Financial data intelligent power data intelligent monitoring and prediction method and system

Abstract

The present application relates to the field of prediction analysis, and more particularly, to a power data intelligent monitoring prediction method and system based on financial digitalization, the method comprising: acquiring a plurality of load data segments, decomposing the load data segments to obtain periodic items and frequency items; calculating regular availability according to periodic analysis and frequency item calculation rules, the regular availability being positively correlated with regular persistence and regular reasoning; the regular persistence being positively correlated with periodic persistence and trend persistence, and the regular reasoning being positively correlated with periodic reasoning and trend reasoning; and adjusting the weighted coefficient of the prediction algorithm using the regular availability to achieve load data prediction. Thus, accurate power load prediction is achieved through adaptive weighted coefficients.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of predictive analysis. More particularly, the present application relates to a power data intelligent monitoring prediction method and system based on financial digitalization. BACKGROUND

[0002] Power load prediction plays an important role in power system operation and energy management. For example, by predicting power load, power companies can adjust production plans in a timely manner to prevent overproduction or underproduction. The accuracy of power load prediction determines the accuracy of power production plans, so how to accurately predict power load becomes a problem to be solved.

[0003] Exponential smoothing method is a commonly used prediction algorithm. The weighting coefficient in this algorithm is used to determine the reference amount of historical data variation rules during prediction analysis. The setting of the weighting coefficient directly affects the prediction accuracy. The patent application file CN112700288A shows a commodity price prediction method using exponential smoothing method. The file uses exponential smoothing method to predict commodity prices. The exponential smoothing method in the file uses a traditional method to set the weighting coefficient, that is, a fixed weighting coefficient is set to complete commodity price prediction at all times. Due to weather and other factors, the variation of power load data in time sequence is large. If a fixed weighting coefficient is used, the reference amount of historical data variation rules during power prediction at each time is constant. This way of setting the weighting coefficient is not suitable for power load prediction scenarios with large variation differences. For example, the future power load data is similar to the current power load data. Because of the inappropriate weighting coefficient, the reference amount of historical variation rules is excessive, which ignores the reference amount of current power load data, and thus cannot accurately predict the future power load data.

[0004] Therefore, how to set an appropriate weighting coefficient is the focus of the present application. SUMMARY

[0005] To solve the problem of how to set an appropriate weighting coefficient, the present application provides a power data intelligent monitoring prediction method and system based on financial digitalization.

[0006] In the first aspect, the present application provides a power data intelligent monitoring prediction method based on financial digitalization, which adopts the following technical solution:

[0007] The power data intelligent monitoring prediction method based on financial digitalization includes the following steps:

[0008] Obtain the time sequence of load data;

[0009] The load data time sequence is segmented to obtain several load data segments, and the load data segments are decomposed to obtain periodic items and trend items;

[0010] The regularity availability is positively correlated with regularity persistence and regularity reasoning;

[0011] The regularity persistence is positively correlated with periodic persistence and trend persistence, the periodic persistence represents the frequency domain information similarity of the periodic items of two adjacent load data segments, and the trend persistence represents the fitting relationship similarity of the trend items of two adjacent load data segments;

[0012] The regularity reasoning is positively correlated with periodic reasoning and trend reasoning, the periodic reasoning represents the autocorrelation of the frequency domain information variation amount of the periodic items of all two adjacent load data segments, and the trend reasoning represents the autocorrelation of the fitting relationship variation amount of the trend items of all two adjacent load data segments;

[0013] The regularity availability is adjusted to adjust the weighting coefficient of the prediction algorithm, and the adjusted weighting coefficient is negatively correlated with the regularity availability, so as to realize load data prediction.

[0014] The present application can accurately describe the content of the historical rules reserved in the data rule at the current time through the regular usability, and further provides a basis for accurately setting the weighting coefficient. Further, when calculating the regular usability, not only the directly reserved historical rule information is considered, but also the historical rule information that can be analyzed by reasoning is considered, so that the present application more comprehensively describes the content of the historical rules reserved in the data rule at the current time. Further, when analyzing the directly reserved historical rule information, the period information and the trend information are respectively researched, so that the directly reserved historical rule information can be more accurately described. Further, when analyzing the historical rule information that can be analyzed by reasoning, the period information and the trend information are respectively researched, so that the historical rule information that can be analyzed by reasoning can be more accurately described. Further, when calculating the period persistence, the frequency domain information is introduced to accurately describe the similarity of the period information of each load data segment, and then by analyzing the similarity of the frequency domain information in the adjacent two load data segments, the situation that the period information of each load data segment reserves the period information of the previous load data segment can be accurately reflected, so as to provide a basis for accurately analyzing the content of the historical rules reserved in the data rule at the current time. Further, when calculating the trend persistence, by analyzing the similarity of the fitting relationship in the trend items in the previous and subsequent load data segments, the situation that the trend information of each load data segment reserves the trend information of the previous load data segment can be accurately described, so as to provide a basis for accurately analyzing the content of the historical rules reserved in the data rule at the current time. Further, when calculating the period reasoning, by analyzing the autocorrelation of the period information variation amount of all adjacent two load data segments, the rule of the period information variation amount can be accurately analyzed, so as to accurately reflect how much rule information in the data rule at the current time can be obtained by reasoning the historical rule information, and thus provide a basis for accurately analyzing the content of the historical rules reserved in the data rule at the current time. Further, when calculating the trend reasoning, by analyzing the autocorrelation of the trend information variation amount of all adjacent two load data segments, the rule of the trend information variation amount can be accurately analyzed, so as to accurately reflect how much rule information in the data rule at the current time can be obtained by reasoning the historical rule information, and thus provide a basis for accurately analyzing the content of the historical rules reserved in the data rule at the current time.

[0015] Preferably, the method for obtaining the period persistence comprises:

[0016] The frequency spectrum data of the period item of the load data segment is obtained, and the vector composed of each frequency and the corresponding amplitude in the frequency spectrum data is denoted as a frequency domain description vector, and all the frequency domain description vectors are arranged according to the frequency size to obtain a frequency domain description vector sequence.

[0017] Calculate the local periodicity of each load data segment, whereby the local periodicity is positively correlated with the similarity of the frequency domain description vector sequences of two adjacent load data segments.

[0018] The average of the local periodicity of all load data segments is taken as the periodicity.

[0019] Preferably, the local periodicity of each load data segment satisfies the following relationship:

[0020] Preferably, the local periodicity of each load data segment satisfies the following relationship:

[0021] ;

[0022] in, The frequency domain description vector sequence representing a load data segment. Each frequency domain describes the frequency in the vector. This represents the first frequency domain descriptor vector sequence of the preceding load data segment. Each frequency domain describes the frequency in the vector. This represents the first frequency domain description vector sequence of the load data segment. Each frequency domain describes the magnitude in the vector. This represents the first frequency domain descriptor vector sequence of the preceding load data segment. Each frequency domain describes the magnitude in the vector. This indicates the number of frequency domain description vectors contained in the frequency domain description vector sequence. This indicates the local periodicity of the load data segment.

[0023] When calculating the periodicity, this invention introduces frequency domain information to accurately describe the similarity of periodic information in each load data segment. Furthermore, by analyzing the similarity of frequency domain information in two adjacent load data segments, it can accurately reflect the retention of periodic information from the previous load data segment in each load data segment. This provides a basis for accurately analyzing the retention of historical patterns in the current data patterns.

[0024] Preferably, the method for obtaining the trend persistence includes:

[0025] The first fitting equation is obtained by fitting a polynomial to the trend components of each load data segment using the least squares method. The constant coefficients in the first fitting equation are then obtained, and all the constant coefficients of the first fitting equation are used to form a coefficient sequence.

[0026] Calculate the cosine similarity of the coefficient sequence of a load data segment with that of the previous load data segment, and denote it as the local trend persistence of the load data segment. The mean of the local trend persistence of all load data segments is denoted as the trend persistence.

[0027] When calculating trend persistence, this invention analyzes the similarity of the fitting relationship in the trend components of two consecutive load data segments to accurately describe the retention of trend information from the previous load data segment in each load data segment. This provides a basis for accurately analyzing the retention of historical patterns in the current data patterns.

[0028] Preferably, the method for obtaining periodic reasoning includes:

[0029] Calculate the periodic variation of each load data segment, and the periodic variation is positively correlated with the difference in the frequency description vector sequence of two adjacent load data segments;

[0030] The autocorrelation value obtained by autocorrelation detection of the periodic variations of all load data segments is used as the periodic inference value.

[0031] When calculating periodic inference, this invention analyzes the autocorrelation of the periodic information variation between all two adjacent load data segments to accurately analyze the regularity of the periodic information variation. This accurately reflects how much regularity information in the current data can be obtained through inference analysis of historical regularity information, thus providing a basis for accurately analyzing the content of historical regularity in the current data.

[0032] Preferably, the periodic variation satisfies the following relationship:

[0033] ;

[0034] in, The frequency domain description vector sequence representing a load data segment. Each frequency domain describes the frequency in the vector. This represents the first frequency domain descriptor vector sequence of the preceding load data segment. Each frequency domain describes the frequency in the vector. This represents the first frequency domain description vector sequence of the load data segment. Each frequency domain describes the magnitude in the vector. This represents the first frequency domain descriptor vector sequence of the preceding load data segment. Each frequency domain describes the magnitude in the vector. This indicates the number of frequency domain description vectors contained in the frequency domain description vector sequence. This indicates the periodic variation of the load data segment.

[0035] Preferably, the method for obtaining the trend inference includes:

[0036] The difference between 1 and the first cosine similarity is the trend change of the load data segment, and the first cosine similarity is the cosine similarity between the coefficient sequence of a load data segment and the coefficient sequence of the previous load data segment.

[0037] The autocorrelation value obtained by performing autocorrelation detection on the trend change of all load data segments is the trend inference.

[0038] In the present application, when calculating the trend inference, the autocorrelation of the trend information change of all adjacent two load data segments is analyzed to accurately analyze the regularity of the trend information change, so that the amount of regular information in the current data regularity can be accurately reflected by inferring and analyzing the historical regular information, thereby providing a basis for accurately analyzing the content of the historical regularity in the current data regularity.

[0039] Preferably, the method for obtaining the load data segment comprises:

[0040] The load data time sequence is segmented to obtain a unit segment; the first unit segment is taken as a first starting segment, a plurality of temporary data segments of the first starting segment are obtained, the temporary data segments are obtained by splicing the unit segments, the prediction error change of the temporary data segments of the first starting segment is calculated, the temporary data segment in which the prediction error change first appears and is greater than a change reference value is taken as the first load data segment, the change reference value is positively correlated with the prediction error changes of all the temporary data segments, in response to the existence of the unit segment behind the first load data segment, the next unit segment of the first load data segment is taken as a second starting segment, the second load data segment is obtained according to the second starting segment, and in response to the non-existence of the unit segment behind the second load data segment, the iteration is ended; and all the load data segments are obtained.

[0041] In the present application, when obtaining the load data segment, the division of the data segment is controlled according to the fitting rule of the subsequent data to the previous data, so that the data with similar rules is divided in a data segment, thereby providing a basis for subsequent analysis of the content of the historical regularity in the current data regularity.

[0042] Preferably, the prediction error change of the temporary data segment is calculated, comprising:

[0043] The fitting relationship obtained by fitting the temporary data segment is taken as a prediction model, the prediction model is used for prediction analysis to obtain load data prediction values of L future time points, the load data prediction values of the L future time points are taken as a prediction value data segment, and the prediction error of the temporary data segment is obtained by dividing the Euclidean distance between the prediction value data segment and the next unit segment of the temporary data segment by L.

[0044] The absolute value of the difference between the prediction error of the temporary data segment and the prediction error of the previous temporary data segment is taken as the prediction error variation.

[0045] In a second aspect, the present application provides a financial digitalization-based power data intelligent monitoring and prediction system, which adopts the following technical solution:

[0046] The financial digitalization-based power data intelligent monitoring and prediction system comprises a processor and a memory, and the memory stores computer program instructions.

[0047] By adopting the above technical solution, the financial digitalization-based power data intelligent monitoring and prediction method is generated into a computer program and stored in the memory to be loaded and executed by the processor, so that a terminal device is manufactured according to the memory and the processor, and the use is facilitated.

[0048] The present application has the following advantages: the present application accurately describes the content of the historical rules retained in the data rule at the current time through the regular usability, thereby providing a basis for accurately setting the weighting coefficient; further, when calculating the regular usability, not only the directly retained historical rule information is considered, but also the historical rule information that can be analyzed by reasoning is considered, so that the present application more comprehensively describes the content of the historical rules retained in the data rule at the current time;

[0049] Further, when analyzing the directly retained historical rule information, the periodic information and the trend information are respectively researched, so as to more accurately describe the directly retained historical rule information; further, when analyzing the historical rule information that can be analyzed by reasoning, the periodic information and the trend information are respectively researched, so as to more accurately describe the historical rule information that can be analyzed by reasoning;

[0050] Further, when calculating the periodic persistence, the frequency domain information is introduced to accurately describe the similarity of the periodic information of each load data segment, and then by analyzing the similarity of the frequency domain information in the adjacent two load data segments, the retention of the periodic information in the previous load data segment in each load data segment can be accurately reflected, thereby providing a basis for accurately analyzing the content of the historical rules retained in the data rule at the current time;

[0051] Further, when calculating the trend persistence, by analyzing the similarity of the fitting relationship in the trend items in the previous and subsequent two load data segments, the retention of the trend information in the previous load data segment in each load data segment can be accurately described, thereby providing a basis for accurately analyzing the content of the historical rules retained in the data rule at the current time;

[0052] Further, when calculating the periodicity reasoning, the application analyzes the autocorrelation of the periodicity information variation of all adjacent two load data segments, accurately analyzes the law of the periodicity information variation, and accurately reflects how much law information in the current time data law can be obtained by reasoning and analyzing the historical law information, thereby providing a basis for accurately analyzing the historical law content in the current time data law.

[0053] Further, when calculating the periodicity reasoning, the application analyzes the autocorrelation of the periodicity information variation of all adjacent two load data segments, accurately analyzes the law of the periodicity information variation, and accurately reflects how much law information in the current time data law can be obtained by reasoning and analyzing the historical law information, thereby providing a basis for accurately analyzing the historical law content in the current time data law.

[0054] Further, when calculating the periodicity reasoning, the application analyzes the autocorrelation of the periodicity information variation of all adjacent two load data segments, accurately analyzes the law of the periodicity information variation, and accurately reflects how much law information in the current time data law can be obtained by reasoning and analyzing the historical law information, thereby providing a basis for accurately analyzing the historical law content in the current time data law. BRIEF DESCRIPTION OF DRAWINGS

[0055] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which a number of embodiments of the application are shown by way of illustration, now embodying the principle of the application, and wherein:

[0056] Figure 1 is a step flow chart of the power data intelligent monitoring and prediction method based on financial digitalization according to the embodiment of the application. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are some but not all of the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the application.

[0058] The specific embodiments of the application will be described in detail below with reference to the drawings.

[0059] The embodiment of the application discloses a power data intelligent monitoring and prediction method and system based on financial digitalization, referring to Figure 1 , including steps S1-S4:

[0060] S1. Obtain a load data time sequence.

[0061] Specifically, the load data of the power company is collected every n hours, and N times of collection are performed. The load data at all times is arranged in time sequence to obtain the load data time sequence. n represents a preset collection interval, and N represents a preset collection number. In this embodiment, n is taken as 1, and N is taken as 2000. Other embodiments can take other values, and the embodiment is not specifically limited.

[0062] S2. Segment the load data time sequence to obtain a plurality of load data segments, and decompose the load data segments to obtain periodic items and trend items.

[0063] S20: Segment the load data time sequence to obtain a plurality of load data segments.

[0064] It should be noted that, in order to set a suitable weighting coefficient, the change of the historical law needs to be analyzed. Due to the influence of factors such as temperature, the power load is unstable in time sequence, and thus the power load has different change laws in different stages. In order to better analyze the change of the historical law of the power load, the data with similar change laws need to be segmented together.

[0065] Optionally, as an embodiment, segmenting the load data time sequence to obtain a plurality of load data segments includes:

[0066] The load data time sequence is uniformly segmented into data segments with a length of M, denoted as load data segments. M represents a preset segment length. In this embodiment, M is taken as 100 for description, and other embodiments can take other values, and the embodiment is not specifically limited.

[0067] It should be noted that, since the predetermined segment length is artificially given, it is not set according to the data change law, and thus the data segments obtained by this segmentation method may segment the data with similar change laws in different segments, and may also segment the data with different change laws in one data segment. Therefore, the segmentation method will affect the accuracy of subsequent setting of the weighting coefficient.

[0068] Preferably, as an example, segmenting the load data time sequence to obtain a plurality of load data segments includes:

[0069] The load data time sequence is evenly divided into several data segments with a length of L, denoted as unit segments; the first unit segment is taken as a first starting segment, several temporary data segments of the first starting segment are obtained, the prediction error variation of the temporary data segments of the first starting segment is calculated, the temporary data segment in which the prediction error variation first appears and is greater than the variation reference value among all the temporary data segments of the first starting segment is taken as a first load data segment, in response to the existence of a unit segment after the first load data segment, the next unit segment of the first load data segment is taken as a second starting segment, several temporary data segments of the second starting segment are obtained, the prediction error variation of the temporary data segments of the second starting segment is calculated, the temporary data segment in which the prediction error variation first appears and is greater than the variation reference value among all the temporary data segments of the second starting segment is taken as a second load data segment, in response to the non-existence of a unit segment after the second load data segment, the iteration is ended, and all the load data segments are obtained. L represents a preset unit segment length, and the embodiment takes 10 as an example for description, and other embodiments can take other values, and the embodiment does not make specific limitation.

[0070] Preferably, as an example, the calculation method of the prediction error variation of the temporary data segment is as follows:

[0071] The least square method is used to fit a polynomial to obtain a fitting relationship, denoted as a prediction model, the load data at future L time points is fitted by using the prediction model, denoted as a load data prediction value, the load data prediction value at the future L time points is used to form a prediction value data segment, and the prediction error of the temporary data segment is obtained by dividing the Euclidean distance between the prediction value data segment and the next unit segment of the temporary data segment by L.

[0072] The absolute value of the difference between the prediction error of the temporary data segment and the prediction error of the previous temporary data segment is taken as the prediction error variation.

[0073] In the above embodiments, the variation reference value is involved, and the determination method of the variation reference value needs to be described as follows, including:

[0074] The twice of the mean of the prediction error variations of all the temporary data segments before the temporary data segment is taken as the variation reference value.

[0075] In the above embodiments, the temporary data segment is involved, and the determination method of the temporary data segment needs to be described as follows, including:

[0076] The starting data segment is spliced with 1 unit segment, 2 unit segments, …, and S unit segments after the starting data segment to obtain several temporary data segments. S represents the number of unit segments after the starting data segment.

[0077] For example, the start data segment is spliced with the following 1 unit segment to obtain a temporary data segment, and the start data segment is spliced with the following S data segments to obtain a temporary data segment.

[0078] It should be noted that the amount of data in the unit segment is small, and therefore a plurality of data segments can have a change rule, and therefore in order to divide the data belonging to the same change rule in a data segment, the unit segment needs to be merged. Since when a data has the same change rule, the fitting relationship of the data is similar, and therefore the fitting relationship of the previous data can better fit the following data, which indicates that the following data can be merged with the previous data. Therefore, the data segment division based on this theory can better divide the data with similar data rules together, and divide the data with large data rule differences in different data segments. To provide a basis for the following rule analysis.

[0079] S21: decompose the load data segment to obtain a period component and a trend component.

[0080] Optionally, as an example, decomposing the load data segment to obtain a period component and a trend component includes: using an ARIMA algorithm to decompose each load data segment to obtain a period component and a trend component of each load data segment.

[0081] It should be noted that the process of using an ARIMA algorithm to decompose each load data segment to obtain a period component and a trend component of each load data segment is a prior art, which will not be described here.

[0082] S3: calculate the rule availability, the rule availability is positively correlated with the rule persistence and the rule reasoning; the rule persistence is positively correlated with the period persistence and the trend persistence, the period persistence indicates the frequency domain information similarity of the period components of two adjacent load data segments, and the trend persistence indicates the fitting relationship similarity of the trend components of two adjacent load data segments; the rule reasoning is positively correlated with the period reasoning and the trend reasoning, the period reasoning indicates the autocorrelation of the frequency domain information variation of the period components of all two adjacent load data segments, and the trend reasoning indicates the autocorrelation of the fitting relationship variation of the trend components of all two adjacent load data segments.

[0083] It should be noted that the weighting coefficient in the exponential smoothing method determines the reference amount of historical rules and the reference amount of current time data in the prediction analysis. In order to set a suitable weighting coefficient, it is necessary to determine how much historical variation rule can be retained in the current time data variation rule. The present embodiment uses the rule availability to reflect the retention of historical variation rules in the current time data variation rule.

[0084] It needs to be further explained that the case of retaining the historical law in the data variation law of the current time includes not only the historical law information retained directly, but also the historical law information obtained through reasoning. Therefore, when analyzing the law availability, the retained historical law case needs to be measured in combination with the information of the two aspects.

[0085] Preferably, as an example, the law availability is calculated, including:

[0086] The normalized value of the cumulative sum of the law persistence and the law reasoning is taken as the law availability.

[0087] It needs to be explained that the embodiment normalizes the cumulative sum of the law persistence and the law reasoning through the sigmoid function, and other embodiments can use other functions for normalization processing.

[0088] In the above embodiments, the law persistence is involved, and the method for obtaining the law persistence needs to be described below.

[0089] It needs to be explained that the retention of the law can be reflected by the remaining situation of the historical law, and the common data law includes the periodic law and the trend law, so the remaining situation of the historical law can be reflected by analyzing the remaining situation of the periodic law and the remaining situation of the trend law. In the embodiment, the periodic persistence is used to reflect the remaining situation of the periodic law, and the trend persistence is used to reflect the remaining situation of the trend law.

[0090] S30: Calculate the law persistence.

[0091] Preferably, as an example, the method for obtaining the law persistence includes steps S301-S303:

[0092] S301: Calculate the periodic persistence.

[0093] Preferably, as an example, the method for obtaining the periodic persistence includes:

[0094] Obtain the frequency spectrum data of the periodic item of the load data segment, and take the vector composed of each frequency and the corresponding amplitude in the frequency spectrum data as the frequency domain description vector, and arrange all the frequency domain description vectors according to the frequency size to obtain the frequency domain description vector sequence.

[0095] Calculate the local periodic persistence of each load data segment.

[0096] Take the mean value of the local periodic persistence of all load data segments as the periodic persistence.

[0097] Preferably, as an example, the local periodic persistence of each load data segment satisfies the relationship:

[0098] ;

[0099] in, The frequency domain description vector sequence representing a load data segment. Each frequency domain describes the frequency in the vector. This represents the first frequency domain descriptor vector sequence of the preceding load data segment. Each frequency domain describes the frequency in the vector. This represents the first frequency domain description vector sequence of the load data segment. Each frequency domain describes the magnitude in the vector. This represents the first frequency domain descriptor vector sequence of the preceding load data segment. Each frequency domain describes the magnitude in the vector. This indicates the number of frequency domain description vectors contained in the frequency domain description vector sequence. This indicates the local periodicity of the load data segment.

[0100] Understandably, the local periodicity of a load data segment can be reflected by how much of the periodicity it inherits from the previous load data segment. Specifically, a high degree of periodic similarity between the current load data segment and the previous one indicates that the current load data segment inherits a significant amount of the periodicity from the previous one. Cosine similarity can measure the similarity between two sequences. The traditional formula for calculating cosine similarity is... This method of calculating cosine similarity only considers the frequency similarity between the current load data segment and the previous load data segment. However, this analysis method does not consider the similarity in the content of corresponding frequencies between the current and previous load data segments. Periodic information contains not only frequency information but also frequency content information; amplitude can better reflect the content of each frequency. Therefore, in this embodiment, to reflect the similarity in the content of corresponding frequencies between the current and previous load data segments, [the following is omitted as it is not relevant to the main point]. As The weights, for Weighting is applied. The larger the value, the higher the frequency. and The higher the similarity in content, the better. This adjustment method allows us to measure not only the frequency similarity between the current load data segment and the previous data segment, but also the frequency content similarity. This results in a more accurate description of the periodicity similarity between the current load data segment and the previous data segment.

[0101] S302: Calculate the persistence of trends.

[0102] Preferably, as an example, methods for obtaining trend persistence include:

[0103] The trend sub-item of each load data segment is fitted by using the least square method to obtain a first fitting formula, and constant coefficients of each sub-item in the first fitting formula are obtained, for example, the first fitting formula is: wherein 4, 6, 3, 5, 9, and 7 are constant coefficients of the second fitting formula, and all constant coefficients of the first fitting formula constitute a coefficient sequence.

[0104] The cosine similarity of the coefficient sequence of a load data segment and a previous load data segment is calculated, denoted as the local trend persistence of the load data segment, and the average of the local trend persistence of all load data segments is denoted as the trend persistence.

[0105] S303: Calculate the rule persistence according to the cycle persistence and the trend persistence.

[0106] Preferably, as an example, the rule persistence is calculated according to the cycle persistence and the trend persistence, including:

[0107] The average of all data in the cycle sub-item is taken as a first average, the average of all data in the trend sub-item is taken as a second average, the first average is taken as the weight of the cycle persistence, and the second average is taken as the weight of the trend persistence, and the cycle persistence and the trend persistence are weighted and summed to obtain the rule persistence.

[0108] In the above embodiments, the rule inference is also involved, and the method for obtaining the rule inference will be described below.

[0109] It should be noted that in addition to the direct rules retained in the data change rule, rules that can be analyzed through inference are also included, for example, the change of the trend sub-item of the power load conforms to the rule of 1, 2, 4, 8, which is a step-by-step doubling rule. Such a change rule can be obtained through inference, and thus such a rule is also a rule available in the historical change rule.

[0110] S31: Calculate the rule inference.

[0111] Preferably, as an example, the method for obtaining the rule inference includes steps S310-S312:

[0112] S310: Calculate the cycle inference.

[0113] Preferably, as an example, the method for obtaining the cycle inference includes:

[0114] The cycle change amount of each load data segment is calculated.

[0115] ​The autocorrelation function is used to detect the periodic variation of all load data segments to obtain the first autocorrelation value, which is then used as the periodic inference value.

[0116] It should be noted that the periodic fluctuations have a strong autocorrelation, indicating that the patterns of the periodic fluctuations are quite obvious. This further suggests that by analyzing the changes in periodic information, we can deduce the following patterns.

[0117] Among them, as a preferred example, the relationship that the periodic variation satisfies is:

[0118] ;

[0119] in, The frequency domain description vector sequence representing a load data segment. Each frequency domain describes the frequency in the vector. This represents the first frequency domain descriptor vector sequence of the preceding load data segment. Each frequency domain describes the frequency in the vector. This represents the first frequency domain description vector sequence of the load data segment. Each frequency domain describes the magnitude in the vector. This represents the first frequency domain descriptor vector sequence of the preceding load data segment. Each frequency domain describes the magnitude in the vector. This indicates the number of frequency domain description vectors contained in the frequency domain description vector sequence. This indicates the periodic variation of the load data segment.

[0120] S311: Calculate trend inference.

[0121] Preferably, as an example, methods for obtaining trend inference include:

[0122] The difference between 1 and the first cosine similarity is taken as the trend change of the load data segment. The first cosine similarity is the cosine similarity between the coefficient sequence of a load data segment and the coefficient sequence of the previous load data segment.

[0123] The autocorrelation function is used to detect the trend changes of all load data segments to obtain a second autocorrelation value, which is then used as the trend inference value.

[0124] S312: Calculate regularity based on periodicity and trend inference.

[0125] Preferably, as an example, methods for obtaining regularity reasoning include:

[0126] The first mean value is taken as a weight of periodic reasoning, the second mean value is taken as a weight of trend reasoning, and the periodic reasoning and the trend reasoning are weighted and summed to obtain rule reasoning.

[0127] S4. Adjust the weighting coefficient of the prediction algorithm by using the rule availability to realize load data prediction.

[0128] S40: Adjust the weighting coefficient of the prediction algorithm by using the rule availability.

[0129] Optionally, as an example, adjusting the weighting coefficient of the prediction algorithm by using the rule availability comprises:

[0130] The adjusted weighting coefficient satisfies a relationship:

[0131]

[0132] wherein, represents a preset weighting coefficient, and in the embodiment, the preset weighting coefficient is 0.5; D represents rule availability, and the greater the rule availability is, the more information that can be referred to from historical rules, so that more historical rules need to be referred to when performing load prediction. Taking 0.5 as an example for description, other embodiments can take other values, and the embodiment is not specifically limited; D represents rule availability, and the greater the rule availability is, the more information that can be referred to from historical rules, so that more historical rules need to be referred to when performing load prediction. represents a function for performing normalization processing, represents a function for performing normalization processing, represents an adjusted weighting coefficient.

[0133] S41: Perform load prediction according to the adjusted weighting coefficient.

[0134] Optionally, as an example, performing load prediction according to the adjusted weighting coefficient comprises:

[0135] Taking the adjusted weighting coefficient as a weighting coefficient, based on the load data in the last load data segment, the load data prediction value at a future time is predicted by using an exponential smoothing method.

[0136] The embodiment of the application further discloses a power data intelligent monitoring and prediction system based on financial digitalization, comprising a processor and a memory, and the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the power data intelligent monitoring and prediction system based on financial digitalization is realized.

[0137] The system further comprises a communication bus and a communication interface and other components familiar to those skilled in the art, and the settings and functions thereof are known in the art, and thus will not be described here.

[0138] In this disclosure, a "storage medium" can be any available medium that can be accessed by a general purpose or special purpose computer system, apparatus, or device to store, retrieve, or store and retrieve information. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store or

[0139] While the present application has been illustrated and described in detail in the drawings and foregoing description, such illustration and description is to be considered illustrative or exemplary and not restrictive; the present application is not limited to the disclosed embodiments. Numerous alternative modifications of the methods, apparatuses, and embodiments of the present application described herein will be apparent in light of this disclosure to one skilled in the art and may be employed without departing from the broad spirit and scope of the present application. Accordingly, the specification and drawings are to be regarded as illustrative and not restrictive.

[0140] The above are only preferred embodiments of the present application, not to limit the protection scope of the present application, therefore: all equivalent changes made on the structure, shape, principle of the present application should be covered in the protection scope of the present application.

Claims

1. A power data intelligent monitoring and prediction method based on financial digitalization, characterized in that, The method comprises the following steps: obtaining a load data time sequence; segmenting the load data time sequence to obtain a plurality of load data segments, and decomposing each load data segment by using an ARIMA algorithm to obtain a periodic component and a trend component; calculating a regularity availability, which is positively correlated with regularity persistence and regularity reasoning; the regularity persistence is positively correlated with periodic persistence and trend persistence, the periodic persistence represents the similarity of frequency domain information of the periodic components of two adjacent load data segments, and the trend persistence represents the similarity of fitting relationship of the trend components of two adjacent load data segments; a periodic persistence obtaining method comprises the following steps: obtaining frequency spectrum data of the periodic component of the load data segment, taking a vector composed of each frequency and the corresponding amplitude in the frequency spectrum data as a frequency domain description vector, and arranging all the frequency domain description vectors according to the frequency to obtain a frequency domain description vector sequence; calculating a local periodic persistence of each load data segment, which is positively correlated with the similarity of the frequency domain description vector sequences of two adjacent load data segments; and taking the average value of the local periodic persistences of all the load data segments as the periodic persistence; a trend persistence obtaining method comprises the following steps: fitting a polynomial to the trend component of each load data segment by using a least square method to obtain a first fitting formula, obtaining a constant coefficient in the first fitting formula, and taking all the constant coefficients of the first fitting formula as a coefficient sequence; calculating a cosine similarity of the coefficient sequence of a load data segment and the coefficient sequence of a previous load data segment, and taking the cosine similarity as the local trend persistence of the load data segment; and taking the average value of the local trend persistences of all the load data segments as the trend persistence; the regularity reasoning is positively correlated with periodic reasoning and trend reasoning, the periodic reasoning represents the autocorrelation of the frequency domain information variation amount of the periodic components of all two adjacent load data segments, and the trend reasoning represents the autocorrelation of the fitting relationship variation amount of the trend components of all two adjacent load data segments; a periodic reasoning obtaining method comprises the following steps: calculating a periodic variation amount of each load data segment, which is positively correlated with the difference of the frequency domain description vector sequences of two adjacent load data segments; and taking an autocorrelation value obtained by autocorrelation detection of the periodic variation amounts of all the load data segments as the periodic reasoning; a trend reasoning obtaining method comprises the following steps: taking the difference between 1 and a first cosine similarity as a trend variation amount of a load data segment, the first cosine similarity being the cosine similarity of the coefficient sequence of the load data segment and the coefficient sequence of a previous load data segment; and taking an autocorrelation value obtained by autocorrelation detection of the trend variation amounts of all the load data segments as the trend reasoning; adjusting a weighting coefficient of a prediction algorithm by using the regularity availability, the adjusted weighting coefficient being negatively correlated with the regularity availability, so as to realize load data prediction.

2. The financial digitalization-based power data intelligent monitoring and prediction method according to claim 1, characterized in that, the local periodic persistences of the load data segments satisfy the following relationship: ； wherein, denotes the frequency in the k-th frequency domain description vector of the sequence of frequency domain description vectors representing the load data segment, denotes the frequency in the k-th frequency domain description vector of the sequence of frequency domain description vectors representing the preceding load data segment of the load data segment, denotes the amplitude in the k-th frequency domain description vector of the sequence of frequency domain description vectors representing the load data segment, denotes the amplitude in the k-th frequency domain description vector of the sequence of frequency domain description vectors representing the preceding load data segment of the load data segment, denotes the number of frequency domain description vectors contained in the sequence of frequency domain description vectors, denotes the local periodicity persistence of the load data segment.​​​​ 3. The financial digitalization-based power data intelligent monitoring and prediction method according to claim 1, characterized in that, the periodic variation amounts satisfy the following relationship: ； wherein, denotes a frequency in the m-th frequency domain description vector of a sequence of frequency domain description vectors representing a load data segment, denotes a frequency in the m-th frequency domain description vector of a sequence of frequency domain description vectors representing a preceding load data segment of the load data segment, denotes an amplitude in the m-th frequency domain description vector of a sequence of frequency domain description vectors representing the load data segment, denotes an amplitude in the m-th frequency domain description vector of a sequence of frequency domain description vectors representing a preceding load data segment of the load data segment, denotes a number of frequency domain description vectors contained in the sequence of frequency domain description vectors, denotes a cyclic variation amount of the load data segment.​​​​ 4. The financial digitalization-based power data intelligent monitoring and prediction method according to claim 1, characterized in that, a load data segment obtaining method comprises the following steps: The unit segment is segmented from the load data time sequence, the first unit segment is taken as a first starting segment, a plurality of temporary data segments of the first starting segment are obtained, the temporary data segments are obtained by splicing the unit segments, a prediction error variation of the temporary data segments of the first starting segment is calculated, a temporary data segment in which a prediction error variation first appears and is greater than a variation reference value is taken as a first load data segment, the variation reference value is positively correlated with prediction error variations of all previous temporary data segments, in response to the existence of a unit segment behind the first load data segment, a next unit segment of the first load data segment is taken as a second starting segment, a second load data segment is obtained according to the second starting segment, in response to the non-existence of a unit segment behind the second load data segment, iteration is ended, and all load data segments are obtained.

5. The financial digitalization-based power data intelligent monitoring and prediction method according to claim 4, characterized in that, The prediction error variation of the temporary data segment comprises: A fitting relationship obtained by fitting the temporary data segment is taken as a prediction model, a prediction analysis is performed by using the prediction model to predict, load data prediction values of L future time points are obtained, the load data prediction values of the L future time points form a prediction value data segment, a prediction error of the temporary data segment is obtained by dividing a Euclidean distance between the prediction value data segment and a next unit segment of the temporary data segment by L. An absolute value of a difference between the prediction error of the temporary data segment and a prediction error of a previous temporary data segment is taken as the prediction error variation.

6. The power data intelligent monitoring and prediction system based on financial digitalization, characterized in that, The method comprises: A processor and a memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, a financial data intelligentized power data intelligent monitoring and prediction method according to any one of claims 1-5 is realized.

Images