the first area, and generate alarm time series data. The alarm time series data is vector data generated by arranging the number of fire equipment alarms in chronological order. The length of the first time period is the first time length.

[0034] The third time period is determined according to the state data in the first region in the second time period, the third time period is the same as the state data, and the time length of the third time period is greater than the first time length. The time of the first time length closest to the current time in the third time period is taken as the first time period, and the second time period is a time period corresponding to the number of fire-fighting equipment alarms predicted based on the Gaussian process.

[0035] Specifically, the time length of the first time period is the first time length. The second time period is a time period corresponding to the number of fire-fighting equipment alarms predicted based on the Gaussian process, which is set according to actual needs, and is not limited in the embodiment. It should be noted that the time length of the second time period is greater than the first time length, and the time length of the third time period is greater than the first time length. The alarm time sequence data is vector data generated after the number of fire-fighting equipment alarms is arranged in chronological order.

[0036] In a specific implementation, the first time period of the number of fire-fighting equipment alarms in the first region is obtained, and the alarm time sequence data is generated, including:

[0037] (1) Based on the first time length, the number of fire-fighting equipment alarms in the first time period in the first region is obtained; wherein the time length of the first time period is the first time length.

[0038] In a specific implementation, a target data document is obtained, and the target data document records the number of fire-fighting equipment alarms in different regions and different time periods. The target data document corresponding to the fire-fighting equipment is obtained by filtering the target data document based on the equipment number of the fire-fighting equipment. Further, the number of fire-fighting equipment alarms in the first time length in the first region in the target data document corresponding to the fire-fighting equipment is read.

[0039] (2) Based on the state data in the first region in the second time period and the first time length, a third time period identical to the state data is determined; wherein the time length of the third time period is greater than the first time length.

[0040] Specifically, according to the above description, the second time period is set according to actual needs, and is not limited in the embodiment.

[0041] In a specific implementation, the state data in the first region in the second time period is obtained, and the state data is analyzed. The target state data identical to the state data in the first region in the second time period and having a time length greater than the first time length is selected from the state data in the first region. The time period corresponding to the target state data is the third time period.

[0042] Optionally, determining the third time period based on the state data in the first region within the second time period and the first time length comprises: determining the number of fire-fighting devices and the working state of the fire-fighting devices in the first region within the second time period as the fire alarm device state data in the first region within the second time period; determining the third time period in the first region that is the same as the fire alarm device state data; correcting the fire alarm device state data in the third time period in the first region according to abnormal activities in the third time period in the first region, the abnormal activities being activities that affect the number of fire-fighting devices and the working state of the fire-fighting devices; and taking the time period in the corrected fire alarm device state data that is the same as the fire alarm device state data in the first region within the second time period as the final third time period.

[0043] Specifically, the abnormal activities refer to activities that affect the number of fire-fighting devices and the working state of the fire-fighting devices in the first region, for example, major events, exercises, entertainment activities, etc.

[0044] The method provided by the embodiment first determines the fire alarm device state data in the first region within the third time period based on the fire alarm device state data in the first region within the second time period, and then identifies and corrects the fire alarm device state data affected by abnormal activities, which can eliminate abnormal fluctuations caused by these abnormal activities. The corrected fire alarm device state data can provide more reliable information and reflect more real fire-fighting device state and fire-fighting device number. The correction ensures that the final selected third time period more truly reflects the working state of the fire-fighting devices and the number of fire-fighting devices under normal circumstances, thereby improving the accuracy of the prediction model and enabling the prediction model to more robustly cope with similar situations that may occur in the future, ensuring that the prediction result is not affected by short-term abnormalities.

[0045] (3) Based on the third time period, determining a first time period in the third time period with a time length of the first time length and closest to the current time.

[0046] In a specific implementation, all time periods in the third time period with a time length of the first time length are determined by traversing the third time period, and the time period with a start time closest to the current time is determined as the first time period.

[0047] Optionally, based on the third time period, after determining the first time period with the time length of the first time length and the closest to the current time, the method further comprises: performing outlier processing on the alarm time sequence data of the first time period in the first region to determine an abnormal point; the difference between the alarm number of the fire-fighting equipment corresponding to the abnormal point and the alarm number of the fire-fighting equipment corresponding to the change time point is greater than a preset value; replacing the alarm number of the fire-fighting equipment corresponding to the abnormal point with first data in the first region to obtain updated alarm time sequence data; the first data comprises the alarm number of the fire-fighting equipment at the time corresponding to the abnormal point.

[0048] Specifically, the abnormal point refers to the alarm number of the fire-fighting equipment that is protruding from the first change trend in the alarm time sequence data, that is, the difference between the alarm number of the fire-fighting equipment corresponding to the abnormal point and the alarm number of the fire-fighting equipment corresponding to the change time point is greater than a preset value, wherein the preset value is set according to actual needs, which is not limited in the embodiment.

[0049] In specific implementation, the alarm time sequence data of the first time period in the first region is traversed, and based on the difference between the alarm number of the fire-fighting equipment under each alarm time sequence data and the change time point, the abnormal point in the alarm time sequence data is determined. Further, the alarm number of the fire-fighting equipment corresponding to each abnormal point is replaced with first data, wherein the first data comprises the alarm number of the fire-fighting equipment at the time corresponding to each abnormal point, and the first data is normal data in the first region, which does not contain abnormal data.

[0050] The method provided in the embodiment can ensure that the generated alarm time sequence data does not contain abnormal data, ensure that the alarm time sequence data can more accurately reflect the latest trend and current mode, provide more relevant information, help to reduce prediction error, improve the accuracy and reliability of fire alarm prediction, and ensure that the prediction and analysis results are consistent with the actual situation.

[0051] Optionally, after acquiring the alarm number of the fire-fighting equipment in the first time period in the first region to generate alarm time sequence data, the method further comprises: traversing the alarm time sequence data to determine the value of the alarm number of the fire-fighting equipment corresponding to each time point; determining the value range of the alarm number of the fire-fighting equipment based on the value of the alarm number of the fire-fighting equipment corresponding to each time point; determining a preset value based on the value range; the preset value is a positive number; adding the alarm number of the fire-fighting equipment corresponding to each time point to the preset value, and performing logarithmic operation on the added value to obtain the alarm time sequence data.

[0052] Specifically, the preset value is set according to actual needs, and this embodiment does not limit it. It should be noted that the preset value is a positive number; for example, in one embodiment, the preset value is 0.5.

[0053] In practice, the alarm time series data is traversed to determine the number of fire equipment alarms at each time point. Based on the number of fire equipment alarms at each time point, a range of values ​​for the number of fire equipment alarms is determined. Based on the minimum value within the range, a preset value is determined, i.e., the sum of the preset value and the minimum value is greater than 0. Further, the number of fire equipment alarms at each time point in the alarm time series data is added to the preset value, and a logarithmic operation is performed to obtain the updated alarm time series data.

[0054] The alarm time series data can be obtained using the following formula:

[0055] ;

[0056] Among them, the The updated alarm time series data; This is the alarm time series data before the update. The The length of the alarm time series data; This is the default value.

[0057] S103. Establish a Gaussian process model based on the alarm time series data.

[0058] Specifically, the Gaussian process model is used to predict the number of fire alarms in a second time period within a first time period within a first region. The Gaussian process model includes a mean function and a covariance function. The mean function characterizes the overall trend of the alarm time series data, while the covariance function characterizes the correlation between adjacent data points in the alarm time series data. The covariance function calculates the correlation between adjacent data points based on the time difference between different times corresponding to adjacent data points in the alarm time series data.

[0059] Furthermore, covariance functions include exponential covariance functions, periodic covariance functions, rational quadratic covariance functions, and noise covariance functions. The exponential covariance function is used to capture the smooth changes in alarm time series data, obtaining the changing trend of the number of fire equipment alarms at different times. The periodic covariance function is used to capture the recurring pattern of fire equipment alarms in alarm time series data. The rational quadratic covariance function is used to capture the number of fire equipment alarms that suddenly change in alarm time series data. The noise covariance function is used to handle random noise in alarm time series data.

[0060] It should be noted that the exponential covariance function is widely used in various prediction tasks due to its smoothness, and is suitable for simulating continuous and smooth changing data. The periodic covariance function is specially used for simulating data with periodic changes, and can capture the repeated patterns in the time series. The periodic covariance function can simulate the periodic alarm situation. The rational quadratic covariance function can reduce the influence of abnormal points in a certain alarm time series (there may be a sudden change in the data at a certain time point). The noise covariance function is used to add independent and identically distributed noise in the Gaussian process model, which helps to process random fluctuations in the data, thereby improving the adaptability of the Gaussian process model to actual data.

[0061] The exponential covariance function is calculated based on the alarm data of two adjacent time points of the alarm time series data, the overall fluctuation intensity of the alarm time series data, and the smoothness parameter of the alarm time series data. The periodic covariance function is calculated based on the alarm data of two adjacent time points of the alarm time series data, the time interval of repeated occurrence of the number of fire-fighting equipment alarms in the alarm time series data, the overall fluctuation intensity of the alarm time series data, and the smoothness parameter of the alarm time series data. The rational quadratic covariance function is calculated based on the alarm data of two adjacent time points of the alarm time series data, the overall fluctuation intensity of the alarm time series data, the smoothness parameter of the alarm time series data, and the weight of the sudden change in the number of fire-fighting equipment alarms in the alarm event sequence data. The noise covariance function is calculated based on the alarm data of two adjacent time points of the alarm time series data and the random noise in the alarm time series data.

[0062] In specific implementation, the alarm time series data is regarded as satisfying a Gaussian random process, that is:

[0063] ;

[0064] The mean function is calculated based on the alarm time series data, the Gaussian process model, and the alarm time series data. The covariance function is calculated based on the alarm time series data, the Gaussian process model, and the alarm time series data.

[0065] ;

[0066] The covariance function is calculated based on the alarm time series data, the Gaussian process model, and the alarm time series data. , , , The weight of the covariance function is calculated based on the alarm time series data, the Gaussian process model, and the alarm time series data. , , ,​​​​ is an internal parameter of the covariance function; the is an exponential covariance function; the is a periodic covariance function; the is a rational quadratic covariance function; the is a noise covariance function.

[0067] Further, it is assumed that , , , is a polynomial coefficient, the is the alarm time series data, then .

[0068] S104, based on the learned Gaussian process model, the number of fire-fighting equipment alarms in the second time period in the first area is predicted, and the time length of the second time period is longer than the time length of the first time period.

[0069] Specifically, the learned Gaussian process model represents the parameter optimization of the Gaussian process model to obtain the optimal model parameters (including the weight , , , internal parameters , , , and polynomial coefficients ) of the Gaussian process model, and the learned Gaussian process model is obtained based on the optimal model parameters.

[0070] In specific implementation, it is assumed that the optimal model parameters are independent of each other, and the parameter optimization problem is converted into a multi-dimensional optimization problem. By maximizing the marginal likelihood function (in the Gaussian process model, the marginal likelihood function evaluates the probability of data generation under the given model parameters and data points), the optimal model parameters can be effectively determined. In order to determine the optimal model parameters, the model parameters are adjusted to maximize the marginal likelihood function, that is, to minimize the negative log-likelihood. The Powell method (a conjugate gradient algorithm suitable for multi-variable optimization problems) is used for optimization, and after setting appropriate iteration parameters, the parameter space will be searched and iterated in detail until the optimal parameter combination that can maximize the marginal likelihood is determined.

[0071] Specifically, the marginal likelihood function can be expressed as:

[0072] ;

[0073] Wherein, the is the alarm time series data; the is an edge likelihood function; the is a time; the is a polynomial coefficient; the is a weight of a covariance function; the is an internal parameter of a covariance function; the is a polynomial matrix, , the is a column of a polynomial matrix; the is a covariance function; the is a length of alarm time series data.

[0074] Further, the edge likelihood function is maximized, that is, the negative log-likelihood is minimized:

[0075] ;

[0076] wherein, the is a polynomial coefficient; the is a weight of a covariance function; the is an internal parameter of a covariance function; the is the alarm time series data; the is an edge likelihood function; the is a time.

[0077] Further, the Powell method is used, and an optimization library such as Adam, l-brgs is used to solve the minimum negative log-likelihood, to obtain the optimal model parameters of the Gaussian process model.

[0078] In specific implementation, the alarm time series data (the number of fire-fighting equipment alarms in a first time period in a first region) is input into the learned Gaussian process model, to obtain the number of fire-fighting equipment alarms in a second time period in the first region, and the first region is managed in fire safety based on the number of fire-fighting equipment alarms in the second time period in the first region.

[0079] The method provided by the embodiment, in the first aspect, determines the number of fire-fighting equipment alarms in the first time period based on the first time length, and determines the third time period based on the second time period and the first time period, and selects the first time period closest to the current time from the third time period. By determining the first time period, the second time period and the third time period, and comparing and analyzing, the pattern and trend in the alarm time series data can be more accurately captured, and the accuracy of the fire alarm prediction is improved. In the second aspect, when the first time period is determined from the third time period, the first time period closest to the current time is selected based on the current time. This can ensure that the alarm time series data of the first time period used to predict the number of fire-fighting equipment alarms in the second time period can more accurately reflect the latest trend and current pattern, provide more relevant information, help reduce prediction errors, improve prediction accuracy and reliability, and ensure that the prediction and analysis results are consistent with the actual situation. In the third aspect, in the alarm time series data, the environment and the alarm pattern will fluctuate over time, for example, the number of fire-fighting equipment alarms may change significantly in a particular season, holiday or during a particular event. By dynamically adjusting the time period of data acquisition, these changes can be adapted to improve the prediction ability of the Gaussian process model. In particular for long-term prediction, the Gaussian process model needs to follow the changes in data patterns and update its historical data. Moreover, due to the particularity of the alarm time series data, the trend in a longer time period is more stable. In general time series prediction, short-term prediction is usually more accurate, but alarm time series data often has potential periodic trends and is significantly affected by seasons, population activities, etc. Because a fixed time period may not effectively adapt to new patterns or long-term trend changes when the Gaussian process model is faced with them, dynamic adjustment of the time period can better capture new data features. By lengthening the time period, the Gaussian process model can better exclude the interference of small fluctuations and grasp the overall trend. This not only improves the robustness of the prediction, but also reduces the dependence on sudden abnormal events or occasional alarms. By continuously updating and lengthening the time period, the Gaussian process model can exclude the influence of small fluctuations in different situations, thereby more accurately predicting the trend of the alarm time series data in the long term. In the fourth aspect, when constructing the Gaussian process model, the application combines four kinds of covariance functions, effectively grasps the direction of fire alarm prediction, and makes the training of data more reasonable. In this way, even with a relatively small data set, effective training can be achieved, avoiding the overfitting problem that traditional neural networks are prone to on small data sets. In the fifth aspect, compared with traditional neural network methods, the application does not need to repeatedly and tediously adjust the hyperparameters of the neural network. Only the covariance function and the mean function in the Gaussian process model need to be adjusted to achieve better fitting effect, greatly simplifying the model parameter adjustment process.This makes the application easier to implement, reduces the influence of human factors on model performance, and improves the accuracy and stability of prediction.

[0080] Corresponding to the foregoing embodiment of the fire alarm prediction method based on the Gaussian process, the application also provides an embodiment of a fire alarm prediction device based on the Gaussian process.

[0081] Figure 2 The structure of the embodiment one of the fire alarm prediction device based on the Gaussian process provided by the application is shown in the figure. Figure 2 The device provided by the embodiment includes a determination module 210, a generation module 220, an establishment module 230, and a prediction module 240.

[0082] The determination module 210 is configured to determine a first time length based on the fire alarm equipment state data in a first region and the state data in the first region.

[0083] The generation module 220 is configured to obtain the number of fire equipment alarms in the first region in a first time period, and generate alarm time sequence data, wherein the alarm time sequence data is vector data generated after the number of fire equipment alarms is arranged in time sequence, and the time length of the first time period is the first time length.

[0084] In the embodiment, a third time period identical to the first region in a second time period is determined based on the state data in the first region, the time length of the third time period is greater than the first time length, the time of the first time length closest to the current time in the third time period is taken as the first time period, and the second time period is a time period corresponding to the number of fire equipment alarms predicted based on the Gaussian process.

[0085] The establishment module 230 is configured to establish a Gaussian process model based on the alarm time sequence data.

[0086] The prediction module 240 is configured to predict the number of fire equipment alarms in a second time period in the first region based on the learned Gaussian process model, and the time length of the second time period is longer than the time length of the first time period.

[0087] The device of the embodiment can be used to execute the steps of the method embodiment shown in the figure, and the specific implementation principle and implementation process are similar, which will not be described here. Figure 1

[0088] The implementation process of the functions and roles of each unit in the device is specifically described in the implementation process of the corresponding steps in the above method, which will not be described here.

[0089] ​For the apparatus embodiment, since it basically corresponds to the method embodiment, the relevant part can be seen from the part of the method embodiment. The apparatus embodiment described above is only illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the application according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0090] The above only describes the preferred embodiments of the present application and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A fire alarm prediction method based on Gaussian processes, characterized in that, The method includes: The first time length is determined based on the status data of the fire alarm equipment in the first area and the status data in the first area. Based on the first time length, the number of fire equipment alarms in the first area during a first time period is obtained; wherein, the time length of the first time period is the first time length; Based on the status data of the first area within the second time period and the first time length, a third time period with the same status data is determined; wherein, the time length of the third time period is longer than the first time length, and the status data is the environmental conditions and external factors of the area where the fire alarm equipment is located; Based on the third time period, a first time period with a time length of the first time length and closest to the current time is determined within the third time period; The alarm count of fire equipment in the first time period within the first area is obtained, and alarm time series data is generated. The alarm time series data is vector data generated by arranging the alarm count of fire equipment in chronological order. Specifically, a third time period is determined based on the status data of the first area within the second time period. The length of the third time period is longer than that of the first time period. The time of the first time period that is closest to the current time within the third time period is taken as the first time period. The second time period is the time period corresponding to the number of fire equipment alarms predicted based on Gaussian process. A Gaussian process model is established based on the alarm time series data; The number of fire alarms in the first area during the second time period is predicted based on the learned Gaussian process model, where the length of the second time period is longer than that of the first time period.

2. The method according to claim 1, characterized in that, The determination of the first time length based on the status data of the fire alarm equipment in the first area and the status data in the first area includes: The status data of fire alarm equipment in the first area are analyzed based on time series analysis method to determine the status changes of fire alarm equipment and the behavior changes related to fire alarm in the first area, and to obtain the first trend of change. The time point of change is determined based on the first trend of change; The first time length is determined based on the first trend of change and the longest interval between the change time points.

3. The method according to claim 1, characterized in that, After determining, based on the third time period, the first time period whose length is the first time period and is closest to the current time, the method further includes: Anomaly processing is performed on the alarm time series data of the first time period in the first area to identify anomaly points; the difference between the number of fire equipment alarms corresponding to the anomaly point and the number of fire equipment alarms corresponding to the time point of change is greater than a preset value; The alarm counts of fire equipment corresponding to the anomaly point are replaced with the first data in the first region to obtain the updated alarm time series data; the first data includes the number of fire equipment alarms at the time corresponding to the anomaly point.

4. The method according to claim 1, characterized in that, The Gaussian process model includes a mean function and a covariance function. The mean function characterizes the overall trend of the alarm time series data, and the covariance function characterizes the correlation between adjacent data in the alarm time series data. The covariance function calculates the correlation between adjacent data based on the time difference between different times corresponding to adjacent data in the alarm time series data.

5. The method according to claim 4, characterized in that, The covariance functions include exponential covariance, periodic covariance, rational quadratic covariance, and noise covariance. The exponential covariance function is used to capture the smooth changes in alarm time series data to obtain the changing trend of the number of fire equipment alarms at different times. The periodic covariance function is used to capture the pattern of repeated occurrences of fire equipment alarms in the alarm time series data. The rational quadratic covariance function is used to capture the number of fire equipment alarms that change abruptly in the alarm time series data. The noise covariance function is used to handle random noise in the alarm time series data.

6. The method according to claim 5, characterized in that, The exponential covariance function is calculated based on the alarm data at two adjacent time points of the alarm time series data, the overall fluctuation intensity of the alarm time series data, and the smoothness parameter of the alarm time series data; the periodic covariance function is calculated based on the alarm data at two adjacent time points of the alarm time series data, the time interval between repeated occurrences of fire equipment alarms in the alarm time series data, the overall fluctuation intensity of the alarm time series data, and the smoothness parameter of the alarm time series data; the rational quadratic covariance function is calculated based on the alarm data at two adjacent time points of the alarm time series data, the overall fluctuation intensity of the alarm time series data, the smoothness parameter of the alarm time series data, and the weight of abrupt fire equipment alarm occurrences in the alarm time series data; the noise covariance function is calculated based on the alarm data at two adjacent time points of the alarm time series data and the random noise in the alarm time series data.

7. The method according to claim 1, characterized in that, After obtaining the number of fire equipment alarms in the first time period within the first area and generating alarm time series data, the method further includes: Iterate through the alarm time series data to determine the value of the number of alarms of the fire equipment corresponding to each time point; Based on the values ​​of the number of fire equipment alarms at each time point, the range of values ​​for the number of fire equipment alarms is determined. Based on the range of values, a preset value is determined; the preset value is a positive number. The alarm counts of fire equipment at each time point are added to the preset value, and the logarithmic operation is performed on the sum to obtain the alarm time series data.

8. The method according to claim 1, characterized in that, The step of determining a third time period that is identical to the state data within a first region during a second time period and the first time length includes: The number and working status of fire-fighting equipment in the first area during the second time period are determined as the fire alarm equipment status data in the first area during the second time period. Determine a third time period within the first area that is identical to the status data of the fire alarm equipment; Based on the abnormal activities in the third time period within the first area, the status data of the fire alarm equipment in the third time period within the first area is corrected. The abnormal activities are those that affect the number of fire-fighting equipment and the working status of the fire-fighting equipment. The time period in the corrected fire alarm equipment status data that is still the same as the fire alarm equipment status data in the second time period in the first area will be used as the final third time period.

9. A fire alarm prediction device based on Gaussian processes, characterized in that, The device includes a determining module, a generating module, an establishing module, and a predicting module; wherein... The determining module is used to determine a first time length based on the status data of the fire alarm equipment in the first area and the status data in the first area; Based on the first time length, the number of fire equipment alarms in the first area during a first time period is obtained; wherein, the time length of the first time period is the first time length; Based on the status data of the first area within the second time period and the first time length, a third time period with the same status data is determined; wherein, the time length of the third time period is longer than the first time length, and the status data is the environmental conditions and external factors of the area where the fire alarm equipment is located; Based on the third time period, a first time period with a time length of the first time length and closest to the current time is determined within the third time period; The generation module is used to obtain the number of fire equipment alarms in the first time period within the first area and generate alarm time series data. The alarm time series data is vector data generated by arranging the number of fire equipment alarms in chronological order. Specifically, a third time period is determined based on the status data of the first area within the second time period. The length of the third time period is longer than that of the first time period. The time of the first time period that is closest to the current time within the third time period is taken as the first time period. The second time period is the time period corresponding to the number of fire equipment alarms predicted based on Gaussian process. The establishment module is used to establish a Gaussian process model based on the alarm time series data; The prediction module is used to predict the number of fire equipment alarms in the first area during a second time period based on the learned Gaussian process model, wherein the length of the second time period is longer than the length of the first time period.

Images