Security risk metering method for supervised object

Abstract

The invention is suitable for the technical field of intelligent nursing and risk assessment, and provides a supervised object security risk measurement method, which comprises the following steps: S1, identifying risk factors, and constructing a risk matrix; s2, constructing a security risk measurement model; s3, data acquisition and risk measurement are carried out; s4, carrying out comprehensive risk assessment; and S5, feeding back an adjustment mechanism. According to the risk measurement method based on data analysis, risk measurement of a supervised object is realized, and the accuracy of risk assessment is improved through a feedback regulation mechanism; and through feedback and learning, self-optimization of the safety risk measurement model is realized, and the risk assessment accuracy is gradually improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent care and risk assessment, and in particular relates to a method for measuring the safety risks of a supervised object. Background Art

[0002] As the scale of centralized supervision sites continues to expand, the cost of care is increasing. Unmanned care modes such as smart care and multi-person care are seen as future trends. Under this circumstance, it is particularly important to control the safety risks of supervised objects in supervision sites, and higher requirements are placed on the safety risk management and control capabilities of supervised objects. At present, the risk assessment methods for supervised objects are mainly subjective assessment, manual assessment, and experience assessment. There is no scientific method to measure risks. However, subjective and empirical manual assessment methods cannot fully reflect the comprehensive risks faced by supervised objects; and the safety risks of supervised objects are not quantified and cannot be measured, making it difficult to reflect the degree and level of safety risks. Although there are some technical contents on risk warning in other fields, the risk assessment methods used for these technical collections lack feedback adjustment mechanisms or have a single feedback adjustment mechanism, which is difficult to meet the needs of this scenario.

[0003] With the development of artificial intelligence and big data technologies, especially the combination of reinforcement learning and probabilistic statistical methods, it is possible to conduct more accurate and dynamic assessments of the health and behavioral risks of regulated entities. Through artificial intelligence and big data, effectively improving the accuracy of risk assessments of regulated entities has become a feasible goal. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a method for measuring the safety risks of regulated objects. The method is a comprehensive risk assessment method for regulated objects based on joint calculation, and is mainly used for scientific and dynamic measurement and assessment of the health risks, behavioral risks and psychological risks of regulated objects. It aims to solve the technical problems that existing risk assessment methods are not accurate enough and cannot be objectively evaluated.

[0005] The present invention adopts the following technical solution:

[0006] The method for measuring the safety risk of the regulated object comprises the following steps:

[0007] Step S1, identifying risk factors through the safety risk case library of the supervised object and constructing a risk matrix, wherein the risk factors include basic risk factors, daily risk factors and interrogation risk factors;

[0008] Step S2: Mark the data features of risk factors, divide them into basic type and frequency type, and build a safety risk measurement model for each risk factor;

[0009] Step S3, collecting basic data of the supervised object, physical sign data and behavioral data during the care process, and physical sign data and behavioral data during the interrogation process, and calculating the basic risk value, daily risk value and inquiry risk value according to the safety risk measurement model;

[0010] Step S4: Calculate the comprehensive risk value based on the three calculated risk values ​​and the corresponding coefficients, and determine the risk level;

[0011] Step S5: Dynamically adjust the risk dimension weights of risk factors through a feedback adjustment mechanism based on a combination of a naive Bayesian network and reinforcement learning to update the comprehensive risk value calculation model.

[0012] The beneficial effects of the present invention are as follows: the present invention provides a multidimensional risk assessment model based on a combination of a naive Bayesian network and reinforcement learning. By comprehensively collecting static information, dynamic behavioral data and psychological data of the supervised objects, a risk assessment matrix of multiple dimensions such as basic risk, daily risk and interrogation risk is established, and different weights can be assigned to risk dimensions of different dimensions respectively; wherein, through the reinforcement learning mechanism, the weight of each risk dimension can be dynamically adjusted according to the actual labeled data to ensure that it can adapt to the personalized characteristics of different cared-for objects; through the naive Bayesian model, the accuracy of risk assessment can be further improved according to the joint probability of each feature, especially when there is a significant correlation between behavioral data and risk level, a multidimensional and multi-level comprehensive risk assessment method is realized, which can perform real-time and accurate assessment of the risks of the cared-for objects, and improve the scientificity and intelligent level of care management. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a flow chart of a method for measuring the safety risk of a supervised object provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0015] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0016] like Figure 1 As shown, the security risk measurement method for the supervised object provided in this embodiment includes the following steps:

[0017] Step S1, identify risk factors and construct a risk matrix: identify risk factors through the safety risk case library of the supervised object and construct a risk matrix, where the risk factors include basic risk factors, daily risk factors and interrogation risk factors.

[0018] The safety risk case library (LRCL) is a collection of cases used to store and analyze various safety accidents, incidents, and potential risks. It aims to help organizations and individuals identify, evaluate, and manage safety risks, thereby preventing accidents and improving the overall safety level. This step identifies risk factors and constructs a risk matrix by analyzing the safety risk case library (LRCL) of the supervised object. The risk matrix can be roughly regarded as a relational table. The risk matrix includes basic risk factors centered on the supervised object's medical history, family relationships, social experience, etc., daily risk factors centered on daily rounds, daily diet, daily behavior, emotional expression, etc., and interrogation risk factors centered on the supervised object's interrogation status, interrogation diet, daily life, behavior records, emotional changes, etc.

[0019] For example, the basic risk factors mainly include hypertension, heart disease, cerebral infarction and cerebral thrombosis, diabetes, limb disability, depression, coronary heart disease, family profile, marital status, etc. Daily risk factors mainly include blood pressure, chest tightness, abdominal pain, dizziness and tinnitus, oral anxiety and fear, wheezing, refusal of drug treatment, urination, drinking water, getting up at night, sleeping, getting up, talking in sleep, food intake, sitting quietly, standing quietly, wandering, cursing, resisting care, beating, self-harm, climbing, toilet timeout, head pounding, crying, sighing, chest beating, frustrated shaking of head, slapping the table, etc. Interrogation risk factors mainly include heart rate, breathing, interrogation emotions, restlessness, drinking water, food intake, getting up at night, sleeping, talking in sleep, wandering, cursing, resisting care, beating, self-harm, etc.

[0020] The corresponding risk matrix is ​​constructed based on these three risk factors. There are three risk matrices in total. Taking the basic risk matrix as an example, a specific example is shown in the following table:

[0021]

[0022] The above risk matrix is ​​just an example. The risk matrices corresponding to daily risk factors and interrogation risk factors are also in similar forms.

[0023] Step S2, constructing a safety risk measurement model: marking the data features of risk factors, dividing them into basic type and frequency type, and constructing a safety risk measurement model for each risk factor.

[0024] The above risk matrix also contains a data feature. After the risk matrix is ​​created, it is marked as basic or frequency type according to the actual situation of each risk dimension. The basic type generally refers to static information, whether it is present or absent. For example, whether there is high blood pressure, heart disease, etc. in the above table. The frequency type mainly refers to the different occurrence frequencies of risk dimensions. For example, the number of times you urinate, drink water, get up at night, talk in your sleep, etc. are all frequency types.

[0025] The last column of the risk matrix is ​​the risk baseline value, which is the initial value of the project risk judgment measurement setting. This step builds a safety risk measurement model for each risk factor. The details are as follows:

[0026] S21. Calculate the risk value of basic risk factors: a 基础型 (i) S 基础型 (x i ), where a 基础型 (i) Risk dimension weight of basic risk factors, S 基础型 (x i )=S 基础基准 (x i ), S 基础基准 (x i ) is the risk benchmark value of the basic risk factor.

[0027] The risk dimensions of each risk factor include both basic and frequency types.

[0028] This step takes the risk measurement model of basic risk factors as an example. The model assesses risk based on the static information of the regulated object, and the evaluation items include past medical history, family relationships, social experience, etc. Most of the data features are common, and the risk dimension x i The risk value is a 基础型 (i) S 基础型 (xi), S 基础基准 (x i ) is the risk benchmark value of the basic risk factor, that is, the value of the last column of the matrix, a 频次型 (i) Risk dimension weights for basic risk factors. Since the risk baseline value is a fixed value, but people of different types and personalities may perform differently, safety risk assessment cannot be completely measured based on performance. Therefore, it is necessary to set the weights of the risk dimensions to adjust the values. This weight needs to be updated in the subsequent model training to obtain a more accurate measurement model.

[0029] S22. Calculate the risk value of frequency risk factors: a 频次型 (i) S 频次型 (x i ), where a 频次型 (i) Risk dimension weights for frequency-based risk factors;

[0030]

[0031] S 频次基准 (x i ) is the risk benchmark value of the frequency risk factor, where f min (i) is the minimum frequency value set, f(i) is the frequency value of the current risk factor, S 上限 (i) is the maximum risk value of the frequency risk factor.

[0032] Frequency risk factors are measured based on the frequency of occurrence of behaviors, such as frequent toileting and frequent mood swings. Different frequencies represent different situations, so they need to be measured separately. The upper limit value S is set in this measurement model. 上限 (x i ), that is, the maximum risk value of the risk dimension of the frequency-type risk factor, to avoid excessive risk values ​​in the local risk dimension, which affects the overall judgment. In order to distinguish the differences, this step uses Multiply by S 频次基准 (x i ) to make adjustments.

[0033] Similarly, the performance of different types of regulated objects will also be different. Risk dimension weights are also set here. 频次型 (i).

[0034] Step S3, data collection and risk measurement: collect basic data of the supervised object, vital sign data and behavioral data during the care process, and vital sign data and behavioral data during the interrogation process, and calculate the basic risk value, daily risk value and inquiry risk value according to the safety risk measurement model.

[0035] This step is data collection and risk measurement. It collects basic data of the supervised object, physical sign data and behavioral data during the care process, and physical sign data and behavioral data during the interrogation process through recording and intelligent terminal devices (such as nursing terminals, smart bracelets, smart mattresses, smart seats, multimodal emotion recognition devices, etc.), and processes the data to provide basic portrait information, nursing portrait information, and interrogation portrait information. The risk values ​​of the three risk factors are calculated through the safety risk measurement model.

[0036] The specific process is as follows:

[0037] S31. Collect basic data of the supervised object, physical sign data and behavior data during the care process, and physical sign data and behavior data during the interrogation process through recording and intelligent terminal devices, and process the data to obtain basic portrait information, care portrait information, and interrogation portrait information of the supervised object;

[0038] S32, calculating the basic risk value S according to the safety risk measurement model 基础风险 、Daily Risk Value S 日常风险 and ask for risk value S 讯问风险 ;

[0039]

[0040]

[0041]

[0042] Among them, m, w, and r are respectively the number of risk dimensions whose data characteristics are basic type in basic risk factors, daily risk factors, and interrogation risk factors; n, h, and g are respectively the number of risk dimensions whose data characteristics are frequency type in basic risk factors, daily risk factors, and interrogation risk factors.

[0043] The three risk value calculations are all based on the accumulation of risk values ​​for the basic and frequency risk dimensions corresponding to each risk factor.

[0044] Step S4, comprehensive risk assessment: Calculate the comprehensive risk value based on the three calculated risk values ​​and the corresponding coefficients, and determine the risk level.

[0045] Based on the risk values ​​of the three portrait information and the set coefficients, the comprehensive risk value is calculated, and the risk level of the regulated object is determined according to the value, such as high, medium, or low risk.

[0046] The specific process of this step is as follows:

[0047] S41. Calculate the comprehensive risk value as follows:

[0048] S 总 =α 基础风险 ·S 基础风险 +α 看护风险 ·S 看护风险 +α 看护风险 ·S 讯问风险

[0049] where α 基础风险 , α 看护风险 , α 看护风险 They are the coefficients corresponding to basic risk factors, daily risk factors and interrogation risk factors, and the coefficients are fixed values.

[0050] S42. Determine the risk level:

[0051] The risk level of the regulated entity is determined according to the comprehensive risk value results, and the risk levels include high risk, medium risk, and low risk.

[0052] Step S5, feedback adjustment mechanism: through the feedback adjustment mechanism based on the combination of naive Bayesian network and reinforcement learning, the risk dimension weights of risk factors are dynamically adjusted to update the comprehensive risk value calculation model.

[0053] In order to meet the accuracy and dynamic adaptability of security risk assessment, a feedback adjustment mechanism based on the combination of naive Bayesian network and reinforcement learning is constructed. The specific process is as follows:

[0054] S51. For each risk dimension x i Set the corresponding initial risk dimension weight α(i) and set the prior probability P(R) for each node in the naive Bayesian network. k ), where R k It is a specific risk level, such as low risk, medium risk, and high risk.

[0055] The present invention needs to update the risk dimension weight, that is, the aforementioned a 基础型 (i) a 频次型 (i), for the sake of uniform representation, it is uniformly denoted as α(i). Each risk dimension has an initial value, and these weights represent the relative importance of each risk dimension. This step initializes the weights and prior probabilities.

[0056] S52. Constructing a Naive Bayesian Network Model:

[0057] According to the risk dimension of the regulated object, the posterior probability of each risk level is calculated through the naive Bayesian network:

[0058]

[0059] Among them, P(R k ) is the risk level R k The prior probability, P(x i ∣R k ) is a given risk level R k When the risk dimension x i The conditional probability of occurrence.

[0060] S53. Dynamically adjust the weights of risk dimensions through reinforcement learning regulation mechanism.

[0061] Reinforcement learning dynamically adjusts the weights of each risk dimension based on environmental feedback (such as the difference between the actual risk level and the predicted level). The specific reinforcement learning adjustment mechanism uses the Q-learning algorithm to update the weights of the risk dimensions to minimize the risk prediction error.

[0062] α(i)←α(i)+α·(R+γ·maxQ(s',a')-Q(s,a))

[0063] in,

[0064] a(i) is the risk dimension x i The current weight of

[0065] α is the learning rate, which controls the step size of each update;

[0066] γ is a discount factor that reflects the impact of future rewards on current actions;

[0067] R is the immediate reward, defined as the difference between the noted risk level and the predicted risk level.

[0068] S54. On the basis of adjusting weights by reinforcement learning, the weight α(i) of the risk dimension is further adjusted in combination with the posterior probability of the naive Bayesian network:

[0069] α(i)←α(i)+β·(P(R k |x1,x2,…,x n )-P 当前 (R k ))

[0070] in,

[0071] Q(s,a) is the Q value when taking action a in state s, which is used to evaluate the effects of different adjustment plans;

[0072] β is the learning rate used for Bayesian probability adjustment;

[0073] P(R k |x1,x2,…,x n ) is the posterior probability of risk level calculated by the naive Bayesian network;

[0074] P 当前 (R k ) is the risk level probability predicted by the current model.

[0075] S54. By continuously collecting new data, the risk dimension weights are adjusted according to reinforcement learning and naive Bayesian network until the accuracy of risk prediction converges to a stable state.

[0076] From S53 and S54, we can see that the weight α(i) needs to be adjusted twice, and new risk data is continuously collected, and the weight a(i) is updated according to the feedback of the actual risk level. Reinforcement learning and Bayesian network are processed alternately. Reinforcement learning adjusts the weight to make the prediction result more consistent with the actual risk level, while Bayesian network further optimizes the influence of each risk dimension through probabilistic reasoning.

[0077] The final weight update after combining reinforcement learning and Bayesian probability can be recorded as:

[0078] α(i)←α(i)+α·(R+γ·maxQ(s′,a′)-Q(s,a))+β·(P(R k |x1,x2,…,x n )-P 当前 (R k ))

[0079] When the accuracy of risk prediction converges to a stable state (i.e. the difference between the actual risk level and the predicted result is reduced to the expected range), the system will stop adjusting the weights significantly. It can be optimized later based on actual usage.

[0080] The safety risk measurement method for regulated objects provided by the present invention is a comprehensive risk measurement and assessment method based on data analysis, reinforcement learning and naive Bayesian joint calculation. It collects static information, behavioral data and psychological data of the regulated objects to conduct multi-dimensional safety risk assessment, and combines reinforcement learning for dynamic adjustment. It further uses the naive Bayesian method to optimize the assessment results through joint probability analysis, improves the accuracy of risk prediction, and evaluates the risks of regulated objects more objectively and accurately.

[0081] In summary, the present invention is mainly used for scientific and dynamic measurement and evaluation of the health risks, behavioral risks and psychological risks of the supervised objects. By introducing the naive Bayesian network and probabilistic reasoning to realize the joint analysis of various risk factors, combined with the reinforcement learning algorithm, the weights of each risk dimension are dynamically adjusted according to the feedback data, thereby improving the prediction accuracy of the risk assessment model; and a weight adaptive optimization mechanism is adopted. Through the Q-learning algorithm in reinforcement learning, the model can adaptively adjust the weights of the risk dimensions according to the risk assessment feedback information, and further optimize the weights in combination with the posterior probability of naive Bayes, and finally achieve a more accurate risk level assessment. The risk assessment mechanism of the present invention includes dynamic feedback adjustment. Through continuous feedback and optimization, the model can gradually improve the prediction effect during multiple evaluations to ensure the dynamic adaptability of risk level assessment.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for measuring safety risks of a regulated object, characterized in that: The method comprises the following steps: Step S1, identifying risk factors through the safety risk case library of the supervised object and constructing a risk matrix, wherein the risk factors include basic risk factors, daily risk factors and interrogation risk factors; Step S2: Mark the data features of risk factors, divide them into basic type and frequency type, and build a safety risk measurement model for each risk factor; Step S3, collecting basic data of the supervised object, physical sign data and behavioral data during the care process, and physical sign data and behavioral data during the interrogation process, and calculating the basic risk value, daily risk value and inquiry risk value according to the safety risk measurement model; Step S4: Calculate the comprehensive risk value based on the three calculated risk values ​​and the corresponding coefficients, and determine the risk level; Step S5: Dynamically adjust the risk dimension weights of risk factors through a feedback adjustment mechanism based on a combination of a naive Bayesian network and reinforcement learning to update the comprehensive risk value calculation model.

2. The method for measuring safety risk of a supervised object according to claim 1, characterized in that: In step S1, basic risk factors include past medical history, family relationships, and social experience; daily risk factors include daily rounds, daily diet and daily life, daily behavior, and emotional expression; interrogation risk factors include interrogation status, interrogation diet and daily life, behavior records, and emotional changes; then a corresponding risk matrix is ​​constructed based on the three risk factors.

3. The method for measuring safety risk of a supervised object according to claim 2, characterized in that: The specific process of step S2 is as follows: S21. Calculate the risk value of basic risk factors: a 基础型 (i) S 基础型 (x i ), where a 基础型 (i) Risk dimension weight of basic risk factors, S 基础型 (x i )=S 基础基准 (x i ), S 基础基准 (x i ) is the risk benchmark value of the basic risk factor; S22. Calculate the risk value of frequency risk factors: a 频次型 (i) S 频次型 (x i ), where a 频次型 (i) Risk dimension weights for frequency-based risk factors; S 频次基准 (x i ) is the risk benchmark value of the frequency risk factor, where f min (i) is the minimum frequency value set, f(i) is the frequency value of the current risk factor, S 上限 (i) is the maximum risk value of the frequency risk factor.

4. The method for measuring safety risk of a supervised object according to claim 3, characterized in that: The specific process of step S3 is as follows: S31. Collect basic data of the supervised object, physical sign data and behavior data during the care process, and physical sign data and behavior data during the interrogation process through recording and intelligent terminal devices, and process the data to obtain basic portrait information, care portrait information, and interrogation portrait information of the supervised object; S32, calculating the basic risk value S according to the safety risk measurement model 基础风险 、Daily Risk Value S 日常风险 and ask for risk value S 讯问风险 ; Among them, m, w, and r are respectively the number of risk dimensions whose data characteristics are basic type in basic risk factors, daily risk factors, and interrogation risk factors; n, h, and g are respectively the number of risk dimensions whose data characteristics are frequency type in basic risk factors, daily risk factors, and interrogation risk factors.

5. The method for measuring safety risk of a supervised object according to claim 4, characterized in that: The specific process of step S4 is as follows: S41. Calculate the comprehensive risk value as follows: S 总 =a 基础风险 ·S 基础风险 +a 看护风险 ·S 看护风险 +a 看护风险 ·S 讯问风险 where α 基础风险 , α 看护风险 , α 看护风险 They are the coefficients corresponding to basic risk factors, daily risk factors, and interrogation risk factors; S42. Determine the risk level: The risk level of the regulated entity is determined according to the comprehensive risk value results, and the risk levels include high risk, medium risk, and low risk.

6. The method for measuring safety risk of a supervised object according to claim 5, characterized in that: The specific process of step S5 is as follows: S51. For each risk dimension x i Set the corresponding initial risk dimension weight α(i) and set the prior probability P(R k ), where R k is the specific risk level; S52. Constructing a Naive Bayesian Network Model: The posterior probability of each risk level is calculated through a naive Bayesian network based on the risk dimension of the regulated object. Among them, P(R k ) is the risk level R k The prior probability, P(x i |R k ) is a given risk level R k When the risk dimension x i The conditional probability of occurrence. S53. Dynamically adjust the weight of the risk dimension through a reinforcement learning adjustment mechanism, and specifically the reinforcement learning adjustment mechanism uses a Q-learning algorithm to update the weight of the risk dimension; α(i)←α(i)+α·(R+γ·maxQ(s′,a′)-Q(s,a)) in, a(i) is the risk dimension x i The current weight of α is the learning rate, which controls the step size of each update; γ is a discount factor that reflects the impact of future rewards on current actions; R is the immediate reward; S54. On the basis of adjusting weights by reinforcement learning, the weight α(i) of the risk dimension is further adjusted in combination with the posterior probability of the naive Bayesian network: α(i)←α(i)+β·(P(R k |x1,x2,…,x n )-P 当前 (R k )) in, Q(s, a) is the Q value when taking action a in state s, which is used to evaluate the effects of different adjustment plans; β is the learning rate used for Bayesian probability adjustment; P(R k |x1,x2,...,x n ) is the posterior probability of risk level calculated by the naive Bayesian network; P 当前 (R k ) is the risk level probability predicted by the current model; S54. By continuously collecting new data, the risk dimension weights are adjusted according to reinforcement learning and naive Bayesian network until the accuracy of risk prediction converges to a stable state.

Images