CN111798984A

CN111798984A - Disease prediction scheme based on Fourier transform

Info

Publication number: CN111798984A
Application number: CN202010648900.2A
Authority: CN
Inventors: 张盛; 章越新
Original assignee: Individual
Current assignee: Individual
Priority date: 2020-07-07
Filing date: 2020-07-07
Publication date: 2020-10-20

Abstract

The invention provides a disease prediction scheme based on Fourier transform, which is characterized by comprising the following steps: collecting historical sick data and health related data of a user, and preprocessing the data; solving a trigonometric function type regression curve by adopting a nonlinear regression method and taking the preprocessed data as a sample set; transforming the regression curve into a disease prediction curve by using a Fourier transform principle; and predicting the disease risk of the user in a future period by using the optimal approximate solution of the disease prediction curve.

Description

Disease prediction scheme based on Fourier transform

Technical Field

The invention belongs to the field of human disease analysis and prediction, and particularly relates to a disease prediction scheme based on Fourier transform.

Background

In recent years, artificial intelligence technology is widely applied to the field of disease prediction, and various prediction schemes based on big data and deep learning are proposed successively, which has the advantages that: compared with traditional disease prediction, the disease prediction scheme based on artificial intelligence has natural performance advantages, but the defects are also obvious. First, the existing solutions are relatively costly due to: the existing scheme needs to identify and train a large number of data samples and needs a large number of calculation force supports, so that the operation cost is high, and the method is difficult to popularize and apply to miniature calculation force equipment such as a mobile intelligent terminal. Second, the prediction accuracy of the existing schemes is not as expected because: the existing scheme usually ignores the complexity and the dynamics of pathogenic mechanisms and physiological laws of diseases, so that the prediction accuracy of the existing scheme is difficult to achieve expectation. Third, the existing approaches have limited kinds of disease prediction due to: most existing schemes can only predict a single disease, and even if a few schemes can predict multiple diseases, the accuracy and reliability of the schemes need to be verified.

In addition, although the disease model based on statistical principles has partial prediction function, the defects are obvious. Firstly, the pertinence is not strong, and the main points are as follows: the existing disease model based on the statistical principle only aims at the special disease species and the epidemic diseases with larger influence range, but the research on common diseases such as cold, fever and the like is less, and the focus is not on disease prediction. Secondly, the prediction function is not strong, and is mainly reflected in: the disease prediction model based on the statistical principle only studies the statistical rules of the diseases, and ignores the pathogenic mechanism and physiological rules of the diseases. Third, the lack of customized services is mainly reflected in: most research objects based on the statistical principle are groups, and few research objects aim at individual modeling analysis. Fourthly, the practicality is not strong, mainly reflects in: most of the existing disease models based on the statistical principle are only used for academic research, and the disease models are characterized by strong universality and weak difference in guiding human practice activities.

The following trends are mainly developed in the field of future disease prediction: one is miniaturization. With the continuous updating and iteration of the 5G technology, the chip technology and the mobile intelligent terminal technology, the future disease prediction is accelerated to be deeply fused with the technologies, and the demand of people for disease prediction is fused with micro equipment such as mobile phones and intelligent wearable equipment. Second, customization. Due to the existence of individual differences, different populations, different individuals, and different age groups have different needs for disease prediction or prevention. Therefore, the disease prediction service with different disease types and different accuracies for different groups, different individuals and different age classes has wide application prospect. And thirdly, real-time implementation. Because the disease has strong correlation with time, the disease prediction inevitably develops towards timeliness, important references such as instant prediction and instant analysis are provided for the user, and the user can master the self health condition in real time. Fourthly, the cost is reduced. On the premise of meeting the accuracy requirement of the disease prediction, the lower the requirement on computational power, the better, and the lower the requirement on software and hardware cost, the better.

Disclosure of Invention

The invention aims to provide a disease prediction scheme based on Fourier transform for solving the defects of the existing disease prediction scheme and meeting the disease prediction development trend, and provides a brand-new disease prediction scheme for medical treatment, scientific research institutions and patients.

In order to achieve the above object, the present invention proposes a fourier transform-based disease prediction scheme, comprising the steps of:

step 1: collecting historical sick data and health related data of a user, and preprocessing the data;

step 2: solving a trigonometric function type regression curve by adopting a nonlinear regression method and taking the preprocessed data as a sample set;

and step 3: transforming the regression curve into a disease prediction curve by using a Fourier transform principle;

and 4, step 4: and predicting the disease risk of the user in a future period by using the optimal approximate solution of the disease prediction curve.

Acquiring historical sick data and health associated data of a user according to the step 1, and preprocessing the data, wherein the historical sick data of the user is a function of time change of the degree of a certain disease of the user in a past period; the user health related data refers to the function of various factors related to human health along with time change, and comprises external environment data, physiological index data and the like.

Acquiring user historical illness data and health associated data according to the step 1, and preprocessing the data, wherein a quantization coding mode is adopted for preprocessing user historical illness records; the health related data is preprocessed in a signal processing mode of derivation, difference, summation, integration and the like so as to extract characteristic information.

And (3) solving a trigonometric function type regression curve by taking the preprocessed data as a sample set according to a nonlinear regression method adopted in the step (2), wherein the nonlinear regression comprises conventional mathematical algorithms such as a machine learning algorithm, a least square method, maximum likelihood estimation, interpolation, fitting and the like.

And (3) transforming the regression curve into a disease prediction curve by utilizing a Fourier transform principle in the step 3, wherein the transformation method is to perform Fourier series expansion or Fourier transform on the disease prediction curve to be solved.

Transforming the regression curve into a disease prediction curve by using the fourier transform principle in the step 3, wherein the following waveform shaping processing is performed on the regression curve before obtaining the disease prediction curve:

firstly, scale transformation is carried out to ensure that time units are uniform;

and secondly, setting corresponding weights to adapt to different climatic regions and individual constitutions.

And (3) transforming the regression curve into a disease prediction curve by using a Fourier transform principle in the step (3), wherein the regression curve corresponds to important frequency spectrum components of the disease prediction curve to be solved, and the importance of the regression curve is realized in that the sum of the energy of the frequency spectrum components corresponding to the regression curve accounts for most of the total energy of the frequency spectrum.

Transforming the regression curve into a disease prediction curve by utilizing a Fourier transform principle in the step 3, wherein after the transformation processing, normalization processing needs to be carried out on a result function so as to describe the disease probability; the function after normalization processing is used as the optimal approximate solution of the disease prediction curve to be solved; in an ideal case, when the number of regression curves is close to infinity, the optimal approximate solution will be equal to the exact solution.

And (3) transforming the regression curve into a disease prediction curve by utilizing a Fourier transform principle in the step 3, wherein the horizontal axis of the disease prediction curve is time, and the vertical axis of the disease prediction curve is the corresponding disease probability of the user at a certain moment.

And (4) predicting the disease risk of the user in a future period by using the optimal approximate solution of the disease prediction curve in the step 4, wherein the prediction method comprises using the disease prediction curve and each order derivative thereof and the like, and a maximum likelihood principle should be followed.

Compared with the prior art, the brand-new disease prediction scheme based on the Fourier transform adopts a nonlinear regression method to obtain the trigonometric function type regression curve of the user historical diseased data and health related data, and converts originally discrete and irregular preprocessed data into a periodically changed rule curve, so that the intrinsic rule of the disease is reflected to a certain extent, and the disease prediction scheme has better robustness.

In addition, compared with the prior art, the brand-new disease prediction scheme based on the Fourier transform provided by the invention utilizes the Fourier transform principle to transform the regression curve into the disease prediction curve, and utilizes the optimal approximate solution of the disease prediction curve to predict the disease risk of the user in a period of the future. Through Fourier transform, frequency domain-time domain conversion analysis of various regression curves can be realized, and the influence of different pathogenic factors on disease development is intuitively reflected. Compared with the existing disease prediction scheme based on the artificial intelligence algorithm and the disease model based on the statistical principle, the practicability and portability of the method are greatly improved.

Drawings

FIG. 1 is a schematic diagram of a Fourier transform-based disease prediction scheme

FIG. 22018 year 1 month user A's disease prediction Curve

FIG. 32020 year 2 month user A's disease prediction Curve

FIG. 42020 disease prediction Curve for user A in month 6

Detailed Description

The following description is only an example of the present invention, and fully follows the concept, principle and flow of the solution in the present invention, and those skilled in the art can reasonably modify the present invention to obtain a disease prediction solution suitable for different application scenarios without departing from the scope defined by the concept of the present invention and the claims. It is also to be understood that these examples are merely illustrative of the principles of the present invention and are not intended to limit the scope of the invention.

Various terms in one example are mentioned throughout this specification for the purpose of describing particular embodiments, and are not intended to limit other examples of the specification. It should also be noted that the terms "and" "or" as used in one example of this specification are meant to encompass one or more of the associated combination of circumstances.

The invention provides a brand-new disease prediction scheme based on Fourier transform, which is characterized in that historical diseased data and health related data of a user are collected, and the data are preprocessed; solving a trigonometric function type regression curve by adopting a nonlinear regression method and taking the preprocessed data as a sample set; transforming the regression curve into a disease prediction curve by using a Fourier transform principle; and predicting the disease risk of the user in a future period by using the optimal approximate solution of the disease prediction curve.

According to the scheme, in step 1, historical sick data and health related data of a user need to be collected and preprocessed. In this example, the disease record for a disease is divided into two distinct events, "diseased" and "not diseased". The "diseased" event refers to a state that exists continuously for a period of time, but in this example, the "diseased" event is defined as a discrete independent value, and if more than one "diseased" event occurs in a sampling period, the disease record data of the sample point is recorded as 1, i.e., the "diseased" event is not accumulated. "non-sick" events include different health states of the user, such as "mild discomfort", "severe discomfort", etc., ordered from small to large in severity. In this example, the cold-type illness condition is taken as historical illness data.

Further, user health related data is collected. Since the user health related data is a function of various factors related to human health along with time, four factors related to colds, such as air temperature, heart rate of the user, respiratory rate, blood pressure and the like, are selected as the user health related data in the example. The sampling interval of all the data is determined according to actual conditions, and includes, but is not limited to, units such as year, month, day, hour, and the like.

In this example, samples of data such as air temperature, heart rate, respiratory rate, blood pressure, etc. are respectively denoted as F_T(t_i)、F_H(t_i)、F_B(t_i) And F_P(t_i) Wherein, t_iIndicating the ith sampling instant.

It should be noted that the air temperature value can be acquired by means of a meteorological department, meteorological software, and the like, and the heart rate, the respiratory rate, and the blood pressure can be recorded and acquired by means of an intelligent wearable device or a medical device. Wherein, F_T(t_i) Is taken as t_iAverage of the maximum and minimum air temperatures at that moment, F_H(t_i) Is taken as t_iHeart rate at rest at all times, F_B(t_i) Is taken as t_iRespiratory rate at time of rest, F_P(t_i) Is taken as t_iThe average of the high and low pressures at rest.

After the health related data is acquired, preprocessing operation is performed on the health related data. The method comprises the steps of firstly preprocessing historical sick data of a user. The pretreatment of the user historical sick data is carried out by adopting a quantization coding mode, and the result is recorded as F_S(t_i) The specific coding mode is as follows: if t_iWhen the user has a cold event, F_S(t_i) Is 1; if t_iIf the user does not have a cold event at all times, F_S(t_i) The value range of (a) is the interval [0,1 ]. This example randomly generates a real number between 0 and 1 using a Gaussian normal distribution N (0,1), which is assigned to F_S(t_i). When F is present_S(t_i) When the value is 0, the user is in an ideal health state; when 0 < F_S(t_i) If < 1, then represent the userHas a disease probability of F_S(t_i). Further, F_S(t_i) The closer the value of (a) is to the true prevalence probability of the user, the higher the prediction accuracy. The random numbers are used in this example to simplify subsequent experimental validation and analysis.

Secondly, the preprocessing of data such as air temperature, heart rate, respiratory rate, blood pressure and the like is carried out in a mode of derivation and absolute value taking. The example adopts a difference approximation to represent the derivative value, and the specific operation is as follows:

|F′_T(t_i)|＝|F_T(t_i)-F_T(t_i-1) L is a preprocessing result of the air temperature data;

|F′_H(t_i)|＝|F_H(t_i)-F_H(t_i-1) I is a preprocessing result of the heart rate data;

|F′_B(t_i)|＝|F_B(t_i)-F_B(t_i-1) L is a preprocessing result of the respiratory frequency data;

|F′_P(t_i)|＝|F_P(t_i)-F_P(t_i-1) And | is a preprocessing result of the blood pressure data.

According to the scheme, in step 2, a trigonometric regression curve needs to be solved. This example uses least squares with minimal cost to perform nonlinear regression. It should be noted that the same purpose can be achieved by using a machine learning method. The regression process is implemented using a Python program.

Firstly, determining a target regression curve, wherein the specific form is as follows:

is F_S(t_i) The regression curve of (d);

is F_T(t_i) The regression curve of (d);

is F_H(t_i) The regression curve of (d);

is F_B(t_i) The regression curve of (d);

is F_P(t_i) The regression curve of (2).

Then, determining the parameter to be regressed as A_S、A_T、A_H、A_B、A_P、f_S、f_T、f_H、f_B、f_P、

And finally, determining the optimal solution of the parameter to be regressed by using a least square method.

According to the scheme, in step 3, a disease prediction curve needs to be solved. Taking the disease prediction curve as y (t), and performing Fourier series expansion to obtain a Fourier series expansion expression of y (t)

Wherein A is_nTo predict the magnitude spectrum of curve y (t),

the phase spectrum of the prediction curve y (t) is obtained.

Order to

Wherein, a_S、a_T、a_H、a_B、a_PFor each regression curve corresponding scale transformation factor, w_s、w_t、w_h、w_b、w_pThe weights corresponding to the regression curves. According to the theorem of Pasval,

can be used as the optimal approximate solution of y (t).

It should be noted that, in the following description,

can be used as the optimal approximate solution of y (t) because: y is_S(t)、y_T(t)、y_H(t)、y_B(t)、y_P(t) is the five most important factors for judging whether the user catches a cold, and the sum of the energies of the five factors accounts for most of the total energy of y (t). In theory, the greater the number of regression curves selected,

the higher the approximation. Since the disease prediction curve is a quantitative index describing the probability of the user's disease, it is necessary to do so

Carrying out normalization treatment to obtain

In addition, the number of regression curves should be selected according to actual needs, and in this example, only five most important pathogenic factors are selected as regression curves, but not limited to five.

According to said scheme, in step 4, use is made of

And (6) predicting the disease risk. In particular, synthesis

And its first derivative

Two important indexes are used for comprehensively studying and judging the disease risk of a user in a certain future time interval. This example provides a reference method by

The disease risk is judged according to the maximum value and gradient change condition of the patient. It should be noted that this reference method is not exclusive, but any other method must follow the maximum likelihood principle.

The detailed implementation and flow of the above example are shown in fig. 1.

And verifying the reliability of the scheme. In the embodiment, the real data of a certain user is randomly selected and calculated by utilizing a Python program to obtain

Then recording the real illness of the user with

Performing comparative analysis to verify

Agreement with actual disease records.

Let the user number be A, and the personal information be: age 26, male, living in fujian province, and data sampling time ranged from 2016 (1 month) to 2020 (6 months), and was recorded as sampling period a 1. Let the sampling interval of the sampling period A1 be month, so that the sampling period A1 has 54 sampling time points in total, and the sampling pre-processing value is denoted as F_S(t_i)。

And (4) preparing data. The description of the history of colds and their health status of the user a at the sampling period a1 is obtained as follows: user A was hospitalized with bronchitis due to cold in 3 months of 2016, in 1 st of 2018, and in 2 months of 2020, with pneumonia. In addition, user a had a cold record in 2016 in 12 th late month and in 2018 in 7 th late monthThere are cold records, and there are cold records from 5 late to 6 late of the year 2020. From the above record, the "sick" event sampling time point of user a can be determined, corresponding to F_S(t_i) The value is assigned to 1. For the rest of sampling time points, a Gaussian normal distribution N (0,1) is adopted to randomly generate a real number between 0 and 1, and the real number is assigned to F_S(t_i). The final pretreatment result F_S(t_i) The details are shown in Table 1.

Table 1 cold history data preprocessing result F of user a at sampling period a1_S(t_i)

Time of day	F_s(t_i)	Time of day	F_s(t_i)	Time of day	F_s(t_i)	Time of day	F_s(t_i)	Time of day	F_s(t_i)
										2016.01	0.9	2017.01	0.3	2018.01	1.0	2019.01	0.2	2020.01	0.8
2016.02	0.9	2017.02	0.2	2018.02	0.9	2019.02	0.3	2020.02	1.0
										2016.03	1.0	2017.03	0.2	2018.03	0.6	2019.03	0.4	2020.03	0.4
2016.04	0.6	2017.04	0.3	2018.04	0.4	2019.04	0.3	2020.04	0.3
										2016.05	0.4	2017.05	0.3	2018.05	0.3	2019.05	0.2	2020.05	1.0
2016.06	0.3	2017.06	0.2	2018.06	0.2	2019.06	0.1	2020.06	1.0
										2016.07	0.2	2017.07	0.3	2018.07	1.0	2019.07	0.2
2016.08	0.1	2017.08	0.4	2018.08	0.5	2019.08	0.3
										2016.09	0.3	2017.09	0.3	2018.09	0.4	2019.09	0.2
2016.10	0.4	2017.10	0.2	2018.10	0.2	2019.10	0.3
										2016.11	0.5	2017.11	0.4	2018.11	0.5	2019.11	0.5
2016.12	1.0	2017.12	0.9	2018.12	0.4	2019.12	0.7

Heart rate data for user a during sampling period a1 is collected and pre-processed. It should be noted that the heart rate of a normal person in a resting state is generally 60 to 100 beats/minute. Therefore, the heart rate data preprocessing result | F 'of the user A in the sampling period A1 is obtained by 60 times/division of the standard heart rate'_H(t_i) See table 2 for details.

Table 2 heart rate data preprocessing result | F of user a at sampling period a1′_H(t_i)|

Time of day	\|F′_H(t_i)\|	Time of day	\|F′_H(t_i)\|	Time of day	\|F′_H(t_i)\|	Time of day	\|F′_H(t_i)\|	Time of day	\|F′_H(t_i)\|
										2016.01	5	2017.01	6	2018.01	2	2019.01	3	2020.01	5
2016.02	3	2017.02	3	2018.02	3	2019.02	2	2020.02	8
										2016.03	4	2017.03	2	2018.03	2	2019.03	1	2020.03	4
2016.04	2	2017.04	1	2018.04	2	2019.04	1	2020.04	2
										2016.05	1	2017.05	1	2018.05	1	2019.05	2	2020.05	5
2016.06	2	2017.06	2	2018.06	1	2019.06	1	2020.06	4
										2016.07	1	2017.07	1	2018.07	3	2019.07	2
2016.08	3	2017.08	1	2018.08	1	2019.08	1
										2016.09	1	2017.09	3	2018.09	2	2019.09	1
2016.10	1	2017.10	2	2018.10	2	2019.10	2
										2016.11	4	2017.11	4	2018.11	1	2019.11	3
2016.12	6	2017.12	5	2018.12	1	2019.12	5

Respiratory rate data for user a at sampling period a1 is acquired and pre-processed. It should be noted that the normal adult respiratory rate is 12-20 per minute. Therefore, with 16 breaths per minute as the standard respiratory rate, the respiratory rate pre-data processing result | F 'of the user A in the sampling period A1 is obtained'_B(t_i) See table 3 for details.

TABLE 3 result of respiratory frequency data preprocessing of user A at sampling period A1 | F'_B(t_i)|

Time of day	\|F′_B(t_i)\|	Time of day	\|F′_B(t_i)\|	Time of day	\|F′_B(t_i)\|	Time of day	\|F′_B(t_i)\|	Time of day	\|F′_B(t_i)\|
										2016.01	8	2016.12	4	2017.11	3	2018.10	2	2019.09	2
2016.02	4	2017.01	5	2017.12	5	2018.11	2	2019.10	1
										2016.03	4	2017.02	3	2018.01	6	2018.12	3	2019.11	2
2016.04	3	2017.03	2	2018.02	3	2019.01	3	2019.12	4
										2016.05	2	2017.04	1	2018.03	1	2019.02	1	2020.01	5
2016.06	1	2017.05	1	2018.04	1	2019.03	1	2020.02	6
										2016.07	1	2017.06	1	2018.05	2	2019.04	1	2020.03	2
2016.08	1	2017.07	2	2018.06	1	2019.05	2	2020.04	1
										2016.09	1	2017.08	1	2018.07	1	2019.06	2	2020.05	4
2016.10	2	2017.09	2	2018.08	2	2019.07	1	2020.06	3
										2016.11	2	2017.10	1	2018.09	1	2019.08	1

Blood pressure data of user a at sampling period a1 is collected and pre-processed. It should be noted that the systolic blood pressure in the resting state of a normal adult ranges from 90 to 139 mmHg. Therefore, with 115mmHg as the blood pressure average value, the blood pressure data preprocessing result | F 'of the user A at the sampling period A1 is obtained'_P(t_i) See table 4 for details.

TABLE 4 blood pressure data preprocessing result | F 'of user A at sampling period A1'_P(t_i)|

Time of day	\|F′_P(t_i)\|	Time of day	\|F′_P(t_i)\|	Time of day	\|F′_P(t_i)\|	Time of day	\|F′_P(t_i)\|	Time of day	\|F′_P(t_i)\|
										2016.01	12	2016.12	6	2017.11	4	2018.10	3	2019.09	3
2016.02	6	2017.01	6	2017.12	8	2018.11	4	2019.10	2
										2016.03	3	2017.02	5	2018.01	7	2018.12	5	2019.11	5
2016.04	4	2017.03	4	2018.02	5	2019.01	6	2019.12	6
										2016.05	2	2017.04	4	2018.03	2	2019.02	5	2020.01	9
2016.06	1	2017.05	3	2018.04	3	2019.03	4	2020.02	10
										2016.07	3	2017.06	2	2018.05	1	2019.04	4	2020.03	4
2016.08	4	2017.07	3	2018.06	1	2019.05	2	2020.04	3
										2016.09	3	2017.08	2	2018.07	2	2019.06	3	2020.05	7
2016.10	2	2017.09	1	2018.08	2	2019.07	1	2020.06	5
										2016.11	3	2017.10	3	2018.09	2	2019.08	1

Consistency verification experiment 1. By usingAnd obtaining a disease prediction curve from data before the month 1 of 2018, and predicting whether the disease risk of the user in the month 1 of 2018 is consistent with the actual disease condition of the user in the month by using the curve. User a was known to be hospitalized with bronchitis due to colds in late 1 month of 2018. Historical data of the air temperature in 2018 of Ningde City in 1 month are collected and detailed in Table 5. Note that, in the following table, the temperature history data and | F'_T(t_i) The units are all in ℃.

TABLE 5 Ningde City 2018 years 1 month temperature historical data

Time of day	Temperature of	Time of day	Temperature of	Time of day	Temperature of	Time of day	Temperature of	Time of day	Temperature of
										01.01	11.5	01.08	9.5	01.15	16	01.22	15.5	01.29	7
01.02	15	01.09	14	01.16	15.5	01.23	13	01.30	7
										01.03	17	01.10	7.5	01.17	15	01.24	10.5	01.31	5.5
01.04	14.5	01.11	6.5	01.18	15.5	01.25	10
										01.05	12.5	01.12	7.5	01.19	16.5	01.26	7.5
01.06	10.5	01.13	10.5	01.20	13	01.27	8.5
										01.07	14	01.14	13.5	01.21	13	01.28	8

The data in table 5 were preprocessed and the results are detailed in table 6.

TABLE 6 Ningde City 2018 year 1 month air temperature historical data preprocessing result | F'_T(t_i)|

Time of day	\|F′_T(t_i)\|	Time of day	\|F′_T(t_i)\|	Time of day	\|F′_T(t_i)\|	Time of day	\|F′_T(t_i)\|
								01.01-01.02	3.5	01.09-01.10	6.5	01.17-01.18	0.5	01.26-01.27	1
01.02-01.03	2	01.10-01.11	1	01.18-01.19	1	01.27-01.28	0.5
								01.03-01.04	2.5	01.11-01.12	1	01.19-01.20	3.5	01.28-01.29	1
01.04-01.05	2	01.12-01.13	3	01.20-01.21	0	01.29-01.30	0
								01.05-01.06	2	01.13-01.14	3	01.21-01.22	2.5	01.30-01.31	1.5
01.06-01.07	3.5	01.14-01.15	2.5	01.22-01.23	2.5
								01.07-01.08	4.5	01.15-01.16	0.5	01.23-01.24	2.5
01.08-01.09	4.5	01.16-01.17	0.5	01.24-01.25	0.5

y_S(t)＝0.23sin(2π×2.96t+1.12)；

y_T(t)＝0.89sin(2π×3.02t-3.56)；

y_H(t)＝1.13sin(2π×4.01t+7.05)；

y_B(t)＝1.90sin(2π×3t+1.41)；

y_P(t)＝1.79sin(2π×2t+1.73)。

carrying out scale transformation on the regression curve, and giving the same weight to obtain the optimal approximate solution of the disease prediction curve of the user A in 1 month in 2018

To pair

Carrying out normalization treatment to obtain

See table 7 and figure 2 for details.

Table 7 user a disease prediction curve values (2018.01)

From table 7 and fig. 2, it can be seen: in 2018, the disease probability of the user A decreases in a gradient from 1 month to 10 months, reaches a minimum value of 0.102 in 10 months in 1 month, and increases in a gradient from 11 days in 1 month to a maximum value of 0.341 in 31 months in 1 month.

In this embodiment, the gradient rising period is defined as a cold latency period, and the gradient falling period is defined as a cold recovery period. According to the record that the user A hospitalizes in late 1 month in 2018 for bronchitis caused by cold, the result better reflects the actual situation, namely the sick period of the user A is just in the cold latent period.

Consistency verification experiment 2. And obtaining a disease prediction curve by using data before 2 months in 2020, and predicting whether the disease risk of the user in 2 months in 2020 is consistent with the actual disease condition of the user in the month. User a was known to be hospitalized with cold in the middle of 2 months of 2020 for pneumonia. Historical data of the air temperature of Ningde City in 2020 and 2 months are collected and shown in Table 8 in detail.

TABLE 8 Ningde City temperature History data of 2020 and 2 months

Time of day	Temperature of	Time of day	Temperature of	Time of day	Temperature of	Time of day	Temperature of	Time of day	Temperature of
										02.01	10.5	02.07	12.5	02.13	16	02.19	11	02.25	20
02.02	11.5	02.08	11	02.14	15.5	02.20	14	02.26	16.5
										02.03	11.5	02.09	9.5	02.15	15.5	02.21	15	02.27	14.5
02.04	11	02.10	12	02.16	8	02.22	13.5	02.28	17
										02.05	10.5	02.11	15	02.17	7	02.23	14.5	02.29	19.5
02.06	10	02.12	17.5	02.18	8	02.24	17

The data in Table 8 were preprocessed and the results are shown in Table 9.

TABLE 9 Ningde City temperature historical data of year 2020 and month 2 | F'_T(t_i)|

Time of day	\|F′_T(t_i)\|	Time of day	\|F′_T(t_i)\|	Time of day	\|F′_T(t_i)\|	Time of day	\|F′_T(t_i)\|
								02.01-02.02	1	02.09-02.10	2.5	02.17-02.18	1	02.25-02.26	3.5
02.02-02.03	0	02.10-02.11	3	02.18-02.19	3	02.26-02.27	2
								02.03-02.04	0.5	02.11-02.12	2.5	02.19-02.20	3	02.27-02.28	2.5
02.04-02.05	0.5	02.12-02.13	1.5	02.20-02.21	1	02.28-02.29	2.5
								02.05-02.06	0.5	02.13-02.14	0.5	02.21-02.22	1.5
02.06-02.07	2.5	02.14-02.15	0	02.22-02.23	1
								02.07-02.08	1.5	02.15-02.16	7.5	02.23-02.24	2.5
02.08-02.09	1.5	02.16-02.17	1	02.24-02.25	3

y_S(t)＝0.20sin(2π×4t+1.15)；

y_T(t)＝-0.56sin(2π×1.99t+0.66)；

y_H(t)＝1.69sin(2π×4.98t+1.85)；

y_B(t)＝0.96sin(2π×7.02t+1.27)；

y_P(t)＝-1.03sin(2π×6t-1.58)。

carrying out scale transformation on the regression curve, and giving the same weight to obtain the optimal approximate solution of the disease prediction curve of the user A in 2 months in 2020

To pair

Carrying out normalization treatment to obtain

See table 10 and figure 3 for details.

TABLE 10 user A disease prediction Curve values (2020.02)

From table 10 and fig. 3, it can be seen that: from 12 days 2 month to 25 days 2 month in 2020, the probability of illness of user a decreases in a gradient manner, reaching a minimum value of 0.185 at 25 days 2 month, and from 4 days 2 month to 11 days 2 month, the probability of illness increases in a gradient manner, reaching a maximum value of 0.605 at 11 days 2 month. According to the record of hospitalization of the user A for pneumonia caused by cold in the middle 2 months of 2020, the result better reflects the actual situation that a part of the sick period of the user A is in the cold latent period and the maximum probability value of the sick period is also in the cold latent period.

Consistency verification experiment 3. And obtaining a disease prediction curve by using data before 6 months in 2020, and predicting whether the disease risk of the user in 6 months in 2020 is consistent with the actual disease condition of the user in the month by using the curve. User a is known to have a cold record in late 6 months of 2020. Now, the temperature data of Ningde City in 2020 and 6 months is collected and shown in Table 11 in detail.

TABLE 11 Ningde City temperature History data in 6 months 2020

Time of day	Temperature of	Time of day	Temperature of	Time of day	Temperature of	Time of day	Temperature of	Time of day	Temperature of
										06.01	28.5	06.07	25	06.13	30.5	06.19	30.5	06.25	29.5
06.02	28	06.08	25.5	06.14	30	06.20	29	06.26	29
										06.03	29	06.09	27	06.15	30.5	06.21	31	06.27	27.5
06.04	26.5	06.10	27	06.16	31.5	06.22	31	06.28	28
										06.05	28.5	06.11	28	06.17	32	06.23	31.5	06.29	28.5
06.06	24.5	06.12	29.5	06.18	32	06.24	31.5	06.30	28.5

The data in Table 11 were preprocessed and the results are shown in Table 12.

Table 12 Ningde City of 6-month-2020 air temperature historical data preprocessing result | F'_T(t_i)|

Time of day	\|F′_T(t_i)\|	Time of day	\|F′_T(t_i)\|	Time of day	\|F′_T(t_i)\|	Time of day	\|F′_T(t_i)\|
								06.01-06.02	0.5	06.09-06.10	0	06.17-06.18	0	06.25-06.26	0.5
06.02-06.03	1	06.10-06.11	1	06.18-06.19	1.5	06.26-06.27	1.5
								06.03-06.04	2.5	06.11-06.12	1.5	06.19-06.20	1.5	06.27-06.28	0.5
06.04-06.05	2	06.12-06.13	1	06.20-06.21	2	06.28-06.29	0.5
								06.05-06.06	4	06.13-06.14	0.5	06.21-06.22	0	06.29-06.30	0
06.06-06.07	0.5	06.14-06.15	0.5	06.22-06.23	0.5
								06.07-06.08	0.5	06.15-06.16	1	06.23-06.24	0
06.08-06.09	1.5	06.16-06.17	0.5	06.24-06.25	2

y_S(t)＝-0.20sin(2π×4.01t-1.89)；

y_T(t)＝0.67sin(2π×2.98t+2.84)；

y_H(t)＝-1.67sin(2π×4.99t+2.13)；

y_B(t)＝0.55sin(2π×7t+0.56)；

y_P(t)＝2.32sin(2π×5.97t+1.82)。

carrying out scale transformation on the regression curve, and giving the same weight to obtain the optimal approximate solution of the disease prediction curve of the user A in 6 months in 2020

To pair

Carrying out normalization treatment to obtain

See table 13 and fig. 4 for details.

Table 13 user a disease prediction curve values (2020.06)

From table 13 and fig. 4, it can be seen that: the disease probability of the user A is increased in a gradient manner in three time periods of 6 month and 1 day to 6 month and 4 days in 2020, 6 month and 9 days to 6 month and 16 days in 6 month and 22 days to 6 month and 28 days in 6 month, the disease probability of the user A is decreased in a gradient manner in the rest time periods, and the prediction result is better in accordance with the cold condition of the user A in the last ten days of 6 months.

Combining the above experiments, it can be concluded that: a disease prediction scheme based on Fourier transform can realize the prediction of the disease probability of a user in a certain time range in the future, through collecting the historical disease data (cold in the example) and health related data (temperature data, heart rate data, respiratory rate data and blood pressure data in the example) of the user, and preprocessing the data (quantitative coding and differential derivation in the example); a nonlinear regression method (a least square method is adopted in the example) is adopted to obtain a trigonometric function type regression curve of the user historical sick data and health related data; transforming the regression curve into a disease prediction curve by using an inverse Fourier transform principle; and predicting the disease risk of the user in a future period by using the optimal approximate solution of the disease prediction curve. The scheme has strong openness, has a wide selection range of user health related data, and can reflect pathogenic mechanisms of various factors.

Claims

1. A fourier transform-based disease prediction scheme, comprising the steps of:

2. A fourier transform-based disease prediction scheme as claimed in claim 1, wherein step 1 collects user history disease data and health related data and preprocesses the data, wherein: the user history illness data refers to a function of the degree of a certain illness suffered by the user in a past period of time with respect to time variation; the user health related data refers to functions of various factors related to human health along with time, and includes but is not limited to external environment data, physiological index data and the like.

3. A fourier transform-based disease prediction scheme as claimed in claim 1, wherein step 1 collects user history disease data and health related data and preprocesses the data, wherein: the method comprises the steps of preprocessing historical sick records of a user in a quantization coding mode; the health-related data is preprocessed by signal processing methods including but not limited to derivation, difference, summation, integration, etc. to extract feature information.

4. The disease prediction protocol based on fourier transform of claim 1, wherein the step 2 employs a non-linear regression method, using the preprocessed data as a sample set, to solve a trigonometric regression curve, and is characterized in that: non-linear regression includes, but is not limited to, conventional mathematical algorithms employing machine learning algorithms, least squares, maximum likelihood estimation, interpolation, fitting, and the like.

5. A fourier transform-based disease prediction scheme as claimed in claim 1 wherein the step 3 transforms the regression curve into a disease prediction curve using fourier transform principle, wherein: the transformation method is to perform Fourier series expansion or Fourier transformation on a disease prediction curve to be solved.

6. A fourier transform-based disease prediction scheme as claimed in claim 1 wherein the step 3 transforms the regression curve into a disease prediction curve using fourier transform principle, wherein: before obtaining the disease prediction curve, the regression curve is subjected to the following wave shaping treatment:

7. A fourier transform-based disease prediction scheme as claimed in claim 1 wherein the step 3 transforms the regression curve into a disease prediction curve using fourier transform principle, wherein: the regression curve corresponds to important frequency spectrum components of the disease prediction curve to be solved, and the importance of the regression curve is that the sum of the energy of the frequency spectrum components corresponding to the regression curve accounts for most of the total energy of the frequency spectrum.

8. A fourier transform-based disease prediction scheme as claimed in claim 1 wherein the step 3 transforms the regression curve into a disease prediction curve using fourier transform principle, wherein: after the transformation process, the result function needs to be normalized in order to describe the prevalence probability. The function after normalization processing is used as the optimal approximate solution of the disease prediction curve to be solved. In an ideal case, when the number of regression curves is close to infinity, the optimal approximate solution will be equal to the exact solution.

9. A fourier transform-based disease prediction scheme as claimed in claim 1 wherein the step 3 transforms the regression curve into a disease prediction curve using fourier transform principle, wherein: the horizontal axis of the disease prediction curve is time, and the vertical axis of the disease prediction curve is the corresponding disease probability of a user at a certain moment.

10. The fourier transform-based disease prediction scheme as claimed in claim 1, wherein the optimal approximate solution of the disease prediction curve is used in step 4 to predict the disease risk of the user in a future period, and the method comprises: including but not limited to, comprehensive prediction of risk using disease prediction curves and their derivatives, and should follow the principle of maximum likelihood.