WO2022222197A1

WO2022222197A1 - Multi-mode personalized monitoring method

Info

Publication number: WO2022222197A1
Application number: PCT/CN2021/092129
Authority: WO
Inventors: 丁贤根; 丁远彤; 肖苑辉
Original assignee: 港湾之星健康生物(深圳)有限公司
Priority date: 2021-04-20
Filing date: 2021-05-07
Publication date: 2022-10-27
Also published as: CN113317782B; CN113317782A

Abstract

A multi-mode personalized monitoring method, comprising: decomposing a monitoring value into a variable monitoring component and a fixed monitoring component; introducing an extended component to monitor the monitoring value and background noise; calculating the variable component, the fixed component, and the extended component by means of a correlation function and a personalized feature set; monitoring a conventional true value by means of historical and current monitoring data; and executing an object algorithm and a differential object algorithm to obtain an object correction value having an error between the correction value and the conventional true value of an object less than an allowable error. In addition, by means of historical monitoring data of a single object and a plurality of objects, the object algorithm and a group algorithm are employed to optimize the personalized feature set. Monitoring data of other objects are employed for further calculation to obtain a group correction value of the object. Modes such as a cloud network big data mode, a local area network mode, and a single point mode are further comprised, and the dilemma brought by measuring all individuals by means of one standard in the prior art is solved.

Description

Methods of Multimodal Personalized Monitoring

technical field

The invention relates to the field of industrial technology and the field of biomedicine, in particular to the measurement and monitoring of medical data and industrial data, in particular to a monitoring method for data obtained by indirect, multi-modal measurement that cannot be directly measured.

Background technique

The measurement of physical and chemical quantities includes direct measurement and indirect measurement. Direct measurement can be accomplished by improving the precision and accuracy of measuring physical and chemical quantities themselves. Indirect measurement is limited to the individual attributes of the person being measured. In many cases, it is difficult to directly measure its physical and chemical quantities. Sometimes even if indirect physical and chemical quantities are measured, it is difficult to monitor and convert them into sufficient precision and accuracy. The error meets the required physical and chemical quantity data.

For example, for the in vitro non-invasive measurement and monitoring of glucose content in human venous blood, for the measurement and monitoring of physical and chemical quantities during optimal energy-saving control of how the cargo hold fans of ocean-going container cargo ships meet the conditions of working with refrigerated containers in the cabin, wind turbines are optimal The measurement and monitoring of the physical and chemical quantities of the wind field in the control is an extremely difficult subject under the current technical means.

For indirect measurement, the inventor also proposes a "multi-mode personalized calibration method", which attempts to measure a plurality of individualized, low-difficulty monitoring factor values associated with the measurement monitoring value by decomposing the measurement monitoring value, or multiple values related to the measurement monitoring value. The monitoring value is related to the monitoring factor value and the measurement monitoring value with low accuracy and difficulty. Using such multi-mode personalized monitoring, the measurement data is obtained, and then the correction of the measurement monitoring value is obtained indirectly by monitoring these data. Value - a reliable measurement result.

How to obtain the credible measurement result of the correction value of the measurement monitoring value based on the indirect measurement data, the inventor found that the following key steps need to be completed:

1. Introduce personalized monitoring

1.1. Indirect measurement often contains a lot of interference factors, which must be deducted, otherwise it is impossible to reduce errors and obtain reliable measurement results.

1.2. The interference factors of object personalization existing in each object are often very different, and must be calculated and deducted one by one according to the personalization of the object.

1.3. Even for the same object, the environment-specific interference factors are often different due to the different environments it is in. Therefore, these environment-based personalized interferences must be deducted.

2. Introduce multi-mode monitoring

In the same measurement object, the physical and chemical quantities (monitoring values) that need to be measured are often related to other physical and chemical quantities (monitoring factor values), and there may be a relatively obvious functional relationship. At this time, introducing sensors of various modes to measure the corresponding physical and chemical quantities, designing corresponding calculation methods, and incorporating the respective measured data into the monitoring calculation will bring obvious effects to the monitoring values.

3. Introduce object algorithm and group algorithm

The object algorithm proposed by the present invention is to monitor the individual of the measurement object according to its historical data. The proposed group algorithm is to monitor each other horizontally for several objects with the same measurement properties. This method has obvious advantages for the monitoring of monitoring values that are related to the environment and to the subdivision classification of objects.

According to the inventor's practice experience, it is found that the current industry status of the measurement industry is mainly based on direct measurement, indirect measurement is relatively rare, and the method of individualization, multi-mode and monitoring has not been found yet. In particular, measurement and monitoring for the following specific applications, the current status is:

1. Current status of human blood glucose monitoring

The proposed human blood glucose monitoring, according to the definition of the International Health Organization, is to monitor the millimolar concentration per liter (mmol/L) of glucose content in human veins. The commonly used methods include:

Invasive single-point method: draw venous blood for glucose testing

Minimally invasive single-point method: puncture the finger to take the capillary blood of the finger and monitor the glucose with a test strip

Minimally invasive continuous method: continuous monitoring of glucose by inserting an indwelling enzyme electrode probe into the arm

Non-invasive continuous methods: electrophoresis is used to measure glucose in tissue fluid on the skin, infrared method to measure glucose through skin, microwave method to measure subcutaneous glucose, and contact lens with microcircuit to measure glucose in tears, etc. However, as of the time of this patent application, no product authorized by the medical authority has been found.

Single-point measurement cannot solve the problem of monitoring blood sugar fluctuations. For patients with type 1 diabetes, because the rapid drop in blood sugar in a short time cannot be alarmed, it will bring the patient's life in danger. Furthermore, for patients with type 2 diabetes, a single point measurement cannot address the optimal management and treatment of diabetes. Both invasive monitoring and minimally invasive monitoring bring pain and inconvenience to patients.

2. Status of monitoring and monitoring of energy conservation in cargo hold

There is no fan energy-saving system found in the existing cargo hold. However, for the current ocean-going container ships, the energy consumption of the fan in the cargo hold is often hundreds of kilowatts. Converted to fuel consumption or LNG natural gas consumption, it is as high as hundreds of thousands of dollars to millions of dollars, and due to high fuel consumption and high gas consumption, sulfur pollution and carbon pollution are also brought about. Therefore, the market needs energy-saving solutions.

Existing methods are insufficient

The inventor believes that the existing monitoring algorithms have the following shortcomings:

1. Direct measurement cannot be achieved in some application scenarios, and indirect measurement is required.

2. From the perspective of measurement principle, the measurement data itself has no information that can monitor itself, and cannot monitor itself.

3. The indirect single-point measurement data that is difficult to measure directly contains interference factors, which cannot be eliminated by traditional methods.

4. In addition to the common content, the interference factors also include some personalized content, and it is impossible to use the same standard to eliminate these personalized interference.

5. Although some measurements cannot be accurately realized, multi-mode monitoring variables can be introduced to increase the measurement accuracy, but no such method has been found so far.

6. Under the premise of big data, there are some commonalities among the objects in the group, and mutual monitoring can be realized through these commonalities, but no such monitoring method has been found so far.

The above-mentioned deficiencies in the existing monitoring methods exist in the actual system, and proper indirect measurement and monitoring methods are needed to solve these problems.

Content and purpose of the invention

Through long-term observation, experiment and research, the inventor proposes a method for multimodal personalized monitoring. The purpose and intent of the present invention are:

1. For measurement items that cannot be directly and accurately measured or are difficult to directly and accurately measure, indirect multi-mode auxiliary monitoring measurement data is introduced to realize monitoring.

2. For continuous real-time measurement, the present invention designs an object algorithm algorithm to carry out personalized monitoring according to historical records and agreed truth values.

3. For a plurality of measurement objects, the present invention proposes a group algorithm algorithm to monitor itself according to the data of other objects.

4. Through the artificial intelligence deep learning algorithm, for the objects that join the group later, the classification monitoring is realized, the collection of the agreed truth value is reduced, and the monitoring is simplified.

Special statement:

1. The agreed truth value referred to in the present invention is not limited to being obtained by measuring equipment with a higher precision, but can also be obtained by artificial intelligence and deep learning.

2. The monitoring functions referred to in the present invention are not limited to these functions and formulas listed in the present application, including other functions and formulas designed by mid-level designers in the industry based on this idea.

3. The step numbers of the present invention, unless otherwise specified, do not exist in the order of the numbers.

The scope of application of the present invention may include monitoring of measurement data and monitoring of other data.

The present invention emphasizes that the types of monitoring values, the types of monitoring factor values, and the division of objects and groups listed in the present invention are all derived from this idea. Due to space limitations and the basic spirit of the invention, the application of the present invention cannot list the types and associations of these data information one by one. The types of information proposed in the invention application are not meant to limit the idea of the present invention.

Beneficial Effects of Invention

1. Provide an innovative method based on multi-mode personalized monitoring, which can realize the invention and its purpose, and solve the shortcomings of the existing monitoring technology.

2. Provide specific methods based on mathematics, statistics, artificial intelligence and deep learning to quickly and efficiently monitor measurement data and realize personalized big data resource sharing.

3. For in vitro continuous monitoring similar to human glucose, it can effectively monitor and improve the processing accuracy.

4. For the energy saving of cargo cabin fans similar to ocean-going ships, energy saving can be achieved through multi-mode measurement, object algorithm and group algorithm.

Description of drawings

Figure 1 is a schematic diagram of multi-mode personalized monitoring;

Figure 2 Schematic diagram of blood glucose monitoring;

Figure 3 monitoring function diagram;

Figure 4 Raman spectrum of mixed substances;

Fig. 5 Raman spectrum of glucose;

Figure 6. The energy-saving monitoring method of the cargo cabin fan;

Figure 7 Cargo compartment reefer layout.

Detailed ways

The purpose and intention of the present invention is to adopt the technical scheme of the following embodiment to realize:

Embodiment 1. External monitoring method of human body glucose data

One of the application embodiments of the present invention is an artificial intelligence-oriented personalized management method for diabetes, which is a typical application example of the present invention. In this embodiment, only the description of the method of the present invention is involved, and it is not regarded as a complete design of an actual system, nor is it a limitation of the present invention.

1. Illustration description

Figure 1 Schematic diagram of multi-mode personalized monitoring

It is extremely difficult to monitor blood glucose levels from outside the human body. According to the World Health Organization (English name: World Health Organization, abbreviated WHO, Chinese abbreviation WHO) statistics, the world's diabetes patients account for about 11.4% of the total population, and the glucose content in human veins (referred to as blood sugar) is the gold for the diagnosis of diabetes. standard. At present, conventional blood glucose monitoring is divided into invasive, minimally invasive and non-invasive methods, among which invasive methods include drawing venous blood in the hospital and using biochemical enzyme methods to monitor blood glucose levels. Most of the minimally invasive methods are monitored by sticking a finger to squeeze out the blood with a test strip and inserting a wireless sensor on the arm. The non-invasive method is limited to the development of monitoring technology because it does not penetrate the skin at all, and no more popular cases have been found so far. In addition, most of the above methods are single-point measurement, which cannot realize continuous blood glucose monitoring.

In Fig. 1, each monitored person is regarded as each object, and in each monitor, the monitored components are monitored by a laser Raman analyzer and an infrared light analyzer, and a PPG/ECG analyzer (PPG: English full name Photoplethysmography, referred to as PPG. ECG: English full name of Electrocardiography, referred to as ECG) as the detection extension component. Taking the individualized interference caused by the biological diversity of the human body as background noise, a series of algorithms in the patent application of the present invention are used to eliminate the background noise, thereby obtaining accurate in vitro blood glucose monitoring values. Among them, as the application of the patent of the present invention in the early stage, the method of venous blood blood test is required to obtain the agreed true value in the early stage, which is used as calibration. According to the historical and current monitoring data, the object algorithm proposed by the present invention is used to calculate Subject-corrected value, that is, blood glucose monitoring results. In addition, the artificial intelligence big data analysis of many objects in the group (that is, the monitors) can finally eliminate the step of venous blood drawing for the monitors and relieve the pain of the monitors.

Figure 2 Schematic diagram of blood glucose monitoring

In Figure 2, 2001 is a laser Raman spectrometer, which is used to irradiate human skin through infrared laser (the wavelengths of 785nm, 830nm, 1063nm, etc. are selected here), and transmit it to a depth of about 1mm to produce Raman effect, according to The characteristic value and amplitude of the Raman spectral shift of glucose, monitor the glucose content, and finally calculate the glucose content in the venous blood by using a series of algorithms of the present invention, that is, the blood glucose value determined by the World Health Organization. 2002 is an infrared light analyzer, which acts as an auxiliary blood glucose monitoring device to monitor another blood glucose monitoring component. 2003 is PPG/ECG analyzer (PPG: English full name Photoplethysmography, referred to as PPG. ECG: English full name Electrocardiography, referred to as ECG), monitoring blood pressure and pulse information, as an extended component of the present invention, participating in a series of algorithms of the present invention , which further enables the precision and accuracy of blood glucose monitoring. 2004 is the real-time unit of the multi-mode personalized monitoring method of the present invention. 2005 is a venous blood analyzer, which is one of the source channels of the agreed truth value of the present invention. 2006 is the monitoring channel for the agreed truth value. It should be noted here that it is not a real-time online monitoring, but is used in the big data accumulation stage of the present invention. Once the big data accumulation is completed, this step can be omitted.

Select in vitro monitoring of blood sugar as the monitoring value, and select the sensor for measuring blood sugar, including the monitoring data of the laser Raman spectrum analyzer as the monitoring component. According to needs, as a preference, the monitoring data of the infrared light analyzer can also be added as the second monitoring. weight. Among them, since it is extremely difficult to monitor blood glucose outside the human body, in order to complete the accurate monitoring, the present invention divides the monitoring components into a fixed component and a variable component. Among them, the fixed component is the inherent part of the blood sugar content contained in the human skin and its subcutaneous tissue, which does not depend on the amount that changes with the blood sugar fluctuation of the human body, and the variable component is sensitive to the fluctuation of the glucose content in the human vein blood vessels. amount of change. As a preferred option, in order to further correlate the blood sugar fluctuations caused by changes in the blood flow rate of the human body, the present invention also introduces a cardiovascular blood pressure and pulse PPG/ECG analyzer to monitor the real-time blood pressure and pulse of the human body, and use this as an extension component.

In addition, as a calibration and sensor calibration measure, a more accurate medical version of the venous blood test method is used to obtain the agreed true value of blood sugar. The monitoring data of the present invention adopts the cloud computing mode, and the blood glucose monitoring data of each monitoring object is stored in the cloud server, forming a group big data mode.

In this embodiment, the human body glucose value is detected based on Raman scattering spectrum, that is, the monitoring value described in the present invention. Since in the circuit of the Raman laser, all glucose will be detected, including at least the content of all glucose in the epidermis, intradermal, interstitial fluid, capillaries, venous blood vessels, etc., and even other The glucose at the unknown site, therefore, the detected glucose value is the sum of these glucose contents. According to the WHO definition, only glucose in the veins is needed. To this end, we design to use the changes in blood pressure and blood lipids as monitoring factor components, and according to the fluctuations in blood pressure and blood lipids, separate the glucose component in the venous blood vessels from the total glucose value, which is proposed by the present invention. Longitudinal multi-mode monitoring - monitor the monitoring value based on the historical data and real-time data of the object to make it a correction value. In addition, according to the statistics of the big data of the group, for the specific situations of different skin color, race, work nature, etc., the present invention also proposes horizontal multi-modal monitoring - monitoring the monitoring of one's own objects based on other historical statistical data in the group value to make it a correction value.

Figure 3 Monitoring function diagram

According to the signal system analysis method of the state equation, this embodiment is shown in FIG. 3 . Among them, after the analysis of Figure 1, PDV is a variable component, which comes from the laser Raman analyzer and the infrared light analyzer in Figure 2, and mainly includes the change value of blood sugar. PDF is a fixed component, also from laser Raman analyzer and infrared light analyzer, mainly including background noise and basic blood sugar content. PDE is the extended component, which comes from a PCG/ECG analyzer. MIF is the subject correction value, that is, the blood glucose value of the monitoring result. DTM ₁ (t), DTM ₂ (t) to DTM _m (t) are the analyzed blood sugar components, including epidermal blood sugar, intradermal blood sugar, interstitial fluid blood sugar, capillary blood sugar, venous blood sugar, etc. .

Figure 4 Raman Spectrum of Mixed Matter

FIG. 4 is a spectrogram obtained by Raman scattering monitoring of mixed substances, that is, the synthesis of Raman waves of all substances detected at the Raman laser detection point. Among them, 4010 is the horizontal axis, which represents the displacement wave of Raman scattering, and 4020 is the vertical axis, which represents the intensity of the Raman wave. , 4001 is the first peak with a displacement of 1003cm ^-1 ^, 4002 is the second peak with a displacement of 1125cm ^-1 , 4003 is the third peak with a displacement of 1450cm-1, these three peaks constitute the characteristics of glucose "fingerprint". Other waveforms are characteristic of mixed non-glucose substances.

The method of the present invention needs to separate the signal quantity of glucose from the Raman scattering wave of the mixed substance in FIG. 4 .

Fig.5 Raman shift spectrum of glucose

FIG. 5 is the Raman shift spectrum of pure glucose, which is also the signal quantity of glucose that needs to be separated from FIG. 4 in the present invention. Among them, 5010 is the horizontal axis, representing the displacement wave of Raman scattering, 5020 is the vertical axis, representing the intensity of the Raman wave, the curve in the figure is the component of the glucose Raman displacement wave, 5001 is the epidermal blood glucose value DTM ₁ (t), 5002 is the intradermal blood glucose value DTM ₂ (t), 5003 is the interstitial fluid blood glucose value DTM ₃ (t), 5004 is the capillary blood glucose value DTM ₄ (t), and 5005 is the venous blood glucose value DTM ₅ (t). According to this classification, the final object correction value MIF(t) = DTM ₅ (t).

2. Basic program steps

2.1: General description

Methods of multimodal personalized monitoring, including:

The monitoring value of the decomposition object is one or more monitoring components, the monitoring components are monitored, and the monitoring components are decomposed including one or more variable components, zero or more fixed components, and zero or more extended components that exist a monitoring function with the monitoring values.

The object algorithm is performed according to the historical and current monitoring components for the object to obtain an object correction value whose agreed true value error with the object is less than the allowable error.

Establishing a group consisting of more than one object with the same monitoring value, using big data to train a personalized feature set, and revising a group algorithm for own objects according to the personalized feature set and other objects, to obtain a group correction value.

In vitro monitoring of blood glucose is selected as the monitoring value, and the sensor for measuring blood glucose includes the monitoring data of the laser Raman spectrum analyzer as the monitoring component. According to needs, as a preference, the monitoring data of the infrared light analyzer can also be added as the monitoring component. Among them, since it is extremely difficult to monitor blood glucose outside the human body, in order to complete the accurate monitoring, the present invention divides the monitoring components into a fixed component and a variable component. Among them, the fixed component is the inherent part of the blood sugar content contained in the human skin and its subcutaneous tissue, which does not depend on the amount that changes with the blood sugar fluctuation of the human body, and the variable component is sensitive to the fluctuation of the glucose content in the human vein blood vessels. amount of change. As a preferred option, in order to further correlate the blood sugar fluctuations caused by changes in the blood flow rate of the human body, the present invention also introduces a cardiovascular blood pressure and pulse PPG/ECG analyzer to monitor the real-time blood pressure and pulse of the human body, and use this as an extension component.

2.2: Decomposing monitoring values

On the basis of the foregoing technical solutions, the present invention includes, but is not limited to, the steps of decomposing the monitoring value, and specifically, the following or various local improvement measures can be adopted:

In step S2010, the variable component is a monitoring component of the monitoring value that changes with environmental changes or a monitoring component specified by a user, and the variable component is obtained by monitoring with a sensor.

In step S2020, the fixed component is a type of the monitoring component that can be separated in the monitoring value, and within a sampling period equal to the variable component, the rate of change of the fixed component is the same as that of the variable component. The ratio of the rate of change of the components is small, for the optimal selection of the biological health category, the ratio of the rate of change is less than 0.20, and for the optimal selection of the industrial category, the ratio of the rate of change is less than 0.10, and, under this condition, the fixed component Depending on the user selection, monitoring is performed using sensors or statistical predictions.

In this embodiment, the other choice of the fixed component is to use the background noise as the fixed component, for example, the background noise of a laser Raman analyzer and the background noise of an infrared light analyzer. Another option for the fixed component is to use the minimum value of the monitored component as the fixed component. The selection of the change rate ratio range depends on the actual characteristics of the system. In some systems with high monitoring accuracy, the change rate ratio range can be smaller, for example, 0.01.

In step S2030, the extended component is obtained by using sensor monitoring, and the monitoring function of the extended component and the monitoring value exists in a monitoring function including:

The change of the monitoring value affects the extended component in one direction, that is, the extended component is a function argument of the monitoring value. In this case, the extended component is used as an auxiliary monitoring of the monitoring value.

In another embodiment, the change of the blood glucose value locally affects the change of the blood oxygen content. In this case, the blood glucose function becomes one of the independent variables of the blood oxygen function.

The change of the monitoring value affects the extended component in both directions, that is, the extended component and the monitoring value are mutually independent variables, and at this time, the extended component is used as the auxiliary monitoring and adjustment of the monitoring value.

In this embodiment, the laser Raman shift amplitude and the change of the infrared wavelength are functions of mutual influence, which can be regarded as independent variables that are functions of each other.

The change of the extended component affects the monitoring value in one direction, that is, the monitoring value is a function argument of the extended component, and at this time, the extended component is used as an auxiliary adjustment of the monitoring value.

For example, when a PPG/ECG analyzer is used as extended component sensing, since according to the definition of the International Health Organization, the blood glucose value is the mmol/L of the glucose content in the blood of human venous blood vessels, so the change of the pulse is important to the instantaneous blood glucose level. The value is influential. In this embodiment, the PPG/ECG pulse function is used as an independent variable of the blood glucose function, so as to more accurately monitor the blood glucose value.

In step S2040, the monitoring value and the monitoring component are continuously monitored according to different continuous time series and specific moments, and the intermediate data and result data are stored in the information base.

In this embodiment, the cloud center will continuously collect various useful data of the object and store it in the database.

2.3: Monitoring functions

On the basis of the foregoing technical solutions, the present invention includes, but is not limited to, the processing steps of the monitoring function. Specifically, the following or various local improvement measures can be adopted:

In step S3010, a monitoring function between the monitoring value and the object is established according to formula (3.1), and the monitoring value is decomposed according to formula (3.2) to monitor components, including the variable component, the fixed component and the extended component component, the monitoring function of the monitoring value and the component is established according to formula (3.3), the variable component set is established according to formula (3.4), the fixed component set is established according to formula (3.5), and the extended component set is established according to formula (3.6).

MI = f _3.1 (PD) (3.1)

PD=PDV∪PDF∪PDE (3.2)

MI=f _3.3 (PDV, PDF, PDE) (3.3)

PDV={PDV _β | PDV _β number β variable component data, 1≤β≤n} (3.4)

PDF = {PDF _γ | PDF _γ fixed component data of number γ, 0≤γ≤p} (3.5)

Wherein, f _3.1 is the monitoring function of the monitoring value and the object, f _3.3 is the monitoring function of the monitoring value and the component, MI is the monitoring value or monitoring value data set, PD is the object or Object data set, PDV is the variable component or variable number component set, PDF is the fixed component or fixed component set, PDE is the extended component or the set of extended components, PFV _β is the set of all the components numbered β. the variable component data or an element of the variable component data set, n is the total number of the variable components, PDV _γ is the fixed component data numbered γ or an element of the fixed component data set, where γ=0 Indicates that the PDF is an empty set, which does not contain the fixed components, p is the total number of the fixed components,

is numbered as

The extended component data of or an element of the extended component data set, where

Indicates that the PDE is an empty set, which means that the fixed component is not included, φ is the total number of the extended components, and the decomposition includes frequency domain decomposition from the perspective of monitoring convenience, and time domain decomposition from the perspective of continuous time.

In this embodiment, f _3.3 is selected to be the glucose monitoring function of the 2001 laser Raman analyzer and the glucose monitoring function of the 2002 infrared light analyzer in FIG. 2, respectively, and the PPG/ECG analyzer monitoring function in 2003 is used as the extended component , in 2004 to do the decomposition of formula (3.4) to formula (3.6).

Since laser Raman analyzers, infrared light analyzers, and PPG/ECG analyzers all use digital sampling to monitor data, the subscripts of their variables identify the sampling time, the same below.

Step S3020, monitor the variable component according to the functional relationship between the variable component determined by the formula (3.7) and the output signal of the sensor, and establish a function set of the variable component according to the formula (3.8), Equation (3.9) is the output signal function of the sensor:

PDV _β = f _3.7β (SS _β ) (3.7)

F _3.8 = {f _3.7β | f _3.7β Variable component data of number β, 1≤β≤n} (3.8)

SS _β = f _3.9β (S _β ) (3.9)

in:

f _3.7β is the function of the variable component numbered β, F _3.8 is the function set of the variable component, f _3.9β is the sensor function, n is the total number of functions of the variable component, n, β All belong to natural numbers, and 1≤β≤n.

PDV _β is the variable component numbered β, SS _β is the signal output by the sensor numbered β, and S _β is the sensor numbered β.

Step S3030, monitor the fixed component according to the functional relationship between the fixed component determined by the formula (3.10) and the output signal of the sensor or the statistical prediction, and establish the function of the fixed component according to the formula (3.11). Set, formula (3.12) is the output signal function of the sensor or the statistical prediction:

PDF _γ = f _3.10 γ (SS _γ ) (3.10)

F _3.11 = {f _3.10γ | f _3.10γ Fixed component data of number β, 0≤γ≤p} (3.11)

SS _γ =f _3.12γ (s _γ ) (3.12)

in:

f _3.10γ is the function of the fixed component numbered γ, F _3.11 is the function set of the fixed component, f _3.12γ is the output signal function of the sensor or the statistical prediction, p is the fixed component The total number of functions, p and γ are both natural numbers, and 0≤γ≤p, p=0 does not include fixed components.

PDF _γ is the fixed component numbered γ, SS _γ is the signal output by the sensor numbered γ or the statistical prediction, S _γ is the sensor numbered γ or the statistical prediction.

Step S3040, monitor the extended component according to the functional relationship between the extended component determined by the formula (3.13) and the output signal of the sensor, and establish a function set of the extended component according to the formula (3.14), the formula (3.15 ) is the output signal function of the sensor:

in:

is numbered as

The function of the extended component, F _3.14 is the function set of the extended component,

is the output signal function of the sensor, n is the total number of functions of the extended component,

φ are all natural numbers,

φ=0 means that the extension component is not included.

is numbered as

The extended component of ,

for the signal output by the sensor numbered β,

is numbered as

of the sensor.

2.4: Background noise

On the basis of the foregoing technical solutions, the present invention includes, but is not limited to, the processing steps of background noise. Specifically, the following or various local improvement measures can be adopted:

Step S4010, for the object with the background noise, calculate according to the following steps:

According to formula (4.1), the monitoring function with background noise is established,

According to formula (4.2), the background noise is set as the fixed component,

The extended component function with background noise is established according to formula (4.3),

The sensor function of the variable component is established according to equation (4.4),

Setting formula (4.5) as a function of the background noise,

Calculate the monitoring value according to formula (4.6),

MI=f _4.1 (PDV, PDF, PDE, PDN) (4.1)

PDF=f _4.2 (PDN) (4.2)

PDE = f _4.3 (SE, PDN) (4.3)

PDV=f _4.4 (SV) (4.4)

PDN=f _4.5 (t) (4.5)

MI = f _4.6 (SV, SE, t) (4.6)

Wherein, f _4.1 is the monitoring function including background noise, f _4.2 is the function with the background noise as the fixed component, f _4.3 is the extended component function including background noise, and f _4.4 is the variable component sensor function, f _4.5 is the function of the background noise, f _4.6 is the monitoring value function, PDN is the background noise, SE is the sensor output signal of the extended component, SV is the sensor output of the variable component signal, t is the time series.

In this embodiment, for the glucose monitoring of the laser Raman analyzer, the background noise is selected as the electrical background noise of the CCD array and the glucose signal unique to the object of in vitro glucose monitoring, including epidermis, subcutaneous tissue, interstitial fluid, etc. Glucose in venous blood vessels as defined by the International Health Organization. For the monitoring of background noise, according to the present invention, the intermediate technicians in the industry can make a detailed combination according to their own understanding, and use a laser Raman analyzer or other monitoring equipment for combined monitoring.

Step S4020, according to the actual application, in the case of single-variable monitoring, and in the case that the range of the variable-component sensor satisfies the application, adopt one of the variable-component sensors, and in the case of the variable-component sensor In the case that the range does not satisfy the application, a plurality of the variable component sensors are used to extend the range.

For blood glucose monitoring, another preferred solution is to divide the blood glucose value based on 4.2 as the boundary. For example, if the blood glucose value is greater than 4.2, one section of monitoring is designed, and if the blood glucose value is less than or equal to 4.2, another section of monitoring is designed.

In step S4030, according to the actual application, in the case of multi-variable monitoring, each variable adopts one sensor of the variable component.

In this embodiment, the use of a laser Raman analyzer and an infrared light analyzer is such a design solution.

Step S4040, according to the actual application, if the output data of the variable component sensor satisfies the application, it is not necessary to use the extended component sensor, and in the case that the output data of the variable component sensor cannot meet the application Next, use more than one sensor of the extended component described above.

2.5: Object Algorithms

On the basis of the foregoing technical solutions, the present invention includes, but is not limited to, the object algorithm. Specifically, the following or multiple local improvement measures can be adopted:

In step S5010, standard or higher-level measurement accuracy monitoring equipment is used to monitor and obtain the monitoring value of the object as the agreed true value, record the monitoring time, and set the agreed true value The error is Δ, and Δ=0, and the allowable error is set as Δ _e .

Step S5020: Set an initial parameter set, and calculate the object correction value, and then calculate the agreed true value according to the step S5010, calculate the error, if the error falls within the allowable error, assign the initial parameter set to the personalized parameter set, if If the error is greater than the allowable error, the initial parameter set is modified to be the personalized parameter set, so that the error falls within the allowable error.

Step S5030, using the personalized parameter set, using the current monitoring component and algorithms including extended Kalman filter algorithm, Monte Carlo particle algorithm, modern Bayesian algorithm, and the personalized parameter set , calculate the object correction value.

In this example, T-test or Z-test was used to analyze errors to remove outliers. The support vector machine SVM and the convolutional neural network CNN algorithm are used to classify the historical values.

Step S5040, based on the setting, using the historical monitoring component, according to the set time interval, execute the step S5020 at the time interval point, and execute the step S5030 outside the time interval point to establish The personalized parameter set and the time series corresponding to the monitoring components are recorded in the information database, and a deep learning algorithm is used to calculate the monitoring components and the personalized parameters of the objects in the information database set and the time series, find out the personalized parameter set with high probability and demarcate the corresponding monitoring component as the optimization point, and the high probability includes the probability specified by the user or the probability is greater than 30%.

In step S5050, based on the personalized parameter set of the optimization point obtained in the step S5040, the object correction value is calculated by adopting a support vector machine algorithm and a convolutional neural network algorithm.

In this embodiment, a medical-grade blood glucose and blood lipid blood test measurement device is used to obtain the measured values of the individual's blood glucose and blood lipids and record the measurement time, which are used as the agreed true values. At the same time, that is, in the same period of time, the values and time series of blood glucose and blood lipid measured by the equipment to be calibrated are used to calculate the vertical synchronization correction value, and the error between the vertical synchronization correction value and the agreed true value is calculated and verified. If the error If it is greater than the allowable error, modify the parameter set and iteratively calculate until the error is less than the allowable error.

The agreed truth value can also be obtained by statistics of other individuals and artificial intelligence algorithms. At this time, it is not necessary to use higher-level medical equipment to measure and obtain.

In this embodiment, the calibration function can select algorithms such as mathematical statistics algorithm, support vector machine SVM, convolutional neural network CNN, etc., with the error smaller than the allowable error as the goal, for example, the convolution kernel or related parameters are trained, which are all summarized and personalized feature set.

2.6: Swarm Algorithms

On the basis of the foregoing technical solutions, the present invention includes, but is not limited to, the group algorithm. Specifically, the following or multiple locally improved measures can be adopted:

Step S6010, for all the objects in the group, calculate the optimization point and the object correction value, and record them in the information database.

Step S6020, according to the objects in the information database, for the optimization point and the object correction value, establish more than one object classification, and calibrate the object classification of the object, the individual classification of the object classification The ratio of the number to the total number of objects is greater than the number specified by the user or greater than 0.2, and is recorded in the information base.

Step S6030, according to the object classification, for the optimization point and the object correction value, according to the principle of minimum error, calculate the personalized feature set of the object classification, and record it in the information database.

Step S6040: Calculate the group correction value of the object according to the personalized feature set of the object classification.

Step S6050, for the self-object, calculate the object classification to which the calibration belongs, and calculate the object correction value and the group correction value of the object according to the personalized feature set of the object classification.

Step S6060, according to the personalized feature set of the object classification, for the newly added object, determine its object classification according to the calculation of the monitoring value of the object, and directly adopt the personalized feature set of the classification to include: said agreed truth value without performing the step of monitoring said agreed truth value.

The personalized feature set includes the parameters of the object algorithm, the parameters of the population algorithm, the classification, and the agreed truth value of the object.

In this embodiment, the use of the algorithm between groups can further improve the calibration efficiency and reduce the error. For the biological diversity of the human body, such as different skin colors, different ages, different occupations, etc., targeted calibration is performed. , in order to obtain better results.

Further, a group classification model is established, and a personalized feature set for classification is established. For individuals newly entering the group, calculation is performed to be included in the classification, and rapid calibration is performed with the personalized feature set corresponding to the classification.

2.7: Differential Object Algorithm

On the basis of the foregoing technical solutions, the present invention includes, but is not limited to, a differential object algorithm, specifically as follows:

Step S7010: Calculate the difference between the monitoring value of the object at the optimization point and the correction value of the object, and record the difference in the information database.

Step S7020, if the difference value of each of the optimization points is a constant difference value, for the object correction value of the subsequent time series, use the monitoring value to subtract the constant difference value to calculate the object correction value .

Step S7030, if the difference value of each optimization point is a function difference value, then for the object correction value of the subsequent time series, subtract the function difference value from the monitoring value to calculate the object correction value .

Step S7040, if the difference value of the optimization point of the time series is an alternating difference value between the constant difference value and the function difference value, subtract the alternating difference value from the monitoring value to calculate the object correction value.

In step S7050, for the Raman spectrum monitoring method, the object is scanned twice or more with excitation light with a slight frequency offset to generate two or more Raman spectra, and according to the Raman scattering characteristic shift of the object, the two or more The Raman scattering spectrum generated by scanning is calculated by differential convolution or linear regression to eliminate the interference of background noise caused by fluorescence, and the wavelength difference of the excitation light with the slight frequency offset is between 0.2 and 100 nm.

In this embodiment, for the laser Raman spectrum analyzer and the infrared spectrum analyzer, the differential object algorithm is used to further counteract the influence of background noise and improve the monitoring accuracy.

2.8: Personalization Feature Sets

On the basis of the foregoing technical solutions, the present invention includes, but is not limited to, the personalized feature set. Specifically, the following or various local improvement measures can be adopted:

Step S8010, set a constraint condition that the error of the corrected value of the object is less than the allowable error, establish the personalized feature set according to formula (8.1), and calculate the personalized feature set:

PF=f _8.1 (MI, MIF, Δ≤Δ _E ) (8.1)

Wherein, f _8.1 is the personalized feature set function, PF is the personalized feature set, MI is the monitoring value or the monitoring value data set, MIF is the object correction value or the object correction value set, Δ is the error, Δ _E is the allowable error.

Step S8020, decompose the personalized feature set:

Based on the specification, the personalization feature set is decomposed into more than one personalized feature set category, and the personalized feature set category is decomposed into one or more personalized feature set category components.

There is a functional relationship determined by formula (8.2) between the personalized feature set and the category of the personalized feature set:

PF=f _8.2 (PFT) (8.2)

Wherein, f _8.2 is the category function of the personalized feature set, PFT is the category of the personalized feature set, and the decomposition includes frequency domain decomposition from the perspective of convenient calculation, and also includes time domain decomposition from the perspective of continuous time.

Step S8030, monitoring the time domain value of the personalized feature set category:

There is a functional relationship between the personalized feature set category and the continuous time series determined by formula (8.3), and the time domain value of the personalized feature set category is monitored according to formula (8.3):

PFT=f _8.3 (t) (8.3)

Among them, f _8.3 is the personalized feature set category time domain function, and t is the continuous time series.

Step S8040, monitor the specific moment value of the personalized feature set category:

There is a functional relationship between the specific moment value of the personalized feature set category and the specific moment, and the specific moment value of the personalized feature set category is monitored according to the formula (8.4):

PFT _T = f _8.3 (t, t = T) (8.4)

Wherein, f _8.3 is the time domain function of the personalized feature set category, T is the specific moment, and PFT _T is the specific moment value of the personalized feature set category at the specific moment.

Step S8050, monitor the time domain value of the category component of the personalized feature set:

The personalized feature set category is decomposed into more than one personalized feature set category component according to formula (8.5), and the personalized feature set category component and the continuous time series have the function determined by formula (8.6) relationship, monitor the time domain value of the category component of the personalized feature set according to formula (8.5):

PFT=f _8.5 (PFT ₁ , PFT ₂ , . . . , PFT _q ) (8.5)

PFT _δ = f _8.6 (t, 1≤δ≤q) (8.6)

Wherein, f _8.5 is the category decomposition function of the personalized feature set, f _8.6 is the time domain function of the category component of the personalized feature set, PFT ₁ , PFT ₂ , ..., PFT _q are the category components of the personalized feature set, and q is the The total number of category components of the personalized feature set, δ is the category component number of the personalized feature set, q and δ are both natural numbers, and 1≤δ≤q, PFT _δ is the category component of the personalized feature set numbered δ The time domain value in the continuous time series.

Step S8060, monitor the specific moment value of the category component of the personalized feature set:

The specific time value of the category component of the personalized feature set at the specific time has a functional relationship determined by the formula (8.7) with the continuous time series, and the category component of the personalized feature set is monitored according to the formula (8.7). The specific moment value at a specific moment:

PFT _δT = f _8.6 (t, t=T, 1≤δ≤q) (8.7)

Wherein, PFT _δT is the specific time value of the category component of the personalized feature set at the specific time.

In this embodiment, the parameters and variables in all the algorithm formulas and the collected time series are taken as the content of the personalized parameter set, and are listed in the database using standardized work for subsequent algorithm use, especially the deep learning algorithm.

2.9: Personalized Feature Set Mathematical Model

On the basis of the foregoing technical solutions, the present invention includes, but is not limited to, the mathematical model of the personalized feature set. Specifically, the following or various local improvement measures can be adopted:

Step S9010, fuzzy optimization method:

According to the following mathematical models, including fuzzy equations, fuzzy partial differential equations, and set equations, the monitoring value and the monitoring component are optimized to obtain the optimal value of the object under the optimized situation, specifically including:

Step S9011, create a set:

A monitoring value set is established with the monitoring value elements, which is recorded as MIset.

A monitoring component set is established with the monitoring components as elements, denoted as MI _α set, where α is the number of the monitoring components.

Create an object set with the object as an element, denoted as PDset;

The monitoring value set is decomposed into a variable value set and a fixed value set.

The variable value set is decomposed into a variable component set, the variable value set is denoted as PDVset, the variable component set is denoted as PDV _β set, and β is the number of the variable component.

The fixed value set is decomposed into a fixed component set, the fixed value set is referred to as PDFset, the fixed component set is referred to as PDF _γ set, and γ is the number of the fixed component.

Decompose the extended value set as the extended component set, denote the extended value set as PDEset, and denote the extended value component set as

is the number of the extended component.

Taking the personalized feature set category as an element to establish a personalized feature set set, the component personalized feature set set is a personalized feature set component set, and the personalized feature set set is recorded as PFTset, and the personalized feature set component set is recorded as PFT _δ set, δ is the number of the individualized feature set component.

The sets include fuzzy sets and non-fuzzy sets.

In this embodiment, for the convenience of management, all sets adopt fuzzy sets.

Step S9012, create a cut set:

According to the mathematical model, the mapping relationship between the sets is established in turn, and the monitoring value set and the monitoring component set are used for sorting as the primary key to become an ordered set, and the first λ is taken after positive sorting from large to small. The elements are used as the ordered head cut set, and the first μ elements are taken as the ordered tail cut set after reverse sorting from small to large, as follows:

MIset _λ = (mi|mi monitoring value positive sequence number θ≤λ} (9.1)

MI _α set _λ = {mi _δ |mi _α monitoring component positive sequence number θ≤λ} (9.2)

PDset _λ = {pd|pd object positive sorting number θ≤λ} (9.3)

PDVset _λ = {pdv|pdv variable value positive sort number θ≤λ} (9.4)

PDV _β set _λ ={pdv _β |pdv _β variable component positive sorting number θ≤λ} (9.5)

PDFset _λ = {pdf|pdf fixed value positive sort number θ≤λ} (9.6)

PDF _γ set _λ = {pdf _γ |pdf _γ fixed component positive sorting number θ≤λ} (9.7)

PDEset _λ = {pde|pde extension value positive sort number θ≤λ} (9.8)

PDE _γ set _λ ={pde _γ |pde _γ extended component positive sorting number θ≤λ} (9.9)

PFTset _λ = {pft|pft personalized feature set positive sorting number θ≤λ} (9.10)

PFT _δ set _λ ={pft _δ |pft _δ Personalized feature set component positive sorting number θ≤λ} (9.11)

MIset _μ = {mi|mi monitoring value inverse order number η≤μ} (9.12)

MI _α set _μ = {mi _δ |mi _α monitoring component reverse order number η≤μ} (9.13)

PDset _μ = {pd|pd object inverse ordering number η≤μ} (9.14)

PDVset _μ = {pdv|pdv variable value reverse order number η≤μ} (9.15)

PDV _β set _μ ={pdv _β |pdv _β variable component inverse order number η≤μ} (9.16)

PDFset _μ = {pdf|pdf fixed value inverse sort number η≤μ} (9.17)

PDF _γ set _μ = {pdf _γ |pdf _γ fixed component inverse ordering number η≤μ} (9.18)

PDEset _μ = {pde|pde extension value inverse sort number η≤μ} (9.19)

PDE _γ set _μ = {pde _γ |pde _γ extended component inverse ordering number η≤μ} (9.20)

PFTset _μ = {pft|pft personalized feature set inverse order number η≤μ} (9.21)

PFT _δ set _μ ={pft _δ |pft _γ personalized feature set component inverse order number η≤μ} (9.22)

Wherein, MIset _λ , MI _α set _λ , PDset _λ , PDVset _λ , PDV _β set _λ , PDFset _λ , PDF _γ set _λ , PDEset _λ , PDE _γ set _λ , PFTset _λ , PFT _δ set _λ are the ordered heads Cut set, MIset _μ , MI _α set _μ , PDset _μ , PDVset _μ , PDV _β set _μ , PDFset _μ , PDF _γ set _μ , PDEset _μ , PDE _γ set _μ , PFTset _μ , PFT _δ set _μ are the ordering For the tail cut set, λ and μ are less than the number of elements in the respective sets, that is, the position of the cut set, which belongs to a natural number, θ is the positive sorting number, and η is the reverse sorting number.

Step S9013, train the cutout:

Continuously monitor and record the monitoring value and the object to the information base, and according to the information in the information base, adopt cyclic and recursive calculation to train the ordered head cut set and the ordered tail cut set, Record the results to the repository.

Step S9014, search for optimization of the interception set:

According to the mathematical model, according to formula (9.1) to formula (9.22), take λ=1, calculate and obtain the optimal value of the monitoring value and the optimal value of the monitoring component, and the corresponding object is also used as the optimal value. Taking μ=1, the worst value of the monitoring value and the worst value of the monitoring component are obtained by calculation, and the corresponding object is simultaneously regarded as the worst value.

In this embodiment, in addition to the above-mentioned data using the interception operation, the characteristic peaks of the laser Raman spectrum and the infrared light spectrum also adopt the interception operation, so as to enhance the monitoring effect.

Step S9020, establish a partial differential model:

Using the principle of partial differential equations, according to formula (9.23) and formula (9.24), establish functions between the monitoring value, the monitoring component, the variable component, the fixed component, and the extended component, and calculate the Monitoring value, the monitoring component:

in:

Both f _9.23 and f _9.24 are partial differential equations, MI′ ₁ is the first derivative of the monitoring component numbered 1,

is the ε order derivative of the monitoring component numbered 1, MI′ _m is the first order derivative of the monitoring component numbered m,

is the ε order derivative of the monitoring component numbered m. _PDV'α is the αth 1st derivative of the variable component that includes more than one associated with the _MIα .

is the ν order derivative of the αth that includes more than one of the variable components associated with the MI _α . PDF' _α is the α-th 1st derivative of the fixed component that includes zero or more of the fixed components associated with the MI _α .

is the ν order derivative of the αth species including zero or more of the fixed components associated with the MI _α . _PDE'α is the αth 1st derivative of the extended component that includes zero or more associated with the _MIα .

is the ξ-th derivative of the variable component that includes zero or more of the variable components associated with the MI _α . PFT' _α is the α-th 1st derivative of the extended component that includes zero or more components associated with the MI _α .

is the α-th derivative of the ε order that includes zero or more of the category components of the personalized feature set associated with the MI _α .

Wherein, as required, the monitoring component in function f _9.23 is also replaced by MI _α in f _9.24 .

Wherein, ε is the highest order of the monitored component derivative, υ is the highest order of the variable component derivative, ξ is the highest order of the extended component derivative, and ε, ν, ξ, and ε are all natural numbers.

In this embodiment, in addition to the partial differential equation operation for the above data, the intercept operation is also used for the characteristic peaks of the laser Raman spectrum and the infrared light spectrum, so as to enhance the monitoring effect.

Step S9030, extreme value optimization method:

According to the partial differential equation and the fuzzy partial differential equation, the method of taking extreme values of the independent variable and the dependent variable in the partial differential equation including formula (9.23) and formula (9.24), and obtaining any When the independent variable and the dependent variable are 0, the method of calculating the monitoring value and the monitoring component is obtained, and the optimal value and the worst value of the monitoring value and the monitoring component are obtained to obtain This gets the optimal value for the object.

In this embodiment, the characteristic peaks of the laser Raman spectrum, the infrared light spectrum, and the instantaneous value of blood glucose monitoring are all optimized according to their extreme values.

In step S9031, according to the information base being continuously updated over time, in timed or irregular conditions, multiple learning and training are implemented, and methods including T test and Z test are used to select the monitoring value and the said monitoring value. The optimal value of the component and the abnormal value of the monitoring value and the worst value of the monitoring component are monitored, and the abnormal value is eliminated, thereby obtaining the abnormal value of the object.

Step S9040, probability optimization method:

In the information base, the object monitoring value and the object extension component collected in different time periods are selected, the monitoring value and the monitoring component are calculated, and the monitoring value and the monitoring component are calculated as the maximum value and the When the object corresponding to the minimum value, the probability calculation method including the Bayesian algorithm is used, and when the statistics of the monitoring value and the monitoring component are the maximum value and the minimum value, the adjustable object in the object is monitored. The probability that the value and the extension component of the adjustable object have similar values, and the high probability verification is verified.

Calibrating the adjustable object monitoring value and the adjustable object extension component in the object when the monitoring value and the monitoring component are maximum values are the optimized adjustable object monitoring value and the optimized adjustable object extension component, The adjustable object monitoring value and the adjustable object extension component in the object when the monitoring value and the monitoring component are minimum values are calibrated to be a degraded adjustable object monitoring value and a degraded adjustable object extension component.

In this embodiment, since the sensitivity of the laser Raman spectrometer is very high, it is affected by the monitoring environment (such as external light, the pollution level of the monitoring point, the movement state of the monitoring individual, etc.), resulting in continuous time series, monitoring The obtained laser Raman spectrum has certain fluctuations, so the method of probability optimization is adopted to eliminate the interference factors.

Step S9050, neural network optimization method:

Step S9051, according to including the mathematical model, for the relational information records in the information base, using the information records as neurons, and establishing a connection function between the neurons with the calculation results including the mathematical model, forming A neural network with more than one layer.

Step S9052, according to the connection function, according to the effect of the optimized adjustable object monitoring value and the optimized adjustable object extended component on the monitored value and the monitored component, divide and establish an excitatory type, an inhibitory type, and an explosive type. , a platform-type linker function, the linker function includes a constant-type weight coefficient and a function-type weight coefficient.

Step S9053, using deep learning algorithms, including supervised learning, unsupervised learning, and reinforcement learning algorithms, to optimize the connection sub-function.

In step S9054, a support vector machine algorithm is used to classify and screen the monitoring value and the monitoring component, and screen out the optimized adjustable object monitoring value and the optimized adjustable object extended component.

Step S9055, using a convolutional neural network algorithm, for the condition of ignoring the association between the objects, implementing convolution, activation, pooling, full connection, and training the connection sub-function to filter out the monitoring including optimization. value and the monitored component.

In step S9056, a cyclic neural network algorithm is used to establish an intra-layer correlation function under the condition that the objects need to be correlated, and the linker function is trained to screen out the monitoring value and the monitoring component including the optimization.

In step S9057, a deep neural network algorithm is used to establish an inter-layer correlation function under the condition that the object, the monitoring value and the monitoring component between the layers of the neural network need to be correlated, and the connection is trained. sub-function to filter out the monitoring value and the monitoring component including optimization.

In step S9058, a feedforward neural network algorithm is used to train the connection sub-function under the condition that each neuron is only connected to the neurons of the previous layer, so as to filter out the monitoring value including the optimized monitoring value and the Monitor quantities.

In step S9059, a feedback neural network algorithm is used to train the connection sub-function under the condition that each neuron is only connected to the neuron of the next layer, so as to filter out the monitoring value including the optimized monitoring value and the monitoring value. weight.

In this embodiment, the neural network optimization method is mainly applied in the calculation of the object algorithm, the group algorithm and the personalized feature set. Among them, in the object algorithm, it can be used for the calibration of the historical data to the current data, in the group algorithm, it can be used for the calibration of other objects to its own object, and in the personalized feature set, it can be used for the optimization of the personalized feature set .

Step S9060, reversible optimization method:

The monitoring value and the When the monitoring component reaches the optimum or the specified value within its interval, the result information can reversibly reproduce the monitoring value and the monitoring component reaches the optimum or the specified value within its interval, and the reversible relationship between the monitoring components .

In this embodiment, in the reverse optimization of the Kalman filter, the reversible optimization method is adopted.

Step S9070, timing repetition method:

According to the mathematical model and the reversible relationship, calculate the timing time series value when the monitoring value and the monitoring component reach the optimal value or the specified value starting from the current moment, that is, starting from the current moment, after the timing time When the time of the sequence value is reached, the monitoring value and the monitoring component reach the optimal value or the specified value.

Step S9080, delay reproduction method:

According to the mathematical model and the reversible relationship, when the timing time series value is less than the predetermined health assessment time domain value, calculate the required delay time difference, and add the delay time difference to the mathematical model. The model is used to ensure that the monitoring value and the monitoring component reach the optimal value or the specified value at the time point of the predetermined health assessment time.

2.10: Network and Big Data

On the basis of the foregoing technical solutions, the present invention includes, but is not limited to, network processing steps. Specifically, the following or multiple local improvement measures can be adopted:

SA100 steps, cloud big data model:

In step SA110, a cloud center based on the Internet model is established, and all information monitored by the present invention, including all the groups, all the information of the objects, the intermediate calculation results, and the information base, are transmitted through the wide area network network in the form of cloud terminals, and all stored One or more cloud servers based on the Internet are used as one or more cloud centers, and cloud computing mode is adopted to manage, calculate and support the cloud terminal.

In step SA120, one or more cloud centers are established in the blockchain mode to store, manage and support the information base and the aforementioned steps. The users use anonymous records, and the information in the information base uses timestamped Chain structure, users access the information base through encryption and decryption communication, the information supports anti-tampering, and supports anti-repudiation, multi-center, and no-center modes.

In step SA130, a secure multi-party computing model is used to establish, manage and support more than one organization. The content of the information base of the organization performs the agreed calculation, and the obtained calculation result is shared by the participating organizations. The organization includes one or more of the cloud centers and manages one or more of the objects. The secure multi-party computation includes: public key mechanisms, hybrid circuits, inadvertent transmission, secret sharing, privacy-preserving set intersection protocols, homomorphic encryption, zero-knowledge proofs, and methods without trusted centers to enhance information security and protection Object Privacy.

In step SA140, the centralized learning mode is used to establish and train the model training when the object privacy protection is not emphasized, and the information database is stored in a cloud center.

In step S9150, the federated learning mode is used to establish and train the model training when the privacy protection of the object needs to be emphasized. At this time, the model training is performed between more than one stored cloud center, and the respective cloud centers do not exchange their respective cloud centers. information.

SA200 steps, LAN mode:

A local area network-based server is established to store and manage the support center, and all information including all groups and objects, intermediate calculation results, and the information base monitored by the present invention are transmitted through the local area network network in the form of network terminals, and all stored in the On the server of the local area network, in order to manage, calculate and support.

SA300 steps, single point mode:

The single point is to monitor the steps of monitoring, storage, management, calculation and support of one of the objects, and all the information of the object monitored by the present invention, the intermediate calculation results, and the information base are all stored in the single point. storage, perform all steps.

In this embodiment, the monitoring object is an individual, and a dedicated monitoring device that supports wireless movement is used. The monitoring device is connected to the individual's smart phone, and then works with the cloud server of the public network. In order to strengthen personal privacy management, in order to adopt the blockchain model. In the initial stage of big data, a large amount of data of individuals with different time periods that have collected the agreed truth value is required, so that after this stage, deep learning can be used to reason and classify, so as to reduce the action of collecting the agreed truth value, and Reduce actual costs and workload. The function fitting here includes arithmetic median, arithmetic mean, curve median, curve mean, least squares method, Gaussian method, neural network method, etc.

Embodiment 2. Monitoring method for energy saving of fan in cargo hold of container ship

1. Introduction to the basic scheme

This embodiment is used as an example of the application of the present invention in the industrial field. Specifically, in ocean-going container ships, the energy-saving control of the high-power fan in the cargo hold is a multi-input and multi-output control system, in which the input includes the ambient temperature, wind speed, and in-box settings of each refrigerated container stacked in the cargo hold. The temperature, the overall wind field of the cargo hold, etc., the output control includes the frequency conversion speed regulation of all the cargo hold fans, the air duct air outlet damper, the cargo hold air inlet, the cargo hold air outlet, the refrigerated container control signal, etc. The overall control index is to ensure that the refrigerated container is in the box Under the condition that the quality of the goods and the temperature in the box set by the user remain unchanged, the energy consumption of the fan in the cargo compartment is reduced as much as possible. According to the actual system operation data, using the multi-mode monitoring method of the present invention, the fan control can be more reasonable and stable, and the energy saving effect is as high as 58%.

2. Illustration description

Figure 6: As shown in the number 6010 is the cargo hold, in which are stacked refrigerated containers (refrigerated containers), and in the refrigerated containers, the refrigerator has a closed-loop temperature control system, such as a PID regulator control temperature system. In this temperature control system, signals including internal set temperature, supply air temperature, return air temperature, and external ambient temperature of the refrigerated container are obtained from the external interface of the refrigerated container through an external communication device, or these signals are obtained through a built-in communication circuit. Signals, which are output as indicated by number 6036 and input to the input of the system as negative feedback signals as indicated by number 6032, together with the system given signal as indicated by number 6031 and the AI given signal as indicated by number 6033 , form a complete feedback signal as shown in No. 6043, drive the fan control subsystem shown in No. 6020 to control the fan, and finally form a closed-loop control method in control theory as shown in No. 5035. That is to say, according to the collected data, the control function is generated as the closed-loop negative feedback data, and the fan or the fan and the gate are directly controlled. However, it should be noted that the closed loop proposed here is for the process time from acquisition to control, and this time difference is relatively short, for example, less than 20 minutes. It should be noted that the 20 minutes here is just an example, this number can be longer or shorter depending on the size of the cabin.

FIG. 3 : The control method associated with FIG. 6 , using the state-space control method with multiple inputs and multiple outputs as shown in FIG. 3 .

Figure 7: The

numbers

7010 and 7020 are the vertical planes where the headers of the refrigerated containers stacked in a cargo hold on the ship are located respectively. The headers of the refrigerated containers refer to the end of the refrigerated containers where the operation panel of the refrigerator is located. In ships, the stacking of containers has a three-dimensional coordinate number, namely row number Bay, column number Row, layer number Tier, referred to as BRT coordinates. In Figure 7, the white squares indicated by

numbers

7012 and 7022 are the positions of the bottom ballast tanks, and the gray squares indicated by

numbers

7011 and 7021 are the headers of the refrigerated containers, where the air at the headers The temperature changes as the reefer container works. For the convenience of description, the patent application of the present invention uses C1 to C9 to indicate the data of the air temperature at the BRT coordinates of the location point of the reefer container or the exhaust air volume and exhaust speed of the cooling fan. It should be noted that this data is a continuous amount , C1 to C9 are used in FIG. 5 to indicate that the continuous quantities vary from small to large, but this does not mean that the continuous quantities are only divided into 9 levels of C1 to C9. The data of one frame picture refers to collecting once for a container in a cargo hold and storing it in the database of environmental data according to the BRT coordinates. And so on for other active containers, passive containers, or non-contained cargo.

3. Differentiation description

The same points with the first embodiment will not be repeated here, and the differences are as follows:

3.1,

The set monitoring value includes at least the position coordinates of the refrigerated container in the cargo hold, the ambient temperature, the wind speed, the set temperature in the container, the temperature of the air outlet, the temperature of the air supply outlet, the operating status of the fan in the cargo hold, and the overall energy consumption of the cargo hold.

The agreed truth value is selected as the historical record value, and the monitoring factor value is set as the control function of the fan in the cargo compartment and the overall energy consumption function of the cargo compartment. The correlation functions are set as refrigerated container control function, cargo hold fan control function, cargo hold temperature control function, and cargo hold energy consumption function.

The set object is the brand and model of the reefer container, and the seed group is set, in which each cargo hold is a group, each ship is a group, and each route is a group.

3.2,

The control of each fan or the control of the damper on each fan is selected as the longitudinal correction value, and the space state equation group shown in Figure 2 is used as the correlation function and the control function. For each ship, use each cargo hold as a group to perform a group algorithm.

3.3,

Artificial intelligence algorithms are used for deep learning and labeling, and training is performed using, for example, convolutional neural network CNN algorithm, Bayesian Bayes algorithm, adversarial neural network GAN algorithm, firefly algorithm, and ant colony algorithm. The optimized indicators include the minimum energy consumption of the fan, the least equipment action (for example, the minimum number of start and stop times, the minimum number of start and stop fans), the least disturbance to the internal temperature field, and the least disturbance to the air output data, and the actual control room. The results are stored in the database according to time, and all relevant data such as ship route data, weather data, container loading and unloading data, and docked terminal data that can be obtained are input into the database. These are used as the content and experience content of deep learning of artificial intelligence algorithms, so that as the system runs, the results of artificial intelligence algorithms become more and more "smarter", and at the same time, due to the reduction of equipment movements, it also prolongs the life of the equipment. service life.

3.4,

Each ship uses data communication satellites to connect to the cloud big data center in real time, and uses blockchain and secure multi-party computing to protect the information of each ship and each cargo owner.

Claims

The method for multimodal personalized monitoring is characterized by comprising:

The monitoring value of the decomposition object is more than one monitoring component, the monitoring component is monitored, and the monitoring component is decomposed to include more than one variable component, more than zero fixed components, and more than zero extended components that have a monitoring function with the monitoring value;

Execute an object algorithm according to the historical and current monitoring components for the object, and obtain an object correction value whose agreed true value error with the object is less than the allowable error; and/or,

Establish a group consisting of more than one object with the same monitoring value, and use big data to train a personalized feature set including the object algorithm parameters, group algorithm parameters, and the agreed truth value, according to the personalized feature set and Other objects modify the group algorithm of their own objects to obtain a group correction value.
method according to claim 1, is characterized in that, comprises the step of following decomposition monitoring value:

Step S2010, the variable component is a monitoring component in the monitoring value that changes with environmental changes or a monitoring component specified by a user, and the variable component is obtained by sensor monitoring;

In step S2020, the fixed component is a type of the monitoring component that can be separated in the monitoring value, and within a sampling period equal to the variable component, the rate of change of the fixed component is the same as that of the variable component. The ratio of the rate of change of the components is small, for the optimal selection of the biological health category, the ratio of the rate of change is less than 0.20, and for the optimal selection of the industrial category, the ratio of the rate of change is less than 0.10, and, under this condition, the fixed component monitoring using sensors or statistical predictions according to said user selection;

In step S2030, the extended component is obtained by using sensor monitoring, and the monitoring function of the extended component and the monitoring value exists in a monitoring function including:

The change of the monitoring value affects the extended component in one direction, that is, the extended component is a function argument of the monitoring value, and at this time, the extended component is used as the auxiliary monitoring of the monitoring value;

The change of the monitoring value affects the extended component in both directions, that is, the extended component and the monitoring value are mutually independent variables, and at this time, the extended component is used as the auxiliary monitoring and adjustment of the monitoring value;

The change of the extended component affects the monitoring value in one direction, that is, the monitoring value is a function argument of the extended component, and at this time, the extended component is used as an auxiliary adjustment of the monitoring value;

In step S2040, the monitoring value and the monitoring component are continuously monitored according to different continuous time series and specific moments, and the intermediate data and result data are stored in the information base.
The method of claim 2, wherein the monitoring function further comprises:

In step S3010, a monitoring function between the monitoring value and the object is established according to formula (3.1), and the monitoring value is decomposed according to formula (3.2) to monitor components, including the variable component, the fixed component and the extended component component, according to formula (3.3) to establish the monitoring function of the monitoring value and the component, according to formula (3.4) to establish a variable component set, according to formula (3.5) to establish a fixed component set, according to formula (3.6) to establish an extended component set;

MI = f 3.1 (PD) (3.1)

PD=PDV∪PDF∪PDE (3.2)

MI=f 3.3 (PDV, PDF, PDE) (3.3)

PDV={PDV β | PDV β number β variable component data, 1≤β≤n} (3.4)

PDF = {PDF γ | PDF γ fixed component data of number γ, 0≤γ≤p} (3.5)

Wherein, f 3.1 is the monitoring function of the monitoring value and the object, f 3.3 is the monitoring function of the monitoring value and the component, MI is the monitoring value or monitoring value data set, PD is the object or Object data set, PDV is the variable component or variable number component set, PDF is the fixed component or fixed component set, PDE is the extended component or the set of extended components, PDV β is the set of all the components numbered β. the variable component data or an element of the variable component data set, n is the total number of the variable components, PDF γ is the fixed component data numbered γ or an element of the fixed component data set, where γ=0 Indicates that the PDF is an empty set, which does not contain the fixed components, p is the total number of the fixed components,
is numbered as
The extended component data of or an element of the extended component data set, where
Indicates that the PDE is an empty set, which means that the fixed component is not included, φ is the total number of the extended components, and the decomposition includes frequency domain decomposition from the perspective of monitoring convenience, and also includes time domain decomposition from the perspective of continuous time; and/or ,

Step S3020, monitor the variable component according to the functional relationship between the variable component determined by the formula (3.7) and the output signal of the sensor, and establish a function set of the variable component according to the formula (3.8), Equation (3.9) is the output signal function of the sensor:

PDV β = f 3.7β (SS β ) (3.7)

F 3.8 = {f 3.7β | f 3.7β Variable component data of number β, 1≤β≤n} (3.8)

SS β = f 3.9β (S β ) (3.9)

in:

f 3.7β is the function of the variable component numbered β, F 3.8 is the function set of the variable component, f 3.9β is the sensor function, n is the total number of functions of the variable component, n, β All belong to natural numbers, and 1≤β≤n;

PDV β is the variable component numbered β, SS β is the signal output by the sensor numbered β, S β is the sensor numbered β;

Step S3030, monitor the fixed component according to the functional relationship between the fixed component determined by the formula (3.10) and the output signal of the sensor or the statistical prediction, and establish the function of the fixed component according to the formula (3.11). Set, formula (3.12) is the output signal function of the sensor or the statistical prediction:

PDF γ = f 3.10 γ (SS γ ) (3.10)

F 3.11 = {f 3.10γ | f 3.10γ Fixed component data of number β, 0≤γ≤p} (3.11)

SS γ =f 3.12γ (S γ ) (3.12)

in:

f 3.10γ is the function of the fixed component numbered γ, F 3.11 is the function set of the fixed component, f 3.12γ is the output signal function of the sensor or the statistical prediction, p is the fixed component The total number of functions, p and γ are both natural numbers, and 0≤γ≤p, p=0 does not include fixed components;

PDF γ is the fixed component numbered γ, SS γ is the signal output by the sensor numbered γ or the statistical prediction, S γ is the sensor numbered γ or the statistical prediction;

Step S3040, monitor the extended component according to the functional relationship between the extended component determined by the formula (3.13) and the output signal of the sensor, and establish a function set of the extended component according to the formula (3.14), the formula (3.15 ) is the output signal function of the sensor:

in:

is numbered as
The function of the extended component, F 3.14 is the function set of the extended component,
is the output signal function of the sensor, n is the total number of functions of the extended component,
φ are all natural numbers,
φ=0 means not including extended components;

is numbered as
The extended component of ,
for the signal output by the sensor numbered β,
is numbered as
of the sensor.
The method according to claim 2 or 3, characterized in that, comprising the processing step of background noise:

Step S4010, for the object with the background noise, calculate according to the following steps:

According to formula (4.1), the monitoring function with background noise is established,

According to formula (4.2), the background noise is set as the fixed component,

The extended component function with background noise is established according to formula (4.3),

The sensor function of the variable component is established according to equation (4.4),

Setting formula (4.5) as a function of the background noise,

Calculate the monitoring value according to formula (4.6),

MI=f 4.1 (PDV, PDF, PDE, PDN) (4.1)

PDF=f 4.2 (PDN) (4.2)

PDE = f 4.3 (SE, PDN) (4.3)

PDV=f 4.4 (SV) (4.4)

PDN=f 4.5 (t) (4.5)

MI = f 4.6 (SV, SE, t) (4.6)

Wherein, f 4.1 is the monitoring function including background noise, f 4.2 is the function with the background noise as the fixed component, f 4.3 is the extended component function including background noise, and f 4.4 is the variable component sensor function, f 4.5 is the function of the background noise, f 4.6 is the monitoring value function, PDN is the background noise, SE is the sensor output signal of the extended component, SV is the sensor output of the variable component signal, t is the time series;

Step S4020, according to the actual application, in the case of single-variable monitoring, and in the case that the range of the variable-component sensor satisfies the application, adopt one of the variable-component sensors, and in the case of the variable-component sensor If the range does not meet the application, use a plurality of the variable component sensors to extend the range;

Step S4030, according to the actual application, in the case of multi-variable monitoring, each variable adopts a sensor of the variable component;

Step S4040, according to the actual application, if the output data of the variable component sensor satisfies the application, it is not necessary to use the extended component sensor, and in the case that the output data of the variable component sensor cannot meet the application Next, use more than one sensor of the extended component described above.
The method according to claim 2, 3 or 4, wherein the object algorithm specifically comprises:

In step S5010, standard or higher-level measurement accuracy monitoring equipment is used to monitor and obtain the monitoring value of the object as the agreed true value, record the monitoring time, and set the agreed true value The error is Δ, and Δ=0, and the allowable error is set as Δ e ;

Step S5020: Set an initial parameter set, and calculate the object correction value, and then calculate the agreed true value according to the step S5010, calculate the error, if the error falls within the allowable error, assign the initial parameter set to the personalized parameter set, if If the error is greater than the allowable error, modify the initial parameter set to be the personalized parameter set, so that the error falls within the allowable error;

Step S5030, using the personalized parameter set, using the current monitoring component and algorithms including extended Kalman filter algorithm, Monte Carlo particle algorithm, modern Bayesian algorithm, and the personalized parameter set , calculating the object correction value; and/or,

Step S5040, based on the setting, using the historical monitoring component, according to the set time interval, execute the step S5020 at the time interval point, and execute the step S5030 outside the time interval point to establish The personalized parameter set and the time series corresponding to the monitoring components are recorded in the information database, and a deep learning algorithm is used to calculate the monitoring components and the personalized parameters of the objects in the information database set and the time series, find out the personalized parameter set with high probability and demarcate the corresponding monitoring component as the optimization point, the high probability includes the user-specified probability or the probability is greater than 30%; and/or,

In step S5050, based on the personalized parameter set of the optimization point obtained in the step S5040, the object correction value is calculated by adopting a support vector machine algorithm and a convolutional neural network algorithm.
The method according to claim 5, wherein the swarm algorithm specifically comprises:

Step S6010, for all the objects in the group, calculate the optimization point and the object correction value, and record them in the information database;

Step S6020, according to the objects in the information database, for the optimization point and the object correction value, establish more than one object classification, and calibrate the object classification of the object, and/or the object classification The ratio of the number of object classifications to the total number of objects is greater than the number specified by the user or greater than 0.2, and is recorded in the information database;

Step S6030, according to the object classification, for the optimization point and the object correction value, according to the principle of minimum error, calculate the personalized feature set of the object classification, and record it in the information database;

Step S6040, calculating the group correction value of the object according to the personalized feature set of the object classification;

Step S6050, for the self-object, calculate the object classification to which the calibration belongs, and calculate the object correction value and the group correction value of the object according to the personalized feature set of the object classification;

Step S6060, according to the personalized feature set of the object classification, for the newly added object, determine its object classification according to the calculation of the monitoring value of the object, and directly adopt the personalized feature set of the classification to include: said agreed truth value without performing the step of monitoring said agreed truth value;

The personalized feature set includes parameters of the object algorithm, parameters of the population algorithm, the classification and/or the agreed truth value for the object.
The method according to claim 5, characterized in that, comprising a differential object algorithm, the details are as follows:

Step S7010, calculating the difference between the monitoring value of the object at the optimization point and the correction value of the object, and recording it in the information database;

Step S7020, if the difference value of each of the optimization points is a constant difference value, for the object correction value of the subsequent time series, use the monitoring value to subtract the constant difference value to calculate the object correction value ;and / or,

Step S7030, if the difference value of each optimization point is a function difference value, then for the object correction value of the subsequent time series, subtract the function difference value from the monitoring value to calculate the object correction value ;and / or,

Step S7040, if the difference value of the optimization point of the time series is an alternating difference value between the constant difference value and the function difference value, subtract the alternating difference value from the monitoring value to calculate the the object correction value;

In step S7050, for the Raman spectrum monitoring method, the object is scanned twice or more with excitation light with a slight frequency offset to generate two or more Raman spectra, and according to the Raman scattering characteristic shift of the object, the two or more The Raman scattering spectrum generated by scanning is calculated by differential convolution or linear regression to eliminate the interference of background noise caused by fluorescence, and the wavelength difference of the excitation light with the slight frequency offset is between 0.2 and 100 nm.
The method according to claim 1, wherein the personalized feature set is as follows:

Step S8010, set a constraint condition that the error of the corrected value of the object is less than the allowable error, establish the personalized feature set according to formula (8.1), and calculate the personalized feature set:

PF=f 8.1 (MI, MIF, Δ≤Δ e ) (8.1)

Wherein, f 8.1 is the personalized feature set function, PF is the personalized feature set including the object algorithm parameters, the population algorithm parameters, and the agreed truth value, MI is the monitoring value or the monitoring value data set , MIF is the object correction value or set of object correction values, Δ is the error, and Δ e is the allowable error;

Step S8020, decompose the personalized feature set:

decomposing the personalized feature set into one or more personalized feature set categories, and decomposing the personalized feature set category into one or more personalized feature set category components based on the specification;

There is a functional relationship determined by formula (8.2) between the personalized feature set and the category of the personalized feature set:

PF=f 8.2 (PFT) (8.2)

Wherein, f 8.2 is the category function of the personalized feature set, PFT is the category of the personalized feature set, and the decomposition includes frequency domain decomposition from the perspective of ease of calculation, and also includes time domain decomposition from the perspective of continuous time;

Step S8030, monitoring the time domain value of the personalized feature set category:

There is a functional relationship between the personalized feature set category and the continuous time series determined by formula (8.3), and the time domain value of the personalized feature set category is monitored according to formula (8.3):

PFT=f 8.3 (t) (8.3)

Wherein, f 8.3 is the personalized feature set category time domain function, and t is the continuous time series;

Step S8040, monitor the specific moment value of the personalized feature set category:

There is a functional relationship between the specific moment value of the personalized feature set category and the specific moment, and the specific moment value of the personalized feature set category is monitored according to the formula (8.4):

PFT T = f 8.3 (t, t = T) (8.4)

Wherein, f 8.3 is the time domain function of the personalized feature set category, T is the specific moment, and PFT T is the specific moment value of the personalized feature set category at the specific moment;

Step S8050, monitor the time domain value of the category component of the personalized feature set:

The personalized feature set category is decomposed into more than one personalized feature set category component according to formula (8.5), and the personalized feature set category component and the continuous time series have the function determined by formula (8.6) relationship, monitor the time domain value of the category component of the personalized feature set according to formula (8.5):

PFT=f 8.5 (PFT 1 , PFT 2 , . . . , PFT q ) (8.5)

PFT δ = f 8.6 (t, 1≤δ≤q) (8.6)

Wherein, f 8.5 is the category decomposition function of the personalized feature set, f 8.6 is the time domain function of the category component of the personalized feature set, PFT 1 , PFT 2 , ..., PFT q are the category components of the personalized feature set, and q is the The total number of category components of the personalized feature set, δ is the category component number of the personalized feature set, q and δ are both natural numbers, and 1≤δ≤q, PFT δ is the category component of the personalized feature set numbered δ the time domain value in the continuous time series;

Step S8060: Monitor the specific moment value of the category component of the personalized feature set.
The specific time value of the category component of the personalized feature set at the specific time has a functional relationship determined by the formula (8.7) with the continuous time series, and the category component of the personalized feature set is monitored according to the formula (8.7). The specific moment value at a specific moment: the method according to claim 8, characterized in that, comprising training a mathematical model of the personalized feature set:

Step S9010, fuzzy optimization method:

According to the following mathematical models, including fuzzy equations, fuzzy partial differential equations, and set equations, the monitoring value and the monitoring component are optimized, and the optimal value of the object under optimization is obtained, specifically including:

Step S9011, create a set:

A set of monitoring values is established as an element with the monitoring value, denoted as MIset;

Taking the monitoring component as an element to establish a monitoring component set, denoted as MI α set, where α is the numbering of the monitoring component;

Create an object set with the object as an element, denoted as PDset;

Decompose the monitoring value set into a variable value set and a fixed value set;

Decomposing the variable value set into a variable component set, denoting the variable value set as PDVset, denoting the variable component set as PDV β set, and β being the number of the variable component;

Decompose the fixed value set into a fixed component set, denote the fixed value set as PDFset, denote the fixed component set as PDF γ set, and γ is the number of the fixed component;

Decompose the extended value set as the extended component set, denote the extended value set as PDEset, and denote the extended value component set as
is the number of the extended component;

Taking the personalized feature set category as an element to establish a personalized feature set set, the component personalized feature set set is a personalized feature set component set, and the personalized feature set set is recorded as PFTset, and the personalized feature set component set is recorded as PFT δ set, δ is the serial number of the individualized feature set component;

the sets include fuzzy sets and non-fuzzy sets;

Step S9012, create a cut set:

According to the mathematical model, the mapping relationship between the sets is established in turn, and the monitoring value set and the monitoring component set are used for sorting as the primary key to become an ordered set, and the first λ is taken after positive sorting from large to small. The elements are used as the ordered head cut set, and the first μ elements are taken as the ordered tail cut set after reverse sorting from small to large, as follows:

MIset λ = {mi|mi monitoring value positive sequence number θ≤λ} (9.1)

MI α set λ = {mi δ |mi α monitoring component positive sequence number θ≤λ} (9.2)

PDset λ = {pd|pd object positive sorting number θ≤λ} (9.3)

PDVset λ = {pdv|pdv variable value positive sort number θ≤λ} (9.4)

PDV β set λ ={pdv β |pdv β variable component positive sorting number θ≤λ} (9.5)

PDFset λ = {pdf|pdf fixed value positive sort number θ≤λ} (9.6)

PDF γ set λ = {pdf γ |pdf γ fixed component positive sorting number θ≤λ} (9.7)

PDEset λ = {pde|pde extension value positive sort number θ≤λ} (9.8)

PDE γ set λ ={pde γ |pde γ extended component positive sorting number θ≤λ} (9.9)

PFTset λ = {pft|pft personalized feature set positive sorting number θ≤λ} (9.10)

PFT δ set λ ={pft δ |pft δ Personalized feature set component positive sorting number θ≤λ} (9.11)

MIset μ = {mi|mi monitoring value inverse order number η≤μ} (9.12)

MI α set μ = {mi δ |mi α monitoring component reverse order number η≤μ} (9.13)

PDset μ = {pd|pd object inverse ordering number η≤μ} (9.14)

PDVset μ = {pdv|pdv variable value reverse order number η≤μ} (9.15)

PDV β set μ ={pdv β |pdv β variable component inverse order number η≤μ} (9.16)

PDFset μ = {pdf|pdf fixed value inverse sort number η≤μ} (9.17)

PDF γ set μ = {pdf γ |pdf γ fixed component inverse ordering number η≤μ} (9.18)

PDEset μ = {pde|pde extension value inverse sort number η≤μ} (9.19)

PDE γ set μ = {pde γ |pde γ extended component inverse ordering number η≤μ} (9.20)

PFTset μ = {pft|pft personalized feature set inverse order number η≤μ} (9.21)

PFT δ set μ ={pft δ |pft γ personalized feature set component inverse order number η≤μ} (9.22)

Wherein, MIset λ , MI α set λ , PDset λ , PDVset λ , PDV β set λ , PDFset λ , PDF γ set λ , PDEset λ , PDE γ set λ , PFTset λ , PFT δ set λ are the ordered heads Cut set, MIset μ , MI α set μ , PDset μ , PDVset μ , PDV β set μ , PDFset μ , PDF γ set μ , PDEset μ , PDE γ set μ , PFTset μ , PFT δ set μ are the ordering Tail cut set, λ and μ are less than the number of elements in their respective sets, that is, the position of the cut set, which belongs to a natural number, θ is the positive sorting number, and η is the reverse sorting number;

Step S9013, train the cutout:

Continuously monitor and record the monitoring value and the object to the information base, and according to the information in the information base, adopt cyclic and recursive calculation to train the ordered head cut set and the ordered tail cut set, record the results to the repository;

Step S9014, search for optimization of the interception set:

According to the mathematical model, according to formula (9.1) to formula (9.22), take λ=1, calculate and obtain the optimal value of the monitoring value and the optimal value of the monitoring component, and the corresponding object is also used as the optimal value; take μ =1, calculate and obtain the worst value of the monitoring value and the worst value of the monitoring component, and the corresponding object is simultaneously regarded as the worst value; and/or,

Step S9020, establish a partial differential model:

Using the principle of partial differential equations, according to formula (9.23) and formula (9.24), establish functions between the monitoring value, the monitoring component, the variable component, the fixed component, and the extended component, and calculate the Monitoring value, the monitoring component:

in:

Both f 9.23 and f 9.24 are partial differential equations, MI′ 1 is the first derivative of the monitoring component numbered 1,
is the ε order derivative of the monitoring component numbered 1, MI′ m is the first order derivative of the monitoring component numbered m,
is the ε-order derivative of the monitoring component numbered m; PDV'α is the α -th first-order derivative of the variable component that includes more than one associated with the MI α ;
is the αth derivative of the v-th order including more than one of the variable components associated with the MI α ; PDF' α is the αth 1 including zero or more of the fixed components associated with the MI α order derivative;
is the α-th derivative of the v-th order including zero or more of the fixed components associated with the MI α ; PDE'α is the α -th 1 including zero or more of the extended components associated with the MI α order derivative;
is the α -th derivative of the variable component including zero or more of the variable components associated with the MI α ; PFT′α is the 1st derivative;
is the αth derivative of the ε order that includes zero or more of the category components of the personalized feature set associated with the MI α ;

Wherein, as required, the monitoring component in function f 9.23 is also replaced by MI α in f 9.24 ;

Wherein, ε is the highest order of the monitored component derivative, v is the highest order of the variable component derivative, ξ is the highest order of the extended component derivative, and ε, v, ξ, and ε are all natural numbers;

Step S9030, extreme value optimization method:

According to the partial differential equation and the fuzzy partial differential equation, the method of taking extreme values of the independent variable and the dependent variable in the partial differential equation including formula (9.23) and formula (9.24), and obtaining any When the independent variable and the dependent variable are 0, the method of calculating the monitoring value and the monitoring component is obtained, and the optimal value and the worst value of the monitoring value and the monitoring component are obtained to obtain This gets the optimal value of the object;

In step S9031, according to the information base being continuously updated over time, in timed or irregular conditions, multiple learning and training are implemented, and methods including T test and Z test are used to select the monitoring value and the said monitoring value. monitoring the optimal value of the component and the abnormal value of the monitoring value and the worst value of the monitoring component, eliminating the abnormal value, thereby obtaining the abnormal value of the object; and/or,

Step S9040, probability optimization method:

In the information base, the object monitoring value and the object extension component collected in different time periods are selected, the monitoring value and the monitoring component are calculated, and the monitoring value and the monitoring component are calculated as the maximum value and the When the object corresponding to the minimum value, the probability calculation method including the Bayesian algorithm is used, and when the statistics of the monitoring value and the monitoring component are the maximum value and the minimum value, the adjustable object in the object is monitored. The probability that the value and the extended component of the adjustable object have similar values, and verify the high probability verification; and/or,

Calibrating the adjustable object monitoring value and the adjustable object extension component in the object when the monitoring value and the monitoring component are maximum values are the optimized adjustable object monitoring value and the optimized adjustable object extension component, Calibrating the adjustable object monitoring value and the adjustable object extension component in the object when the monitoring value and the monitoring component are minimum values to be a degraded adjustable object monitoring value and a degraded adjustable object extension component; and / or,

Step S9050, neural network optimization method:

Step S9051, according to including the mathematical model, for the relational information records in the information base, using the information records as neurons, and establishing a connection function between the neurons with the calculation results including the mathematical model, forming More than one layer of neural network;

Step S9052, according to the connection function, according to the effect of the optimized adjustable object monitoring value and the optimized adjustable object extended component on the monitored value and the monitored component, divide and establish an excitatory type, an inhibitory type, and an explosive type. , a platform-type linker function, and the linker function includes a constant-type weight coefficient and a function-type weight coefficient;

Step S9053, using deep learning algorithms, including supervised learning, unsupervised learning, and reinforcement learning algorithms, to optimize the connection sub-function; and/or,

Step S9054, adopting a support vector machine algorithm to classify and screen the monitoring value and the monitoring component, and screen out the optimized adjustable object monitoring value and the optimized adjustable object extended component; and/or,

Step S9055, using a convolutional neural network algorithm, for the condition of ignoring the association between the objects, implementing convolution, activation, pooling, full connection, and training the connection sub-function to filter out the monitoring including optimization. value and said monitored component; and/or,

Step S9056, adopting a cyclic neural network algorithm, under the condition that the objects need to be correlated, establish an intra-layer correlation function, and train the linker function to filter out the monitoring value and the monitoring component including optimization; and / or,

In step S9057, a deep neural network algorithm is used to establish an inter-layer correlation function under the condition that the object, the monitoring value and the monitoring component between the layers of the neural network need to be correlated, and the connection is trained. a sub-function to filter out said monitoring value and said monitoring component including optimization; and/or,

In step S9058, a feedforward neural network algorithm is used to train the connection sub-function under the condition that each neuron is only connected to the neurons of the previous layer, so as to filter out the monitoring value including the optimized monitoring value and the monitor the amount; and/or,

In step S9059, a feedback neural network algorithm is used to train the connection sub-function under the condition that each neuron is only connected to the neuron of the next layer, so as to filter out the monitoring value including the optimized monitoring value and the monitoring value. amount; and/or,

Step S9060, reversible optimization method:

The monitoring value and When the monitoring component reaches the optimum or the specified value within its interval, the result information can reversibly reproduce the monitoring value and the monitoring component reaches the optimum or the specified value within the interval, and the difference between the monitoring components is reversible. reversible relationship; and/or,

Step S9070, timing repetition method:

According to the mathematical model and the reversible relationship, calculate the timing time series value when the monitoring value and/or the monitoring component reach the optimal value or the specified value starting from the current moment, that is, starting from the current moment, after the When the time of the time series value is timed, the monitored value and/or the monitored component reaches an optimized value or a specified value; and/or,

Step S9080, delay reproduction method:

According to the mathematical model and the reversible relationship, when the timing time series value is less than the predetermined health assessment time domain value, calculate the required delay time difference, and add the delay time difference to the mathematical model. A model is used to ensure that the monitoring value and/or the monitoring component reaches an optimal value or a specified value at the time point of the predetermined health assessment time.

PFT δT = f 8.6 (t, t=T, 1≤δ≤q) (8.7)

Wherein, PFT δT is the specific time value of the category component of the personalized feature set at the specific time.
The method of claim 1, comprising:

SA100 steps, cloud big data model:

In step SA110, a cloud center based on the Internet model is established, and all information monitored by the present invention, including all the groups, all the information of the objects, the intermediate calculation results, and the information base, are transmitted through the wide area network network in the form of cloud terminals, and all stored to one or more Internet-based cloud servers as one or more cloud centers to manage, calculate and support the cloud terminals in a cloud computing mode; and/or,

In step SA120, one or more cloud centers are established in the blockchain mode to store, manage and support the information base and the aforementioned steps. The users use anonymous records, and the information in the information base uses timestamped Chain structure, users access the information base through encryption and decryption communication, the information supports anti-tampering, supports anti-repudiation, multi-center, and no-center modes; and/or,

In step SA130, a secure multi-party computing model is used to establish, manage and support more than one organization. The content of the information base of the organization performs the agreed calculation, and the obtained calculation result is shared by the participating organizations; the organization includes one or more of the cloud centers and manages more than one of the objects; the secure multi-party computing includes : public key mechanisms, hybrid circuits, inadvertent transmission, secret sharing, privacy-preserving set intersection protocols, homomorphic encryption, zero-knowledge proofs, methods without trusted centers to enhance information security and protect object privacy; and/or ,

In step SA140, the centralized learning mode is used to establish and train the model training when the object privacy protection is not emphasized, and the information database is stored in a cloud center; and/or,

In step S9150, the federated learning mode is used to establish and train the model training when the privacy protection of the object needs to be emphasized. At this time, the model training is performed between more than one stored cloud center, and the respective cloud centers do not exchange their respective cloud centers. information; and/or,

SA200 steps, LAN mode:

A local area network-based server is established to store and manage the support center, and all information including all groups and objects, intermediate calculation results, and the information base monitored by the present invention are transmitted through the local area network network in the form of network terminals, and all stored in the servers in the local area network for management, computing and support; and/or,

SA300 steps, single point mode:

The single point is to monitor the steps of monitoring, storage, management, calculation and support of one of the objects, and all the information of the object monitored by the present invention, the intermediate calculation results, and the information base are all stored in the single point. storage, perform all steps.