Fire-fighting early warning method based on intelligent lock and combining temperature and smoke sensing data

Abstract

The invention relates to the technical field of data processing, and particularly discloses a fire-fighting early warning method combining temperature and smoke sensing data based on an intelligent lock, and the method comprises the steps: collecting temperature data and smoke sensing data from an environment; if the variation amplitude of the temperature data exceeds a first preset threshold value and the peak value of the smoke sensing data exceeds a second preset threshold value, constructing a fusion feature vector through the temperature data and the smoke sensing data, and obtaining a potential fire risk level by adopting a classification algorithm according to the fusion feature vector; acquiring a temperature change trend sequence and a smoke change trend sequence, performing tracking analysis, and determining an early warning response score in combination with the potential fire risk level; if the early warning response score exceeds an early warning preset threshold value, an early warning signal is output, and an unlocking instruction is sent to the owner APP through the Internet platform. According to the fire-fighting early warning method based on the intelligent lock and combining the temperature and smoke sensing data, the problem that the early warning precision and the rescue efficiency are difficult to improve in a current fire-fighting early warning mode is solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a fire early warning method based on smart locks that combines temperature and smoke detection data. Background Technology

[0002] In modern society, fire safety, as a crucial component of public safety, is paramount, concerning people's lives and property. With rapid urbanization and the widespread adoption of smart technologies, fire early warning systems play a vital role in disaster prevention and loss reduction. However, existing fire early warning methods often suffer from significant deficiencies in data integration and real-time response. Many systems rely solely on data from a single type of sensor, failing to comprehensively reflect the diversity and complexity of fire hazards.

[0003] Temperature changes serve as a crucial indicator of fire, requiring extremely high sensitivity and real-time monitoring. However, relying solely on temperature data often fails to distinguish the subtle differences between normal high temperatures and potential fire hazards. While the addition of smoke detector data provides more information when smoke or gas leaks are present, it also introduces new challenges—accurately assessing the true extent of fire risk amidst interference from multiple data sources. This complex interrelationship between data often leads to misjudgments or delays in real-world scenarios due to the system's inability to effectively integrate information. Furthermore, current alarm systems struggle to integrate with smart locks after an alarm is triggered, leaving interior doors locked and hindering firefighters' rescue efforts. Therefore, current fire alarm methods fall short in improving both warning accuracy and rescue efficiency. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the present invention provides a highly adaptable track power box structure, which aims to solve the problems in the above-mentioned existing technology.

[0005] The technical solution adopted by this invention to solve its technical problem is: a fire early warning method based on smart locks that combines temperature and smoke detection data, comprising the following steps: S1: Collect a multi-source environmental dataset from the environment, including temperature data and smoke sensing data; S2: If the change in temperature data in the multi-source environmental dataset exceeds the first preset threshold and the peak value of smoke detection data exceeds the second preset threshold, then a fusion feature vector is constructed using temperature data and smoke detection data, and a classification algorithm is used to obtain the potential fire hazard level based on the fusion feature vector. S3: Obtain the temperature change trend sequence and the smoke detection change trend sequence and perform tracking analysis, and determine the early warning response score in combination with the potential fire hazard level; S4: If the warning response score exceeds the preset warning threshold, output a warning signal and send an unlocking request instruction to the owner's APP through the Internet platform.

[0006] Preferably, in step S1, temperature data and smoke detection data in the environment are acquired from the sensor network; The temperature data and smoke sensing data were timestamped using the Pandas library, and data with uniform format were filtered out by consistency checks to generate temperature change sequences and smoke sensing change sequences. The median absolute deviation method is used to detect outliers in the temperature change sequence and the smoke sensation change sequence, respectively. If the detected outlier exceeds a preset threshold range, the outlier is marked as interference data and excluded, resulting in a multi-source environmental dataset containing the cleaned temperature change sequence and the smoke sensation change sequence.

[0007] Optionally, in step S2, the temperature change sequence corresponding to the temperature data and the smoke sensing change sequence corresponding to the smoke sensing data are obtained from the multi-source environmental dataset, the change amplitude is calculated for the temperature change sequence, and the peak value is calculated for the smoke sensing change sequence. If the change magnitude of the temperature change sequence exceeds a preset first threshold and the peak value of the smoke sensation change sequence exceeds a preset second threshold, then a fused feature vector is constructed based on the temperature change sequence and the smoke sensation change sequence.

[0008] Specifically, in step S2, the Noise Robust Decrease Rate (NRD) of the fused feature vector is obtained, and the Noise Robust Decrease Rate (NRD), the change amplitude of the temperature change sequence, and the peak value of the smoke sensation change sequence are input into a pre-established random forest classification model to obtain preliminary hazard classification results. By mapping the preliminary hazard classification results to different levels, the potential fire hazard level is determined.

[0009] It is worth noting that in step S2, Gaussian noise is used to add noise to the fused feature vector to obtain a noisy feature vector; by comparing the performance of the fused feature vector and the noisy feature vector, the noise robustness descent rate (NRD) is calculated.

[0010] Preferably, in step S2, the step of constructing the fused feature vector includes: A convolutional layer is used to extract local features from the temperature change sequence to obtain a temperature feature sub-vector; A convolutional layer is used to extract local features from the smoke sensation change sequence to obtain smoke sensation feature sub-vectors; The temperature feature subvector and the smoke sensing feature subvector are concatenated to obtain a preliminary fused subvector. The initial fused subvector is transformed using a fully connected layer to obtain the transformed fused subvector. The hidden state of the Long Short-Term Memory network is initialized based on the transformed fused sub-vectors; The smoke-sensing change sequence is processed using a Long Short-Term Memory (LSTM) network, where the current hidden state is determined by ht=tanh(W_hh). ht-1+W_xh xt) is calculated to obtain the enhanced smoke sensing feature sub-vector, tanh() is the hyperbolic tangent function, ht is the current hidden state, ht-1 is the previous hidden state, W_hh is the weight matrix from hidden state to hidden state, W_xh is the weight matrix from input to hidden state, and xt is the element of the smoke sensing change sequence at the current time. The enhanced smoke-sensing feature vector and the transformed fusion vector are weighted and fused to obtain a weighted fusion result; the weighted fusion result is then normalized to obtain a fused feature vector.

[0011] Optionally, in step S3, temperature change data and smoke detection change data from the current time to a preset future time window are obtained from the sensor network to obtain the temperature change trend sequence and the smoke detection change trend sequence. For the temperature change trend sequence and the smoke sensation change trend sequence, data at each time point are extracted, the difference between each time point and the previous time point is calculated, and all differences are summed to obtain the total temperature difference and the total smoke sensation difference, respectively. The temperature increase index value and the smoke sensation increase index value are determined by normalizing the ratio of the sum of the differences to the maximum value of the sequence. The warning response score is obtained by weighting the temperature increase index value, the smoke detection increase index value, and the fire hazard level using preset weight values.

[0012] Specifically, in step S4, the warning response score is compared with a preset threshold. If the warning response score exceeds the preset warning threshold, a warning signal is obtained and the warning signal is output. The current lock status is obtained from the smart lock device using the API interface. If the current lock status is locked, an unlock request command is sent to the owner's APP through the Internet platform.

[0013] The beneficial effects of this invention are as follows: In the fire early warning method based on smart locks and combining temperature and smoke detection data, temperature and smoke detection data are first collected from the environment. When the temperature change exceeds a first preset threshold and the smoke detection peak exceeds a second preset threshold, a fusion feature vector is constructed and a classification algorithm is used to determine the potential fire hazard level. Then, the temperature and smoke detection change trend sequences are obtained for tracking and analysis, and an early warning response score is calculated based on the hazard level. If the score exceeds the preset early warning threshold, an early warning signal is immediately output, and an unlocking request is sent to the owner's APP via an internet platform. This method achieves an organic combination of deep fusion of multi-source data, hazard classification, and dynamic trend assessment, effectively improving the accuracy and timeliness of fire early warning, significantly reducing the false alarm rate, supporting rapid emergency response, and improving the efficiency of fire rescue through alarm and unlocking linkage. Attached Figure Description

[0014] Figure 1 This is a flowchart of a fire early warning method based on smart locks that combines temperature and smoke detection data.

[0015] Figure 2 This is a flowchart of the steps in step S2.

[0016] Figure 3 This is a flowchart of the steps in step S3. Detailed Implementation

[0017] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0018] Combination Figures 1 to 3 The fire early warning method based on smart locks and combining temperature and smoke detection data includes the following steps: S1: Collect a multi-source environmental dataset from the environment, including temperature data and smoke sensing data; S2: If the change in temperature data in the multi-source environmental dataset exceeds the first preset threshold and the peak value of smoke detection data exceeds the second preset threshold, then a fusion feature vector is constructed using temperature data and smoke detection data, and a classification algorithm is used to obtain the potential fire hazard level based on the fusion feature vector. S3: Obtain the temperature change trend sequence and the smoke detection change trend sequence and perform tracking analysis, and determine the early warning response score in combination with the potential fire hazard level; S4: If the warning response score exceeds the preset warning threshold, output a warning signal and send an unlocking request instruction to the owner's APP through the Internet platform.

[0019] In the fire early warning method based on smart locks and combining temperature and smoke detection data, temperature and smoke detection data are first collected from the environment. When the temperature change exceeds a first preset threshold and the smoke peak exceeds a second preset threshold, a fused feature vector is constructed and a classification algorithm is used to determine the potential fire hazard level. Then, the temperature and smoke detection change trend sequences are acquired and tracked for analysis, and an early warning response score is calculated based on the hazard level. If the score exceeds the preset early warning threshold, an early warning signal is immediately output, and an unlocking request is sent to the owner's APP via an internet platform. This method achieves an organic combination of deep fusion of multi-source data, hazard classification, and dynamic trend assessment, effectively improving the accuracy and timeliness of fire early warning, significantly reducing the false alarm rate, supporting rapid emergency response, and improving the efficiency of fire rescue through alarm and unlocking linkage.

[0020] Preferably, in step S1, temperature data and smoke detection data in the environment are acquired from the sensor network; The temperature data and smoke sensing data were timestamped using the Pandas library, and data with uniform format were filtered out by consistency checks to generate temperature change sequences and smoke sensing change sequences. The median absolute deviation method is used to detect outliers in the temperature change sequence and the smoke sensation change sequence, respectively. If the detected outlier exceeds a preset threshold range, the outlier is marked as interference data and excluded, resulting in a multi-source environmental dataset containing the cleaned temperature change sequence and the smoke sensation change sequence.

[0021] Consider a smart furniture monitoring system as an example of acquiring temperature and smoke detection data from a sensor network. Assume the system's sensor network comprises temperature sensors and smoke detectors distributed across various locations, transmitting data in real time via an internet platform. Specifically, temperature sensors record ambient temperature values, while smoke detectors record smoke concentration values. The system retrieves this raw data from the sensor network nodes, ensuring that the data packets include timestamps and sensor IDs to prevent information loss due to transmission delays.

[0022] When using the Pandas library to align temperature and smoke sensing data with timestamps, the data must first be imported into a DataFrame structure. Pandas is a powerful Python data analysis library that allows for efficient processing of tabular and time series data. For example, the temperature and smoke sensing data can be loaded into two separate DataFrames, and then the merge function can be used to perform an inner join based on the timestamp column, ensuring data alignment at each time point. For instance, if the temperature data has a timestamp of 2023-10-01 12:00:00, while the smoke sensing data has a slight deviation such as 2023-10-01 12:00:05, the resample method can be used to unify them to the minute level.

[0023] Median absolute deviation (MAD) is a robust statistical method used to identify outliers that deviate from the median. Specifically, it first calculates the median of the temperature or smoke perception change series, then calculates the absolute deviation of each data point from the median, and finally takes the median of these absolute deviations as the MAD value. Preferably, the outlier threshold is the median of the change series ±3. MAD. For example, for a temperature variation series, if the median is 25 degrees Celsius and the MAD is 2, then values ​​exceeding 31 or falling below 19 are considered outliers. This method is more resistant to extreme values ​​than standard deviation and is particularly effective in environmental monitoring because it can distinguish between normal fluctuations and interference caused by sensor malfunctions.

[0024] Optionally, in step S2, temperature change sequences corresponding to temperature data and smoke sensing change sequences corresponding to smoke sensing data are obtained from the multi-source environmental dataset. The magnitude of the temperature change sequence is calculated, and the peak value is calculated for the smoke sensing change sequence. The magnitude of the temperature change sequence can be determined by comparing the maximum and minimum temperature values ​​within a specified time window. The peak value of the smoke sensing change sequence can be calculated by identifying the maximum smoke sensing reading within the specified time window. If the change magnitude of the temperature change sequence exceeds a preset first threshold and the peak value of the smoke sensation change sequence exceeds a preset second threshold, then a fused feature vector is constructed based on the temperature change sequence and the smoke sensation change sequence.

[0025] When constructing the fused feature vector, key indicators from the temperature change sequence and the smoke perception change sequence can be combined to form a multidimensional feature representation. This approach can comprehensively reflect the changing trends of the environmental state, and this vector can provide more comprehensive data support for subsequent analysis.

[0026] Specifically, in step S2, the Noise Robust Decrease Rate (NRD) of the fused feature vector is obtained, and the Noise Robust Decrease Rate (NRD), the change amplitude of the temperature change sequence, and the peak value of the smoke sensation change sequence are input into a pre-established random forest classification model to obtain preliminary hazard classification results. By mapping the preliminary hazard classification results to different levels, the potential fire hazard level is determined.

[0027] Specifically, when the NRD, temperature change amplitude, and smoke detection peak are input into the random forest classification model, the random forest classification model consists of multiple decision trees, each trained based on a subset of samples. For example, the input to the random forest classification model is a noise robust descent rate NRD of 0.15, a temperature change amplitude of 12 degrees Celsius, and a smoke detection peak of 400 ppm. The preliminary hazard classification result, such as "high risk" or "low risk", is obtained through tree voting. The advantage of random forest is that ensemble learning reduces overfitting and can handle nonlinear relationships in environmental hazard detection.

[0028] In one possible implementation, by performing a hierarchical mapping on the preliminary hazard classification results, the classification output can be mapped to a specific level. For example, "high risk" is mapped to a level value of 0.8 through the mapping table, and "low risk" is mapped to a level value of 0.4 through the mapping table. Then, the count value is used as the potential fire hazard level.

[0029] It is worth noting that in step S2, Gaussian noise is used to add noise to the fused feature vector to obtain a noisy feature vector; by comparing the performance of the fused feature vector and the noisy feature vector, the noise robustness descent rate (NRD) is calculated.

[0030] To address the noise reduction process for fused feature vectors, Gaussian noise can be introduced to simulate data interference in real-world environments, resulting in noisy feature vectors. By comparing the performance differences of the feature vectors before and after noise reduction in classification tasks, the Noise Robust Reduction Rate (NRD) can be calculated. Assuming the accuracy is 90% before noise reduction and drops to 85% after, the NRD is 5%, reflecting the stability of the data in noisy environments. This analysis helps assess the robustness of feature vectors against interference.

[0031] Specifically, in step S2, the step of constructing the fused feature vector includes: A convolutional layer is used to extract local features from the temperature change sequence to obtain a temperature feature sub-vector; A convolutional layer is used to extract local features from the smoke sensation change sequence to obtain smoke sensation feature sub-vectors; The temperature feature subvector and the smoke sensing feature subvector are concatenated to obtain a preliminary fused subvector. The initial fused subvector is transformed using a fully connected layer to obtain the transformed fused subvector. The hidden state of the Long Short-Term Memory network is initialized based on the transformed fused sub-vectors; The smoke-sensing change sequence is processed using a Long Short-Term Memory (LSTM) network, where the current hidden state is determined by ht=tanh(W_hh). ht-1+W_xh xt) is calculated to obtain the enhanced smoke sensing feature sub-vector, tanh() is the hyperbolic tangent function, ht is the current hidden state, ht-1 is the previous hidden state, W_hh is the weight matrix from hidden state to hidden state, W_xh is the weight matrix from input to hidden state, and xt is the element of the smoke sensing change sequence at the current time. The enhanced smoke-sensing feature vector and the transformed fusion vector are weighted and fused to obtain a weighted fusion result; the weighted fusion result is then normalized to obtain a fused feature vector.

[0032] Specifically, when using convolutional layers to extract local features from a temperature change sequence to obtain a temperature feature subvector, the convolutional layer captures local patterns by scanning the sequence through a sliding window. For example, if the temperature change sequence shows a gradual increase from 22 degrees Celsius to 28 degrees Celsius, the convolutional operation extracts the edge features of the upward trend, forming a low-dimensional subvector. Similarly, when using convolutional layers to extract local features from a smoke perception change sequence to obtain a smoke perception feature subvector, the convolutional layer focuses on the peak region of the smoke perception change sequence. For example, if the smoke perception concentration increases from 10 ppm to 150 ppm, the local texture of the concentration gradient is extracted, forming a smoke perception subvector, thus highlighting the dynamic pattern of smoke diffusion. When concatenating the temperature feature subvector and the smoke perception feature subvector to obtain a preliminary fused subvector, the two are integrated through a simple concatenation method. For example, the dimensions of the temperature subvector and the smoke perception subvector are connected into a larger vector. This can capture the correlation between temperature increase and smoke increase, avoiding the limitations of a single sequence. When the initial fused subvector is transformed using a fully connected layer to obtain the transformed fused subvector, the fully connected layer applies a linear transformation to adjust the vector weights. For example, after inputting the fused subvector, a more compact representation is generated through matrix multiplication, which improves the non-linear expressive power of the features.

[0033] When initializing the hidden state of the Long Short-Term Memory (LSTM) network based on the transformed fused sub-vector, it is used as the initial input to the hidden layer, ensuring that the network starts with fused information and captures long-term dependencies. When processing the smoke perception change sequence using the LSM network to obtain the enhanced smoke perception feature sub-vector, the LSM network processes elements in the sequence through a gating mechanism, such as updating the hidden state step by step, enhancing the temporal characteristics of the sequence, which highlights the persistent accumulation pattern of smoke. When weighted fusion of the enhanced smoke perception feature sub-vector and the transformed fused sub-vector, the weight coefficients are multiplied and summed. In this embodiment, the enhanced smoke perception feature sub-vector is given a higher weight; by integrating multi-source information, the overall robustness is improved. Finally, when normalizing the weighted fusion result to obtain the fused feature vector, the values ​​are scaled to the 0-1 range to standardize the vector, thereby bringing a more reliable hazard identification effect in detection.

[0034] Preferably, in step S3, temperature change data and smoke detection change data from the current time to a preset future time window are obtained from the sensor network to obtain the temperature change trend sequence and the smoke detection change trend sequence. For the temperature change trend sequence and the smoke sensation change trend sequence, data at each time point are extracted, the difference between each time point and the previous time point is calculated, and all differences are summed to obtain the total temperature difference and the total smoke sensation difference, respectively. The temperature increase index value and the smoke sensation increase index value are determined by normalizing the ratio of the sum of the differences to the maximum value of the sequence. The warning response score is obtained by weighting the temperature increase index value, the smoke detection increase index value, and the fire hazard level using preset weight values.

[0035] For example, if the preset time window is set to one hour, when acquiring temperature and smoke detection data from the sensor network from the current time to a future preset time window, data can be collected from the sensor network starting at the current time, such as 10:00 AM, and ending at 11:00 AM. This data can then form a corresponding trend sequence ordered by time, such as a temperature trend sequence showing a gradual increase from 25 degrees Celsius to 28 degrees Celsius, or a smoke detection trend sequence showing a change from a low concentration, such as 50 ppm, to a high concentration, such as 150 ppm. This acquisition process emphasizes real-time performance.

[0036] Specifically, after extracting data for each time point from the temperature change trend series and the smoke sensation change trend series, the difference between each time point and the previous time point is calculated, and then summed to obtain the total. For example, for the temperature change trend series, the time point data of the series are 25, 26, and 28 degrees Celsius, with differences of 1 and 2, and the sum is 3 as the total temperature difference. For the smoke sensation change trend series, the time point data of the series are 50, 100, and 150 ppm, with differences of 50 and 50, and the sum is 100 as the total smoke sensation difference. This calculation method captures cumulative changes and helps to quantify the overall trend.

[0037] When determining the temperature increase index and smoke sensation increase index by normalizing the ratio of the sum of the differences to the maximum value of the sequence, the ratio of the sum of the temperature differences (3) to the maximum value of the sequence (28) is approximately 0.107, which can be used as the temperature increase index. Similarly, the ratio of the sum of the smoke sensation differences (100) to the maximum value (150) is approximately 0.667, which can be used as the smoke sensation increase index. This normalization makes data of different dimensions comparable and facilitates integrated analysis.

[0038] Finally, the process of calculating the early warning response score by weighting the temperature increase index value, smoke detection increase index value, and fire hazard level using preset weight values ​​can be used. For example, the temperature weight can be set to 0.4, the smoke detection weight to 0.4, and the hazard level weight to 0.2. If the temperature increase is 0.107, the smoke detection increase is 0.667, and the hazard level is 0.8, then the score is calculated as (0.4...). 0.107 + 0.4 0.667 + 0.2 The result is a score of 0.8548 (0.8). This weighted average integrates multi-source information and provides a comprehensive early warning basis. In factory safety management, it can improve response efficiency and avoid misjudgment caused by a single indicator.

[0039] The connection between the above steps lies in the fact that data acquisition provides input for calculating the difference, while the calculation of the sum of the differences lays the foundation for normalization. Finally, the weighted calculation achieves the overall assessment. From obtaining the sequence to calculating the score, the entire process forms a closed loop, ensuring that the early warning response score accurately reflects the fire risk, thereby guiding timely intervention measures.

[0040] It is worth noting that in step S4, the warning response score is compared with a preset threshold. If the warning response score exceeds the preset warning threshold, a warning signal is obtained and the warning signal is output. The current lock status is obtained from the smart lock device using the API interface. If the current lock status is locked, an unlock request command is sent to the owner's APP through the Internet platform.

[0041] When comparing the early warning response score with a preset threshold, for example, if the calculated early warning response score is 0.8548 and the preset threshold is set to 0.5, the comparison determines that 0.8548 is greater than 0.5, thus confirming the existence of a potential risk. This comparison mechanism is based on a simple threshold judgment principle, that is, the early warning response score represents the quantitative result of comprehensive fire hazard, and the preset threshold value is a boundary value predefined based on historical data and safety standards. If the score exceeds the threshold, subsequent response actions are triggered.

[0042] The steps for obtaining the current lock / unlock status from a smart lock device using an API interface can be implemented in an access control system linked to fire alarms. An API is an application programming interface used for data exchange between devices. For example, by calling the smart lock's endpoint URL, such as " / lock / status," via a RESTful API, status data is obtained from the device server. The current lock / unlock status refers to the lock's locked or unlocked state; for example, "locked" indicates a locked door. If the current lock / unlock status is locked, during the process of sending an unlock request command to the homeowner's mobile app via an internet platform, the internet platform, such as a cloud service platform, acts as an intermediary. After receiving the locked status, it generates a command data packet and pushes it to the homeowner's mobile app. For example, the command content might be "Potential fire detected, do you agree to remote unlocking?" Upon receiving the request, the app displays a confirmation button. This request mechanism introduces human intervention to avoid security risks associated with automatic unlocking. Simultaneously, the platform uses the HTTPS protocol to ensure secure transmission.

[0043] In this solution, the connection between early warning comparison and smart lock acquisition is that the early warning signal acts as a trigger to initiate a lock status query. After the signal is output, the system automatically calls the API to check the door lock. If the door is locked, a request is initiated, forming a complete chain from risk detection to response authorization, ensuring that the door lock can respond quickly in an emergency.

[0044] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A fire early warning method based on smart locks, combining temperature and smoke detection data, characterized in that, Includes the following steps: S1: Collect a multi-source environmental dataset from the environment, including temperature data and smoke sensing data; S2: If the change in temperature data in the multi-source environmental dataset exceeds the first preset threshold and the peak value of smoke detection data exceeds the second preset threshold, then a fusion feature vector is constructed using temperature data and smoke detection data, and a classification algorithm is used to obtain the potential fire hazard level based on the fusion feature vector. S3: Obtain the temperature change trend sequence and the smoke detection change trend sequence and perform tracking analysis, and determine the early warning response score in combination with the potential fire hazard level; S4: If the warning response score exceeds the preset warning threshold, output a warning signal and send an unlocking request instruction to the owner's APP through the Internet platform.

2. The fire early warning method based on smart locks and combining temperature and smoke detection data as described in claim 1, characterized in that: In step S1, temperature data and smoke detection data in the environment are acquired from the sensor network; The temperature data and smoke sensing data were timestamped using the Pandas library, and data with uniform format were filtered out by consistency checks to generate temperature change sequences and smoke sensing change sequences. The median absolute deviation method is used to detect outliers in the temperature change sequence and the smoke sensation change sequence, respectively. If the detected outlier exceeds a preset threshold range, the outlier is marked as interference data and excluded, resulting in a multi-source environmental dataset containing the cleaned temperature change sequence and the smoke sensation change sequence.

3. A fire early warning method based on a smart lock and combining temperature and smoke detection data, as described in claim 1, is characterized in that: In step S2, temperature change sequences corresponding to temperature data and smoke sensing change sequences corresponding to smoke sensing data are obtained from the multi-source environmental dataset. The change amplitude is calculated for the temperature change sequence, and the peak value is calculated for the smoke sensing change sequence. If the change magnitude of the temperature change sequence exceeds a preset first threshold and the peak value of the smoke sensation change sequence exceeds a preset second threshold, then a fused feature vector is constructed based on the temperature change sequence and the smoke sensation change sequence.

4. A fire early warning method based on a smart lock and combining temperature and smoke detection data, as described in claim 3, is characterized in that: In step S2, the Noise Robust Decrease Rate (NRD) of the fused feature vector is obtained. The Noise Robust Decrease Rate (NRD), the magnitude of the temperature change sequence, and the peak value of the smoke sensation change sequence are input into a pre-established random forest classification model to obtain preliminary hazard classification results. By mapping the preliminary hazard classification results to different levels, the potential fire hazard level is determined.

5. A fire early warning method based on a smart lock and combining temperature and smoke detection data, as described in claim 4, characterized in that: In step S2, Gaussian noise is used to add noise to the fused feature vector to obtain a noisy feature vector; the noisy robust descent rate (NRD) is calculated by comparing the performance of the fused feature vector and the noisy feature vector.

6. A fire early warning method based on a smart lock and combining temperature and smoke detection data, as described in claim 3, is characterized in that: In step S2, the step of constructing the fused feature vector includes: A convolutional layer is used to extract local features from the temperature change sequence to obtain a temperature feature sub-vector; A convolutional layer is used to extract local features from the smoke sensation change sequence to obtain smoke sensation feature sub-vectors; The temperature feature subvector and the smoke sensing feature subvector are concatenated to obtain a preliminary fused subvector. The initial fused subvector is transformed using a fully connected layer to obtain the transformed fused subvector. The hidden state of the Long Short-Term Memory network is initialized based on the transformed fused sub-vectors; The smoke-sensing change sequence is processed using a Long Short-Term Memory (LSTM) network, where the current hidden state is determined by ht=tanh(W_hh). ht-1+W_xh xt) is calculated to obtain the enhanced smoke sensing feature sub-vector, tanh() is the hyperbolic tangent function, ht is the current hidden state, ht-1 is the previous hidden state, W_hh is the weight matrix from hidden state to hidden state, W_xh is the weight matrix from input to hidden state, and xt is the element of the smoke sensing change sequence at the current time. The enhanced smoke-sensing feature vector and the transformed fusion vector are weighted and fused to obtain a weighted fusion result; the weighted fusion result is then normalized to obtain a fused feature vector.

7. A fire early warning method based on a smart lock and combining temperature and smoke detection data, as described in claim 1, characterized in that: In step S3, temperature change data and smoke detection change data from the current time to a preset future time window are obtained from the sensor network to obtain the temperature change trend sequence and the smoke detection change trend sequence. For the temperature change trend sequence and the smoke sensation change trend sequence, data at each time point are extracted, the difference between each time point and the previous time point is calculated, and all differences are summed to obtain the total temperature difference and the total smoke sensation difference, respectively. The temperature increase index value and the smoke sensation increase index value are determined by normalizing the ratio of the sum of the differences to the maximum value of the sequence. The warning response score is obtained by weighting the temperature increase index value, the smoke detection increase index value, and the fire hazard level using preset weight values.

8. A fire early warning method based on a smart lock and combining temperature and smoke detection data, as described in claim 1, characterized in that: In step S4, the warning response score is compared with a preset threshold. If the warning response score exceeds the preset warning threshold, a warning signal is obtained and output. The current lock status is obtained from the smart lock device using the API interface. If the current lock status is locked, an unlock request command is sent to the owner's APP through the Internet platform.

Images