Input data normalization method and device for data prediction model and equipment

Abstract

The application provides an input data normalization method and device and equipment of a data prediction model, the method comprising: obtaining input data to be normalized, and limiting the value of the input data to a preset range when it is determined that the input data exceeds the preset range; determining the unit data length corresponding to each dimension vector according to the data length of the preset range and a preset dimension N, wherein the N is not less than 2; mapping the difference beta between the input data and the minimum threshold of the preset range into an N-dimensional vector with the absolute value of the first n dimensions being not greater than 1 and the N-n dimensions being 0, wherein the sum of the product of each dimension code value in the N-dimensional vector and the unit data length is the difference beta. By using the method provided by the application, the dimension normalization of the input data of the data prediction model can be realized, and the convergence speed and convergence effect of the data prediction model using the dimension normalized input data can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, and in particular to an input data normalization method and device for a data prediction model and equipment. BACKGROUND

[0002] With the rapid development of big data technology, data centers have also rapidly increased. A data center is usually composed of cabinets, Internet Technology (IT) servers, Uninterruptible Power Supply (UPS), air conditioners, temperature and humidity sensors, and other equipment. The above-mentioned equipment will generate a large amount of heat when running, and consume a large amount of electric energy. Therefore, reducing the Power Usage Effectiveness (PUE) through artificial intelligence has become a current development trend.

[0003] A data prediction model can be constructed using various machine learning methods such as artificial intelligence to reduce PUE, for example, inputting the equipment running data and environmental state data of a data center into a PUE prediction model to obtain the PUE data predicted by the PUE prediction model to determine an equipment running scheme that has a lower PUE; or inputting the equipment running data and environmental state data of a data center into a temperature prediction model to obtain the temperature predicted by the temperature prediction model to adjust the equipment in the data center. When using the above-mentioned machine learning method, in order to improve the accuracy of the machine learning model, the data input into the machine learning model will be normalized first. The standard 0-1 normalization method is to compare the difference between the value to be normalized and the minimum threshold value with the difference between the maximum threshold value and the minimum threshold value to calculate the normalized data, wherein the value to be normalized is required to be between the maximum threshold value and the minimum threshold value.

[0004] The range of variation of the input data of the above-mentioned data prediction model is often large, for example, IT power. The numerical range of IT power of different data centers is different, and the numerical range of IT power of the same data center at different times is also different. Therefore, when using the above-mentioned standard 0-1 normalization method, two problems will be encountered: (1) If the preset range between the maximum threshold value and the minimum threshold value is set to be large, for the data with a small range of variation in the input data to be normalized, the normalized value will be concentrated around a certain value, and the effect is not ideal; (2) If the preset range between the maximum threshold value and the minimum threshold value is set to be small, the above-mentioned standard 0-1 normalization method cannot process the input data outside the preset range. SUMMARY

[0005] The application provides an input data normalization method, device and equipment for a data prediction model, which solves the problem of the existing 0-1 normalization method for processing input data of a data prediction model, the normalization result being concentrated around a value, and the input data outside a preset range being unable to be processed.

[0006] In a first aspect, the application provides an input data normalization method for a data prediction model, which comprises:

[0007] obtaining input data to be normalized, and limiting the value of the input data to a preset range when it is determined that the input data is outside the preset range;

[0008] determining the unit data length corresponding to each dimension vector according to the data length of the preset range and a preset dimension N, wherein the N is not less than 2;

[0009] mapping the difference β between the input data and the minimum threshold of the preset range to an N-dimensional vector with the first n dimensions adopting non-zero encoding values with absolute values not greater than 1 and the N-n dimensions adopting encoding value 0, wherein the sum of the product of each dimension encoding value in the N-dimensional vector and the unit data length is the difference β.

[0010] Optionally, mapping the difference β between the input data and the minimum threshold of the preset range to an N-dimensional vector with the first n dimensions adopting non-zero encoding values with absolute values not greater than 1 and the N-n dimensions adopting encoding value 0 comprises:

[0011] dissolving the difference β between the input data and the minimum threshold of the preset range according to the unit data length to obtain n-1 unit data lengths and the difference m between the difference β and the n-1 unit data lengths;

[0012] the N-dimensional vector comprises n-1 dimensions with encoding value 1, one dimension with encoding value being the ratio of the difference m and the unit data length, and N-n dimensions with encoding value 0.

[0013] Optionally, dissolving the difference β between the input data and the minimum threshold of the preset range according to the unit data length comprises:

[0014] iteratively decreasing the product of the non-zero encoding value and the unit data length according to the product, and ending when the decreasing result is less than the unit data length after n iterations; or

[0015] performing a modulo operation on the product of the non-zero encoding value and the unit data length.

[0016] Optionally, when it is determined that the input data is out of the preset range, the value of the input data is limited to be within the preset range, comprising:

[0017] When it is determined that the input data is greater than the maximum threshold of the preset range, the input data is updated to the maximum threshold;

[0018] When it is determined that the input data is less than the minimum threshold of the preset range, the input data is updated to the minimum threshold;

[0019] When it is determined that the input data is not less than the minimum threshold of the preset range and not greater than the maximum threshold of the preset range, the input data is kept unchanged.

[0020] Optionally, according to the data length of the preset range and the preset dimension N, the unit data length corresponding to each dimension vector is determined, comprising:

[0021] The data length of the preset range is compared with the preset dimension N to obtain the unit data length corresponding to each dimension vector.

[0022] Optionally, the method further comprises:

[0023] The N-dimensional vector is taken as the input of a data prediction model, and the parameter of the data prediction model is adjusted to output the actual data of the data center corresponding to the N-dimensional vector as the target.

[0024] Optionally, the data prediction model comprises a data center power usage effectiveness (PUE) prediction model and / or a data center temperature prediction model.

[0025] In a second aspect, the present application provides an input data normalization device of a data prediction model, comprising a memory and a processor, wherein:

[0026] The memory is used to store a computer program;

[0027] The processor is used to read the program in the memory and perform the following steps:

[0028] The input data to be normalized is obtained, and when it is determined that the input data is out of the preset range, the value of the input data is limited to be within the preset range;

[0029] According to the data length of the preset range and the preset dimension N, the unit data length corresponding to each dimension vector is determined, and the N is not less than 2;

[0030] mapping the difference β between the input data and the minimum threshold of the preset range into an N-dimensional vector with non-zero encoding values not greater than 1 in the first n dimensions and 0 in the N-n dimensions, wherein a sum of products of each dimension encoding value in the N-dimensional vector and the unit data length is the difference β.

[0031] Optionally, the processor mapping the difference β between the input data and the minimum threshold of the preset range into an N-dimensional vector with non-zero encoding values not greater than 1 in the first n dimensions and 0 in the N-n dimensions comprises:

[0032] decomposing the difference β between the input data and the minimum threshold of the preset range according to the unit data length to obtain n-1 unit data lengths and a difference m between the difference β and the n-1 unit data lengths;

[0033] the N-dimensional vector comprises n-1 dimensions with an encoding value of 1, one dimension with an encoding value of a ratio of the difference m and the unit data length, and N-n dimensions with an encoding value of 0.

[0034] Optionally, the processor decomposing the difference β between the input data and the minimum threshold of the preset range according to the unit data length comprises:

[0035] iteratively decreasing a product of the non-zero encoding value and the unit data length by 1 for n times until the decreasing result is less than the unit data length; or

[0036] performing a modulo operation on the product of the non-zero encoding value and the unit data length.

[0037] Optionally, when the processor determines that the input data is out of the preset range, the processor limiting the value of the input data to the preset range comprises:

[0038] determining that the input data is greater than the maximum threshold of the preset range, and updating the input data to the maximum threshold;

[0039] determining that the input data is less than the minimum threshold of the preset range, and updating the input data to the minimum threshold;

[0040] determining that the input data is not less than the minimum threshold of the preset range and not greater than the maximum threshold of the preset range, and keeping the input data unchanged.

[0041] Optionally, the processor determining the unit data length corresponding to each dimension vector according to the data length of the preset range and the preset dimension N comprises:

[0042] The preset range of data length is compared with a preset dimension N to obtain a unit data length corresponding to each dimension vector.

[0043] Optionally, the processor is further configured to:

[0044] The N-dimensional vector is taken as an input of a data prediction model to output an actual data target of a data center corresponding to the N-dimensional vector, and parameters of the data prediction model are adjusted.

[0045] Optionally, the data prediction model comprises a data center power usage effectiveness (PUE) prediction model and / or a data center temperature prediction model.

[0046] In a third aspect, the present application provides an input data normalization device of a data prediction model, comprising:

[0047] A data adjustment unit is configured to obtain input data to be normalized, and limit a value of the input data to a preset range when it is determined that the input data exceeds the preset range;

[0048] A length calculation unit is configured to determine a unit data length corresponding to each dimension vector according to a data length of the preset range and a preset dimension N, and the N is not less than 2.

[0049] A vector mapping unit is configured to map a difference β between the input data and a minimum threshold of the preset range into an N-dimensional vector in which the first n dimensions adopt non-zero encoding values with absolute values not greater than 1 and the N-n dimensions adopt an encoding value 0, and a sum of products of each dimension encoding value in the N-dimensional vector and the unit data length is the difference β.

[0050] Optionally, the vector mapping unit maps the difference β between the input data and the minimum threshold of the preset range into the N-dimensional vector in which the first n dimensions adopt non-zero encoding values with absolute values not greater than 1 and the N-n dimensions adopt the encoding value 0, and the sum of the products of each dimension encoding value in the N-dimensional vector and the unit data length is the difference β, comprising:

[0051] The difference β between the input data and the minimum threshold of the preset range is decomposed according to the unit data length to obtain n-1 unit data lengths and a difference m between the difference β and the n-1 unit data lengths.

[0052] The N-dimensional vector comprises n-1 dimensions with an encoding value 1, one dimension with an encoding value being a ratio of the difference m and the unit data length, and N-n dimensions with an encoding value 0.

[0053] Optionally, the vector mapping unit decomposes the difference β between the input data and the minimum threshold of the preset range according to the unit data length, comprising:

[0054] The difference β between the input data and the minimum threshold of the preset range is iteratively decreased by a product of a non-zero encoding value and a unit data length, and the iteration is performed n times until the decreased result is less than the unit data length.

[0055] The difference β between the input data and the minimum threshold of the preset range is subjected to a modulo operation on the product of the non-zero encoding value and the unit data length.

[0056] Optionally, when the data adjustment unit determines that the input data is out of the preset range, the data adjustment unit limits the value of the input data to be within the preset range, including:

[0057] When the input data is greater than the maximum threshold of the preset range, the input data is updated to the maximum threshold.

[0058] When the input data is less than the minimum threshold of the preset range, the input data is updated to the minimum threshold.

[0059] When the input data is not less than the minimum threshold of the preset range and not greater than the maximum threshold of the preset range, the input data is kept unchanged.

[0060] Optionally, the length calculation unit determines the unit data length corresponding to each dimension vector according to the data length of the preset range and a preset dimension N, including:

[0061] The data length of the preset range is divided by the preset dimension N to obtain the unit data length corresponding to each dimension vector.

[0062] Optionally, the vector mapping unit is further configured to:

[0063] The N-dimensional vector is taken as an input of a data prediction model, and parameters of the data prediction model are adjusted to output actual data corresponding to the N-dimensional vector in a data center as a target.

[0064] Optionally, the data prediction model includes a data center power usage effectiveness (PUE) prediction model and / or a data center temperature prediction model.

[0065] In a fourth aspect, the present application provides a computer program medium, which stores a computer program, and the program is executed by a processor to implement the steps of the input data normalization method of the data prediction model provided in the first aspect.

[0066] The input data normalization method and device and equipment of the data prediction model provided by the present application have the following beneficial effects:

[0067] The input data of the data prediction model is transformed into an N-dimensional vector with non-zero coding values with absolute values not greater than 1 in the first n dimensions and coding value 0 in the N-n dimensions, so that the dimension increasing normalization of the input data is realized, and the reasonable normalization of the input data is completed; a larger preset range can be set to expand the numerical range of the input data that can be processed, and the degree of concentration of the input data with a smaller change range around a certain value after normalization is reduced, so that the coding value after normalization has good precision; and the convergence speed and convergence effect of the data prediction model using the dimension increasing normalized input data are improved. BRIEF DESCRIPTION OF DRAWINGS

[0068] Figure 1 A flowchart of an input data normalization method of a data prediction model provided for an embodiment of the present application is shown in the figure.

[0069] Figure 2 An effect diagram of an input data normalization method of a data prediction model provided for an embodiment of the present application is shown in the figure.

[0070] Figure 3 A flowchart of mapping input data into an N-dimensional vector provided for an embodiment of the present application is shown in the figure.

[0071] Figure 4 A flowchart of a specific implementation of an input data normalization method of a data prediction model provided for an embodiment of the present application is shown in the figure.

[0072] Figure 5 A schematic diagram of an input data normalization device of a data prediction model provided for an embodiment of the present application is shown in the figure.

[0073] Figure 6 A schematic diagram of an input data normalization device of a data prediction model provided for an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, and not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0075] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-described drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein.

[0076] The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present disclosure. Instead, they are merely examples of apparatuses and methods consistent with some aspects of the present disclosure as detailed in the appended claims. Based on the implementations in this application, all other implementations obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present application.

[0077] In the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in the text only means a description of the relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. In addition, in the description of the embodiments of the present application, "multiple" means two or more than two, and other quantifiers similar thereto should be understood. The preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application, and in the case of no conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0078] In the following, some terms in the embodiments of the present application are explained and described, so as to facilitate the understanding of those skilled in the art.

[0079] (1) In the embodiments of the present application, the term "power usage effectiveness (PUE)" is an index for evaluating the energy efficiency of a data center, PUE = total energy consumption of data center / IT equipment energy consumption, wherein the total energy consumption of the data center includes the energy consumption of the IT equipment and the energy consumption of the cooling, power distribution and other equipment, and its value is greater than 1. The closer PUE is to 1, the less non-IT equipment energy consumption, that is, the better the energy efficiency level of the data center.

[0080] (2) In the embodiments of the present application, the term "dimensional disaster" refers to a phenomenon that the amount of calculation increases exponentially with the increase of the dimension of the vector in the problem involving vector calculation.

[0081] In view of the existing scheme of using the standard 0-1 normalization method to process the input data of the data prediction model, the normalization result is concentrated around a value, and the input data outside the above preset range cannot be processed, the present application provides a data prediction model input data normalization method and device and equipment.

[0082] In the following, a data prediction model input data normalization method and device and equipment in the embodiments of the present application are described in detail.

[0083] Embodiment 1

[0084] The embodiment of the present application provides a flow chart of an input data normalization method of a data prediction model, as shown in the figure, comprising: Figure 1

[0085] In step S101, input data to be normalized is acquired, and when it is determined that the input data is out of a preset range, the value of the input data is limited to be within the preset range.

[0086] The input data can be data of any data type used by the data prediction model, including but not limited to device running data such as IT power, air conditioner power, power distribution equipment power, and environmental state data such as temperature and humidity.

[0087] As an optional implementation, the input data is IT power.

[0088] The preset range can be determined by setting a minimum threshold to a maximum threshold.

[0089] The specific size of the preset range can be set according to specific implementation conditions, for example, the data range during normal operation of the data center is multiplied by a preset proportion to serve as the preset range, or the preset range is determined according to the expected size of the input data, and the embodiment of the present application does not make any limitation in this regard.

[0090] For example, when the input data is IT power, the upper limit MaxL of the preset range is set to 200 (kw), and the lower limit MinL of the preset range is set to 10 (kw).

[0091] For input data out of the preset range, that is, input data less than the minimum threshold of the preset range or greater than the maximum threshold of the preset range, processing is performed to limit the input data out of the preset range to be within the preset range.

[0092] The processing method can be any method that can achieve the limiting effect, for example, input data greater than the maximum threshold of the preset range is truncated to the maximum threshold, and input data less than the minimum threshold of the preset range is truncated to the minimum threshold; or, according to a preset proportion, the input data out of the preset range is mapped to data within the preset range, and the embodiment of the present application does not make any limitation in this regard.

[0093] In step S102, according to the data length of the preset range and a preset dimension N, the unit data length corresponding to each dimension vector is determined, and the N is not less than 2.

[0094] It should be noted that, in order to avoid the curse of dimensionality, the preset dimension N is usually set to be within 10.

[0095] ​The preset dimension N can be any value not less than 2 and not greater than 10, and the specific value of the preset dimension N can be set according to specific implementation, and the embodiment of the application does not make any limitation on this.

[0096] The unit data length interval corresponding to each dimension vector can be understood as precision, which can be adjusted by the maximum threshold MaxL of the preset range, the minimum threshold MinL of the preset range, and the preset dimension N.

[0097] The smaller the unit data length interval is, the higher the precision is, but the larger the dimension N is.

[0098] In step S103, the difference β between the input data and the minimum threshold of the preset range is mapped into an N-dimensional vector with the first n dimensions using non-zero encoding values with absolute values not greater than 1 and the last N-n dimensions using encoding value 0, wherein the sum of the products of the encoding values of each dimension in the N-dimensional vector and the unit data length interval is the difference β.

[0099] It should be noted that the specific value of the non-zero encoding value with the absolute value not greater than 1 can be set in advance, and the specific value can be set according to the specific implementation, and the embodiment of the application does not make any limitation on this.

[0100] For example, the encoding value is set to 1, or -1, or a decimal number with an absolute value not greater than 1.

[0101] The N-dimensional vector is [n1, n2, n3, n4], wherein N=4, n1, n2, n3, and n4 are the encoding values of each dimension in the N-dimensional vector, and the input data = unit data length interval*(n1+n2+n3+n4)+minimum threshold MinL of the preset range.

[0102] According to the difference β between the input data and the minimum threshold of the preset range, and the unit data length interval, the difference β is mapped into an N-dimensional vector with the first n dimensions using non-zero encoding values with absolute values not greater than 1 and the last N-n dimensions using encoding value 0, and the specific encoding value of each dimension is determined to ensure that the sum of the products of the encoding values of each dimension in the N-dimensional vector and the unit data length interval is the difference β.

[0103] As an optional implementation, the non-zero encoding value with the absolute value not greater than 1 is set to 1 or a value between 0 and 1.

[0104] As shown in FIG. 1, Figure 2 The embodiment of the application provides an effect diagram of an input data normalization method of a data prediction model.

[0105] In Figure 2In example 101, the input data is 51, the configuration is (0, 60, 5), which means the preset range is 0-60 and the preset dimension N is 5. Then the difference between the input data and 0 is mapped to a 5-dimensional vector, and the encoding values of each dimension in the 5-dimensional vector are determined as [1, 1, 1, 1, 0.25], wherein 51-0=(60-0) / 5*(1+1+1+1+0.25).

[0106] In Figure 2 In example 102, the input data is 51, the configuration is (20, 60, 5), which means the preset range is 20-60 and the preset dimension N is 5. Then the difference between the input data and 20 is mapped to a 5-dimensional vector, and the encoding values of each dimension in the 5-dimensional vector are determined as [1, 1, 1, 0.875, 0], wherein 51-20=(60-20) / 5*(1+1+1+0.875).

[0107] In Figure 2 In example 103, the input data is 51, the configuration is (30, 60, 3), which means the preset range is 30-60 and the preset dimension N is 3. Then the difference between the input data and 30 is mapped to a 3-dimensional vector, and the encoding values of each dimension in the 3-dimensional vector are determined as [1, 1, 0.1], wherein 51-30=(60-30) / 3*(1+1+0.1).

[0108] As an optional implementation, when it is determined that the input data is out of the preset range, the value of the input data is limited to the preset range, including:

[0109] determining that the input data is greater than the maximum threshold of the preset range, and updating the input data to the maximum threshold;

[0110] determining that the input data is less than the minimum threshold of the preset range, and updating the input data to the minimum threshold;

[0111] determining that the input data is not less than the minimum threshold of the preset range and not greater than the maximum threshold of the preset range, and keeping the input data unchanged.

[0112] According to the maximum threshold MaxL of the preset range and the minimum threshold MinL of the preset range, the input data α is processed by truncation: when α>MaxL, α is corrected to MaxL; when α<MinL, α is corrected to MinL; when MinL<=α<=MaxL, α is not modified.

[0113] As an optional implementation, according to the data length of the preset range and the preset dimension N, the unit data length corresponding to each dimension vector is determined, including:

[0114] The preset range of data length is divided by the preset dimension N to obtain a unit data length corresponding to each dimension vector.

[0115] According to the maximum threshold MaxL of the preset range and the minimum threshold MinL of the preset range, and the preset dimension N, a unit data length interval corresponding to each dimension vector is calculated: interval=(MaxL-MinL) / N.

[0116] As shown in Figure 3 The embodiment of the present application provides a flowchart for mapping input data into an N-dimensional vector, which comprises:

[0117] In step S301, the difference β between the input data and the minimum threshold of the preset range is decomposed according to the unit data length to obtain n-1 unit data lengths and the difference m between the difference β and the n-1 unit data lengths.

[0118] The difference β between the input data α and the minimum threshold MinL of the preset range is calculated, i.e. β=α-MinL.

[0119] The difference β is decomposed into n-1 unit data lengths interval and the difference m between the difference β and the n-1 unit data lengths, i.e. β=(n-1)*interval+m.

[0120] As an optional implementation, the decomposing of the difference β between the input data and the minimum threshold of the preset range according to the unit data length comprises:

[0121] The difference β between the input data and the minimum threshold of the preset range is iteratively decreased according to the product of a non-zero encoding value and the unit data length, and the iteration is performed n times until the decreasing result is less than the unit data length.

[0122] That is, the difference β and the unit data length interval are iterated, and the iteration rule is:

[0123] (1) If β>interval, β is updated as β=β-interval*1, and then the updated β is compared with interval for the next iteration;

[0124] Wherein, the 1 is a non-zero encoding value.

[0125] (2) If β<=interval, the iteration is stopped.

[0126] The iteration number n is reduced by 1 as the number n-1 of unit data lengths, and the difference β after the iteration is stopped is taken as the difference m.

[0127] Or, the difference β of the input data and the minimum threshold of the preset range is taken modulo the product of the non-zero encoding value and the unit data length.

[0128] That is, the difference β and the unit data length interval*1 are taken modulo, the modulus of the calculation is taken as the number n-1 of the unit data length, and the remainder of the calculation is taken as the difference m: β=(n-1)*interval+m.

[0129] Wherein, the 1 is a non-zero encoding value.

[0130] Step S302, the N-dimensional vector includes n-1 dimensions with an encoding value of 1, one dimension with an encoding value of the ratio of the difference m and the unit data length, and N-n dimensions with an encoding value of 0.

[0131] That is, the N-dimensional vector [n1, n2, …, nk, nk+1, …] = [1, 1, …, 1, m / interval, …].

[0132] It should be noted that when the input data to be normalized is equal to the maximum threshold of the preset range, the N-dimensional vector is composed of k+1 dimensions with an encoding value of 1, that is, the N-dimensional vector [n1, n2, …, nk+1] = [1, 1, …, 1].

[0133] The difference β of the input data and the minimum threshold of the preset range is iteratively subtracted from the product of the non-zero encoding value and the unit data length until the iteration result is less than the unit data length; the N-dimensional vector includes n-1 dimensions with an encoding value of 1, one dimension with an encoding value of the ratio of the difference m and the unit data length, and N-n dimensions with an encoding value of 0.

[0134] That is, the difference β and the unit data length interval are iterated, and the iteration rule is:

[0135] (1) If β>interval, update β=β-interval*1, and then compare the updated β with interval again for the next iteration;

[0136] When the iteration is performed, the encoding result output is calculated simultaneously: add a dimension to output and set the encoding value of the dimension to 1.

[0137] (2) If β<=interval, stop iteration.

[0138] When the iteration is performed, the encoding result output is calculated simultaneously: a dimension is added to the output and the encoding value of the dimension is set to β / interval.

[0139] It should be noted that after the iteration is completed, it is checked whether the dimension of the encoding result output is the preset dimension N, and if not, the dimension is increased to the preset dimension N and the encoding value of the added dimension is set to 0.

[0140] As an optional implementation, the method further comprises:

[0141] The N-dimensional vector is taken as input of a data prediction model, and parameters of the data prediction model are adjusted to output actual data of the data center corresponding to the N-dimensional vector as a target.

[0142] As an optional implementation, the data prediction model comprises a data center power usage effectiveness (PUE) prediction model and / or a data center temperature prediction model.

[0143] It should be noted that the data prediction model can be any machine learning model with PUE prediction capability, and is not limited to the data center PUE prediction model and the data center temperature prediction model, and the embodiments of the present application do not make any limitation thereon.

[0144] As shown in Figure 4 the embodiments of the present application provide a flowchart of a specific implementation of an input data normalization method of a data prediction model, which comprises:

[0145] For example, the input data of the data prediction model is IT power α=45kw, the maximum threshold MaxL of the preset range is 160kw, the minimum threshold MinL of the preset range is 0kw, and the preset dimension N is 8 dimensions.

[0146] In step S401, input data to be normalized is obtained;

[0147] The input data to be normalized is IT power α=45kw.

[0148] In step S402, it is judged whether the input data exceeds the preset range, and if so, step S403 is executed, otherwise, step S404 is executed.

[0149] The IT power α=45kw is within the preset range [MinL=0kw, MaxL=160kw], so the input data does not exceed the preset range, and step S404 is executed.

[0150] If the IT power α=170kw is greater than the MaxL=160kw, step S403 is executed.

[0151] Step S403, limit the value of the input data to the preset range, execute step S404;

[0152] If the IT power a = 170kw, the IT power a is corrected to MaxL = 160kw.

[0153] Step S404, according to the data length of the preset range and the preset dimension N, determine the unit data length corresponding to each dimension vector;

[0154] Calculate the unit data length interval corresponding to each dimension vector: interval = (160-0) / 8 = 20.

[0155] Step S405, the difference between the input data and the minimum threshold of the preset range β, subtract the product of the non-zero encoding value and the unit data length;

[0156] Calculate the difference β between the IT power a = 45kw and the minimum threshold MinL = 0kw of the preset range: β = 45-0 = 45.

[0157] First: the difference β minus the product of the non-zero encoding value 1 and the unit data length interval: β-1*interval = 45-1*20 = 25;

[0158] Second: the updated difference β minus the product of the non-zero encoding value 1 and the unit data length interval: β-1*interval = 25-1*20 = 5.

[0159] Step S406, determine whether the calculation result is less than the unit data length, if yes, execute step S408, otherwise, execute step S407;

[0160] First: the calculation result is 25, which is greater than the unit data length interval, execute step S407.

[0161] Second: the calculation result is 5, which is less than the unit data length interval, execute step S408.

[0162] Step S407, update the calculation result to the difference β, and increase the output by one dimension and set the encoding value of the dimension to 1, execute step S405;

[0163] First: the difference β is updated to β = 25, and the output becomes [1].

[0164] Second time: the above difference β is updated to β = 5, and the output becomes [1, 1].

[0165] Step S408, the output is increased by one dimension, and the encoding value of the above dimension is set to β / interval;

[0166] The output becomes [1, 1, 5 / 20], that is, [1, 1, 0.25].

[0167] Step S409, it is judged whether the dimension of the above output is less than the preset dimension N, if yes, step S410 is executed, otherwise the process is ended.

[0168] The dimension of the above output = 3 is less than the preset dimension N = 8.

[0169] Step S410, the dimension of the above output is increased to the above preset dimension N, and the encoding value of the above newly added dimension is set to 0, and the process is ended.

[0170] The dimension of the above output is increased to 8, and the encoding value of the above newly added dimension is set to 0, and the output becomes [1, 1, 0.25, 0, 0, 0, 0, 0].

[0171] The process is ended, the encoding is completed, the IT power is normalized before encoding to 45, and after encoding to [1, 1, 0.25, 0, 0, 0, 0, 0].

[0172] The input data of the data prediction model is converted into an N-dimensional vector including constant and decimal encoding values, a larger preset range can be set, the numerical range of the input data that can be processed is expanded, and the degree of concentration of the input data with a smaller change range after normalization around a certain value is reduced.

[0173] That is, the input data normalization method of the data prediction model implemented by the application can use larger upper and lower limits of the input data, so as to be suitable for different input data, and the change range of the normalized input data is larger.

[0174] For example, a preset range of 0 to 200 is set, and a standard 0-1 normalization formula is (a-0) / (200-0), where a represents input data to be normalized. Assuming that a certain input data set includes multiple input data between 40 and 60, the result of normalizing the input data set using the above standard 0-1 normalization formula is a value between 0.2 and 0.3; if N=10 is set, the input data of the data prediction model of the application is normalized using the input data normalization method, and an N-dimensional vector is obtained, the first 2 positions are always 1, the fourth to tenth positions are always 0, and the third position is an effective position, the change range of which is [0.0, 1.0], the change range of the effective position is wider, and the normalization effect is better than the above standard 0-1 normalization.

[0175] The normalized input data of the embodiment of the application can be used in AI algorithms of data centers such as air conditioner group control and temperature control, and can improve the convergence speed and effect of the AI algorithm.

[0176] The method uses the dimension increasing method, and after use, some dimensions achieve better normalization, and other dimensions become constants. The normalized value range of IT power is reasonable, and when combined with machine learning, neural network and other algorithms, the convergence speed and effect can be improved, especially when combined with air conditioner AI group control, the convergence effect of the algorithm can be improved.

[0177] Embodiment 2

[0178] The embodiment of the application provides a schematic diagram of an input data normalization device 500 of a data prediction model, which includes a memory 501 and a processor 502, as shown in the figure. Figure 5 As shown in the figure,

[0179] The memory is used to store a computer program;

[0180] The processor is used to read the program in the memory and perform the following steps:

[0181] Obtain input data to be normalized, and limit the value of the input data to a preset range when it is determined that the input data exceeds the preset range;

[0182] According to the data length of the preset range and the preset dimension N, determine the unit data length corresponding to each dimension vector, and the N is not less than 2;

[0183] Map the difference β between the input data and the minimum threshold of the preset range to an N-dimensional vector using non-zero encoding values with absolute values not greater than 1 for the first n dimensions and encoding value 0 for the N-n dimensions, wherein the product sum of the dimension encoding values in the N-dimensional vector and the unit data length is the difference β.

[0184] Optionally, the processor maps the difference β between the input data and the minimum threshold of the preset range into an N-dimensional vector with the first n dimensions having non-zero encoding values not greater than 1 in absolute value and the remaining N-n dimensions having an encoding value of 0, including:

[0185] The difference β between the input data and the minimum threshold of the preset range is decomposed according to the unit data length to obtain n-1 unit data lengths and a difference m between the difference β and the n-1 unit data lengths.

[0186] The N-dimensional vector includes n-1 dimensions with an encoding value of 1, one dimension with an encoding value of a ratio of the difference m and the unit data length, and N-n dimensions with an encoding value of 0.

[0187] Optionally, the processor decomposes the difference β between the input data and the minimum threshold of the preset range according to the unit data length, including:

[0188] The difference β between the input data and the minimum threshold of the preset range is iteratively decreased according to a product of a non-zero encoding value and the unit data length, and the iteration is performed n times until the decreased result is less than the unit data length; or

[0189] The difference β between the input data and the minimum threshold of the preset range is subjected to a modulo operation on the product of the non-zero encoding value and the unit data length.

[0190] Optionally, when the processor determines that the input data is out of the preset range, the processor limits the value of the input data to be within the preset range, including:

[0191] When the input data is greater than the maximum threshold of the preset range, the input data is updated to the maximum threshold;

[0192] When the input data is less than the minimum threshold of the preset range, the input data is updated to the minimum threshold;

[0193] When the input data is not less than the minimum threshold of the preset range and not greater than the maximum threshold of the preset range, the input data is kept unchanged.

[0194] Optionally, the processor determines a unit data length corresponding to each dimension vector according to a data length of the preset range and a preset dimension N, including:

[0195] The data length of the preset range is divided by the preset dimension N to obtain the unit data length corresponding to each dimension vector.

[0196] Optionally, the processor is further configured to:

[0197] The N-dimensional vector is taken as an input of a data prediction model, and parameters of the data prediction model are adjusted to output actual data of the data center corresponding to the N-dimensional vector as a target.

[0198] Optionally, the data prediction model comprises a data center power usage effectiveness (PUE) prediction model and / or a data center temperature prediction model.

[0199] An embodiment of the present application provides a schematic diagram of an input data normalization device of a data prediction model, as shown in the figure, comprising: Figure 6 A data adjustment unit 601 is configured to acquire input data to be normalized, and limit a value of the input data to a preset range when it is determined that the input data exceeds the preset range;

[0200] A length calculation unit 602 is configured to determine a unit data length corresponding to each dimension vector according to a data length of the preset range and a preset dimension N, wherein the N is not less than 2;

[0201] A vector mapping unit 603 is configured to map a difference β between the input data and a minimum threshold of the preset range into an N-dimensional vector with absolute values of non-zero encoding values of the first n dimensions not greater than 1 and encoding values of the N-n dimensions being 0, wherein a sum of products of each dimension encoding value in the N-dimensional vector and the unit data length is the difference β.

[0202] Optionally, the vector mapping unit maps the difference β between the input data and the minimum threshold of the preset range into an N-dimensional vector with absolute values of non-zero encoding values of the first n dimensions not greater than 1 and encoding values of the N-n dimensions being 0, comprising:

[0203] The difference β between the input data and the minimum threshold of the preset range is decomposed according to the unit data length to obtain n-1 unit data lengths and a difference m between the difference β and the n-1 unit data lengths;

[0204] The N-dimensional vector comprises n-1 dimensions with an encoding value of 1, one dimension with an encoding value of a ratio of the difference m and the unit data length, and N-n dimensions with an encoding value of 0.

[0205] Optionally, the vector mapping unit decomposes the difference β between the input data and the minimum threshold of the preset range according to the unit data length, comprising:

[0206]

[0207] ​The difference β between the input data and the minimum threshold of the preset range is iteratively decreased by a product of a non-zero encoding value and a unit data length, and the iteration is performed n times until the decreased result is less than the unit data length; or

[0208] The difference β between the input data and the minimum threshold of the preset range is subjected to a modulo operation on the product of the non-zero encoding value and the unit data length.

[0209] Optionally, when the data adjustment unit determines that the input data is out of the preset range, the data adjustment unit limits the value of the input data to be within the preset range, including:

[0210] When the input data is greater than the maximum threshold of the preset range, the input data is updated to the maximum threshold;

[0211] When the input data is less than the minimum threshold of the preset range, the input data is updated to the minimum threshold;

[0212] When the input data is not less than the minimum threshold of the preset range and not greater than the maximum threshold of the preset range, the input data is kept unchanged.

[0213] Optionally, the length calculation unit determines the unit data length corresponding to each dimension vector according to the data length of the preset range and a preset dimension N, including:

[0214] The data length of the preset range is divided by the preset dimension N to obtain the unit data length corresponding to each dimension vector.

[0215] Optionally, the vector mapping unit is further configured to:

[0216] The N-dimensional vector is taken as the input of a data prediction model, and the parameter of the data prediction model is adjusted to output actual data corresponding to the N-dimensional vector as the target of the data center.

[0217] Optionally, the data prediction model includes a data center power usage effectiveness (PUE) prediction model and / or a data center temperature prediction model.

[0218] The application further provides a computer program medium, which stores a computer program, and the program is executed by a processor to implement the steps of the data prediction model input data normalization method provided in the above embodiment 1.

[0219] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented by other manners. For example, the above-described device embodiments are merely illustrative, for example, the division of the modules is merely a logical function division, and actual implementation can have another division manner, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection can be indirect coupling or communication connection through some interfaces, devices or modules, and can be electrical, mechanical or other forms.

[0220] The modules described as separate components can or can not be physically separated, and the components shown as modules can or can not be physical modules, that is, can be located in one place, or can be distributed to a plurality of network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0221] In addition, the functional modules in each embodiment of the present application can be integrated in one processing module, or each module can exist physically, or two or more modules can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of software functional module. The integrated module, if realized in the form of software functional module and sold or used as an independent product, can be stored in a computer readable storage medium.

[0222] In the above embodiments, all or part can be realized by software, hardware, firmware or any combination thereof. When realized by software, all or part can be realized in the form of a computer program product.

[0223] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be stored by the computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)), etc.

[0224] The above describes the technical solutions provided by the present application in detail. The principles and implementation manners of the present application are described by applying specific examples. The above examples are only used to help understand the method and core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range will be changed; therefore, the content of the specification should not be understood as a limitation of the present application.

[0225] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0226] The present application is described with reference to flowcharts and / or block diagrams according to the method, device (system), and computer program product of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of the flows and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a means for implementing the functions described in the flowchart and / or block diagram. The computer program instructions can also be stored in a computer readable storage medium that can guide the computer to operate in a specific manner. The computer readable storage medium can be a computer readable storage medium, a computer readable storage medium, or a computer readable storage medium. Figure 1apparatus for performing the functions specified in the flow or flows and / or blocks. Figure 1

[0227] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 apparatus for performing the functions specified in the flow or flows and / or blocks. Figure 1

[0228] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 apparatus for performing the functions specified in the flow or flows and / or blocks. Figure 1

[0229] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the apparatus for performing the functions specified in the flow or flows and / or blocks. Figure 1

[0227] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 apparatus for performing the functions specified in the flow or flows and / or blocks. Figure 1

[0228] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 apparatus for performing the functions specified in the flow or flows and / or blocks. Figure 1

[0229] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the apparatus for performing the functions specified in the flow or flows and / or blocks.

Claims

1. A method for normalizing input data of a data prediction model, characterized by, The method comprises the following steps: acquiring input data to be normalized, and limiting the value of the input data to a preset range when it is determined that the input data exceeds the preset range, wherein the input data comprises device operation data, and the device operation data comprises IT power, air conditioner power, power distribution equipment power, and environmental state data, and the environmental state data comprises temperature and humidity; determining the unit data length corresponding to each dimension vector according to the data length of the preset range and a preset dimension N, wherein the N is not less than 2; mapping the difference β between the input data and the minimum threshold of the preset range into an N-dimensional vector in which the first n dimensions adopt non-zero encoding values with absolute values not greater than 1, and the N-n dimensions adopt an encoding value 0, wherein the sum of the products of each dimension encoding value in the N-dimensional vector and the unit data length is the difference β; using the N-dimensional vector as the input of a data center power usage effectiveness (PUE) prediction model and / or a data center temperature prediction model, and using an artificial intelligence method to construct the PUE prediction model for predicting PUE data and / or the data center temperature prediction model for predicting temperature.

2. The method of claim 1, wherein, mapping the difference β between the input data and the minimum threshold of the preset range into an N-dimensional vector in which the first n dimensions adopt non-zero encoding values with absolute values not greater than 1, and the N-n dimensions adopt an encoding value 0, comprising: decomposing the difference β between the input data and the minimum threshold of the preset range according to the unit data length to obtain n-1 unit data lengths and the difference m between the difference β and the n-1 unit data lengths; the N-dimensional vector comprises n-1 dimensions with an encoding value 1, one dimension with an encoding value being the ratio of the difference m and the unit data length, and N-n dimensions with an encoding value 0.

3. The method of claim 2, wherein, decomposing the difference β between the input data and the minimum threshold of the preset range according to the unit data length, comprising: iteratively decreasing the product of the non-zero encoding value and the unit data length n times until the decreasing result is less than the unit data length; or performing a modulo operation on the product of the non-zero encoding value and the unit data length.

4. The method of claim 1, wherein, limiting the value of the input data to the preset range when it is determined that the input data exceeds the preset range, comprising: determining that the input data is greater than the maximum threshold of the preset range, and updating the input data to the maximum threshold; determining that the input data is less than the minimum threshold of the preset range, and updating the input data to the minimum threshold; determining that the input data is not less than the minimum threshold of the preset range and not greater than the maximum threshold of the preset range, and keeping the input data unchanged.

5. The method of claim 1, wherein, determining the unit data length corresponding to each dimension vector according to the data length of the preset range and a preset dimension N, comprising: dividing the data length of the preset range by the preset dimension N to obtain the unit data length corresponding to each dimension vector.

6. The method of claim 1, wherein, The method further comprises the following steps: The N-dimensional vector is taken as input of a data prediction model, and parameters of the data prediction model are adjusted to output actual data of the data center corresponding to the N-dimensional vector.

7. The method of claim 1, wherein, The data prediction model comprises a data center power usage effectiveness (PUE) prediction model and / or a data center temperature prediction model.

8. An input data normalization device of a data prediction model, characterized by, The data prediction model comprises a data center power usage effectiveness (PUE) prediction model and / or a data center temperature prediction model. The memory is configured to store a computer program. The processor is configured to read the program in the memory and execute the input data normalization method of the data prediction model according to any one of claims 1-7.

9. An input data normalization apparatus of a data prediction model, characterized by, The data adjustment unit is configured to obtain input data to be normalized, and limit the value of the input data to a preset range when it is determined that the input data exceeds the preset range, wherein the input data comprises device operation data, and the device operation data comprises IT power, air conditioner power, power distribution equipment power and environmental state data, and the environmental state data comprises temperature and humidity. The length calculation unit is configured to determine a unit data length corresponding to each dimension vector according to a data length of the preset range and a preset dimension N, wherein the N is not less than 2. The vector mapping unit is configured to map a difference β between the input data and a minimum threshold of the preset range into an N-dimensional vector in which the first n dimensions adopt non-zero encoding values with absolute values not greater than 1, and the N-n dimensions adopt an encoding value 0, wherein a sum of products of each dimension encoding value in the N-dimensional vector and the unit data length is the difference β. The N-dimensional vector is taken as input of a data center power usage effectiveness (PUE) prediction model and / or a data center temperature prediction model, and a PUE prediction model for predicting PUE data and / or a data center temperature prediction model for predicting temperature is constructed using an artificial intelligence method. The computer program is stored on the memory and is executed by the processor to implement the steps of the input data normalization method of the data prediction model according to any one of claims 1-7.

10. A computer program medium, characterized in that, ​

Images