Kitchen safety detection method, system and equipment based on data perception and medium

Abstract

The invention relates to a kitchen safety detection method, system and device based on data perception and a medium. The method comprises the following steps: acquiring multi-source heterogeneous time sequence monitoring data in a kitchen; inputting the multi-source heterogeneous time sequence monitoring data into a time sequence risk feature extraction model to obtain a multi-dimensional time sequence feature vector; the multi-dimensional time sequence feature vector is used for representing the comprehensive safety state of the kitchen; generating a kitchen comprehensive risk score at the current moment based on the multi-dimensional time sequence feature vector; generating a hidden danger result of the kitchen at the current moment according to the comprehensive risk score of the kitchen at the current moment and a dynamic safety threshold value; the hidden danger result comprises existence of potential safety hazards and absence of potential safety hazards; and if the hidden danger result shows that the potential safety hazard exists, differential early warning information is generated according to the hidden danger result, and the differential early warning information is sent to each collaborative terminal. By adopting the method, quantitative evaluation and dynamic grading early warning of the multi-source safety risk of the kitchen can be realized.

Description

Technical Field

[0001] This invention belongs to the field of kitchen safety monitoring technology, and specifically relates to a data-sensing-based kitchen safety detection method, system, equipment, and medium. Background Technology

[0002] With the development of smart home and restaurant kitchen intelligent technologies, various kitchen safety monitoring technologies have gradually entered the application stage. Most of these technologies have single-dimensional risk perception or behavior control capabilities in specific scenarios, thus forming a traditional kitchen safety protection method based on single sensor monitoring and manual inspection, supplemented by machine vision and human behavior management in some scenarios.

[0003] In traditional technologies, most residential and commercial kitchens monitor specific safety indicators using single sensing devices such as gas sensors and smoke sensors, while also conducting regular manual inspections to check for potential hazards such as the status of stoves and the food storage environment. In addition, some technical solutions attempt to use multimodal AI models to fuse and analyze information such as images and sounds in the kitchen to monitor human behavior, or to achieve intelligent control of range hoods based on odor and smoke signals collected by electronic noses and smoke sensors.

[0004] However, current kitchen safety protection methods have many significant drawbacks: First, monitoring methods relying on single sensors cannot comprehensively perceive multiple dynamic data such as temperature, smoke, gas concentration, flame status, and personnel presence in real time, making it prone to problems such as delayed risk detection, frequent false alarms and missed alarms, and failing to provide effective early warnings before danger occurs; Second, personnel operation behavior management solutions based on machine vision are completely unable to detect progressive hazards caused by non-human behaviors in scenarios where no one is on duty or personnel are briefly away from their posts, such as stoves not being turned off, minor gas leaks, continuously rising oil temperatures, and refrigerator power outages; Third, existing multimodal AI monitoring solutions mostly focus on analyzing personnel operation behaviors, and do not adequately cover hidden risks such as equipment failures and environmental anomalies that are not related to personnel. Furthermore, detection systems based on odor and smoke signals can only control range hoods and cannot take into account early warnings of core safety hazards such as gas leaks and food spoilage, ultimately making it impossible to avoid a large number of kitchen safety accidents. Summary of the Invention

[0005] Therefore, it is necessary to provide a data-driven kitchen safety detection method, system, equipment, and medium to address the aforementioned technical issues.

[0006] Firstly, this application provides a data-driven kitchen safety detection method, comprising:

[0007] Acquire multi-source heterogeneous time-series monitoring data in the kitchen; the multi-source heterogeneous time-series monitoring data includes at least one of the following: water leakage status data, gas concentration time-series data, smoke particle concentration data, and ambient temperature time-series data;

[0008] Multi-source heterogeneous time-series monitoring data are input into the time-series risk feature extraction model to obtain a multi-dimensional time-series feature vector; the multi-dimensional time-series feature vector is used to characterize the overall safety status of the kitchen.

[0009] Based on multi-dimensional temporal feature vectors, a comprehensive risk score for the kitchen at the current moment is generated;

[0010] Based on the kitchen's overall risk score and dynamic safety threshold at the current moment, the potential hazard results for the kitchen at the current moment are generated; the hazard results include whether there are safety hazards or not.

[0011] If the hazard result indicates the existence of a safety hazard, differentiated early warning information is generated based on the hazard result and sent to each collaborative terminal.

[0012] In one embodiment, a comprehensive kitchen risk score for the current moment is generated based on a multi-dimensional temporal feature vector, including:

[0013] Dynamic importance weights are assigned to each dimension of the multidimensional time-series feature vector to obtain the dynamic weight coefficients of each dimension.

[0014] Based on the features of each dimension and the dynamic weight coefficients of each feature, weighted features of each dimension are generated.

[0015] Based on the features of each weighted dimension and combined with the preset feature-hazard probability mapping rules, the single-dimensional safety hazard probability corresponding to each weighted dimension feature is obtained, and a multi-dimensional safety hazard probability set for the kitchen is generated based on the single-dimensional safety hazard probability.

[0016] By performing time-dimensional correlation correction on the probability set of multidimensional safety hazards in the kitchen, a corrected probability set of multidimensional safety hazards in the kitchen is obtained.

[0017] Based on the corrected probability set of multidimensional kitchen safety hazards, a comprehensive kitchen risk score is generated at the current moment.

[0018] In one embodiment, a comprehensive kitchen risk score is generated at the current moment based on a multi-dimensional kitchen safety hazard correction probability set, including:

[0019] The expression for the overall kitchen risk score is:

[0020]

[0021] in, This indicates the overall risk score for the kitchen. This represents the total number of dimensions representing the types of safety hazards. Indicates the first The dynamic weighting coefficients corresponding to the types of hidden dangers Indicates the first The probability of correcting a single-dimensional safety hazard after adjusting for time-related correlation. Indicates the first The cumulative duration of such potential hazards exceeding the normal threshold Indicates the first Risk triggering time threshold for this type of hidden danger This represents the penalty coefficient for uneven distribution of hidden dangers. The Gini coefficient represents the corrected probability of multidimensional hidden dangers.

[0022] In one embodiment, the dynamic security threshold is obtained by the following method:

[0023] Obtain a pre-defined set of risk assessment rules; the set of risk assessment rules includes the target safety risk type, associated risk assessment factors, and assessment conditions; the assessment conditions are the logical conditions that the associated risk assessment factors must meet within a pre-defined continuous monitoring time window.

[0024] Extract the target hazard probability corresponding to each risk judgment rule from the multidimensional safety hazard correction probability set in the kitchen, and calculate the condition satisfaction degree of the target hazard probability based on the judgment conditions to obtain the rule trigger intensity value;

[0025] Based on the rule trigger strength value, the dynamic weight at the current moment is obtained;

[0026] The dynamic security threshold is obtained based on the dynamically calculated weights and the preset baseline security threshold.

[0027] In one embodiment, the target hazard probability corresponding to each risk judgment rule is extracted from the multi-dimensional safety hazard correction probability set of the kitchen, and the condition satisfaction degree of the target hazard probability is calculated based on the judgment conditions to obtain the rule triggering intensity value, including:

[0028] For each risk assessment rule in the risk assessment rule set, match the target hazard probability in the kitchen multidimensional safety hazard correction probability set;

[0029] Based on the judgment criteria, the satisfaction index of the target hidden danger probability is obtained;

[0030] Based on the satisfaction index and combined with the preset intensity mapping function, the rule trigger intensity value is obtained.

[0031] In one embodiment, if the hazard result indicates the existence of a safety hazard, differentiated early warning information is generated based on the hazard result and sent to each collaborative terminal, including:

[0032] Based on the results of the potential hazards and the current overall risk score of the kitchen, an early warning level is determined;

[0033] Select differentiated early warning information corresponding to the early warning level from the preset early warning template library, and send the differentiated early warning information to each collaborative terminal.

[0034] In one embodiment, the parameters of the time-series risk feature extraction model are obtained through federated learning collaborative optimization between the kitchen safety inspection center server and each separate monitoring node.

[0035] Secondly, this application also provides a data-driven kitchen safety detection system, comprising:

[0036] The data acquisition module is used to acquire multi-source heterogeneous time-series monitoring data in the kitchen; the multi-source heterogeneous time-series monitoring data includes at least one of the following: water leakage status data, gas concentration time-series data, smoke particle concentration data, and ambient temperature time-series data.

[0037] The feature extraction module is used to input multi-source heterogeneous time-series monitoring data into the time-series risk feature extraction model to obtain a multi-dimensional time-series feature vector; the multi-dimensional time-series feature vector is used to characterize the overall safety status of the kitchen.

[0038] The risk quantification module is used to generate a comprehensive kitchen risk score for the current moment based on a multi-dimensional time-series feature vector.

[0039] The risk assessment module is used to generate the current kitchen hazard results based on the current kitchen comprehensive risk score and dynamic safety threshold; the hazard results include whether there are safety hazards or not.

[0040] The early warning push module is used to generate differentiated early warning information based on the result of the potential hazard, and send the differentiated early warning information to each collaborative terminal if the result indicates that there is a potential safety hazard.

[0041] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0042] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0043] The aforementioned data-driven kitchen safety detection method, system, equipment, and media acquire multi-source heterogeneous time-series monitoring data within the kitchen. This data includes at least one of the following: water leakage status data, gas concentration time-series data, smoke particle concentration data, and ambient temperature time-series data. The multi-source heterogeneous time-series monitoring data is input into a time-series risk feature extraction model to obtain a multi-dimensional time-series feature vector. This vector characterizes the overall safety status of the kitchen. Based on the multi-dimensional time-series feature vector, a comprehensive kitchen risk score is generated for the current moment. Based on the current comprehensive kitchen risk score and a dynamic safety threshold, a potential hazard result for the kitchen at the current moment is generated. This result includes whether a safety hazard exists or not. If the result indicates a safety hazard exists, differentiated early warning information is generated and sent to each collaborative terminal. This method enables quantitative assessment and dynamic hierarchical early warning of multi-source kitchen safety risks. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A schematic diagram illustrating the implementation environment of a data-aware kitchen safety detection method provided in an exemplary embodiment of this application;

[0046] Figure 2 A schematic flowchart of a data-aware kitchen safety detection method provided as an exemplary embodiment of this application;

[0047] Figure 3 This is a schematic diagram of the structure of a data-aware kitchen safety detection system provided as an exemplary embodiment of this application. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0049] The implementation environment of the embodiments of this application will be described. For illustrative purposes, please refer to... Figure 1 The implementation environment includes a detection terminal 101 and various collaborative terminals 102. The detection terminal 101 and the collaborative terminals 102 are connected via a network to achieve data interaction and communication. The network connection can be a wired network or a wireless network.

[0050] The detection terminal 101 can be a computer device, such as a desktop computer or a laptop computer. The detection terminal 101 can be used to receive and process multi-source heterogeneous time-series monitoring data from multiple separate monitoring nodes in the kitchen in real time.

[0051] Each collaborative terminal 102 can be a computer device, a mobile device, an IoT device, or a portable wearable device. Among them, the computer device can be a desktop computer or a laptop computer; the mobile device can be a smartphone or a tablet computer; the IoT device can be a smart speaker, a smart TV, etc.; and the portable wearable device can be a smartwatch, a smart bracelet, or a head-mounted device, etc.

[0052] The application scenarios of the embodiments of this application will be described in conjunction with the implementation environment.

[0053] The kitchen safety detection method based on data perception provided in this application embodiment can be used for intelligent kitchen safety proactive protection, home / commercial kitchen safety early warning, intelligent monitoring of kitchen facility status, and prevention of kitchen safety accidents.

[0054] For example, such as Figure 2 As shown in the embodiment of this application, a data-aware kitchen safety detection method is provided, which is applied to... Figure 1 Taking the detection terminal 101 as an example, in this embodiment, the method includes the following steps:

[0055] Step S201: Obtain multi-source heterogeneous time-series monitoring data in the kitchen; the multi-source heterogeneous time-series monitoring data includes at least one of the following: water leakage status data, gas concentration time-series data, smoke particle concentration data, and ambient temperature time-series data.

[0056] The multi-source heterogeneous time-series monitoring data in the kitchen can be a collection of monitoring data collected by the detection terminal 101 through different types of sensors or monitoring nodes in the kitchen over a continuous time series. These data come from diverse sources and have different formats and types, but are all used to characterize the safety status of the kitchen. Leakage status data can reflect the status of water-using equipment and pipes such as water pipes, faucets, and water purifiers in the kitchen, indicating whether there are leaks or seepage. Combustible gas concentration time-series data can be the concentration values ​​of combustible gases in the kitchen. Smoke and particulate matter concentration data can be the concentration data of smoke, fumes, and suspended particulate matter in the kitchen. Ambient temperature time-series data can be the temperature values ​​of the overall kitchen environment or key areas recorded in a continuous time series.

[0057] Optionally, combustible gases in the kitchen may include, but are not limited to, natural gas and liquefied petroleum gas. The overall kitchen environment or key areas may include, but are not limited to, the stove area, around the refrigerator, and near gas pipes.

[0058] Specifically, the detection terminal 101 can perform data acquisition operations through different types of sensors or separate monitoring nodes deployed in the kitchen. These sensors or separate monitoring nodes are respectively deployed near water-using equipment and pipes, around gas pipes and stoves, throughout the kitchen space, and around key equipment. The detection terminal continuously collects data at preset time intervals, generating multi-source heterogeneous time-series monitoring data.

[0059] Step S202: Input the multi-source heterogeneous time-series monitoring data into the time-series risk feature extraction model to obtain a multi-dimensional time-series feature vector; the multi-dimensional time-series feature vector is used to characterize the overall safety status of the kitchen.

[0060] The temporal risk feature extraction model can be a bidirectional long short-term memory network model based on an attention mechanism in the detection terminal 101. Its core layers include an input layer, an embedding layer, a bidirectional LSTM layer, an attention mechanism layer, and an output layer. For example, the temporal risk feature extraction model can perform feature fusion and deep mining on the input multi-source heterogeneous temporal data, filter redundant noise data, extract temporal evolution features and cross-dimensional correlation features of each monitoring dimension, and generate a feature vector that comprehensively represents the kitchen's safety status. The input to the temporal risk feature extraction model can be a multi-source heterogeneous temporal data sequence, and the output can be a multi-dimensional temporal feature vector. Each dimension in the multi-dimensional temporal feature vector corresponds to a safety-related feature after fusion processing, and the multi-dimensional temporal feature vector as a whole is used to quantitatively represent the comprehensive safety status of the kitchen.

[0061] Optionally, the temporal evolution characteristics of each monitoring dimension may include, but are not limited to, the rate of concentration change and the frequency of abnormal fluctuations. Cross-dimensional correlation characteristics may include, but are not limited to, the coordinated changes in gas concentration and ambient temperature.

[0062] The multidimensional time series feature vector can be a numerical vector with fixed dimensions generated by the detection terminal 101 after fusing, denoising, and deeply mining multi-source heterogeneous time series monitoring data through the time series risk feature extraction model.

[0063] Specifically, the detection terminal 101 can input the collected multi-source heterogeneous time-series monitoring data into the time-series risk feature extraction model. The time-series risk feature extraction model can fuse, denoise, and deeply mine the multi-source heterogeneous time-series monitoring data, filter redundant noise, extract the time-series evolution features and cross-dimensional correlation features of each monitoring dimension, and generate a multi-dimensional time-series feature vector with fixed dimensions.

[0064] Step S203: Generate the kitchen comprehensive risk score at the current moment based on the multi-dimensional time-series feature vector.

[0065] Among them, the overall kitchen risk score at the current moment can be a numerical result obtained by the detection terminal 101 based on multi-dimensional time-series feature vectors and through formulaic calculation, which is used to quantitatively characterize the overall safety risk level of the kitchen at the current moment.

[0066] Specifically, the detection terminal 101 can integrate and calculate the multi-dimensional time-series feature vector through preset quantitative calculation logic, and convert the multi-dimensional features into a single numerical score, namely the kitchen comprehensive risk score at the current moment.

[0067] Step S204: Based on the current kitchen comprehensive risk score and dynamic safety threshold, generate the current kitchen hazard results; the hazard results include whether there are safety hazards or not.

[0068] Among them, the dynamic safety threshold can be a numerical threshold dynamically calculated by the detection terminal 101 based on the preset baseline safety threshold, combined with the pre-set risk judgment rule set, the condition satisfaction degree of the real-time hidden danger probability, and the rule triggering intensity value.

[0069] The hazard result can be a binary judgment result obtained by the detection terminal 101 by comparing the current kitchen comprehensive risk score with the dynamic safety threshold, which is used to clarify the current kitchen safety status. The hazard result includes whether there is a safety hazard or not.

[0070] Specifically, the detection terminal 101 can compare the current kitchen comprehensive risk score with the dynamic safety threshold. If the comprehensive risk score is higher than or equal to the dynamic safety threshold, the result is determined to be a safety hazard; if the comprehensive risk score is lower than the dynamic safety threshold, the detection terminal 101 determines that there is no safety hazard.

[0071] Step S205: If the hazard result indicates that there is a safety hazard, then generate differentiated early warning information based on the hazard result and send the differentiated early warning information to each collaborative terminal.

[0072] Among them, differentiated early warning information can be targeted and hierarchical early warning content generated and pushed by the detection terminal 101 based on the different risk types, severity and impact range of the hidden danger results.

[0073] Specifically, after determining that a safety hazard exists, the detection terminal 101 can immediately retrieve the currently generated comprehensive kitchen risk score. Through preset risk level mapping rules, combined with the score value, the number of monitoring dimensions involved in the hazard, and the degree of data anomaly in each dimension, it comprehensively determines the severity and scope of impact of the safety hazard. Based on the determined severity and scope of impact, the detection terminal 101 can automatically fill in relevant information from the built-in preset early warning template library with appropriate content, generating differentiated early warning information.

[0074] Optionally, the severity of a safety hazard may include, but is not limited to, minor, moderate, severe, and urgent. The scope of impact of a safety hazard may include, but is not limited to, local equipment risks and global environmental risks. Real-time monitoring data may include, but is not limited to, abnormal data types and their duration.

[0075] The aforementioned data-driven kitchen safety detection method involves a detection terminal 101 acquiring multi-source heterogeneous time-series monitoring data within the kitchen. This data includes at least one of the following: water leakage status data, gas concentration time-series data, smoke particle concentration data, and ambient temperature time-series data. The multi-source heterogeneous time-series monitoring data is input into a time-series risk feature extraction model to obtain a multi-dimensional time-series feature vector. This vector characterizes the overall safety status of the kitchen. Based on the multi-dimensional time-series feature vector, a comprehensive kitchen risk score is generated for the current moment. Based on the current comprehensive kitchen risk score and a dynamic safety threshold, a potential hazard result for the kitchen at the current moment is generated. This result includes whether a safety hazard exists or not. If the result indicates a safety hazard exists, differentiated early warning information is generated and sent to each collaborative terminal. This method enables quantitative assessment and dynamic hierarchical early warning of multi-source kitchen safety risks.

[0076] In one embodiment, generating a comprehensive kitchen risk score for the current moment based on a multi-dimensional temporal feature vector may include the following steps:

[0077] Step S301: Dynamic importance weight allocation is performed on each dimension of the multidimensional time-series feature vector to obtain the dynamic weight coefficients of each dimension.

[0078] Among them, the dynamic weight coefficients of each dimension feature can be used by the detection terminal to quantify the degree of influence of each dimension feature on the overall safety risk of the kitchen.

[0079] For example, the detection terminal can obtain a preset feature risk contribution assessment rule. This rule defines how to combine the real-time values, historical fluctuation trends, anomaly frequency within the current time window, and statistical correlation with other feature dimensions to assess their risk contribution. Subsequently, the detection terminal can calculate the initial contribution assessment value for each feature in the multi-dimensional time-series feature vector according to the preset feature risk contribution assessment rule. Finally, the detection terminal can call a preset dynamic weight allocation algorithm to calculate and generate normalized weight coefficients, i.e., the dynamic weight coefficients of each feature dimension, based on the initial contribution assessment values ​​of each feature dimension.

[0080] Optionally, the dynamic weight allocation algorithm may be, but is not limited to, the entropy weight method or the analytic hierarchy process.

[0081] Step S302: Generate weighted features for each dimension based on the features of each dimension and the dynamic weight coefficients of the features of each dimension.

[0082] The detection terminal can perform a weighted operation on the original feature data of each dimension in the multi-dimensional time-series feature vector, along with the corresponding dynamic weight coefficients, to obtain the weighted dimensional features. These weighted dimensional features can be used as enhanced feature data to strengthen high-risk association features and weaken low-risk association features.

[0083] For example, the detection terminal can traverse each dimension of the multi-dimensional temporal feature vector to obtain its current feature value and its corresponding dynamic weight coefficient. Then, the detection terminal can perform a multiplication operation (i.e., weighting) on ​​the value of each dimension feature and its dynamic weight coefficient. The detection terminal can perform this weighting operation sequentially on all dimensions of the multi-dimensional temporal feature vector. Finally, the detection terminal can recombine the weighted numerical sequence into a new feature vector with the same dimensions as the original vector. Each element in this new feature vector is the weighted result, i.e., each weighted dimension feature.

[0084] Step S303: Based on the features of each weighted dimension and combined with the preset feature-hazard probability mapping rules, obtain the single-dimensional safety hazard probability corresponding to each weighted dimension feature, and generate a multi-dimensional safety hazard probability set for the kitchen based on the single-dimensional safety hazard probability.

[0085] The preset feature-hazard probability mapping rule can be a set of standardized rules pre-configured in the detection terminal to establish a quantitative correlation between weighted dimensional features and single-dimensional safety hazard probabilities. The feature-hazard probability mapping rule can be constructed by the detection terminal based on the evolution patterns of kitchen safety hazards, historical monitoring data, and industry safety standards. The detection terminal can set the correspondence between weighted feature value ranges and single-dimensional safety hazard probabilities for different types of weighted dimensional features.

[0086] Optionally, the weighted dimensional features may include, but are not limited to, gas concentration weighted features and leakage status weighted features.

[0087] The single-dimensional safety hazard probability corresponding to each weighted dimension feature can be a quantitative probability value that corresponds only to the safety hazard type associated with the corresponding dimension feature, calculated by the detection terminal after substituting the specific value of each weighted dimension feature into the preset feature-hazard probability mapping rule. The value range of the single-dimensional safety hazard probability can be [0,1].

[0088] The multidimensional safety hazard probability set in the kitchen can be a set of data that includes the probabilities of various safety hazards in the kitchen, generated by the detection terminal after structuring and organizing the single-dimensional safety hazard probabilities corresponding to all weighted dimensional features.

[0089] For example, the detection terminal can acquire a preset feature-hazard probability mapping rule. This rule defines the hazard probability value corresponding to different feature value ranges in the form of a lookup table or mathematical function. Subsequently, the detection terminal can iterate through each weighted dimension feature: for each weighted dimension feature, the terminal can match its specific weighted value with the feature-hazard probability mapping rule to obtain the probability of occurrence of the specific safety hazard associated with that dimension, i.e., the single-dimensional safety hazard probability. Finally, the detection terminal can summarize the single-dimensional safety hazard probabilities calculated for all dimensions and organize them in a unified format to generate a multi-dimensional kitchen safety hazard probability set that includes the hazard probabilities corresponding to all monitoring dimensions in the kitchen.

[0090] Optionally, the uniform format can be, but is not limited to, SON objects, database records, or arrays.

[0091] Step S304: Perform time-dimensional correlation correction on the multidimensional safety hazard probability set in the kitchen to obtain the corrected multidimensional safety hazard probability set in the kitchen.

[0092] Among them, the kitchen multidimensional safety hazard correction probability set can be a structured probability set data generated by the detection terminal after performing time-dimensional correlation analysis and quantitative correction on the probability of each single-dimensional safety hazard in the kitchen multidimensional safety hazard probability set.

[0093] For example, the detection terminal can acquire the time-series data sequence of the probability of each single-dimensional hazard in the multi-dimensional kitchen safety hazard probability set within the current and preset historical time windows. Subsequently, the detection terminal can analyze the changing trend of each single-dimensional hazard probability in the time dimension and its temporal correlation with the changes in hazard probabilities of other dimensions. Further, the detection terminal can invoke a preset time dimension correction function. This function comprehensively considers historical trends, time decay effects, and cross-dimensional time correlation weights to quantitatively adjust the probability values ​​of each single-dimensional hazard at the current moment, thereby eliminating probability deviations caused by instantaneous data fluctuations and enhancing the temporal stability and predictive rationality of the probability values. Finally, the detection terminal can recombine the corrected single-dimensional hazard probabilities into a structured set, namely, the corrected probability set of the multi-dimensional kitchen safety hazard.

[0094] Step S305: Based on the multidimensional safety hazard correction probability set of the kitchen, generate the comprehensive risk score of the kitchen at the current moment.

[0095] For example, the detection terminal can first retrieve the multi-dimensional safety hazard correction probability set for the kitchen, extracting the corrected probabilities of each dimension and parameters such as the corresponding dynamic weight coefficients and cumulative over-threshold duration. Then, the detection terminal can substitute these parameters into a preset comprehensive risk score formula, fusing the probabilities of each dimension and the time decay factor through a weighted summation operation. Subsequently, the detection terminal can combine the hazard probability Gini coefficient and the imbalance penalty coefficient for correction, ultimately obtaining the current comprehensive kitchen risk score with a value range of [0,1].

[0096] In this embodiment, the detection terminal improves the real-time performance, accuracy, and scenario adaptability of the comprehensive risk score through a step-by-step quantification process.

[0097] In one embodiment, a comprehensive kitchen risk score is generated at the current moment based on a multi-dimensional kitchen safety hazard correction probability set, including:

[0098] The expression for the overall kitchen risk score is:

[0099]

[0100] in, This indicates the overall risk score for the kitchen. This represents the total number of dimensions representing the types of safety hazards. Indicates the first The dynamic weighting coefficients corresponding to the types of hidden dangers Indicates the first The probability of correcting a single-dimensional safety hazard after adjusting for time-related correlation. Indicates the first The cumulative duration of such potential hazards exceeding the normal threshold Indicates the first Risk triggering time threshold for this type of hidden danger This represents the penalty coefficient for uneven distribution of hidden dangers. The Gini coefficient represents the corrected probability of multidimensional hidden dangers.

[0101] For example, Indicates the first The dynamic weight coefficients corresponding to the types of hidden dangers are derived from the calculation results of step S301. The detection terminal uses a preset feature risk contribution evaluation function to assess the importance of each dimension feature in the multi-dimensional time-series feature vector, combining the real-time numerical range, historical fluctuation level, abnormal frequency, and statistical correlation between features of each dimension feature. This allows for the allocation of the corresponding features for each dimension according to the risk contribution ratio. And all dimensions It meets the normalization requirements; Indicates the first The single-dimensional safety hazard correction probability, after time-dimensional correlation correction, is derived from the correction result of step S303. It can be adjusted in step S302 to... The specific numerical values ​​of the class-dimensional features are matched with the preset feature-hazard probability mapping rules, combined with The initial single-dimensional safety hazard probability is calculated. Then, the detection terminal retrieves historical time-series data of this probability, analyzes the changing trends, fluctuation frequency, and correlation response patterns within a continuous time window, and, based on the time decay coefficient, cross-time correlation weight, and historical abnormal time-series characteristics, eliminates single-time-moment data fluctuation bias through a time-dimensional correlation correction algorithm, ultimately obtaining... ; Indicates the first The cumulative duration of a potential hazard exceeding the normal threshold originates from the detection terminal's monitoring of the first... Real-time statistics of various monitoring data. The detection terminal can continuously track monitoring data from various dimensions. When the data exceeds the preset normal threshold, a timer automatically starts, accumulating the duration of this state; if the data returns to within the normal threshold, the timer stops, and the current accumulated duration is the monitoring data. ; Indicates the first The risk triggering time threshold for potential hazards is derived from the preset safety standard configuration, which is set by the detection terminal based on industry safety standards, equipment operating parameters, and historical accident data. The Gini coefficient, representing the corrected probability of multi-dimensional hidden dangers, is used by the detection terminal in accordance with the standard calculation method of the Gini coefficient obtained in step S303. indivual Perform distribution balance analysis to quantify the dispersion of the probability of hidden dangers in each dimension. The closer the coefficient is to 1, the more unbalanced the distribution of the probability of hidden dangers is, and the closer it is to 0, the more balanced the distribution is.

[0102] In this embodiment, the detection terminal integrates the dynamic weighting coefficient of step S301 and the single-dimensional safety hazard correction probability of step S303, and combines the cumulative over-threshold duration, preset risk trigger time threshold and Gini coefficient in real time. By fusing multi-dimensional risk factors and distribution balance penalty through formula, it realizes the quantitative calculation of comprehensive kitchen risk, improves the accuracy and comprehensiveness of risk assessment, and provides a reliable quantitative basis for hazard judgment.

[0103] In one embodiment, the dynamic security threshold is obtained through the following steps:

[0104] Step S401: Obtain a pre-set set of risk assessment rules; the set of risk assessment rules includes target safety risk type, associated risk assessment factors and assessment conditions; the assessment conditions are the logical conditions that the associated risk assessment factors must meet within a preset continuous monitoring time window.

[0105] The pre-defined risk assessment rule set can be a pre-configured standardized rule set used to dynamically calculate safety thresholds. It consists of multiple independent risk assessment rules, each corresponding to a specific kitchen safety management scenario.

[0106] For example, the detection terminal can parse the pre-stored risk assessment rule configuration file, and then load all the rules and their elements defined in the risk assessment rule configuration file into memory, and finally generate a set of currently available risk assessment rules. The set of risk assessment rules includes the target security risk type, associated risk assessment factors and assessment conditions. The assessment conditions are the logical conditions that the associated risk assessment factors must meet within a preset continuous monitoring time window.

[0107] Step S402: Extract the target hazard probability corresponding to each risk judgment rule from the multidimensional safety hazard correction probability set of the kitchen, and calculate the condition satisfaction degree of the target hazard probability based on the judgment conditions to obtain the rule trigger intensity value.

[0108] Among them, the target hazard probability can be the single-dimensional safety hazard correction probability extracted by the detection terminal from the multi-dimensional safety hazard correction probability set in the kitchen, after matching the target safety risk type corresponding to each rule in the pre-set risk judgment rule set. This probability is corrected for time-dimensional correlation.

[0109] The rule trigger strength value can be used as a quantitative value to characterize the degree to which a single risk assessment rule is triggered.

[0110] For example, the detection terminal can iterate through each rule in the risk assessment rule set. For each rule, the detection terminal can search within the kitchen multi-dimensional safety hazard correction probability set according to its defined target safety risk type, and extract the single-dimensional safety hazard correction probability that perfectly matches the risk type; this value is the target hazard probability. Subsequently, based on the judgment conditions of this rule, and combined with the time-series data of the target hazard probability within the current and historical time windows, the detection terminal can obtain an index indicating the degree to which the judgment conditions are met (i.e., the satisfaction index). Further, the detection terminal can convert the satisfaction index into a normalized value, which is the rule trigger strength value of this rule.

[0111] Step S403: Based on the rule trigger strength value, obtain the dynamic calculation weight at the current moment.

[0112] The dynamic weights calculated at the current moment can be a set of weight coefficients used to dynamically adjust the baseline safety threshold.

[0113] For example, the detection terminal can obtain the rule trigger strength values ​​calculated by all risk assessment rules at the current moment, thus obtaining a trigger strength vector; subsequently, the detection terminal can call a preset dynamic weight generation algorithm to process all strength values ​​in the trigger strength vector. Finally, the detection terminal obtains the dynamically calculated weights at the current moment.

[0114] Optionally, the preset dynamic weight generation algorithm may be, but is not limited to, the Softmax function or a normalized weighting method. The dynamic weight generation algorithm converts the trigger strength value of each rule into a weight coefficient between 0 and 1.

[0115] Step S404: Obtain the dynamic security threshold based on the dynamically calculated weights and the preset baseline security threshold.

[0116] The preset baseline security threshold can be a fixed numerical threshold used as the basis for calculating the dynamic security threshold.

[0117] For example, the detection terminal can retrieve a preset baseline safety threshold from pre-stored system configuration information. This threshold is a fixed benchmark value calibrated based on industry safety standards, equipment operating parameters, and historical accident data. Subsequently, the detection terminal can load the dynamic calculation weight set obtained in step S403, call the preset threshold dynamic adjustment algorithm, and perform a weighted summation operation on the dynamic weights corresponding to each risk assessment rule and the baseline safety threshold.

[0118] Optionally, the preset threshold dynamic adjustment algorithm can be a calculation rule that linearly weights and adjusts the baseline safety threshold based on dynamically calculated weights. For example, the algorithm can be expressed as:

[0119]

[0120] The weighting factors for each risk assessment rule are adjustment coefficients pre-set according to the rule type and scenario; or, they can be non-linear functions, such as those combining exponential smoothing or piecewise functions.

[0121] In this embodiment, the detection terminal achieves dynamic calculation of the safety threshold by adapting to real-time risk scenarios through risk rule matching, intensity value quantification, dynamic weight allocation, and nonlinear threshold adjustment, thereby improving the real-time performance and accuracy of risk identification.

[0122] In one embodiment, extracting the target hazard probability corresponding to each risk judgment rule from the multi-dimensional kitchen safety hazard correction probability set, and calculating the condition satisfaction degree of the target hazard probability based on the judgment conditions to obtain the rule triggering intensity value may include the following steps:

[0123] Step S501: For each risk judgment rule in the risk judgment rule set, match the target hazard probability in the kitchen multidimensional safety hazard correction probability set.

[0124] For example, the detection terminal can traverse each rule in the risk assessment rule set and parse the target safety risk type identifier defined therein. Then, the detection terminal matches the target safety risk type identifier in the kitchen multi-dimensional safety hazard correction probability set according to the hazard type label associated with each single-dimensional hazard probability entry. If a hazard probability entry is found that completely matches the target safety risk type identifier of the current rule, the detection terminal extracts the value of this hazard probability entry as the target hazard probability corresponding to the current rule.

[0125] Step S502: Based on the judgment conditions, obtain the satisfaction index of the target hidden danger probability.

[0126] Among them, the satisfaction index of the target hidden danger probability can be an intermediate quantitative index obtained by quantitatively analyzing the matched target hidden danger probability based on the judgment conditions in the risk judgment rules, which is used to characterize the degree to which the target hidden danger probability meets the judgment conditions.

[0127] For example, the detection terminal can parse the judgment conditions in the risk judgment rules to obtain the logical requirements included in the judgment conditions and the preset continuous monitoring time window parameters. Subsequently, the detection terminal can retrieve the complete time series data of the target hazard probability within the time window and perform targeted quantitative analysis in combination with the judgment condition type: if it is a single probability threshold condition, calculate the proportion of the target hazard probability meeting the standard and the peak compliance degree within the window; if it is a multi-factor correlation condition, quantify the degree of synergistic satisfaction of each factor through logical operations. Finally, the detection terminal can convert the above analysis results into an intermediate quantitative index with a value range of [0,1], that is, the satisfaction index of the target hazard probability.

[0128] Optionally, the judgment criteria may include, but are not limited to, probability thresholds, duration, and multi-factor associations.

[0129] Step S503: Based on the satisfaction index and combined with the preset intensity mapping function, the rule trigger intensity value is obtained.

[0130] The preset intensity mapping function can be a pre-configured nonlinear calculation function used to convert the satisfaction index of the target hidden danger probability into the rule trigger intensity value.

[0131] For example, the detection terminal can obtain a pre-configured intensity mapping function, which is defined in the form of a mathematical formula or a mapping table. Subsequently, the detection terminal can use the satisfaction index calculated in step S502 as input, substituting it into the intensity mapping function for calculation. If the function is in formula form, the function value is calculated directly; if it is in mapping table form, the corresponding output value is obtained through table lookup or interpolation. Through this function calculation, the satisfaction index is non-linearly converted into a normalized value, which is the quantified result representing the intensity of the risk assessment rule being triggered at the current moment, i.e., the rule triggering intensity value.

[0132] In this embodiment, the detection terminal achieves a quantitative assessment of the trigger intensity of each risk judgment rule through hazard probability matching, multi-dimensional satisfaction quantification based on judgment conditions, and nonlinear intensity mapping, thereby providing high-quality input for dynamic weight synthesis.

[0133] In one embodiment, if the hazard result indicates the existence of a safety hazard, differentiated early warning information is generated based on the hazard result and sent to each collaborative terminal. This may include the following steps:

[0134] Step S601: Based on the hazard results and the current kitchen comprehensive risk score, obtain the warning level.

[0135] The warning level can be the result of classifying and quantifying safety risks based on the severity, urgency, and potential impact of current kitchen safety hazards. In this embodiment, the warning levels include a first warning level, a second warning level, a third warning level, and a fourth warning level; the first warning level corresponds to alert-type risks; the second warning level corresponds to caution-type risks; the third warning level corresponds to alarm-type risks; and the fourth warning level corresponds to emergency-type risks.

[0136] For example, if the detection terminal confirms that a safety hazard exists, it can retrieve the currently generated comprehensive risk score for the kitchen and load a preset risk level classification standard. The detection terminal then compares the current comprehensive risk score with a preset range: if the score is in the lowest range, it is classified as a first warning level (alert risk); if the score is in a lower range, it is classified as a second warning level (attention risk); if the score is in a higher range, it is classified as a third warning level (alarm risk); and if the score is in the highest range, it is classified as a fourth warning level (emergency risk).

[0137] Step S602: Select differentiated early warning information corresponding to the early warning level from the preset early warning template library, and send the differentiated early warning information to each collaborative terminal.

[0138] The preset warning template library can be a structured database pre-configured in the detection terminal, which stores warning information templates corresponding to different warning levels, and is used to quickly generate targeted warning content for various safety hazard scenarios.

[0139] For example, the detection terminal can retrieve a preset warning template library based on the determined warning level, match the template corresponding to the level, and automatically fill in the real-time hidden danger information extracted from the monitoring data to generate warning information with differentiated content and description adapted to the risk level. Subsequently, the detection terminal can synchronously push the differentiated warning information to each bound collaborative terminal through a preset communication protocol.

[0140] In this embodiment, the detection terminal achieves automatic classification of safety hazards and push of differentiated early warning information by quickly mapping and matching the current kitchen comprehensive risk score with the standard range and automatically generating templated information.

[0141] In one embodiment, the parameters of the time-series risk feature extraction model are obtained through federated learning collaborative optimization between the kitchen safety inspection center server and each separate monitoring node.

[0142] In this embodiment, the detection terminal uses a federated learning mechanism to achieve distributed collaborative optimization of model parameters while ensuring the data privacy and security of each node, thereby improving the global generalization ability and local adaptability of the time-series risk feature extraction model.

[0143] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0144] Based on the same inventive concept, this application also provides a data-aware kitchen safety detection system for implementing the aforementioned method. The solution provided by this system is similar to the implementation described in the above method. Therefore, the specific limitations of one or more data-aware kitchen safety detection system embodiments provided below can be found in the limitations of the data-aware kitchen safety detection method described above, and will not be repeated here.

[0145] In one exemplary embodiment, such as Figure 3 As shown, a data-aware kitchen safety detection system 700 is provided, comprising:

[0146] The data acquisition module 701 is used to acquire multi-source heterogeneous time-series monitoring data in the kitchen; the multi-source heterogeneous time-series monitoring data includes at least one of the following: water leakage status data, gas concentration time-series data, smoke particle concentration data, and ambient temperature time-series data.

[0147] The feature extraction module 702 is used to input multi-source heterogeneous time-series monitoring data into the time-series risk feature extraction model to obtain a multi-dimensional time-series feature vector; the multi-dimensional time-series feature vector is used to characterize the overall safety status of the kitchen;

[0148] The risk quantification module 703 is used to generate a comprehensive kitchen risk score for the current moment based on a multi-dimensional time-series feature vector.

[0149] The risk assessment module 704 is used to generate the current kitchen hazard results based on the current kitchen comprehensive risk score and dynamic safety threshold; the hazard results include whether there are safety hazards or not.

[0150] The early warning push module 705 is used to generate differentiated early warning information based on the result of the hidden danger, and send the differentiated early warning information to each collaborative terminal if the result of the hidden danger is that there is a safety hazard.

[0151] In one embodiment, the risk quantification module includes:

[0152] The weight allocation unit is used to dynamically allocate the importance weights of each dimension of the multidimensional time-series feature vector to obtain the dynamic weight coefficients of each dimension of the feature.

[0153] The dimensional feature generation unit is used to generate weighted dimensional features based on each dimensional feature and its dynamic weight coefficients.

[0154] The hazard probability set generation unit is used to obtain the single-dimensional safety hazard probability corresponding to each weighted dimension feature based on each weighted dimension feature and combined with the preset feature-hazard probability mapping rule, and to generate a multi-dimensional safety hazard probability set for the kitchen based on each single-dimensional safety hazard probability.

[0155] The probability time series correction unit is used to correct the temporal correlation of the probability set of multidimensional safety hazards in the kitchen, so as to obtain the corrected probability set of multidimensional safety hazards in the kitchen.

[0156] The scoring generation unit is used to generate a comprehensive kitchen risk score for the current moment based on the corrected probability set of multidimensional kitchen safety hazards.

[0157] In one embodiment, the risk quantification module includes:

[0158] The expression for the overall kitchen risk score is:

[0159]

[0160] in, This indicates the overall risk score for the kitchen. This represents the total number of dimensions representing the types of safety hazards. Indicates the first The dynamic weighting coefficients corresponding to the types of hidden dangers Indicates the first The probability of correcting a single-dimensional safety hazard after adjusting for time-related correlation. Indicates the first The cumulative duration of such potential hazards exceeding the normal threshold Indicates the first Risk triggering time threshold for this type of hidden danger This represents the penalty coefficient for uneven distribution of hidden dangers. The Gini coefficient represents the corrected probability of multidimensional hidden dangers.

[0161] In one embodiment, the risk quantification module obtains the dynamic safety threshold through the following unit:

[0162] The judgment rule acquisition unit is used to acquire a pre-set set of risk judgment rules. The set of risk judgment rules includes the target safety risk type, associated risk judgment factors, and judgment conditions. The judgment conditions are the logical conditions that the associated risk judgment factors must meet within a preset continuous monitoring time window.

[0163] The hazard probability extraction unit is used to extract the target hazard probability corresponding to each risk judgment rule from the multi-dimensional safety hazard correction probability set of the kitchen, and calculate the condition satisfaction degree of the target hazard probability based on the judgment conditions to obtain the rule trigger intensity value.

[0164] The dynamic weight calculation unit is used to obtain the dynamic weight at the current moment based on the rule-triggered intensity value.

[0165] The dynamic security threshold generation unit is used to obtain the dynamic security threshold based on the dynamically calculated weights and the preset baseline security threshold.

[0166] In one embodiment, the hazard probability extraction unit includes:

[0167] The probability matching subunit is used to match the probability of a target hazard in the kitchen multidimensional safety hazard correction probability set with each risk judgment rule in the risk judgment rule set.

[0168] The satisfaction calculation subunit is used to obtain the satisfaction index of the target hidden danger probability based on the judgment conditions.

[0169] The trigger intensity generation subunit is used to obtain the rule trigger intensity value based on the satisfaction index and a preset intensity mapping function.

[0170] In one embodiment, the early warning push module includes:

[0171] The level determination unit is used to determine the warning level based on the hazard results and the overall risk score of the kitchen;

[0172] The early warning push unit is used to select differentiated early warning information corresponding to the early warning level from the preset early warning template library and send the differentiated early warning information to each collaborative terminal.

[0173] In one embodiment, the parameters of the time-series risk feature extraction model are obtained through federated learning collaborative optimization between the kitchen safety inspection center server and each separate monitoring node.

[0174] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a data-aware kitchen safety detection method as described above.

[0175] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0176] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0177] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A kitchen safety detection method based on data perception, characterized in that, The method includes: Acquire multi-source heterogeneous time-series monitoring data in the kitchen; the multi-source heterogeneous time-series monitoring data includes at least one of water leakage status data, gas concentration time-series data, smoke particle concentration data, and ambient temperature time-series data. The multi-source heterogeneous time-series monitoring data is input into the time-series risk feature extraction model to obtain a multi-dimensional time-series feature vector; the multi-dimensional time-series feature vector is used to characterize the overall safety status of the kitchen. Based on the multidimensional temporal feature vector, a comprehensive risk score for the kitchen at the current moment is generated; Based on the kitchen's comprehensive risk score and dynamic safety threshold at the current moment, a potential hazard result for the kitchen at the current moment is generated; the potential hazard result includes whether there is a safety hazard or not. If the hazard result indicates the existence of a safety hazard, then differentiated early warning information is generated based on the hazard result and sent to each collaborative terminal.

2. The method according to claim 1, characterized in that, The process of generating a comprehensive kitchen risk score at the current moment based on the multi-dimensional time-series feature vector includes: Dynamic importance weights are assigned to each dimension of the multidimensional temporal feature vector to obtain the dynamic weight coefficients of each dimension feature. Based on each of the dimensional features and the dynamic weight coefficients of each of the dimensional features, each weighted dimensional feature is generated; Based on each weighted dimension feature and combined with the preset feature-hazard probability mapping rule, the single-dimensional safety hazard probability corresponding to each weighted dimension feature is obtained, and a kitchen multi-dimensional safety hazard probability set is generated based on each single-dimensional safety hazard probability. The time-dimensional correlation of the aforementioned multidimensional kitchen safety hazard probability set is corrected to obtain the corrected multidimensional kitchen safety hazard probability set. Based on the aforementioned multidimensional kitchen safety hazard correction probability set, a comprehensive kitchen risk score is generated for the current moment.

3. The method according to claim 2, characterized in that, The process of generating a comprehensive kitchen risk score at the current moment based on the multi-dimensional kitchen safety hazard correction probability set includes: The expression for the comprehensive risk score of the kitchen is: in, This represents the overall risk score of the kitchen. This represents the total number of dimensions representing the types of safety hazards. Indicates the first The dynamic weighting coefficients corresponding to the types of hidden dangers, Indicates the first The probability of correcting a single-dimensional safety hazard after adjusting for time-related correlation. Indicates the first The cumulative duration of such potential hazards exceeding the normal threshold Indicates the first Risk triggering time threshold for this type of hidden danger This represents the penalty coefficient for uneven distribution of hidden dangers. The Gini coefficient represents the corrected probability of multidimensional hidden dangers.

4. The method according to claim 2, characterized in that, The dynamic security threshold is obtained through the following method: Obtain a pre-defined set of risk assessment rules; the set of risk assessment rules includes target security risk types, associated risk assessment factors, and assessment conditions; The determination criteria are the logical conditions that the associated risk determination factors must meet within a preset continuous monitoring time window. Extract the target hazard probability corresponding to each of the risk judgment rules from the multidimensional kitchen safety hazard correction probability set, and calculate the condition satisfaction degree of the target hazard probability based on the judgment conditions to obtain the rule triggering intensity value; Based on the trigger strength value of the aforementioned rule, the dynamic calculation weight at the current moment is obtained; The dynamic security threshold is obtained based on the dynamically calculated weights and the preset baseline security threshold.

5. The method according to claim 4, characterized in that, The step of extracting the target hazard probability corresponding to each of the risk judgment rules from the multi-dimensional kitchen safety hazard correction probability set, and calculating the condition satisfaction degree of the target hazard probability based on the judgment conditions to obtain the rule triggering intensity value includes: For each risk determination rule in the set of risk determination rules, match the target hazard probability in the set of corrected multidimensional safety hazards in the kitchen; Based on the aforementioned judgment criteria, a satisfaction index for the probability of the target hidden danger is obtained; Based on the satisfaction index and combined with the preset intensity mapping function, the rule trigger intensity value is obtained.

6. The method according to any one of claims 1 to 5, characterized in that, If the hazard result indicates the existence of a safety hazard, then differentiated early warning information is generated based on the hazard result, and the differentiated early warning information is sent to each collaborative terminal, including: Based on the identified hazards and the overall kitchen risk score, a warning level is determined. Select differentiated early warning information corresponding to the early warning level from the preset early warning template library, and send the differentiated early warning information to each of the collaborative terminals.

7. The method according to claim 1, characterized in that, The parameters of the time-series risk feature extraction model are obtained through federated learning collaborative optimization between the kitchen safety inspection center server and each separate monitoring node.

8. A kitchen safety detection system based on data perception, characterized in that, The system includes: The data acquisition module is used to acquire multi-source heterogeneous time-series monitoring data in the kitchen; the multi-source heterogeneous time-series monitoring data includes at least one of the following: water leakage status data, gas concentration time-series data, smoke particle concentration data, and ambient temperature time-series data. The feature extraction module is used to input the multi-source heterogeneous time-series monitoring data into the time-series risk feature extraction model to obtain a multi-dimensional time-series feature vector; the multi-dimensional time-series feature vector is used to characterize the overall safety status of the kitchen. The risk quantification module is used to generate a comprehensive kitchen risk score at the current moment based on the multi-dimensional time-series feature vector. The risk assessment module is used to generate a hazard assessment result for the kitchen at the current moment based on the kitchen's comprehensive risk score and dynamic safety threshold; the hazard assessment result includes whether there is a safety hazard or not. The early warning push module is used to generate differentiated early warning information based on the result of the potential hazard if the result indicates that there is a potential safety hazard, and then send the differentiated early warning information to each collaborative terminal.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images