An intelligent comprehensive control system and medium for bedsore prevention

Abstract

The present invention discloses an intelligent comprehensive control system for beds that prevent pressure sores and a medium, and relates to the field of medical beds; the system comprises: a bed; at least one processor is configured to implement an intelligent comprehensive control method for beds that prevent pressure sores when executing a computer program, and the specific steps are as follows: obtaining information related to the patient's physical characteristics and information related to the condition; using the patient's physical characteristics and information related to the condition as input variables for fuzzy processing, and then performing reasoning through a preset fuzzy rule library to obtain a fuzzy value of the output variable, and finally performing defuzzification to obtain a specific bed adjustment plan; executing the specific bed adjustment plan; the application has the following effects: formulating a personalized bed adjustment plan in combination with the patient's physical characteristics and condition, improving the patient's comfort and rehabilitation effect, effectively preventing the occurrence of pressure sores, and reducing the workload of medical staff.

Description

Technical Field

[0001] The present invention relates to the field of medical beds, and in particular to an intelligent comprehensive management and control system for beds that prevent pressure sores and a medium. Background Art

[0002] With the continuous development of medical technology and the improvement of people's requirements for the quality of medical services, the application of intelligent medical equipment in clinical nursing is becoming more and more extensive. For patients who are bedridden for a long time, the anti-pressure sore bed is an important medical device that can effectively prevent the occurrence of pressure sores and improve the comfort and rehabilitation effect of patients.

[0003] Traditional anti-pressure sore beds mainly prevent pressure sores through timed turning, air cushion beds, etc., or some smart anti-pressure sore beds can automatically adjust according to simple physical characteristics such as the patient's weight.

[0004] With respect to the above-mentioned related technologies, the inventors found that there are the following defects: on the one hand, there is a lack of personalized consideration of the patient's physical characteristics and condition, and the adjustment plan is relatively simple. Moreover, for patients with complex conditions, traditional beds are difficult to meet the nursing needs of multiple diseases at the same time. Summary of the invention

[0005] In order to effectively prevent the occurrence of pressure sores, reduce the workload of medical staff, and improve the intelligence level of bed management, the present application provides an intelligent anti-pressure sore bed comprehensive management and control system and medium.

[0006] In the first aspect, the present application provides an intelligent comprehensive management and control system for beds that prevent pressure sores, which adopts the following technical solutions:

[0007] An intelligent comprehensive management and control system for beds that prevent pressure sores, comprising:

[0008] Hospital beds;

[0009] at least one processor;

[0010] At least one memory storing a computer program, and at least one processor configured to implement a comprehensive management and control method for an intelligent anti-pressure sore bed when executing the computer program, the specific steps are as follows:

[0011] Obtaining information related to the patient's physical characteristics and medical conditions, where the physical characteristics include height, weight, and body shape, and the medical conditions include specific disease types and the severity of the disease;

[0012] The patient's physical characteristics and condition-related information are used as input variables for fuzzy processing, and then reasoning is performed through the preset fuzzy rule base to obtain the fuzzy value of the output variable. Finally, the fuzzification is performed to obtain a specific bed adjustment plan, where the output variables are the bed angle, height and support position.

[0013] Implement specific bed adjustment plans;

[0014] Analyze whether the patient's feeling description information is received within the preset time range;

[0015] If not, maintain the original setting;

[0016] If yes, then use the natural language processing algorithm to analyze the patient's description of feelings and extract key information, and use the naive Bayes algorithm to classify the patient's description of feelings and determine the specific problem type;

[0017] Based on the correspondence between problems and adjustment strategies, analyze and determine the adjustment strategies and implement the corresponding strategies.

[0018] By adopting the above technical solutions, a personalized bed adjustment plan can be formulated in combination with the patient's physical characteristics and condition to improve the patient's comfort and rehabilitation effect. Natural language processing can respond to patients' feelings in a timely manner, and the adjustment strategy is more accurate, which improves the quality of medical services and patient satisfaction. At the same time, it effectively prevents the occurrence of pressure sores, reduces the workload of medical staff, improves the intelligence level of bed management, and provides patients with a better medical care environment.

[0019] Optionally, the method further includes steps after obtaining the patient's physical characteristic-related information and the disease-related information and before using the patient's physical characteristic-related information and the disease-related information as input variables for fuzzy processing, which are as follows:

[0020] Analyze whether the patient's condition involves multiple specific disease types;

[0021] If not, maintain the original setting;

[0022] If yes, the combination of the bed angle, height and support position is encoded as a chromosome of the genetic algorithm, and the multiple specific disease types involved in the patient's condition are defined as additional genes and added to the chromosome;

[0023] Taking the preset target and the multiple specific disease types involved as targets, and constructing a fitness function according to the preset target weight allocation method;

[0024] The preset fitness ratio selection method is used to select individuals for reproduction according to the fitness value of the chromosome;

[0025] Perform crossover operation on selected individuals to form new individuals, and perform mutation operation on chromosomes with preset probability;

[0026] The chromosome algorithm is continuously iterated. When the algorithm converges or reaches the preset stop condition, the iteration is stopped and the chromosome with a higher fitness value is obtained, that is, the optimal combination of bed angle, height, support position and disease, as the final bed adjustment plan.

[0027] By adopting the above technical solutions, genetic algorithms are used to optimize bed adjustment plans when dealing with patients with multiple specific diseases. It can comprehensively consider the needs of multiple diseases and improve the accuracy and adaptability of the plan. It can find the optimal combination through iteration, improve patient comfort and treatment effects, and reduce the risk of complications. At the same time, it can improve the intelligence level of bed management, provide better medical care for patients with complex conditions, reduce the workload of medical staff, and help improve the utilization efficiency of medical resources and the overall quality of medical services.

[0028] Optionally, the preset target and the multiple specific disease types involved are taken as targets, and according to the preset target weight allocation method, a fitness function is constructed to include:

[0029] The preset goals are to reduce patient discomfort and reduce the risk of pressure ulcers. The common goals for multiple diseases include maintaining stable body functions and promoting recovery. The quantitative value of the patient's discomfort is defined as , the goal is to minimize patient discomfort, i.e. The smaller the better, the risk index of pressure ulcers is defined as , the goal is to make The smaller the value, the better. The body function stability index is defined as , the goal is to make Stay within the preset range and define the recovery index as , the goal is to make Continuously improve;

[0030] According to the initial weights determined by the preset different goals, the weight of alleviating patient discomfort is set to , the weight for reducing the risk of pressure ulcers is , the weight to maintain stable body function is , the weight for promoting recovery is ;

[0031] The fitness function is expressed as the weighted sum of multiple objective functions, as follows:

[0032] ;

[0033] in, is a function related to the goal of alleviating patient discomfort, is a function related to the goal of reducing pressure ulcer risk, is a function related to the goal of maintaining stable body functions. is a function related to promoting rehabilitation goals and is defined according to rehabilitation indicators. is the fitness value.

[0034] By adopting the above technical solutions and taking into account multiple important goals, we can accurately balance patient discomfort, pressure sore risk, body function stability and rehabilitation needs. Through clear goal quantification and weight allocation, the bed adjustment plan is more scientific and reasonable. It helps to provide personalized care for patients with multiple diseases, minimize discomfort and reduce the risk of pressure sores, while promoting body function stability and rehabilitation progress, improving the quality of medical services and patient satisfaction, improving the intelligence level of bed management, and creating a more favorable treatment environment for patients with complex conditions.

[0035] Optionally, a natural language processing algorithm is used to analyze the patient's description of feelings and extract key information. The patient's description of feelings is classified by a naive Bayes algorithm to determine the specific problem types, including:

[0036] The form of expression to obtain information describing the patient's feelings;

[0037] If the patient's feeling description information is expressed in the form of voice information, adaptive voice recognition technology is used to obtain the patient's voice description signal, and the voice description signal is preprocessed;

[0038] The preprocessed speech description signal is converted into a spectrogram, and a convolutional neural network is used to extract key features of the spectrogram, including the frequency distribution, time domain characteristics, and harmonic structure of the speech.

[0039] Match the key features extracted by the convolutional neural network with the question type features in the preset speech feature library, use the naive Bayes algorithm to calculate the probability of each question type, and determine the specific question type according to the probability. The speech feature library stores the mapping relationship between known question types and specific feature expressions.

[0040] If the patient's feeling description information is expressed in text information, the text description information provided by the patient is pre-processed and word segmented;

[0041] Extract key features of text information, including lexical features, grammatical features and semantic features;

[0042] Matching the extracted text information features with the question type features in a preset text feature library, which stores the mapping relationship between known question types and specific text feature expressions;

[0043] The naive Bayes algorithm is used to calculate the probability of each question type, and the specific question type is determined based on the probability.

[0044] By adopting the above technical solutions, natural language processing and naive Bayes algorithm are used to analyze the patient's description of feelings, and key information can be accurately extracted and the type of problem can be determined regardless of voice or text form. Preprocessing and spectrum analysis of voice signals are performed to extract multi-dimensional features to improve accuracy. The extraction of vocabulary, grammar and semantic features of text information also enhances the accuracy of judgment. It can quickly identify patient problems, provide a basis for timely adjustment of beds and medical strategies, improve the quality and efficiency of medical services, meet patient needs, and provide strong support for the comprehensive management and control of intelligent anti-pressure sore beds.

[0045] Optionally, the key features extracted by the convolutional neural network are matched with the question type features in the preset voice feature library, and the probability of each question type is calculated using the naive Bayes algorithm. The specific question type is determined according to the probability, including:

[0046] Extracting prosodic features from speech description signals, including stress, rhythm, and duration;

[0047] Analyze the speech description signal for specific prosodic features or specific keywords;

[0048] If not, maintain the original setting;

[0049] If yes, then according to the prosodic feature quantitative evaluation method, the overall effective evaluation value of the prosodic feature is calculated, and according to the number of occurrences of the specific keyword and the evaluation value of the specific keyword, the evaluation value of the specific keyword is analyzed and determined. In addition, the key features extracted by the convolutional neural network are matched with the question type features in the preset speech feature library, and the probability of each question type is calculated using the naive Bayes algorithm. According to the corresponding relationship between the probability interval into which the probability of each question falls and the evaluation score, the deterministic evaluation value of the deep learning classification result is analyzed and determined;

[0050] The urgency score is calculated based on the evaluation value of the specific keyword determined by the analysis, the overall effective evaluation value of the prosodic feature, the certainty evaluation value of the deep learning classification result, and the preset weight distribution ratio;

[0051] If the urgency score exceeds the preset urgency score, the new problem type probability is recalculated according to the preset problem type probability weight distribution scheme, and the specific problem type is determined according to the new problem probability;

[0052] If the urgency score is less than or equal to the preset urgency score, the original setting is maintained.

[0053] By adopting the above technical solution, the type and urgency of the problem can be accurately judged through multi-dimensional analysis of the voice description signal. The accuracy is improved by extracting rhythmic features and keyword evaluation values ​​and combining them with deep learning classification. The urgency scoring mechanism helps to quickly adjust strategies in emergency situations. The probability weight of the problem type is adjusted according to the score to make the judgment more accurate. It provides strong support for the management and control of intelligent anti-pressure sore beds, can respond to patient needs in a timely manner, improve the quality and efficiency of medical services, and ensure patient safety. It is an effective comprehensive management and control method.

[0054] Optionally, the overall effective evaluation value of the prosodic feature is calculated to include:

[0055] According to the preset stress and rhythm intensity quantitative analysis method, the stress and rhythm intensity quantitative evaluation value is calculated. At the same time, according to the preset duration quantitative analysis method, the duration quantitative evaluation value is calculated. Based on the stress and rhythm intensity quantitative evaluation value and the duration quantitative evaluation value, the overall effective evaluation value of the rhythmic feature is calculated.

[0056] By adopting the above technical solutions, by quantifying the stress rhythm and duration respectively, the key information in the voice can be accurately grasped. The overall evaluation value obtained by combining the two more comprehensively reflects the patient's status and provides a strong basis for determining the type and urgency of the problem. It helps to adjust the bed settings and medical strategies more accurately, improve patient comfort and treatment effects. It improves the scientificity and effectiveness of the comprehensive management and control of intelligent anti-pressure sore beds, better meets the needs of patients, and contributes to the improvement of medical service quality.

[0057] Optionally, recalculate the probability of obtaining a new question type including:

[0058] Obtain the difference between the urgency score and the preset urgency score;

[0059] According to the corresponding relationship between the difference and the dynamic adjustment factor value, the dynamic adjustment factor value is analyzed and determined;

[0060] According to the preset question type probability weight distribution scheme, the new question type probability is recalculated and obtained, and the new question type probability is re-determined based on the product of the obtained new question type probability and the dynamic adjustment factor value.

[0061] By adopting the above technical solution, the dynamic adjustment factor is determined according to the difference in the urgency score, making the probability calculation more flexible. Combining the preset weight distribution scheme and the dynamic factor, it can more accurately reflect the impact of emergency situations on the problem type. It helps to quickly adjust strategies in emergency situations and provide patients with more timely and effective medical services. It improves the adaptability and accuracy of the intelligent anti-pressure sore bed management, ensures patient safety, improves the quality and efficiency of medical services, and provides a scientific and effective method for the treatment of complex conditions.

[0062] Optionally, according to the correspondence between the problem and the adjustment strategy, analyzing and determining the adjustment strategy and executing the corresponding strategy include:

[0063] Analyze whether the urgency score exceeds a preset urgency score;

[0064] If yes, then according to the correspondence between the conditions and problems exceeding the preset urgency score and the adjustment strategy, analyze and determine the adjustment strategy and execute the corresponding strategy;

[0065] If not, keep the original setting.

[0066] By adopting the above technical solutions, when the urgency is high, timely analysis and implementation of corresponding strategies can quickly respond to the patient's emergency and improve the efficiency of treatment. Avoid excessive adjustments under low urgency and maintain the original settings to save resources. It helps to provide patients with precise medical services and improve the scientificity and effectiveness of bed management. It improves the efficiency of medical resource utilization, ensures patient safety, enhances the adaptability of comprehensive management of intelligent anti-pressure sore beds, and creates better conditions for patient rehabilitation.

[0067] Optionally, steps are also included in parallel with analyzing and determining the adjustment strategy and executing the corresponding strategy, as follows:

[0068] The expression form of obtaining information describing the patient's feelings;

[0069] The expression form of the patient's feeling description information is used as the expression method of the adjustment strategy, and the specific adjustment strategy is sent to the terminal held by the person in charge.

[0070] By adopting the above technical solution, the expression of the adjustment strategy is determined according to the expression form of the patient's feeling description information, which enhances the pertinence and effectiveness of information transmission. The specific adjustment strategy is sent to the terminal held by the person in charge to ensure that the person in charge is aware of the situation and takes action in a timely manner, thereby improving the response speed. It helps to better meet the needs of patients, improve the quality and efficiency of medical services, make the comprehensive management and control of smart anti-pressure sore beds more humane and intelligent, provide patients with a better medical care environment, and promote the patient's recovery process.

[0071] In a second aspect, the present application provides a computer storage medium, which adopts the following technical solution:

[0072] A computer storage medium includes a program that can be loaded and executed by a processor to implement a comprehensive control method for an anti-pressure sore bed in an intelligent anti-pressure sore bed comprehensive control system as described in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1It is a schematic diagram of the overall process of a smart anti-pressure sore bed comprehensive management and control method in a smart anti-pressure sore bed comprehensive management and control system in an embodiment of the present application;

[0074] Figure 2 It is a flowchart of another embodiment of the present application, which is a step after obtaining the patient's physical characteristic related information and the disease related information, and before using the patient's physical characteristic related information and the disease related information as input variables for fuzzy processing;

[0075] Figure 3 This is a flowchart of another embodiment of the present application, which uses a natural language processing algorithm to analyze the patient's feeling description information, extract key information, and classify the patient's feeling description through a naive Bayes algorithm to determine the specific problem type;

[0076] Figure 4 This is a flowchart of another embodiment of the present application that matches the key features extracted by the convolutional neural network with the question type features in a preset speech feature library, uses the naive Bayes algorithm to calculate the probability of each question type, and determines the specific question type based on the probability. DETAILED DESCRIPTION

[0077] The present application is further described in detail below in conjunction with the accompanying drawings.

[0078] Reference Figure 1 , is a smart anti-pressure sore bed comprehensive management and control system disclosed in the present application, comprising a bed, at least one processor and at least one memory storing a computer program, at least one processor is configured to implement a smart anti-pressure sore bed comprehensive management and control method when executing the computer program, and the specific steps are as follows:

[0079] Step S100, obtaining information related to the patient's physical characteristics and condition.

[0080] Among them, information related to physical characteristics includes height, weight, and body shape, and information related to the disease includes specific disease types and the severity of the disease. The body shape is obtained as follows: It is determined by calculating the body mass index (BMI) and measuring body dimensions. First, the BMI value is calculated based on the weight (kg) divided by the square of the height (m). Then, use a soft tape measure to measure the patient's waist circumference, hip circumference and other dimensions. For example, a BMI below 18.5 can be judged as thin, 18.5-23.9 is normal, and 24 and above is obese. At the same time, the body type is further determined by combining the ratio of waist circumference to hip circumference. For example, a larger waist-to-hip ratio may tend to be centrally obese.

[0081] The method of obtaining specific disease types is as follows: obtain the patient's diagnosis report information through the hospital's electronic medical record system. After diagnosis, the doctor enters the patient's disease type such as heart disease, diabetes, hypertension, etc. into the electronic medical record system, and the system can automatically extract this information. For complex disease conditions, a comprehensive judgment can also be made by combining multi-source information such as test reports in the laboratory information system and examination results in the imaging information system. Consider the requirements of different diseases for body positions, as follows: Patients with hypertension, diabetes, etc. generally have no special requirements for body positions. Patients with cardiopulmonary failure are generally forced to sit because this can alleviate symptoms such as dyspnea. Patients with acute peritonitis are forced to lie supine with their knees bent, which helps relieve abdominal pain. Patients with spinal diseases are forced to lie prone, which can reduce spinal pressure. Shock patients generally adopt a concave position to increase the amount of blood returning to the heart. Patients with intracranial hypertension adopt a head-high, feet-low position to help reduce intracranial pressure.

[0082] The severity of the disease is obtained as follows: the evaluation is carried out in combination with medical examination indicators and symptom manifestations, and the APACHEII scoring system (disease severity score) can be referred to. For example, for patients with heart disease: a comprehensive evaluation is carried out by detecting cardiac function indicators such as ejection fraction, electrocardiogram abnormality, etc., as well as the patient's symptoms such as dyspnea and palpitation frequency. Different indicator intervals can be set to correspond to different disease severity levels, and the system automatically determines the severity of the disease based on the specific values ​​detected. For example, an ejection fraction of more than 50% and no symptoms is mild, an ejection fraction of 30%-50% and mild symptoms are moderate, and an ejection fraction of less than 30% and obvious symptoms are severe. For critically ill patients who meet the scope of application of the APACHEII scoring system, the patient's physiological indicators (such as body temperature, heart rate, respiratory rate, mean arterial pressure, arterial oxygen partial pressure, pH value, serum sodium, serum potassium, serum creatinine, hematocrit, white blood cell count, etc.), age and chronic health status information are collected, and a comprehensive score is performed to determine the severity of the disease.

[0083] Step S200, the patient's physical characteristics and condition-related information are used as input variables for fuzzy processing, and then reasoning is performed through a preset fuzzy rule base to obtain the fuzzy value of the output variable, and finally defuzzification is performed to obtain a specific bed adjustment plan, where the output variables are the bed angle, height and support position.

[0084] Specifically, for height, it can be divided according to the pre-set fuzzy interval. For example, height is divided into three fuzzy language variables: "short", "medium" and "tall". The division interval can be determined based on statistical data, such as height below 160cm is "short", 160cm-180cm is "medium", and above 180cm is "tall". The system automatically determines the fuzzy language variable to which the patient belongs based on the actual height value. The same is true for weight, which can be divided into "light", "medium" and "heavy". For example, weight below 50kg is "light", 50kg-70kg is "medium", and above 70kg is "heavy". Body shape can be divided into "thin", "normal" and "obese". It can be divided according to the range of body mass index (BMI), with BMI below 18.5 being "thin", 18.5-23.9 being "normal", and 24 and above being "obese".

[0085] Specific disease types can be fuzzified according to the severity of the disease and the degree of impact on bed adjustment. For example, heart disease can be divided into fuzzy language variables such as "mild heart problem", "moderate heart problem" and "severe heart problem". It can be divided according to heart function indicators, symptoms, etc. The severity of the disease can be divided into fuzzy language variables such as "mild", "moderate" and "severe". It can be divided according to the specific numerical range of medical examination indicators.

[0086] In addition, the fuzzy rule base is a collection of fuzzy rules, which are formulated based on expert experience and medical knowledge and are used to derive the fuzzy value of the output variable based on the input fuzzy variable value. The fuzzy rule base is obtained as follows: medical experts, engineers and other professionals formulate fuzzy rules based on a large amount of clinical experience and research data. For example, a fuzzy rule can be "If the patient is obese and the severity of the disease is severe, then the angle of the bed should be increased, the height should be increased, and the support part should be strengthened." These rules are stored in the system in the form of conditional statements. When the input variable meets the condition part of the rule, the system derives the fuzzy value of the output variable according to the conclusion part of the rule.

[0087] Defuzzification to obtain a specific bed adjustment plan refers to converting the fuzzy value of the output variable into a specific numerical value, that is, determining the adjustment values ​​of the specific angle, height and support part of the bed.

[0088] The specific acquisition method is as follows: Defuzzification methods such as the centroid method and the maximum membership method can be used. For example, when the centroid method is used, the centroid position of the output variable fuzzy set is calculated and output as a specific numerical value. Assuming that the fuzzy set of the output variable "bed angle" is {small angle: 0.3, medium angle: 0.5, large angle: 0.2}, by calculating the centroid position, the specific bed angle adjustment value can be obtained. For example, assuming that the small angle corresponds to 30 degrees, the medium angle corresponds to 45 degrees, and the large angle corresponds to 60 degrees, the centroid position is calculated to be (30×0.3+45×0.5+60×0.2) / (0.3+0.5+0.2)=43.5 degrees, that is, it is determined that the bed angle should be adjusted to 43.5 degrees. The defuzzification process of height and support position is similar.

[0089] Step S300, executing a specific bed adjustment plan.

[0090] Step S400, analyzing whether the patient's feeling description information is received within a preset time range. If not, executing step S500; if yes, executing step S600.

[0091] Whether the patient's feeling description information has been received can be automatically detected in the following ways: If the patient's bed is equipped with a voice input device, the system can continuously monitor the voice signal. When a voice signal is input, the system automatically determines whether it is the patient's feeling description information through voice recognition technology. For example, using automatic voice recognition software, the input voice is analyzed in real time. If specific keywords such as "uncomfortable", "painful", "uncomfortable", etc. are recognized, it is determined to be the patient's feeling description information. If the bed is equipped with a touch screen or input device, the patient can provide a feeling description by text input. The system periodically scans the text content on the input device. If new text input is detected and the content meets the characteristics of the feeling description (such as containing words expressing physical feelings), it is determined that the patient's feeling description information has been received.

[0092] The preset time range is determined as follows: The preset time range can be set according to clinical experience and actual needs. For example, it can be set to detect once every hour, that is, the system checks whether there is any patient's feeling description information at a fixed time point every hour. Or the time range can be dynamically adjusted according to the severity of the patient's condition and the specific disease type. For patients with more serious conditions or unstable conditions, the detection time interval can be shortened, such as once every half an hour; for patients with relatively stable conditions, the time interval can be appropriately extended, such as once every two hours.

[0093] Therefore, the steps for determining the preset time range may be as follows: analyzing and determining the detection time interval according to the corresponding relationship between the severity of the disease, the specific disease type and the detection time interval, and analyzing and obtaining the preset time range according to the determined detection time interval.

[0094] Step S500, maintaining the original settings.

[0095] Step S600, using a natural language processing algorithm to analyze the patient's feeling description information and extract key information, and using a naive Bayes algorithm to classify the patient's feeling description to determine the specific problem type.

[0096] Natural language processing algorithm is a technology that can process human natural language, including operations such as word segmentation, part-of-speech tagging, named entity recognition, and sentiment analysis to extract key information and semantic content from the text.

[0097] If the patient's description of his feelings is provided in the form of speech, the speech is first converted into text through speech recognition technology. Then, the text is processed using mature natural language processing toolkits, such as NLTK (Natural Language Toolkit) or SpaCy. For example, for the patient's words "my back hurts a lot, and I feel a little short of breath", after word segmentation, the words "my", "back", "very", "painful", "feeling", "a little", "breathing", and "difficult" are obtained; part-of-speech tagging can determine the part of speech of each word, such as "back" is a noun and "pain" is an adjective; named entity recognition can determine whether there is a specific medical entity, and there may be no obvious named entity here. Through these processes, the key information "back pain" and "difficulty breathing" can be extracted.

[0098] The Naive Bayes algorithm is a classification algorithm based on the Bayesian theorem and the conditional independence assumption of features. It determines which category the input text belongs to by calculating the probability of the features appearing in different categories. In this step, the patient's description of his feelings is used as input, and different types of questions are used as categories. The specific type of question is determined by the Naive Bayes algorithm.

[0099] The specific acquisition method is as follows:

[0100] First, we need to collect a large number of patient description samples, annotate them, and determine the corresponding problem types. For example, we can collect different descriptions of feelings such as "back pain", "leg pain", and "difficulty breathing", and annotate them as different problem types such as "pain problem" and "breathing problem".

[0101] Then, these labeled samples are used to train the naive Bayes classifier. During the training process, the algorithm learns the probability of various features appearing under different problem types. When a new description of a patient's feelings is input, the algorithm calculates the probability that the description belongs to different problem types based on the learned probability distribution, and the category with the highest probability is the specific problem type determined. For example, for a new description of feelings, "My arm is sore", after calculation by the algorithm, if the probability of "pain problem" is 0.6 and the probability of "other problems" is 0.4, then it is determined that the description belongs to "pain problem".

[0102] Step S700: Analyze and determine the adjustment strategy based on the corresponding relationship between the problem and the adjustment strategy, and execute the corresponding strategy.

[0103] Specifically, a database corresponding to problems and adjustment strategies is established: For "pain problems": quantitative indicators: adjustment angle range of support parts, increase in support strength. For example, when it is determined to be back pain, if the pain is mild, the adjustment angle of the back support part of the bed is within the range of 5-10 degrees, and the support strength is increased by 10%-20%; if the pain is severe, the adjustment angle is within the range of 10-15 degrees, and the support strength is increased by 20%-30%. For "respiratory problems": quantitative indicators: head of bed elevation angle. For example, for mild respiratory problems, the head of the bed is elevated by 15-25 degrees; for moderate respiratory problems, it is elevated by 25-35 degrees; for severe respiratory problems, it is elevated by 35-45 degrees.

[0104] The system implements the following adjustment strategy: The bed control system uses the motor drive to adjust the angle and support position. For example, for angle adjustment, the motor rotates at a speed of X degrees per minute until the target angle is reached. For support force adjustment, the air pressure or hydraulic system is used to increase the pressure by Y units per second until the target force is reached.

[0105] Reference Figure 2 The intelligent anti-pressure sore bed comprehensive management and control system further includes steps after obtaining the patient's physical characteristics related information and the disease condition related information, and before using the patient's physical characteristics related information and the disease condition related information as input variables for fuzzy processing, which are as follows:

[0106] Step SA00, analyzing whether the patient's condition involves multiple specific disease types. If not, executing step SB00; if yes, executing step SC00.

[0107] Among them, step SA00 mainly determines whether the patient's condition involves multiple specific disease types. This step is crucial for determining the subsequent bed adjustment strategy. If the patient only suffers from a single disease, a relatively simple adjustment method may be used; if multiple diseases are involved, a more complex algorithm is required to comprehensively consider various factors to find the optimal bed adjustment plan.

[0108] The sources of information acquisition are as follows: 1. Electronic medical record system: The hospital's electronic medical records record the patient's diagnosis information, medical history, test results, etc. in detail. The system can automatically search the electronic medical records to find out whether there are multiple diagnostic records for specific diseases. For example, if the electronic medical record records the diagnosis of different diseases such as "diabetes", "hypertension", and "heart disease" at the same time, it can be determined that the patient's condition involves multiple specific disease types. 2. Medical testing equipment: Various medical testing equipment such as blood glucose meters, blood pressure monitors, electrocardiographs, etc. can provide real-time data about the patient's physical condition. If multiple different types of testing equipment show abnormal results, it may suggest that the patient has multiple diseases. For example, the blood glucose meter detects high blood sugar, the blood pressure monitor shows high blood pressure, and the electrocardiogram is abnormal, which may indicate that the patient suffers from diabetes, hypertension, and heart disease at the same time.

[0109] There are two methods for judging whether multiple specific disease types are involved, as follows: 1. Disease code matching: set a specific disease code for each specific disease. The system determines whether there are multiple specific diseases by comparing the diagnosis code in the patient's electronic medical record with the preset disease code list. For example, the preset diabetes code is "D01", the hypertension code is "H02", and the heart disease code is "C03". If multiple such codes appear in the electronic medical record, it is judged that the patient's condition involves multiple specific disease types. 2. Comprehensive indicator judgment: judge by combining the test results of multiple medical testing equipment and clinical indicators. For example, set the normal range of indicators such as blood sugar, blood pressure, and heart rate. If the patient's blood sugar and blood pressure are both outside the normal range, and the electrocardiogram shows abnormalities, it is judged that the patient may have multiple diseases.

[0110] Step SB00, maintain the original settings.

[0111] Step SC00, encoding the bed angle, height and support position combination into chromosomes of the genetic algorithm, and defining various specific disease types involved in the patient's condition as additional genes and adding them to the chromosomes.

[0112] Specifically, the bed parameters are encoded as follows: The bed angle can be encoded according to the actual angle range. For example, if the adjustment range of the bed angle is 0-90 degrees, it can be encoded as an integer, such as 0 represents 0 degrees, 90 represents 90 degrees, and the angles in between are encoded proportionally. The bed height can also be encoded according to the actual height range. Assuming the height range is 50-100 cm, different height values ​​can be encoded as integers. The support parts can be encoded in binary to represent different support states, such as 0 for disabled support and 1 for enabled support.

[0113] Disease type coding is as follows: a unique code is assigned to each specific disease type. For example, the numbers 1, 2, 3, etc. can represent different diseases, such as diabetes, hypertension, heart disease, etc. If a patient suffers from multiple diseases, there will be multiple coding combinations at the corresponding gene position.

[0114] The structure of chromosome is as follows: the chromosome can be composed of the bed angle, height, support site code and disease type code in sequence. For example, a chromosome may be "angle code-height code-support site code-disease type code 1-disease type code 2-...".

[0115] Suppose a patient suffers from diabetes and heart disease, and the bed angle can be adjusted from 0 to 60 degrees, the height range is 60 to 80 cm, and there are two support parts: back and legs.

[0116] First, encode the bed parameters. Suppose the angle is encoded as an integer between 0 and 60, the height is encoded as an integer between 60 and 80, the back support is represented by 0 (not turned on) and 1 (turned on), and the leg support is also represented by 0 and 1.

[0117] Then, code the disease, say diabetes is coded as 1 and heart disease is coded as 2.

[0118] Then the chromosomes of this patient may be "30 (coding for angle) -70 (coding for height) -10 (back open, legs not open) -1 (coding for diabetes) -2 (coding for heart disease)".

[0119] Step SD00, taking the preset target and the multiple specific disease types involved as targets, and constructing a fitness function according to the preset target weight allocation method.

[0120] Step SE00, using the preset fitness ratio selection method, select individuals for reproduction according to the fitness value of the chromosome.

[0121] In this step, the preset fitness proportional selection method is used to select individuals for reproduction according to the fitness value of the chromosome. The fitness proportional selection method is also called the roulette wheel selection method. Its basic idea is that the probability of an individual being selected is proportional to its fitness value. The higher the fitness value, the greater the probability of the individual being selected.

[0122] Specifically, the process is as follows:

[0123] Calculate the total fitness: First, calculate the total fitness of all chromosomes. Assuming there are n chromosomes, their fitness values ​​are Fitness1, Fitness2, ..., Fitnessn, then the total fitness SumFitness = Fitness1 + Fitness2 + ... + Fitnessn.

[0124] Calculate the selection probability: For each chromosome, calculate its probability of being selected. The selection probability of chromosome i is Pi = Fitnessi / SumFitness.

[0125] Selection operation: You can use the roulette wheel method to select. For example, generate a random number r between 0 and 1, and then accumulate the selection probability starting from the first chromosome until the accumulated value is greater than r, at which point the corresponding chromosome is selected. Repeat this process until a sufficient number of individuals are selected for reproduction.

[0126] For example:

[0127] Assume that there are 5 chromosomes, and their fitness values ​​are 80, 60, 70, 90, and 50 respectively. Calculate the sum of fitness: SumFitness=80+60+70+90+50=350. Calculate the selection probability: The selection probability of chromosome 1 is P1=80 / 350≈0.229. The selection probability of chromosome 2 is P2=60 / 350≈0.171. The selection probability of chromosome 3 is P3=70 / 350=0.2. The selection probability of chromosome 4 is P4=90 / 350≈0.257. The selection probability of chromosome 5 is P5=50 / 350≈0.143.

[0128] The selection operation is performed as follows: Assume that the first random number generated is r1=0.4. Starting from the first chromosome, the selection probability is accumulated, P1=0.229, which is less than r1; continue to accumulate P1+P2=0.229+0.171=0.4, and the accumulated value is greater than r1, so chromosome 2 is selected. Assume that the second random number generated is r2=0.7. Continue to accumulate the selection probability starting from chromosome 3, P3=0.2, which is less than r2; P3+P4=0.2+0.257=0.457, which is less than r2; P3+P4+P5=0.2+0.257+0.143=0.6, which is less than r2; P1+P2+P3+P4+P5=1, which is greater than r2, so chromosome 5 is selected.

[0129] Step SF00, perform a crossover operation on the selected individuals to form new individuals, and perform a mutation operation on the chromosomes with a preset probability.

[0130] In this step, crossover operations on selected individuals can increase the diversity of the population and help explore a wider solution space. Mutation operations on chromosomes with a preset probability can introduce new gene combinations and prevent the algorithm from falling into a local optimal solution.

[0131] Specifically, the settings are as follows:

[0132] 1. Crossover operation: Select two parent chromosomes and generate two daughter chromosomes according to a certain crossover method. Common crossover methods include single-point crossover and two-point crossover. For example, for single-point crossover, randomly select a crossover point position, truncate the two parent chromosomes at the crossover point, and then exchange the truncated parts to form two daughter chromosomes. Assume that parent chromosome 1 is "angle coding 1-height coding 1-support site coding 1-disease type coding 1", and parent chromosome 2 is "angle coding 2-height coding 2-support site coding 2-disease type coding 2", and randomly select the crossover point at the third gene position. After crossover, daughter chromosome 1 is "angle coding 1-height coding 1-support site coding 2-disease type coding 1", and daughter chromosome 2 is "angle coding 2-height coding 2-support site coding 1-disease type coding 2".

[0133] 2. Mutation operation: mutate a gene of a chromosome with a preset probability. The mutation method can be to randomly change the value of the gene. For example, for the gene of the bed angle, if the mutation probability is 0.1, then in each generation, each chromosome has a 0.1 probability of mutating the angle gene. Assuming that the coding range of the angle gene is 0-90, a new value can be randomly generated within this range to replace the original angle value during mutation.

[0134] For example, suppose there are the following four parent chromosomes:

[0135] Chromosome 1: "30 (angle coding) -70 (height coding) -10 (support site coding) -1 (disease type coding)"; Chromosome 2: "40 (angle coding) -80 (height coding) -11 (support site coding) -2 (disease type coding)"; Chromosome 3: "25 (angle coding) -65 (height coding) -01 (support site coding) -1 (disease type coding)"; Chromosome 4: "35 (angle coding) -75 (height coding) -10 (support site coding) -2 (disease type coding)"; The crossover operation is as follows: Randomly select chromosome 1 and chromosome 2 for single-point crossover, assuming that the crossover point is at the second gene position. After crossover, the offspring chromosome 1 is "30 (angle coding) -80 (height coding) -10 (support site coding) -1 (disease type coding)", and the offspring chromosome 2 is "40 (angle coding) -70 (height coding) -11 (support site coding) -2 (disease type coding)".

[0136] The mutation operation is as follows: Assuming the mutation probability is 0.05, all offspring chromosomes are judged for mutation. If offspring chromosome 1 is selected for mutation, and it is assumed that the angle gene mutation is a random new angle value in the range of 0-90, such as 32. Then the offspring chromosome 1 after mutation is "32 (angle encoding) -80 (height encoding) -10 (support site encoding) -1 (disease type encoding)".

[0137] Step SG00, continuously iterate the chromosome algorithm. When the algorithm converges or reaches the preset stop condition, the iteration stops and the chromosome with a higher fitness value is obtained, that is, the optimal combination of bed angle, height, support position and disease, as the final bed adjustment plan.

[0138] The continuous iteration process is as follows: starting from the initial population, execute step SE00 (selecting individuals for reproduction using the fitness proportional selection method) and step SF00 (performing crossover and mutation operations on the selected individuals) in sequence to continuously generate new populations. In each generation of the population, calculate the fitness value of each chromosome, and evaluate the quality of the current population based on the fitness value. Repeat the above process until the convergence condition is met or the preset stop condition is reached.

[0139] The convergence condition can be that the fitness value of the population changes very little in several consecutive generations, or reaches a preset threshold. For example, if the average fitness value of the population changes less than a certain value for five consecutive generations, the algorithm is considered to have converged.

[0140] The stopping condition may be reaching a preset maximum number of iterations. For example, the maximum number of iterations is set to 100, and when the number of iterations reaches 100, the iteration is stopped regardless of whether the algorithm converges or not.

[0141] Assume that the initial population has 10 chromosomes, representing different bed angles, heights, support sites, and disease combinations.

[0142] First generation iteration: Calculate the fitness value of each chromosome. Use the fitness proportional selection method to select individuals for reproduction. Perform crossover and mutation operations on the selected individuals to generate a new population. Second and subsequent iterations: Repeat the above process, calculating fitness values, selection, crossover and mutation in each generation. Observe the changes in the fitness value of the population. If the average fitness value of the population changes by less than 5% for three consecutive generations, the algorithm is considered to have converged and the iteration is stopped. Or if the number of iterations reaches the preset maximum number of 80 times, the iteration is also stopped.

[0143] Assume that the algorithm converges at the 60th iteration. At this time, select the chromosome with the highest fitness value from the population. Assume that the bed angle represented by this chromosome is 45 degrees, the height is 80 cm, the support part is a specific combination, and the various types of diseases of the patient are taken into account. The bed adjustment plan represented by this chromosome is the final optimal plan.

[0144] Taking the preset target and the various specific disease types involved as the target, and according to the preset target weight allocation method, the fitness function is constructed to include:

[0145] The preset goals are to reduce patient discomfort and reduce the risk of pressure ulcers. The common goals for multiple diseases include maintaining stable body functions and promoting recovery. The quantitative value of the patient's discomfort is defined as , the goal is to minimize patient discomfort, i.e. The smaller the better, the risk index of pressure ulcers is defined as , the goal is to make The smaller the value, the better. The body function stability index is defined as , the goal is to make Stay within the preset range and define the recovery index as , the goal is to make Continuously improve;

[0146] According to the initial weights determined by the preset different goals, the weight of alleviating patient discomfort is set to , the weight for reducing the risk of pressure ulcers is , the weight to maintain stable body function is , the weight for promoting recovery is ;

[0147] The fitness function is expressed as the weighted sum of multiple objective functions, as follows:

[0148] ;

[0149] in, is a function related to the goal of alleviating patient discomfort, is a function related to the goal of reducing pressure ulcer risk, is a function related to the goal of maintaining stable body functions. is a function related to promoting rehabilitation goals and is defined according to rehabilitation indicators. is the fitness value.

[0150] Specifically, the reduction of patient discomfort is assumed to be measured by the patient's pain rating, which is divided into 1-10 levels, with 1 indicating the most comfortable and 10 indicating the most uncomfortable. It can be defined as 10-pain rating, that is, the lower the pain rating, The higher the value of .

[0151] To reduce the risk of pressure sores, a pressure sore risk assessment model is used to calculate the risk value based on factors such as the patient's bed rest time and skin condition. The risk value is between 0 and 1. It can be defined as 1-pressure ulcer risk value, the lower the risk, The higher the value of .

[0152] To maintain the stability of body functions, it is assumed that by monitoring the patient's blood pressure, heart rate and other physiological indicators, if the indicators are within the normal range, then The value is 1, otherwise the value is 0.5.

[0153] For promoting recovery, patients are rated based on their disease recovery, with ratings ranging from 1 to 5, with 1 indicating the worst recovery and 5 indicating the best recovery. Can be defined as rehabilitation rating , that is, the higher the rating, The higher the value of .

[0154] Reference Figure 3 , using natural language processing algorithms to analyze the patient's description of feelings and extract key information, and using the naive Bayes algorithm to classify the patient's description of feelings and determine the specific types of problems, including:

[0155] Step S610, obtaining the expression form of the patient's feeling description information.

[0156] The information is obtained from the following sources: the patient's feeling description information can be obtained through the interactive device equipped with the smart bed. For example, if the bed is equipped with a voice input device (such as a microphone) and a text input device (such as a touch screen or keyboard), the system can automatically detect how the patient inputs the feeling description. When the patient speaks through the voice input device, the system can determine that the expression is voice information; when the patient enters text through the text input device, the system can determine that the expression is text information.

[0157] The method for determining the patient's feeling description information is as follows: the expression form can be determined by detecting the type of input signal. For example, if the system receives an audio signal, it is determined to be voice information; if it receives a text signal, it is determined to be text information. Different input options can also be set on the interactive interface to allow the patient to clearly select whether to input voice or text, thereby determining the expression form.

[0158] Step S620: If the patient's feeling description information is in the form of voice information, adaptive voice recognition technology is used to perform voice recognition to obtain the patient's voice description signal, and the voice description signal is pre-processed.

[0159] Adaptive speech recognition technology: This technology usually includes acoustic models, language models, and decoders. The acoustic model is used to convert speech signals into acoustic features, the language model is used to predict the vocabulary and grammatical structure in the language, and the decoder searches based on the output of the acoustic model and language model to find the most likely text sequence.

[0160] The preprocessing of speech description signal is as follows: Denoising: A filter can be used to remove background noise. For example, the Wiener filter is used to process the speech signal to reduce the interference of environmental noise. Assuming that the original speech signal contains a lot of background noise, after being processed by the Wiener filter, the noise is effectively suppressed and the signal-to-noise ratio of the signal is improved. Endpoint detection: Determine the starting point and end point of the speech signal and remove the silent part. For example, use features such as short-time energy and zero-crossing rate for endpoint detection. If a signal is detected to have low short-time energy and low zero-crossing rate, it can be judged as a silent part and removed. Pre-emphasis: Enhance the high-frequency part of the speech signal to improve the resolution of the signal. For example, the speech signal is pre-emphasized through a first-order high-pass filter. Assuming that the high-frequency part of the original speech signal is relatively weak, after pre-emphasis, the high-frequency part is enhanced, which is conducive to subsequent feature extraction.

[0161] Step S630, converting the preprocessed speech description signal into a spectrogram, and using a convolutional neural network to extract key features of the spectrogram, the key features including the frequency distribution, time domain features, and harmonic structure of the speech.

[0162] The process of converting the preprocessed speech description signal into a spectrogram is as follows: the speech description signal in the time domain can be converted into a spectrogram in the frequency domain using methods such as the Fast Fourier Transform (FFT). FFT can convert the signal from the time domain to the frequency domain and display the components of the signal at different frequencies. For example, for a speech description signal of duration T, it is divided into several small time periods, and each time period is subjected to an FFT transform to obtain the corresponding spectrogram. These spectrograms can form a three-dimensional spectrum matrix, in which two dimensions represent frequency and time, and the third dimension represents the energy intensity of the signal.

[0163] The process of using convolutional neural networks to extract key features from spectrograms is as follows: Convolutional neural networks are composed of multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layer extracts local features by convolving the convolution kernel with the input data; the pooling layer is used to reduce the resolution of the feature map and reduce the amount of calculation; the fully connected layer integrates the extracted features and outputs the final feature vector. For spectrograms, convolutional neural networks can learn feature patterns in different frequency ranges and changing features in the time domain. For example, the convolutional layer can learn the energy distribution characteristics of speech within a specific frequency range, the pooling layer can extract important frequency and time regions, and the fully connected layer integrates these features into a feature vector of fixed length.

[0164] Frequency distribution characteristics: reflect the energy distribution of the speech signal at different frequencies. For example, some types of problems may be related to increased or decreased energy in a specific frequency range.

[0165] Time domain features: including the duration of the speech signal, amplitude changes, etc. For example, rapid speech may be related to urgent question types.

[0166] Harmonic structure features: The harmonic components in speech signals can provide information about the timbre and quality of the sound. Different types of problems may have different harmonic structure features.

[0167] Step S640, matching the key features extracted by the convolutional neural network with the question type features in the preset speech feature library, using the naive Bayes algorithm to calculate the probability of each question type, and determining the specific question type based on the probability, wherein the speech feature library stores a mapping relationship between known question types and specific feature expressions.

[0168] The feature matching process is as follows: The preset speech feature library stores the mapping relationship between known question types and specific feature expressions. These features can be obtained by analyzing and summarizing a large number of speech samples. For the key feature vectors extracted by the convolutional neural network, they are compared and matched with the features of each question type in the speech feature library. Distance metrics, such as Euclidean distance, cosine similarity, etc., can be used to measure the similarity between features. For example, if the distance metric between the extracted feature vector and the feature of a certain question type is less than a preset threshold, the two are considered to have a high similarity.

[0169] The Naive Bayes algorithm calculates the probability as follows: The Naive Bayes algorithm is based on Bayes’ theorem and the conditional independence assumption of features, and is used to calculate the probability of each question type given the features.

[0170] Suppose there are N question types in the speech feature library, namely T1, T2,..., TN. For an extracted feature vector X, according to the Naive Bayes algorithm, calculate P(Ti|X), that is, the probability that the question type is Ti under the condition that the feature vector X appears.

[0171] According to Bayes' theorem, P(Ti|X) = P(X|Ti) * P(Ti) / P(X). Among them, P(X|Ti) represents the probability of the feature vector X appearing when the question type is Ti, P(Ti) represents the prior probability of the question type Ti, and P(X) is a normalization constant.

[0172] In actual calculation, since P(X) is the same for all question types, it can be ignored. It is usually assumed that the features are conditionally independent, that is, P(X|Ti) = P(x1|Ti) * P(x2|Ti) *... * P(xm|Ti), where x1, x2,..., xm are the elements of the feature vector X.

[0173] Determine the question type as follows: After calculating the probability of each question type, determine the specific question type according to the probability size. Usually, the question type with the largest probability is selected as the final classification result.

[0174] For example, if it is calculated that P(T1|X) = 0.3, P(T2|X) = 0.4, P(T3|X) = 0.3, then the probability of the question type T2 is the largest, and it is determined as the specific question type corresponding to the speech description information.

[0175] Step S650, if the expression form of the patient's feeling description information is text information, preprocess the text description information provided by the patient and perform word segmentation.

[0176] The preprocessing of the text description information includes the following: 1. Remove special characters: Delete special characters such as punctuation marks and emojis in the text. For example, process "I feel very painful!" as "I feel very painful". 2. Unify case: Convert all letters in the text to uppercase or lowercase. For example, convert "Ifeelpain." to "ifeelpain.". 3. Remove stop words: Stop words are some common words that are not very helpful for text classification, such as "of", "is", "in", etc. By removing stop words, the computational complexity of subsequent processing can be reduced.

[0177] Step S660, extract the key features of the text information, including lexical features, syntactic features, and semantic features.

[0178] The extraction of vocabulary features includes the following: Word frequency statistics: Count the frequency of each word in the text. High-frequency words may play an important role in determining the type of question. 2. Keyword extraction: Use keyword extraction algorithms, such as the TF-IDF algorithm, to extract key words in the text. These keywords can highlight the core content of the text. For example, for "My legs are numb and I have difficulty moving", keywords such as "legs", "numbness", and "impossible movement" may be extracted.

[0179] The extraction of grammatical features includes the following: 1. Part-of-speech tagging: Tag each word in the text to determine whether it is a noun, verb, adjective, etc. The combination of words with different parts of speech can reflect the grammatical structure of the text. This combination of parts of speech may be related to a specific type of question. 2. Sentence structure analysis: Analyze the sentence structure of the text, such as subject-verb-object structure, attributive-adverbial-complement structure, etc. The sentence structure can provide information about the subject, action, and status of the question.

[0180] The extraction of semantic features includes the following: 1. Semantic similarity calculation: Calculate the semantic similarity between words in the text. You can use word vector models, such as Word2Vec or GloVe, to convert words into vector representations, and then calculate the similarity between vectors. Semantic similarity can help discover implicit relationships in texts. For example, "pain" and "uncomfortable" are semantically similar, and they may both be related to the patient's discomfort. 2. Topic model: Use topic models, such as LDA, to extract the topic of the text. Topics can reflect the main content of the text and potential problem types. For example, for a series of texts describing pain problems, the topic model may extract topics such as "pain management" and "physical discomfort".

[0181] Step S670, matching the extracted text information features with question type features in a preset text feature library, wherein the text feature library stores a mapping relationship between known question types and specific text feature expressions.

[0182] The text feature library is constructed as follows: First, collect a large number of text description samples of known problem types. For example, collect patient descriptions of various problem types such as pain problems, breathing problems, psychological problems, etc. Then, analyze these samples and extract typical text feature expressions for each problem type. These features can include specific vocabulary, phrases, grammatical structures, and semantic patterns. Finally, these problem types and corresponding text feature expressions are stored in the text feature library to form a mapping relationship.

[0183] The process of feature matching is as follows: for the key features of the extracted patient text description information, compare them one by one with the features of each question type in the text feature library. The similarity calculation method can be used to measure the degree of matching between features. For example, the overlap of vocabulary features, the similarity of grammatical features, and the correlation of semantic features can be calculated. If the feature of a question type has a high degree of matching with the feature of the patient's text description information, it is considered that the question type is more relevant to the patient's description.

[0184] Step S680, using the naive Bayes algorithm to calculate the probability of each question type, and determining the specific question type according to the probability.

[0185] Reference Figure 4 , match the key features extracted by the convolutional neural network with the question type features in the preset voice feature library, use the naive Bayes algorithm to calculate the probability of each question type, and determine the specific question type based on the probability, including:

[0186] Step S641, extracting prosodic features from the speech description signal, where the prosodic features include stress, rhythm, and duration.

[0187] Accent extraction: Accent refers to the part of speech that is pronounced more prominently and emphasized. The position of the accent can be determined by analyzing the energy distribution, frequency change and other characteristics of the speech signal. For example, the short-time energy analysis method can be used to calculate the energy value of the speech signal in different time periods. If the energy of a certain time period is significantly higher than that of the surrounding time periods, then this time period may contain an accent. The energy distribution of the speech signal at different frequencies can also be observed through spectrum analysis. If the energy of certain frequencies is significantly enhanced in a specific time period, it may also indicate the presence of an accent.

[0188] Rhythm extraction: Rhythm refers to the time intervals and repetitive patterns in speech signals. Rhythm can be determined by analyzing the duration, pauses and other features of speech signals. For example, the time intervals between adjacent syllables in a speech signal can be calculated and the regularity of these time intervals can be analyzed. If the time intervals are relatively stable, the rhythm is relatively regular; if the time intervals vary greatly, the rhythm is relatively irregular. The changing points of the rhythm can also be determined by detecting the position and duration of the pauses in the speech signal.

[0189] Duration extraction: Duration refers to the duration of each syllable, word, or sentence in a speech signal. Duration can be determined by measuring the time length of a speech signal. For example, the duration of each syllable can be calculated, or the duration of an entire sentence can be calculated. Duration can reflect information such as the speaker's speaking speed and emphasis.

[0190] Step S642, analyzing whether the speech description signal has specific prosodic features or specific keywords. If not, executing step S643; if yes, executing step S644.

[0191] The specific rhythmic features are determined as follows:

[0192] Accent judgment: Accents can be determined by comparing the energy intensity of different parts of the speech signal. If the energy of a part is significantly higher than the surrounding parts, it can be considered that there is a strong accent. For example, an energy threshold is set, and when the energy of a syllable exceeds the threshold, it is judged that there is an accent.

[0193] Rhythm judgment: Analyze the time interval and repetition pattern of the speech signal. If the rhythm is fast, that is, the time interval between adjacent syllables is short, the rhythm can be considered strong. For example, the average time interval between adjacent syllables is calculated. If it is less than a preset value, the rhythm is judged to be strong.

[0194] Duration judgment: observe the duration of certain parts of the speech signal. If the pronunciation duration of a word or phrase is significantly shortened or prolonged, it may have a specific prosodic meaning. For example, when the pronunciation duration of a keyword is less than a certain percentage of the normal duration, it can be considered that there is a specific prosodic feature.

[0195] The determination of specific keywords is as follows:

[0196] Build a keyword library: Collect keywords related to various types of problems in advance. For example, for pain problems, you can include "pain", "pain", "uncomfortable", etc.; for breathing problems, you can include "difficulty breathing", "shortness of breath", etc.

[0197] Speech recognition and matching: Perform speech recognition on the speech description signal and match the recognized text with the keyword library. You can use string matching algorithms such as exact matching and fuzzy matching. If the recognized text contains a keyword in the keyword library, it is determined that the specific keyword exists.

[0198] Step S643, maintaining the original setting;

[0199] Step S644, according to the prosodic feature quantitative evaluation method, the overall effective evaluation value of the prosodic feature is calculated, and the evaluation value of the specific keyword is analyzed and determined based on the number of occurrences of the specific keyword and the evaluation value of the specific keyword. In addition, the key features extracted by the convolutional neural network are matched with the question type features in the preset speech feature library, and the probability of each question type is calculated using the naive Bayes algorithm. According to the corresponding relationship between the probability interval into which the probability of each question falls and the evaluation score, the deterministic evaluation value of the deep learning classification result is analyzed and determined.

[0200] Calculation of the overall effective evaluation value of rhythmic features:

[0201] Quantitative analysis of stress and rhythm intensity: For stress, the intensity of the stress can be determined by analyzing the energy distribution and frequency changes of the speech signal. For example, the ratio of the energy of the stress part to the average energy of the entire speech signal can be calculated as a quantitative indicator of the stress intensity. If the ratio is greater than a preset threshold, the stress intensity is considered to be high. For rhythm, the standard deviation of the time interval between adjacent syllables in the speech signal can be calculated. The larger the standard deviation, the stronger the rhythm change. Different rhythm intensity levels can be divided according to the size of the standard deviation and assigned corresponding quantitative evaluation values.

[0202] Duration quantification analysis: Analyze the duration of each syllable, word or sentence in the speech signal. Indicators such as average duration, maximum duration and minimum duration can be calculated. For example, if the pronunciation duration of a word is significantly longer or shorter than the average duration, it can be considered that the duration of the word is special and a corresponding quantitative evaluation value is given.

[0203] Comprehensive calculation: Based on the quantitative evaluation values ​​of the intensity of stress and rhythm and the quantitative evaluation value of duration, the overall effective evaluation value of the rhythmic features is calculated by weighted average or other methods. For example, the weight of the intensity of stress and rhythm can be set to 0.6, and the weight of duration can be set to 0.4, and the weighted sum of the quantitative evaluation values ​​of the two can be used to obtain the overall effective evaluation value.

[0204] Determination of specific keyword evaluation value: Count the number of times a specific keyword appears in the speech description signal. The more times it appears, the greater the possibility that the keyword represents the type of question.

[0205] Set a rating range for each specific keyword. For example, common keywords can be assigned a lower rating, while keywords related to emergency situations can be assigned a higher rating. Determine the rating of a specific keyword based on the number of times the keyword appears and its rating range. For example, if a keyword appears twice, and the rating range for the keyword is 1-5, then a specific rating can be determined within this range based on the number of times it appears, such as a rating of 3 for two occurrences.

[0206] Determination of the deterministic evaluation value of deep learning classification results: After calculating the probability of each question type using the naive Bayes algorithm, divide the probability into different intervals. For example, it can be divided into a high probability interval (0.8-1), a medium probability interval (0.5-0.8), and a low probability interval (0-0.5). Set an evaluation score for each probability interval. The evaluation score for the high probability interval is higher, and the evaluation score for the low probability interval is lower. Determine the deterministic evaluation value of the deep learning classification results based on the probability interval into which the calculated question type probability falls. For example, if the probability of a certain question type falls in the high probability interval, then the deterministic evaluation value of the question type is higher.

[0207] Calculation of the overall effective evaluation value of rhythmic features: Quantitative analysis of stress and rhythm intensity: By analyzing the speech signal, it is found that the stress intensity of the two parts "severe headache" and "difficult" is relatively high, and the ratio of the energy of the stress part to the average energy of the entire speech signal is 1.5 (the preset threshold is 1.2), indicating that the stress intensity is relatively high, and the quantitative evaluation value is given as 4 (the full score is 5). In terms of rhythm, the standard deviation of the time interval between adjacent syllables is 0.3 (the preset standard deviation is greater than 0.2, indicating a high rhythm intensity), and the quantitative evaluation value of the rhythm intensity is 3.

[0208] Duration quantification analysis: The average duration of the entire speech description is 5 seconds, among which the pronunciation duration of "I have a severe headache" is slightly longer than the average duration, and the duration quantification evaluation value is given as 2.

[0209] Comprehensive calculation: the weight of stress and rhythm intensity is 0.6, the weight of duration is 0.4, and the overall effective evaluation value of rhythmic features is (4*0.6+3*0.6+2*0.4) / (0.6+0.6+0.4)=3.4.

[0210] Determine the evaluation value of a specific keyword: Count the number of times a specific keyword appears. "Headache" appears once, and "Dyspnea" appears once. Assume that the evaluation value range of "Headache" is 1-3, and the evaluation value range of "Dyspnea" is 2-4. Since the two keywords appear once each, they can be assigned evaluation values ​​of 2 and 3 respectively.

[0211] The total evaluation value of a specific keyword is 2+3=5.

[0212] Determination of the deterministic evaluation value of deep learning classification results: Use the naive Bayes algorithm to calculate the probability of the problem type, assuming that the probability of pain problems is 0.7 and the probability of breathing problems is 0.6. Divide the probabilities into intervals, 0.5-0.8 is the medium probability interval, and the evaluation score is 4 (out of 5). The probabilities of pain problems and breathing problems both fall within the medium probability interval, and the deterministic evaluation value of the deep learning classification results is 4.

[0213] Step S645, calculate the urgency score based on the evaluation value of the specific keyword determined by analysis, the overall effective evaluation value of the rhythmic feature, the certainty evaluation value of the deep learning classification result, and the preset weight distribution ratio.

[0214] Determine the weight distribution ratio: The weight distribution ratio can be determined based on experience, expert opinion or data analysis. For example, a large number of speech samples can be analyzed to observe the impact of different factors on the degree of urgency, so as to determine the reasonable weight. Assume that the weight of a specific keyword is w1, the weight of the prosodic feature is w2, and the weight of the certainty of the deep learning classification result is w3, and w1+w2+w3=1.

[0215] Calculate the urgency score: First, obtain the evaluation value of the specific keyword (denoted as V_keyword), the overall effective evaluation value of the rhythmic feature (denoted as V_rhythm), and the deterministic evaluation value of the deep learning classification result (denoted as V_classification). Then, calculate the urgency score (denoted as S_emergency) by weighted summation according to the weight distribution ratio, that is, S_emergency=w1*V_keyword+w2*V_rhythm+w3*V_classification.

[0216] Assume that after the previous steps, the following evaluation values ​​are obtained: the evaluation value of a specific keyword V_keyword=4 (assuming that the specific keyword appears frequently and is of high importance). The overall effective evaluation value of the rhythmic feature V_rhythm=3 (the rhythmic feature shows a certain urgency). The evaluation value of the certainty of the deep learning classification result V_classification=2 (the certainty of the classification result is general).

[0217] Set the weight distribution ratio to w1=0.4, w2=0.3, w3=0.3. Then the urgency score S_emergency=0.4*4+0.3*3+0.3*2=1.6+0.9+0.6=3.1. The urgency of the voice description can be determined by the calculated urgency score. If the preset urgency score threshold is 3, then this score exceeds the threshold, and step S646 will be executed to recalculate the question type probability and determine the specific question type. If the score is less than or equal to the threshold, the original setting is maintained.

[0218] Step S646: If the urgency score exceeds the preset urgency score, a new problem type probability is recalculated according to a preset problem type probability weight distribution scheme, and a specific problem type is determined according to the new problem probability.

[0219] Preset question type probability weight distribution scheme: This scheme can be designed according to different emergency situations and question type characteristics. For example, the weight of features related to emergency situations in the question type probability calculation can be increased, and the weight of less relevant features can be reduced. Assume that in the original question type probability calculation, the weights of specific keywords, prosodic features, and the certainty of deep learning classification results are w1, w2, and w3, respectively. In an emergency, these weights can be adjusted to w1', w2', and w3', so that the weights of factors that are more relevant to the emergency are increased. For example, if a specific keyword can more accurately indicate the question type in an emergency, the weight of the specific keyword can be increased to w1'>w1.

[0220] Recalculate the probability of question types: Use the adjusted weights and apply the naive Bayes algorithm again to calculate the probability of each question type. Recalculate the posterior probability of each question type based on the new weights and existing data (such as the number of occurrences of specific keywords, the evaluation value of rhythmic features, the evaluation value of the certainty of deep learning classification results, etc.). Assuming that the question types are T1, T2, ..., Tn, for question type Ti, the new probability calculation formula is P(Ti|X)=P(X|Ti)*P(Ti) / P(X), where X represents the extracted feature set, P(X|Ti) is calculated based on the new weights and features, P(Ti) can be determined based on prior knowledge or historical data, and P(X) is a normalization constant that can be ignored in the calculation.

[0221] Determine the specific problem type: Based on the recalculated problem type probabilities, select the problem type with the highest probability as the final specific problem type. If the new probability distribution is significantly different from the previous one, the specific problem type may be changed to better suit the emergency situation.

[0222] Step S647: if the urgency score is less than or equal to the preset urgency score, the original setting is maintained.

[0223] Among them, calculating the overall effective evaluation value of the rhythmic feature includes: calculating the quantitative evaluation value of the stress and rhythm intensity according to the preset stress and rhythm intensity quantitative analysis method, and calculating the quantitative evaluation value of the duration according to the preset duration quantitative analysis method, and calculating the overall effective evaluation value of the rhythmic feature based on the quantitative evaluation value of the stress and rhythm intensity and the quantitative evaluation value of the duration.

[0224] Recalculating the probability of getting new question types includes:

[0225] Step S646.1, obtaining the difference between the urgency score and the preset urgency score.

[0226] The urgency score is calculated in step S645, which combines the evaluation value of the specific keyword, the overall effective evaluation value of the prosodic feature, and the deterministic evaluation value of the deep learning classification result, and is calculated according to a preset weight distribution ratio. For example, the urgency score can be a score between 0 and 10, and a higher score indicates a higher urgency.

[0227] The preset urgency score is a preset threshold used to determine whether the urgency of the voice description requires recalculation of the problem type probability. This preset value can be determined based on actual application scenarios and requirements. For example, a score value that can better distinguish between emergency situations and non-emergency situations can be determined by analyzing a large amount of historical data.

[0228] The difference is calculated by subtracting the preset urgency score from the urgency score to obtain the difference between the two.

[0229] Step S646.2: Analyze and determine the dynamic adjustment factor value based on the corresponding relationship between the difference value and the dynamic adjustment factor value.

[0230] First, it is necessary to determine the dynamic adjustment factor values ​​corresponding to different difference ranges. This correspondence can be determined through experience, data analysis or experiments.

[0231] For example, the difference value may be divided into several intervals, each interval corresponding to a specific dynamic adjustment factor value. A smaller difference value may correspond to a smaller adjustment factor, and a larger difference value may correspond to a larger adjustment factor.

[0232] Assume that the difference is between 0 and 1, the dynamic adjustment factor is 1.1; the difference is between 1 and 2, the dynamic adjustment factor is 1.2; the difference is between 2 and 3, the dynamic adjustment factor is 1.3, and so on.

[0233] Once the corresponding relationship between the difference and the dynamic adjustment factor value is determined, the corresponding dynamic adjustment factor value can be determined according to the difference calculated in step S646.1.

[0234] If the difference falls within a certain interval, the dynamic adjustment factor value corresponding to the interval is selected.

[0235] Assume that the difference between the urgency score calculated in step S646.1 and the preset urgency score is 1.8.

[0236] Establish a corresponding relationship between the difference value and the dynamic adjustment factor value: As mentioned above, assuming that the difference value is between 1 and 2, the dynamic adjustment factor is 1.2.

[0237] Analyze and determine the dynamic adjustment factor value: Since the difference 1.8 falls within the range of 1 to 2, the dynamic adjustment factor value is determined to be 1.2. Next, in step S646.3, the dynamic adjustment factor value is multiplied by the recalculated problem type probability to determine the new problem type probability. In this way, the problem type probability can be appropriately adjusted according to the different urgency levels, thereby improving the accuracy of determining the correct problem type in emergency situations.

[0238] Step S646.3, according to the preset question type probability weight distribution scheme, recalculate and obtain a new question type probability, and re-determine a new question type probability based on the product of the obtained new question type probability and the dynamic adjustment factor value.

[0239] Assuming that under normal circumstances, the weight of a specific keyword is 0.3, the weight of a prosodic feature is 0.3, and the weight of the certainty of a deep learning classification result is 0.4, after steps S646.1 and S646.2, it is determined that the dynamic adjustment factor value is 1.2.

[0240] Recalculate the probability of new problem types: Assume there are three types of problems: pain problems, breathing problems, and psychological problems. The original calculated probabilities are P(pain problem)=0.4, P(breathing problem)=0.3, P(psychological problem)=0.3. Recalculate based on the new weights and features, and get the new probabilities of P(pain problem)=0.45, P(breathing problem)=0.35, P(psychological problem)=0.2.

[0241] Multiplying with the dynamic adjustment factor value: the adjusted probability is P'(pain problem)=0.45*1.2=0.54, P'(breathing problem)=0.35*1.2=0.42, P'(psychological problem)=0.2*1.2=0.24.

[0242] Redetermine the new problem type probability: Comparing the adjusted probabilities, P'(pain problem) is the largest, so in an emergency, determine the specific problem type as a pain problem.

[0243] According to the corresponding relationship between the problem and the adjustment strategy, analyzing and determining the adjustment strategy and implementing the corresponding strategy include the following steps:

[0244] Step S710, analyzing whether the urgency score exceeds a preset urgency score. If yes, execute step S720; if no, execute step S730.

[0245] Step S720, analyzing and determining the adjustment strategy and executing the corresponding strategy according to the condition exceeding the preset urgency score, the problem and the adjustment strategy.

[0246] Determine the specific situation of exceeding the preset urgency score: First, analyze the degree to which the urgency score exceeds the preset score. The degree of excess can be divided into different levels, such as slight excess, moderate excess, and large excess. Consider other relevant factors, such as the type of specific keywords, the specific performance of prosodic features, and the certainty evaluation value of deep learning classification results. These factors can provide more detailed information about the emergency situation.

[0247] Establish a correspondence between problems and adjustment strategies: Predetermine a series of adjustment strategies for different types of problems and urgency. For example, if it is a pain problem and the urgency is high, consider notifying medical staff immediately and adjusting the comfort settings of the bed to relieve pain. If it is a breathing problem and the urgency is high, you can start emergency oxygen supply equipment, adjust the angle of the bed to help the patient breathe, etc. If it is a psychological problem and the urgency is high, you can arrange for a psychologist to conduct remote consultation, play soothing music, etc.

[0248] Analyze and determine the adjustment strategy and execute it: Based on the specific situation that exceeds the preset urgency score and the correspondence between the problem and the adjustment strategy, a comprehensive analysis is performed to determine the most appropriate adjustment strategy. Once the adjustment strategy is determined, the corresponding measures are immediately implemented. This may include sending notifications to medical staff, automatically adjusting equipment settings, starting specific treatment programs, etc.

[0249] Step S730, maintaining the original settings.

[0250] An intelligent anti-pressure sore bed comprehensive management and control system also includes steps in parallel with analyzing and determining adjustment strategies and executing corresponding strategies, as follows:

[0251] Step Sa00, obtaining the expression form of the patient's feeling description information.

[0252] Step Sb00, using the expression form of the patient's feeling description information as the expression method of the adjustment strategy, and sending the specific adjustment strategy to the terminal held by the person in charge.

[0253] If the patient's feeling description is in the form of voice information, the voice can be converted into text, and the converted text can be sent together with the adjustment strategy. For example, the message sent can be marked with "patient voice description: [converted text content], adjustment strategy: [specific strategy content]".

[0254] If the patient's feeling description information is expressed in text information, the text content can be directly sent together with the adjustment strategy. For example, "patient's text description: [text content], adjustment strategy: [specific strategy content]".

[0255] Determine the contact information of the terminal held by the responsible person, such as mobile phone number, email address or specific application notification channel.

[0256] Use appropriate communication methods to send the adjustment strategy to the terminal held by the person in charge. You can use SMS, email, application push notification, etc. to ensure that the person in charge can receive the information in time.

[0257] Based on the same inventive concept, an embodiment of the present invention provides a computer storage medium, including a computer program product that can be loaded and executed by a processor to implement the following Figures 1 to 4 Procedures for each step involved in the process.

[0258] The embodiments of this specific implementation method are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Therefore, all equivalent changes made based on the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. An intelligent comprehensive management and control system for beds that prevent pressure sores, characterized in that: include: Hospital beds; at least one processor; At least one memory storing a computer program, and at least one processor configured to implement a comprehensive management and control method for an intelligent anti-pressure sore bed when executing the computer program, the specific steps are as follows: Obtaining information related to the patient's physical characteristics and medical conditions, where the physical characteristics include height, weight, and body shape, and the medical conditions include specific disease types and the severity of the disease; The patient's physical characteristics and condition-related information are used as input variables for fuzzy processing, and then reasoning is performed through the preset fuzzy rule base to obtain the fuzzy value of the output variable. Finally, the fuzzification is performed to obtain a specific bed adjustment plan, where the output variables are the bed angle, height and support position. Implement specific bed adjustment plans; Analyze whether the patient's feeling description information is received within the preset time range; If not, maintain the original setting; If yes, the patient's description of feelings is analyzed using a natural language processing algorithm, and key information is extracted. The patient's description of feelings is classified using a naive Bayes algorithm to determine the specific type of problem. According to the corresponding relationship between the problem and the adjustment strategy, analyze and determine the adjustment strategy and implement the corresponding strategy; The method also includes steps after obtaining the patient's physical characteristic-related information and the disease-related information and before using the patient's physical characteristic-related information and the disease-related information as input variables for fuzzy processing, which are as follows: Analyze whether the patient's condition involves multiple specific disease types; If not, maintain the original setting; If yes, the combination of the bed angle, height and support position is encoded as a chromosome of the genetic algorithm, and the multiple specific disease types involved in the patient's condition are defined as additional genes and added to the chromosome; Taking the preset target and the multiple specific disease types involved as targets, and constructing a fitness function according to the preset target weight allocation method; The preset fitness ratio selection method is used to select individuals for reproduction according to the fitness value of the chromosome; Perform crossover operation on selected individuals to form new individuals, and perform mutation operation on chromosomes with preset probability; The chromosome algorithm is continuously iterated. When the algorithm converges or reaches the preset stop condition, the iteration is stopped and the chromosome with a higher fitness value is obtained, that is, the optimal combination of bed angle, height, support position and disease, as the final bed adjustment plan; Taking the preset target and the various specific disease types involved as the target, and according to the preset target weight allocation method, the fitness function is constructed to include: The preset goals are to reduce patient discomfort and reduce the risk of pressure ulcers. The common goals for multiple diseases include maintaining stable body functions and promoting recovery. The quantitative value of the patient's discomfort is defined as , the goal is to minimize patient discomfort, i.e. The smaller the better, the risk index of pressure ulcers is defined as , the goal is to make The smaller the value, the better. The body function stability index is defined as , the goal is to make Stay within the preset range and define the recovery index as , the goal is to make Continuously improve; According to the initial weights determined by the preset different goals, the weight of alleviating patient discomfort is set to , the weight for reducing the risk of pressure ulcers is , the weight to maintain stable body function is , the weight for promoting recovery is ; The fitness function is expressed as the weighted sum of multiple objective functions, as follows: ； in, is a function related to the goal of alleviating patient discomfort, is a function related to the goal of reducing pressure ulcer risk, is a function related to the goal of maintaining stable body functions. is a function related to promoting rehabilitation goals and is defined according to rehabilitation indicators. is the fitness value.

2. According to claim 1, the intelligent anti-pressure sore bed comprehensive management and control system is characterized in that: The natural language processing algorithm is used to analyze the patient's description of feelings and extract key information. The naive Bayes algorithm is used to classify the patient's description of feelings and determine the specific types of problems, including: The expression form of obtaining information describing the patient's feelings; If the patient's feeling description information is expressed in the form of voice information, adaptive voice recognition technology is used to obtain the patient's voice description signal, and the voice description signal is preprocessed; The preprocessed speech description signal is converted into a spectrogram, and a convolutional neural network is used to extract key features of the spectrogram, including the frequency distribution, time domain characteristics, and harmonic structure of the speech. Match the key features extracted by the convolutional neural network with the question type features in the preset speech feature library, use the naive Bayes algorithm to calculate the probability of each question type, and determine the specific question type according to the probability. The speech feature library stores the mapping relationship between known question types and specific feature expressions. If the patient's feeling description information is expressed in text information, the text description information provided by the patient is pre-processed and word segmented; Extract key features of text information, including lexical features, grammatical features and semantic features; Matching the extracted text information features with the question type features in a preset text feature library, which stores the mapping relationship between known question types and specific text feature expressions; The naive Bayes algorithm is used to calculate the probability of each question type, and the specific question type is determined based on the probability.

3. According to claim 2, the intelligent anti-pressure sore bed comprehensive management and control system is characterized in that: The key features extracted by the convolutional neural network are matched with the question type features in the preset voice feature library, and the probability of each question type is calculated using the naive Bayes algorithm. The specific question types are determined based on the probability, including: Extracting prosodic features from speech description signals, including stress, rhythm, and duration; Analyze the speech description signal for specific prosodic features or specific keywords; If not, maintain the original setting; If yes, then according to the prosodic feature quantitative evaluation method, the overall effective evaluation value of the prosodic feature is calculated, and according to the number of occurrences of the specific keyword and the evaluation value of the specific keyword, the evaluation value of the specific keyword is analyzed and determined. In addition, the key features extracted by the convolutional neural network are matched with the question type features in the preset speech feature library, and the probability of each question type is calculated using the naive Bayes algorithm. According to the corresponding relationship between the probability interval into which the probability of each question falls and the evaluation score, the deterministic evaluation value of the deep learning classification result is analyzed and determined; The urgency score is calculated based on the evaluation value of the specific keyword determined by the analysis, the overall effective evaluation value of the prosodic feature, the certainty evaluation value of the deep learning classification result, and the preset weight distribution ratio; If the urgency score exceeds the preset urgency score, the new problem type probability is recalculated according to the preset problem type probability weight distribution scheme, and the specific problem type is determined according to the new problem probability; If the urgency score is less than or equal to the preset urgency score, the original setting is maintained.

4. According to claim 3, the intelligent anti-pressure sore bed comprehensive management and control system is characterized in that: The overall effective evaluation value of the calculated prosodic features includes: According to the preset stress and rhythm intensity quantitative analysis method, the stress and rhythm intensity quantitative evaluation value is calculated. At the same time, according to the preset duration quantitative analysis method, the duration quantitative evaluation value is calculated. Based on the stress and rhythm intensity quantitative evaluation value and the duration quantitative evaluation value, the overall effective evaluation value of the rhythmic feature is calculated.

5. According to claim 3, the intelligent anti-pressure sore bed comprehensive management and control system is characterized in that: Recalculating the probability of getting new question types includes: Obtain the difference between the urgency score and the preset urgency score; According to the corresponding relationship between the difference and the dynamic adjustment factor value, the dynamic adjustment factor value is analyzed and determined; According to the preset question type probability weight distribution scheme, the new question type probability is recalculated and obtained, and the new question type probability is re-determined based on the product of the obtained new question type probability and the dynamic adjustment factor value.

6. An intelligent anti-pressure sore bed comprehensive management and control system according to any one of claims 3 to 5, characterized in that: According to the corresponding relationship between the problem and the adjustment strategy, analyze and determine the adjustment strategy and implement the corresponding strategy, including: Analyze whether the urgency score exceeds a preset urgency score; If yes, then according to the correspondence between the conditions and problems exceeding the preset urgency score and the adjustment strategy, analyze and determine the adjustment strategy and execute the corresponding strategy; If not, keep the original setting.

7. The intelligent comprehensive management and control system for preventing bedsores according to claim 6 is characterized in that: It also includes steps that run in parallel with analyzing and determining adjustment strategies and executing the corresponding strategies, as follows: The expression form of obtaining information describing the patient's feelings; The expression form of the patient's feeling description information is used as the expression method of the adjustment strategy, and the specific adjustment strategy is sent to the terminal held by the person in charge.

8. A computer storage medium, characterized in that: It includes a program that can be loaded and executed by a processor to implement a comprehensive control method for an anti-pressure sore bed in an intelligent anti-pressure sore bed comprehensive control system as described in any one of claims 1 to 7.

Images