Archive room environment comprehensive evaluation method and evaluation system based on artificial intelligence

Abstract

The application discloses an archive room environment comprehensive evaluation method and system based on artificial intelligence. The steps of the method include: obtaining multi-dimensional environment data to be evaluated in the archive room at a regular time, the multi-dimensional environment data including temperature value, humidity value, air pressure value, safety state level and time stamp; preliminarily processing the multi-dimensional environment data to generate a first processing sequence and a second processing sequence, and extracting a first key feature and a second key feature from the first processing sequence and the second processing sequence respectively, introducing the first processing sequence and the first key feature into a first calculation model constructed in advance to calculate a first evaluation coefficient; introducing the second processing sequence and the second key feature into a second calculation model constructed in advance to calculate a second evaluation coefficient; constructing a multi-dimensional feature space based on the first evaluation coefficient and the second evaluation coefficient; performing deep analysis on the multi-dimensional feature space, overlapping the multi-dimensional feature space and a standard feature space with each other to generate an evaluation factor, constructing a dynamic change curve based on the evaluation factor, comparing the dynamic change curve with a preset standard threshold curve to obtain an abnormal period and an abnormal area, marking the abnormal multi-dimensional feature space, generating an archive room environment multi-dimensional heat map through visual analysis and uploading and displaying the archive room environment multi-dimensional heat map; and the safety state level is determined based on a machine learning model established in advance. The application can effectively fuse and intelligently judge the monitoring data, and perform abnormal identification on the archive room environment to perform early warning processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of archive room evaluation, in particular to an archive room environment comprehensive evaluation method and system based on artificial intelligence. BACKGROUND

[0002] As a place for saving and managing various important documents, the environmental conditions of the archive room play a crucial role in the long-term preservation and effective management of archives. However, the application of archive room environment monitoring and evaluation is still in its infancy.

[0003] The existing application publication number CN110986323A, titled "Archive room interval constant temperature control method", proposes to predict temperature changes and start or stop temperature control equipment in advance, thereby extending the continuous operation and continuous stop time of the temperature control equipment, and reducing the number of state alternations of the temperature control equipment. The technical solution is a single parameter (temperature parameter) monitoring and alarm, which lacks comprehensive analysis and deep evaluation of multi-dimensional environmental data and real-time intelligent optimization.

[0004] In recent years, artificial intelligence has made significant progress in the field of archive management. Through precise evaluation of the environment of archives, especially timely detection of archive preservation environment, it can reduce the reduction or loss of archive life caused by environmental problems, and intervene through intelligent prediction and automatic response system. The automatic response system usually measures with the help of various monitoring devices (such as temperature and humidity sensors, smoke detectors), but the monitoring devices work independently, lack effective fusion and intelligent judgment of monitoring data, and most evaluation systems cannot provide early warning before the environment deteriorates, lacking sufficient foresight and intelligence.

[0005] Based on the above documents and existing technology, an archive room environment comprehensive evaluation method and system based on artificial intelligence is proposed. SUMMARY

[0006] (I) Technical problems solved

[0007] To address the shortcomings of existing technologies, this invention provides a comprehensive evaluation method and system for the environment of archive storage facilities based on artificial intelligence. First, a machine learning model is built to analyze and determine the state data. By incorporating relevant data such as temperature, humidity, and air pressure, multi-dimensional environmental data is formed, enabling effective comprehensive monitoring of the archive storage facility. A first processing sequence and a first key feature are imported into a pre-constructed first computational model to calculate a first evaluation coefficient. A second processing sequence and a second key feature are imported into a pre-constructed second computational model to calculate a second evaluation coefficient, forming a multi-dimensional feature space. This multi-dimensional feature space is then overlapped with a standard feature space to generate evaluation factors. This constructs a dynamic change curve, obtaining abnormal time periods and abnormal areas to mark the abnormal multi-dimensional feature space, generating a multi-dimensional heat map of the archive storage facility environment. This improves the accuracy of the model to a certain extent, providing timely early warning and solving the problems mentioned in the background technology.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] Firstly, this application provides a comprehensive evaluation method for the environment of an archive storage facility based on artificial intelligence, the method comprising:

[0011] The system periodically acquires multidimensional environmental data to be evaluated from the archive storage. This multidimensional environmental data includes temperature, humidity, air pressure, safety status level, and timestamp. The safety status level is determined based on a pre-established machine learning model.

[0012] The multidimensional environmental data is initially processed to generate a first processing sequence and a second processing sequence. A first key feature and a second key feature are extracted from the first processing sequence and the second processing sequence, respectively. The first processing sequence and the first key feature are imported into a pre-constructed first calculation model to calculate a first evaluation coefficient. The second processing sequence and the second key feature are imported into a pre-constructed second calculation model to calculate a second evaluation coefficient. Based on the first evaluation coefficient and the second evaluation coefficient, a multidimensional feature space is constructed.

[0013] In-depth analysis of the multidimensional feature space is performed, and the multidimensional feature space and the standard feature space are overlapped to generate evaluation factors. Based on the evaluation factors, dynamic change curves are constructed. The dynamic change curves are compared with the preset standard threshold curves to obtain abnormal time periods and abnormal areas, and the abnormal multidimensional feature space is marked. Through visualization analysis, a multidimensional heat map of the archive storage environment is generated and uploaded for display.

[0014] Further, the machine learning model is built-in with a first determining unit and a second determining unit, the first determining unit comprises a first input layer, a first hidden layer, a second hidden layer and a first output layer, and the second determining unit comprises a second input layer, a third hidden layer and a second output layer, and the steps of the determining process comprise:

[0015] The first determining unit comprises:

[0016] A first feature vector is generated based on the state data to be determined through the first input layer;

[0017] A first output vector is generated based on the first feature vector through the first hidden layer;

[0018] A second output vector is generated based on the first output vector through the second hidden layer;

[0019] A first determining result is generated based on the second output vector through the first output layer, the first determining result comprises a first data set meeting the safe state, a second data set not meeting the safe state and a third data set unable to determine whether to meet the safe state obtained based on the first determining unit;

[0020] The second determining unit comprises:

[0021] A second feature vector is generated based on the state data to be determined through the second input layer;

[0022] A third output vector is generated based on the second feature vector through the third hidden layer;

[0023] A second determining result is generated based on the third output vector through the third output layer, the second determining result comprises a first data set meeting the safe state and a second data set not meeting the safe state obtained based on the second determining unit;

[0024] A first state identifier meeting the safe state, a second state identifier not meeting the safe state and a third state identifier unable to determine whether to meet the safe state are assigned at the same time;

[0025] The state data at least comprises the collected fire water tank water level and the spray pipe network pressure;

[0026] The machine learning model is trained based on the first determining result and the second determining result to obtain an updated machine learning model for next time determining analysis.

[0027] Further, the machine learning model is at least pre-trained twice, and the trained machine learning model is used for grading determination of the state data to be determined to obtain the corresponding safety state level of the archive room, comprising:

[0028] First level determination: obtain data that does not exist in the third data set in the first data set, obtain a fourth data set; obtain data that does not exist in the first data set in the third data set, obtain a fifth data set; import the fourth data set and the first state identifier, and the fifth data set and the second state identifier as input information into the machine learning model for training, obtain an updated first training model; input the state data into the first training model, perform first level prediction, and obtain a first prediction coefficient;

[0029] Second level determination: obtain data that does not exist in the second data set in the third data set, obtain a sixth data set; obtain data that does not exist in the third data set in the second data set, obtain a seventh data set; import the sixth data set and the first state identifier, and the seventh data set and the second state identifier as input information into the machine learning model for training, obtain an updated second training model; input the state data into the second training model, perform second level prediction, and obtain a second prediction coefficient;

[0030] Comprehensive determination: based on the combination of the first prediction coefficient and the second prediction coefficient, a comprehensive prediction coefficient is generated, and the comprehensive prediction coefficient is compared and analyzed with the standard prediction interval [g1, g2] to obtain the corresponding safety state level;

[0031] The formula based on the comprehensive prediction coefficient is:

[0032] In the formula, forecast represents the comprehensive prediction coefficient, c1 represents the first prediction coefficient, c2 represents the second prediction coefficient, θ1 and θ2 are both weight coefficients, and θ1 and θ2 are both greater than 0.

[0033] Further, the calculation model of the first determination unit is:

[0034]

[0035] F1 = sigmoid (S1 o w2) + b2;

[0036] F2 = sigmoid (S2 o w3) + b3;

[0037] In the formula, H1 represents the first determination result, F1 represents the first intermediate vector, F2 represents the second intermediate vector, S1 represents the first output vector, and S2 represents the second output vector; softmax() represents a normalized exponential activation function, sigmoid() represents a nonlinear activation function; w1 represents a first weight parameter, w2 represents a second weight parameter, and w3 represents a third weight parameter; b1 represents a first bias parameter, b2 represents a second bias parameter, and b3 represents a third bias parameter; o represents a multiplication operation, represents a complexity metric operation, represents a complexity, represents a range of complexity, and takes values in [Θ / 2, Θ].

[0038] Further, the calculation model of the second determination unit is:

[0039] H2 = softmax(G1 ³ F3) * w4 + b4;

[0040] F3 = 2tanh(w5 * 2S3 + b5);

[0041] G1 = [G2 * F3 + (1 - G2) * G3] * w6;

[0042] In the formula, H2 represents a second determination result, S3 represents a third output vector, and F3 represents a third intermediate vector; w4 represents a fourth weight parameter, w5 represents a fifth weight parameter, and w6 represents a sixth weight parameter; b4 represents a fourth bias parameter, and b5 represents a fifth bias parameter; G1 represents a first gating value, G2 represents a second gating value, and G3 represents a third gating value; and tanh() represents a nonlinear activation function.

[0043] Further, the preliminary processing of the multi-dimensional environment data comprises the following steps.

[0044] First processing: removing noise from the multi-dimensional environment data through a band-pass filter to obtain a first processing sequence, and extracting a first key feature based on the first processing sequence, including an average value and a variance;

[0045] Importing a first calculation model: constructing a first parameter matrix based on the first processing sequence, calculating a mean factor based on the average value of each column vector of the first parameter matrix, calculating a difference factor based on the variance of each column vector, and combining the mean factor and the difference factor to obtain a first evaluation coefficient of the column vector;

[0046] Second processing: performing secondary noise reduction on the first processing sequence through fast Fourier transform, and restoring to the time domain through inverse Fourier transform to obtain a second processing sequence, and extracting a second key feature based on the second processing sequence, including a difference value, a standard deviation, and a quartile difference;

[0047] Importing a second calculation model: constructing a second parameter matrix based on the second processing sequence, calculating an increase rate factor based on the difference value between adjacent vectors of each column vector of the second parameter matrix and combining the standard deviation, calculating a fluctuation factor based on the quartile difference of each column vector, and combining the increase rate factor and the fluctuation factor to obtain a second evaluation coefficient of the column vector;

[0048] Comprehensive processing: constructing a corresponding multi-dimensional feature space with a time collection period as the X-axis, the first evaluation coefficient as the Y-axis, and the second evaluation coefficient as the Z-axis, and the multi-dimensional feature space is in the form of a three-dimensional graph.

[0049] Further, the deep analysis of the multi-dimensional feature space, the multi-dimensional feature space and the standard feature space are overlapped with each other to generate an evaluation factor, a dynamic change curve is constructed based on the evaluation factor, the dynamic change curve is compared with a preset standard threshold curve to obtain an abnormal period and an abnormal area, comprising:

[0050] Space overlap: a preset standard feature space, the multi-dimensional feature space and the standard feature space are overlapped to obtain a horizontal overlap amount and a vertical overlap amount, and the overlap area and the non-overlap area are obtained based on the comparison of the horizontal overlap amount and the vertical overlap amount; the evaluation factor is generated based on the combination of the overlap area and the non-overlap area;

[0051] Curve comparison: a two-dimensional rectangular coordinate system is established, and a dynamic change curve of the evaluation factor-time is drawn; at the same time, a standard threshold curve is drawn in the two-dimensional rectangular coordinate system, and the dynamic change curve and the standard threshold curve are compared and analyzed, the period in which the dynamic change curve is above the standard threshold curve and the area surrounded by the standard threshold curve are counted, and are marked as the abnormal period and the abnormal area.

[0052] In a second aspect, the present application provides an archive room environment comprehensive evaluation system based on artificial intelligence, the system comprising:

[0053] A data acquisition module is configured to acquire multi-dimensional environment data to be evaluated in an archive room at a regular time, the multi-dimensional environment data comprising temperature value, humidity value, air pressure value, safety state grade and time stamp, wherein the safety state grade is obtained based on a pre-established machine learning model;

[0054] A preliminary processing module is configured to preliminarily process the multi-dimensional environment data to generate a first processing sequence and a second processing sequence, and extract a first key feature and a second key feature from the first processing sequence and the second processing sequence respectively, and introduce the first processing sequence and the first key feature into a pre-established first calculation model to calculate a first evaluation coefficient; introduce the second processing sequence and the second key feature into a pre-established second calculation model to calculate a second evaluation coefficient; and construct a multi-dimensional feature space based on the first evaluation coefficient and the second evaluation coefficient;

[0055] A deep analysis module is configured to overlap the multi-dimensional feature space and a standard feature space with each other to generate an evaluation factor, construct a dynamic change curve based on the evaluation factor, compare the dynamic change curve with a preset standard threshold curve to obtain an abnormal period and an abnormal area, mark the abnormal multi-dimensional feature space, and generate and upload a multi-dimensional heat map of the archive room environment through visual analysis and display.

[0056] (Three) beneficial effects

[0057] The application provides an archive room environment comprehensive evaluation method and system based on artificial intelligence, which has the following beneficial effects:

[0058] 1、The application first trains data such as the water level of a fire water tank and the pressure of a sprinkler pipe network by building a machine learning model, identifies first and second determination results by setting a first and a second determination unit, and uses the trained machine learning model to make a hierarchical determination on the state data to be determined, so as to obtain the safety state grade of the archive room; to a certain extent, the accuracy of the model is improved; further, related data such as temperature, humidity and air pressure are introduced to form multi-dimensional environmental data, which effectively monitors the archive room comprehensively;

[0059] 2、The first and second evaluation coefficients are calculated by the first and second calculation models built in advance to construct a multi-dimensional feature space, which is overlapped with a preset standard feature space, an offset factor is defined based on the horizontal and vertical overlap amounts, and a formula is set to generate an evaluation factor; by curve comparison, the abnormal period and area are identified, and the multi-dimensional feature space of the anomaly is identified, so that the accuracy of subsequent anomaly analysis is improved, and timely warning is facilitated. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 is a flowchart of an archive room environment comprehensive evaluation method according to an example embodiment.

[0061] Figure 2 is a module diagram of an archive room environment comprehensive evaluation system according to an example embodiment. DETAILED DESCRIPTION

[0062] 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 only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0063] Embodiment 1

[0064] The application embodiment provides an archive room environment comprehensive evaluation method based on artificial intelligence. Figure 1 is a flowchart of an archive room environment comprehensive evaluation method according to an example embodiment. Please refer to Figure 1 , the method comprises the following steps:

[0065] S1, a timing acquisition archive room under the multi-dimensional environment data to be evaluated, multi-dimensional environment data including temperature value, humidity value, air pressure value, safety state level and time stamp, wherein, the safety state level is based on the pre-established machine learning model to obtain the judgment;

[0066] Wherein, the machine learning model is built-in with a first determination unit and a second determination unit, the first determination unit includes a first input layer, a first hidden layer, a second hidden layer and a first output layer, the second determination unit includes a second input layer, a third hidden layer and a second output layer, and the determination process includes:

[0067] The first determination unit includes:

[0068] Through the first input layer, the first feature vector is generated based on the state data to be determined, wherein the state data includes at least the collected fire water tank water level and the spray pipe network pressure, and also includes the residual flow of the electrical circuit and the cable temperature;

[0069] Through the first hidden layer, the first output vector is generated based on the first feature vector;

[0070] Through the second hidden layer, the second output vector is generated based on the first output vector;

[0071] Through the first output layer, the first determination result is generated based on the second output vector, the first determination result including the first data set conforming to the safety state, the second data set not conforming to the safety state and the third data set unable to determine whether it conforms to the safety state obtained based on the first determination unit;

[0072] It should be noted that the above steps are usually assisted by a neural network architecture, in which the input layer is responsible for receiving the original data, which in this embodiment is the state data to be determined, and the state data is usually measured by means of flow sensor and pressure sensor. The flow sensor is used to measure the water level of the fire water tank, and the pressure sensor is used to measure the pressure of the spray pipe network. The dimension of each state data is 1 (i.e. the water level of the fire water tank and the pressure of the spray pipe network correspond to a scalar respectively). In this embodiment, only the water level of the fire water tank and the pressure of the spray pipe network are taken as examples;

[0073] Assuming that the first hidden layer has 128 neurons, the dimension of the first output vector is 128; this dimension indicates that the first feature vector is input in this layer, and the feature is deep transformed and learned, which can capture more data patterns; assuming that the second hidden layer has 256 neurons, the dimension of the second output vector is 256; by setting a higher dimension, the neural network can build more abstract and complex feature representations between the two layers, thereby improving the modeling ability of the data; finally, a 3-dimensional output vector is generated using the softmax layer to represent the probability of each class (i.e., the three cases of conforming to the safety state, not conforming to the safety state, and being unable to judge whether to conform to the safety state);

[0074] Conforming to the safety state: indicating that the current state data conforms to all preset safety standards, rules or specifications, and there is no known security risk; not conforming to the safety state: indicating that the current state data violates certain safety standards or specifications, and there may be a security risk or potential harm; unable to judge: indicating that it is unable to judge whether the current state data violates the safety standards or specifications, for example: during the training process, the complexity formed according to the extracted features is too high, resulting in unstable system state or incomplete data;

[0075] The calculation model of the first determination unit is:

[0076]

[0077] F1=sigmoid(S1⊙w2)+b2;

[0078] F2=sigmoid(S2⊙w3)+b3;

[0079] In the formula, H1 represents the first determination result, F1 represents the first intermediate vector, F2 represents the second intermediate vector, S1 represents the first output vector, and S2 represents the second output vector; softmax() represents a normalized exponential activation function, and sigmoid() represents a nonlinear activation function; w1 represents a first weight parameter, w2 represents a second weight parameter, and w3 represents a third weight parameter; b1 represents a first bias parameter, b2 represents a second bias parameter, and b3 represents a third bias parameter; and represents a multiplication operation, represents a complexity measurement operation, which does not directly affect the specific value of each feature or data, but only exists as a measure, Θ represents complexity, θ represents the range of complexity, and θ takes a value of [Θ / 2, Θ], and the operation reflects the dynamic interaction relationship between F1 and F2, which can be nonlinearly compressed or expanded to adaptively select the weight or adjustment factor of the calculation process;

[0080] ​It should be noted that the first hidden layer and the second hidden layer are usually the core part of the neural network for processing data, responsible for extracting features or learning certain patterns from input information, through the transformation of these layers, the model complexity gradually increases, and more non-linear relationships can be captured, improving the modeling ability of data, so that the input data will be mapped to a new representation, which is the first output vector;

[0081] In this embodiment, the first intermediate vector and the second intermediate vector are two intermediate step vectors generated in the above process, and their complexity can be quantified by information entropy. The general idea is to convert the elements or some features in the vector into a probability distribution, and then calculate the entropy value of the distribution.

[0082] Specifically, it includes:

[0083] The first intermediate vector and the second intermediate vector are discretized, the elements or features in the vector are compressed into several intervals [0, 1], forming discrete categories, and the frequency of elements in each interval is calculated.

[0084] For example:

[0085] Data 1: Suppose a vector is A1 = [1.2, 2.4, 3.1, 4.2, 2.3, 3.5, 2.9];

[0086] The normalized vector is [0.12, 0.24, 0.31, 0.42, 0.23, 0.35, 0.29];

[0087] Map it to different categories, assuming the interval is set as:

[0088] Interval 1: [0.10, 0.20): Category A; Interval 2: [0.20, 0.30): Category B;

[0089] Interval 3: [0.30, 0.40): Category C; Interval 4: [0.40, 0.50): Category D;

[0090] Interval 5: [0.50, 0.60): Category E; Interval 6: [0.60, 0.70): Category F;

[0091] Interval 7: [0.70, 0.80): Category G; Interval 8: [0.80, 0.90): Category H;

[0092] Interval 9: [0.90, 1.00]: Category I;

[0093] Then the discretized vector is [A, B, C, D, B, C, B];

[0094] The information entropy is: - ((1 / 7) log2 (1 / 7) + (3 / 7) log2 (3 / 7) + (2 / 7) log2 (2 / 7) + (1 / 7) log2 (1 / 7)) approximately equal to 1.842.

[0095] Data 2: Assuming there is a vector: A2 = [1.2, 3.1, 2.8, 3.4, 2.2, 2.8, 1.2];

[0096] The normalized vector is: [0.12, 0.31, 0.28, 0.34, 0.22, 0.28, 0.12];

[0097] Based on the interval category of data 1, the discretized vector is: [A, C, B, C, B, B, A];

[0098] The information entropy is: - ((2 / 7) log2 (2 / 7) + (3 / 7) log2 (3 / 7) + (2 / 7) log2 (2 / 7) + 0) approximately equal to 1.5575.

[0099] Then data 1 is greater than data 2, which means that the information amount of the vector corresponding to data 1 is higher and more complex;

[0100] The second determination unit comprises:

[0101] Through the second input layer, a second feature vector is generated based on the state data to be determined;

[0102] Through the third hidden layer, a third output vector is generated based on the second feature vector;

[0103] Through the third output layer, a second determination result is generated based on the third output vector, and the second determination result includes a first data set meeting the safe state and a second data set not meeting the safe state obtained based on the second determination unit;

[0104] At the same time, the first state identifier meeting the safe state is assigned, the second state identifier not meeting the safe state is assigned, and the third state identifier unable to determine whether to meet the safe state is assigned;

[0105] It should be noted that the first state identifier, the second state identifier and the third state identifier are combined with the corresponding information, and a hash value is generated by processing, which can be stored in a database, a file or other storage medium, and can be used for subsequent data verification, searching and other operations. The specific steps are not described here;

[0106] The calculation model of the second determination unit is:

[0107] H2 = sofmtax (G1 * F3) * w4 + b4;

[0108] F3 = 2tanh(w5*2S3+b5);

[0109] G1 = [G3*F3+(-G2)*G3]*w6;

[0110] In the formula, H2 represents the second determination result, S3 represents the third output vector, and F3 represents the third intermediate vector; w4 represents the fourth weight parameter, w5 represents the fifth weight parameter, and w6 represents the sixth weight parameter; b4 represents the fourth bias parameter, and b5 represents the fifth bias parameter; G1 represents the first gating value, G2 represents the second gating value, and G3 represents the third gating value; tanh() represents a nonlinear activation function;

[0111] It should be noted that, generally, the gating value is calculated based on some activation function (such as sigmoid or tanh), and the value thereof is between 0 and 1, and is used to "control" the information flow of other tensors; the complexity of defining the second determination unit is much smaller than that of the first determination unit, and the features obtained in the second determination unit process can be regarded as shallow-level features, and the features obtained in the first determination unit can be regarded as deep-level features; in the case of deep level (too high complexity), it is easy to increase the training difficulty, and in the embodiment, it is impossible to determine whether the safety state is met, so the application fuses the shallow level (lower complexity) with the deep level for training to ensure the accuracy of the model, which is beneficial to subsequent training optimization;

[0112] The machine learning model is trained based on the first determination result and the second determination result to obtain an updated machine learning model for next determination analysis; at least two pre-trained machine learning models are used to perform hierarchical determination on the state data to be determined to obtain the safety state level corresponding to the archive room, including:

[0113] First-level determination: obtaining data that does not exist in the third data set from the first data set to obtain a fourth data set; obtaining data that does not exist in the first data set from the third data set to obtain a fifth data set; inputting the fourth data set and the first state identifier and the fifth data set and the second state identifier into the machine learning model for training to obtain an updated first training model; inputting the state data into the first training model to perform first-level prediction to obtain a first prediction coefficient;

[0114] Second level determination: obtain data in the third data set that is not in the second data set to obtain a sixth data set; obtain data in the second data set that is not in the third data set to obtain a seventh data set; input the sixth data set and the first state identifier and the seventh data set and the second state identifier into the machine learning model for training to obtain an updated second training model; input the state data into the second training model for second level prediction to obtain a second prediction coefficient;

[0115] It should be noted that by inputting the state data to be determined into the first training model or the second training model, the prediction result obtained will undergo a conversion analysis, the purpose of which is to convert the output result of the first training model or the second training model into a prediction value; it can be processed by weighted average, weighted voting, etc., which will not be described here;

[0116] Comprehensive determination: based on the combination of the first prediction coefficient and the second prediction coefficient, a comprehensive prediction coefficient is generated, and the formula based on which the comprehensive prediction coefficient is generated is:

[0117] In the formula, forecast represents the comprehensive prediction coefficient, c1 represents the first prediction coefficient, c2 represents the second prediction coefficient, and θ1 and θ2 are both weight coefficients, and θ1 and θ2 are both greater than 0;

[0118] The comprehensive prediction coefficient is compared and analyzed with the standard prediction interval [g1, g2]:

[0119] When forecast < g1, a first level identifier is assigned, and the value is combined with the first level identifier to generate a first safety level, and it is determined that the safety value of the environment at the time is low;

[0120] When g1≤forecast≤g2, a second level identifier is assigned, and the value is combined with the second level identifier to generate a first safety level, and it is determined that the safety value of the environment at the time is general;

[0121] When forecast > g2, a third level identifier is assigned, and the value is combined with the third level identifier to generate a third safety level, and it is determined that the safety value of the environment at the time is high;

[0122] S2, preliminarily process the multi-dimensional environment data to generate a first processing sequence and a second processing sequence, and extract a first key feature and a second key feature from the first processing sequence and the second processing sequence respectively, input the first processing sequence and the first key feature into a pre-constructed first calculation model to calculate a first evaluation coefficient; input the second processing sequence and the second key feature into a pre-constructed second calculation model to calculate a second evaluation coefficient; based on the first evaluation coefficient and the second evaluation coefficient, a multi-dimensional feature space is constructed;

[0123] The preliminary processing of the multi-dimensional environment data includes:

[0124] Primary processing: removing noise from the multi-dimensional environment data through a band-pass filter to obtain a first processing sequence, and extracting first key features including mean and variance based on the first processing sequence;

[0125] Importing a first calculation model:

[0126] Constructing a first parameter matrix based on the first processing sequence:

[0127] In the formula, Ja1 represents the first parameter matrix, j represents the first processing sequence, j 1,1 , j 1,2 , j 1,3 , and j 1,4 respectively represent the temperature value, humidity value, air pressure value, and safety state level at the first time, j 2,1 , j 2,2 , j 2,3 , and j 2,4 respectively represent the temperature value, humidity value, air pressure value, and safety state level at the second time, j N0,1 , j N0,2 , j N0,3 , and j N0,4 respectively represent the temperature value, humidity value, air pressure value, and safety state level at the N0th time;

[0128] Calculating a mean factor based on the mean of each column vector of the first parameter matrix, calculating a difference factor based on the variance of each column vector, and combining the mean factor and the difference factor to obtain a first evaluation coefficient of the column vector;

[0129] The formula for calculating the first evaluation coefficient is:

[0130]

[0131] In the formula, assess1 represents the first evaluation coefficient, μ represents the mean factor, σ represents the difference factor, and α1 and α1 are weight coefficients;

[0132] For example, taking the first group of column vectors:

[0133] The formula for calculating the mean factor is:

[0134] The formula for calculating the difference factor is:

[0135] In the formula, m represents the number of column vectors, and its value range is [1, N0];

[0136] Secondary processing: secondary denoising on the first processing sequence by fast Fourier transform, and obtaining the second processing sequence by using inverse Fourier transform to restore to the time domain, and extracting the second key features including difference, standard deviation and quartile deviation based on the second processing sequence;

[0137] Importing the second calculation model:

[0138] Constructing the second parameter matrix based on the second processing sequence:

[0139] In the formula, Ja2 represents the second parameter matrix, y represents the second processing sequence, each row vector represents the second processing sequence obtained in N0 time measurement periods, and each column vector represents the temperature value, humidity value, air pressure value and safety state grade corresponding to the second processing sequence respectively;

[0140] Based on the difference between adjacent vectors of each column vector of the second parameter matrix and combined with the standard deviation, the rate factor is calculated, and based on the quartile deviation of each column vector, the fluctuation factor is calculated, and the second evaluation coefficient of the column vector is obtained by combining the rate factor and the fluctuation factor;

[0141] The formula for calculating the second evaluation coefficient is:

[0142]

[0143] In the formula, assess2 represents the second evaluation coefficient, rate represents the rate factor, fluct represents the fluctuation factor, and β1 and β1 are weight coefficients;

[0144] For example, taking the first group of column vectors:

[0145] The formula for calculating the rate factor is:

[0146] The formula for calculating the fluctuation factor is:

[0147] In the formula, IOR represents the quartile deviation, which is calculated as the difference between the third quantile and the first quantile, and is used to measure the dispersion degree or volatility of the data; represents the average value of the first group of vectors, y n,1 represents a column vector;

[0148] Comprehensive processing: constructing the corresponding multi-dimensional feature space with time measurement period N0 as the X axis, first evaluation coefficient assess1 as the Y axis, and second evaluation coefficient assess2 as the Z axis, and the multi-dimensional feature space is in the form of a three-dimensional graph;

[0149] It should be noted that by combining temperature values, humidity values, air pressure values and safety state levels with each other for analysis, the accuracy and precision of system analysis are further improved in the process of judgment;

[0150] S3, deep analysis of multi-dimensional feature space, overlapping the multi-dimensional feature space and the standard feature space with each other to generate evaluation factors, constructing a dynamic change curve based on the evaluation factors, comparing the dynamic change curve with a preset standard threshold curve to obtain abnormal time periods and abnormal areas, marking the abnormal multi-dimensional feature space, generating an archive room environment multi-dimensional thermal map through visual analysis and uploading and displaying;

[0151] Deep analysis of multi-dimensional feature space, overlapping the multi-dimensional feature space and the standard feature space with each other to generate evaluation factors, constructing a dynamic change curve based on the evaluation factors, comparing the dynamic change curve with a preset standard threshold curve to obtain abnormal time periods and abnormal areas, including:

[0152] Spatial overlap: preset standard feature space, overlapping the multi-dimensional feature space and the standard feature space to obtain horizontal overlap and vertical overlap, obtaining overlapping area and non-overlapping area based on the comparison of horizontal overlap and vertical overlap; based on the combination of overlapping area and non-overlapping area, evaluation factors are generated;

[0153] The formula for calculating the evaluation factor is:

[0154] In the formula, Evaluation represents the evaluation factor, fmj represents the non-overlapping area, cmj represents the overlapping area, and lambda represents the offset factor;

[0155] The formula for calculating the offset factor is:

[0156] In the formula, hx represents the horizontal overlap, zx represents the vertical overlap, bz represents the multi-dimensional feature space, and Bz represents the standard feature space;

[0157] Curve comparison: a two-dimensional rectangular coordinate system is established, and a dynamic change curve of the evaluation factor-time is drawn; at the same time, a standard threshold curve is drawn in the two-dimensional rectangular coordinate system, and the dynamic change curve and the standard threshold curve are compared and analyzed, the time period in which the dynamic change curve is above the standard threshold curve and the area surrounded by the standard threshold curve are counted, and are marked as abnormal time period and abnormal area;

[0158] It should be noted that the abnormal period represents the time period when the environment of the archive room appears abnormal during the evaluation process, and the abnormal area reflects the severity of the environment of the archive room during the evaluation process, and the larger the abnormal area, the worse the environment, and the greater the risk of fire; by analyzing the abnormal time period and the abnormal area, the abnormal label is marked in the multi-dimensional feature space, and the multi-dimensional heat map of the environment of the archive room is drawn through visual analysis, so as to accurately track the changes of the archive room and improve the accuracy of subsequent abnormal analysis; based on the multi-dimensional heat map of the environment of the archive room, clustering algorithms such as K-means or DBSCAN can be used to group the abnormal points in the map, and by analyzing which areas in the heat map are most concentrated, combined with the sensors or devices involved in the system, the mode or reason of the abnormality can be identified; the specific process will not be described here.

[0159] In summary of the above technical solutions: the present application first determines and analyzes the state data by building a machine learning model, and forms multi-dimensional environmental data by introducing temperature, humidity and air pressure values and other related data, which effectively monitors the archive room; by introducing the first processing sequence and the first key feature into the first calculation model, the first evaluation coefficient is calculated, and by introducing the second processing sequence and the second key feature into the second calculation model, the second evaluation coefficient is calculated, forming a multi-dimensional feature space, and overlapping it with a standard feature space, generating an evaluation factor, to build a dynamic change curve, obtain an abnormal period and an abnormal area, and mark the multi-dimensional feature space of the abnormality, and generate a multi-dimensional heat map of the environment of the archive room, which improves the accuracy of the model to a certain extent and plays a timely warning role.

[0160] Example 2

[0161] The present application provides an archive room environment comprehensive evaluation system based on artificial intelligence. Figure 2 is a module schematic diagram of the archive room environment comprehensive evaluation system according to an exemplary embodiment. Please refer to Figure 2 The system comprises a data acquisition module, a preliminary processing module and a deep analysis module, and the data acquisition module, the preliminary processing module and the deep analysis module are in communication connection;

[0162] The data acquisition module is used for acquiring multi-dimensional environmental data of the archive room to be evaluated at regular intervals, and the multi-dimensional environmental data includes temperature value, humidity value, air pressure value, safety state level and time stamp, wherein the safety state level is obtained based on a pre-established machine learning model;

[0163] The preliminary processing module is configured to preliminarily process the multi-dimensional environment data, generate a first processing sequence and a second processing sequence, extract a first key feature and a second key feature from the first processing sequence and the second processing sequence respectively, input the first processing sequence and the first key feature into a first calculation model constructed in advance, and calculate a first evaluation coefficient; input the second processing sequence and the second key feature into a second calculation model constructed in advance, and calculate a second evaluation coefficient; and construct a multi-dimensional feature space based on the first evaluation coefficient and the second evaluation coefficient.

[0164] The deep analysis module is configured to overlap the multi-dimensional feature space and a standard feature space with each other, generate an evaluation factor, construct a dynamic change curve based on the evaluation factor, compare the dynamic change curve with a preset standard threshold curve, obtain an abnormal period and an abnormal area, mark the multi-dimensional feature space that is abnormal, and generate a multi-dimensional thermal map of the archive room environment through visual analysis and upload and display the multi-dimensional thermal map.

[0165] In the present application, the weight coefficients involved are determined by using the coefficient of variation method. The coefficient of variation method is a method of weighting each index according to the variation degree of the current value and the target value of each evaluation index. If the numerical difference of an index is large and each evaluated object can be clearly distinguished, it means that the index has rich distinguishing information, and thus the index should be given a larger weight. Conversely, if the numerical difference of each evaluated object on an index is small, the index has weak ability to distinguish each evaluated object, and thus the index should be given a smaller weight. This method directly uses the information contained in each index to obtain the weight of the index by calculation, and thus is objective.

[0166] In the present application, the several formulas involved are calculated by taking the numerical values after dimensionless, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the most recent real situation. The parameters in the formulas are set by a person skilled in the art according to the actual situation.

[0167] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized in the form of a computer program product wholly or partially. A person of ordinary skill in the art can realize that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software depends on the specific application and design constraints of the technical solutions.

[0168] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, and may be located in one place, or distributed on multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiment of the present application according to actual needs.

[0169] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A comprehensive evaluation method for the environment of an archive storage facility based on artificial intelligence, characterized in that: The method includes: The system periodically acquires multidimensional environmental data to be evaluated from the archive storage. This multidimensional environmental data includes temperature, humidity, air pressure, safety status level, and timestamp. The safety status level is determined based on a pre-established machine learning model. The multidimensional environmental data is initially processed to generate a first processing sequence and a second processing sequence. A first key feature and a second key feature are extracted from the first processing sequence and the second processing sequence, respectively. The first processing sequence and the first key feature are imported into a pre-constructed first calculation model to calculate a first evaluation coefficient. The second processing sequence and the second key feature are imported into a pre-constructed second calculation model to calculate a second evaluation coefficient. Based on the first evaluation coefficient and the second evaluation coefficient, a multidimensional feature space is constructed. In-depth analysis of the multidimensional feature space is performed, and the multidimensional feature space and the standard feature space are overlapped to generate evaluation factors. Based on the evaluation factors, dynamic change curves are constructed. The dynamic change curves are compared with the preset standard threshold curves to obtain abnormal time periods and abnormal areas. The abnormal multidimensional feature space is marked, and a multidimensional heat map of the archive storage environment is generated through visualization analysis and uploaded for display. The machine learning model includes a first decision unit and a second decision unit. The first decision unit comprises a first input layer, a first hidden layer, a second hidden layer, and a first output layer. The second decision unit comprises a second input layer, a third hidden layer, and a second output layer. The decision process includes the following steps: First determination unit: The first feature vector is generated based on the state data to be determined through the first input layer. A first output vector is generated based on the first feature vector through the first hidden layer; The second output vector is generated based on the first output vector through the second hidden layer; Through the first output layer, based on the second output vector, a first determination result is generated. The first determination result includes a first data set that conforms to the safe state, a second data set that does not conform to the safe state, and a third data set that cannot be determined to conform to the safe state, all obtained based on the first determination unit. Second determination unit: The second input layer generates a second feature vector based on the state data to be determined. A third output vector is generated based on the second feature vector through the third hidden layer; Through the third output layer, based on the third output vector, a second determination result is generated. The second determination result includes a first data set that conforms to the safe state and a second data set that does not conform to the safe state, obtained based on the second determination unit. At the same time, a first state identifier is assigned to meet the safety status, a second state identifier is assigned to not meet the safety status, and a third state identifier is assigned to which it is impossible to determine whether the safety status is met. The status data includes at least the collected fire water tank level and sprinkler network pressure; A machine learning model is trained based on the first and second decision results, and an updated machine learning model is obtained for the next decision analysis.

2. The comprehensive evaluation method for the archive storage environment based on artificial intelligence according to claim 1, characterized in that, The machine learning model is pre-trained at least twice, and the trained machine learning model is used to classify the state data to be judged in order to obtain the security status level of the archive storage, including: First-level determination: Obtain data that does not exist in the third data set from the first data set, and obtain the fourth data set; obtain data that does not exist in the first data set from the third data set, and obtain the fifth data set; input the fourth data set and the first state identifier, and the fifth data set and the second state identifier as input information into the machine learning model for training, and obtain the updated first training model; input the state data into the first training model, perform the first-level prediction, and obtain the first prediction coefficient; Second-level determination: Obtain data from the third data set that does not exist in the second data set to obtain the sixth data set; obtain data from the second data set that does not exist in the third data set to obtain the seventh data set; import the sixth data set and the first state identifier, and the seventh data set and the second state identifier as input information into the machine learning model for training to obtain an updated second training model; input the state data into the second training model to perform second-level prediction and obtain the second prediction coefficient; Comprehensive judgment: Based on the combination of the first and second prediction coefficients, a comprehensive prediction coefficient is generated, and the comprehensive prediction coefficient is compared and analyzed with the standard prediction interval [g1, g2] to obtain the corresponding safety status level; The formula used for the comprehensive prediction coefficient is as follows: ; In the formula, forecast represents the overall prediction coefficient, c1 represents the first prediction coefficient, c2 represents the second prediction coefficient, θ1 and θ2 are both weighting coefficients, and both θ1 and θ2 are greater than 0.

3. The comprehensive evaluation method for the archive storage environment based on artificial intelligence according to claim 1, characterized in that, The calculation model of the first determination unit is as follows: ； ； ； In the formula, H1 represents the first decision result, F1 represents the first intermediate vector, F2 represents the second intermediate vector, S1 represents the first output vector, and S2 represents the second output vector; softmax() represents the normalized exponential activation function, and sigmoid() represents the nonlinear activation function; w1 represents the first weight parameter, w2 represents the second weight parameter, and w3 represents the third weight parameter; b1 represents the first bias parameter, b2 represents the second bias parameter, and b3 represents the third bias parameter; ⊙ represents the successive multiplication operation. This represents the complexity measurement operation, where Θ represents the complexity, θ represents the range of the complexity, and θ takes the value [Θ / 2, Θ].

4. The comprehensive evaluation method for the archive storage environment based on artificial intelligence according to claim 1, characterized in that, The calculation model of the second determination unit is as follows: ； ； ； In the formula, H2 represents the second decision result, S3 represents the third output vector, F3 represents the third intermediate vector, g1 represents the first gate value, w4 represents the fourth weight parameter, w5 represents the fifth weight parameter, and w6 represents the sixth weight parameter. b4 represents the fourth bias parameter, b5 represents the fifth bias parameter; G1 represents the first gating value, G2 represents the second gating value, G3 represents the third gating value; tanh() represents the nonlinear activation function.

5. The comprehensive evaluation method for the archive storage environment based on artificial intelligence according to claim 1, characterized in that, The preliminary processing of multidimensional environmental data includes: First processing: The multidimensional environmental data is removed by a bandpass filter to obtain the first processing sequence, and the first key features, including the mean and variance, are extracted based on the first processing sequence; Import the first calculation model: Construct the first parameter matrix based on the first processing sequence, calculate the mean factor based on the average value of each column vector of the first parameter matrix, calculate the difference factor based on the variance of each column vector, and combine the mean factor and the difference factor to obtain the first evaluation coefficient of the column vector. Secondary processing: The first processed sequence is denoised a second time by fast Fourier transform, and the time domain is restored by inverse Fourier transform to obtain the second processed sequence. The second key features, including difference, standard deviation and interquartile range, are extracted based on the second processed sequence. Import the second calculation model: Construct a second parameter matrix based on the second processing sequence, calculate the growth rate factor based on the difference between adjacent vectors of each column vector of the second parameter matrix and the standard deviation, calculate the volatility factor based on the interquartile range of each column vector, and combine the growth rate factor and the volatility factor to obtain the second evaluation coefficient of the column vector. Comprehensive processing: Using the time acquisition period as the X-axis, the first evaluation coefficient as the Y-axis, and the second evaluation coefficient as the Z-axis, a corresponding multi-dimensional feature space is constructed, and the multi-dimensional feature space is represented in the form of a three-dimensional graph.

6. The comprehensive evaluation method for the archive storage environment based on artificial intelligence according to claim 1, characterized in that, The deep analysis of the multidimensional feature space involves overlapping the multidimensional feature space and the standard feature space to generate evaluation factors. A dynamic change curve is constructed based on these evaluation factors, and the dynamic change curve is compared with a preset standard threshold curve to obtain abnormal time periods and abnormal areas. This includes: Spatial overlap: A standard feature space is preset, and the multi-dimensional feature space is overlapped with the standard feature space to obtain the horizontal overlap and vertical overlap. The overlapping area and non-overlapping area are obtained based on the comparison of the horizontal and vertical overlap. Evaluation factors are generated based on the combination of overlapping and non-overlapping areas. Curve Comparison: Establish a two-dimensional rectangular coordinate system and plot the dynamic change curve of the evaluation factor-time; at the same time, plot a standard threshold curve in the two-dimensional rectangular coordinate system, compare and analyze the dynamic change curve and the standard threshold curve, and count the time periods when the dynamic change curve is above the standard threshold curve and the area enclosed by the standard threshold curve, and mark them as abnormal time periods and abnormal areas.

7. An AI-based comprehensive evaluation system for the environment of an archive storage facility, characterized in that: include: The data acquisition module is used to periodically acquire multi-dimensional environmental data to be evaluated in the archive storage room. The multi-dimensional environmental data includes temperature, humidity, air pressure, safety status level and timestamp. The safety status level is determined based on a pre-established machine learning model. The preliminary processing module is used to perform preliminary processing on multidimensional environmental data, generate a first processing sequence and a second processing sequence, and extract a first key feature and a second key feature from the first processing sequence and the second processing sequence, respectively. The first processing sequence and the first key feature are imported into a pre-built first calculation model to calculate a first evaluation coefficient. The second processing sequence and the second key feature are imported into a pre-built second calculation model to calculate a second evaluation coefficient. Based on the first evaluation coefficient and the second evaluation coefficient, a multidimensional feature space is constructed. The deep analysis module overlaps the multidimensional feature space and the standard feature space to generate evaluation factors. Based on the evaluation factors, a dynamic change curve is constructed. The dynamic change curve is compared with the preset standard threshold curve to obtain abnormal time periods and abnormal areas. The abnormal multidimensional feature space is marked. Through visualization analysis, a multidimensional heat map of the archive storage environment is generated and uploaded for display. The machine learning model includes a first decision unit and a second decision unit. The first decision unit comprises a first input layer, a first hidden layer, a second hidden layer, and a first output layer. The second decision unit comprises a second input layer, a third hidden layer, and a second output layer. The decision process includes the following steps: First determination unit: The first feature vector is generated based on the state data to be determined through the first input layer. A first output vector is generated based on the first feature vector through the first hidden layer; The second output vector is generated based on the first output vector through the second hidden layer; Through the first output layer, based on the second output vector, a first determination result is generated. The first determination result includes a first data set that conforms to the safe state, a second data set that does not conform to the safe state, and a third data set that cannot be determined to conform to the safe state, all obtained based on the first determination unit. Second determination unit: The second input layer generates a second feature vector based on the state data to be determined. A third output vector is generated based on the second feature vector through the third hidden layer; Through the third output layer, based on the third output vector, a second determination result is generated. The second determination result includes a first data set that conforms to the safe state and a second data set that does not conform to the safe state, obtained based on the second determination unit. At the same time, a first state identifier is assigned to meet the safety status, a second state identifier is assigned to not meet the safety status, and a third state identifier is assigned to which it is impossible to determine whether the safety status is met. The status data includes at least the collected fire water tank level and sprinkler network pressure; A machine learning model is trained based on the first and second decision results, and an updated machine learning model is obtained for the next decision analysis.

Images