A decision-making method, device, equipment and storage medium for multi-data fusion

Abstract

The present invention relates to a decision-making method, device, equipment and storage medium for multi-data fusion. The method includes: collecting multi-dimensional target data, where the multi-dimensional target data includes normal data and / or abnormal data; detecting the abnormal data, and based on an RBF neural network model of the cuckoo search algorithm, correcting the abnormal data to obtain corrected data; locally fusing the corrected data and the normal data of the same dimension to obtain feature data of the same dimension; based on the improved D-S evidence theory, globally fusing the feature data to determine a target decision-making scheme. The present invention relates to a decision-making method, device, equipment and storage medium for multi-data fusion. After obtaining the target data, the abnormal data therein is corrected, the corrected data and the normal data are locally fused, and a decision-making scheme is determined according to the fused data, thus completing the state of things and improving the accuracy of decision-making.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent decision-making, and particularly to a decision-making method, device, equipment and storage medium for multi-data fusion. Background Art

[0002] With the rapid development of industrial Internet of Things technology, the integration degree of the system is getting higher and higher, and the data space system based on big data technology is gradually improving. The traditional decision-making process relies on a single data source for analysis, and the decision-making accuracy and trust are limited by the data source, making it difficult to meet the basic requirements of the integrated system. In the data space system, there are correlations and couplings between data sources. By using information fusion means to extract the data feature set between multi-source heterogeneous data, each data feature provides strong evidence for the final decision after data cleaning, improving the accuracy of system decision-making. Therefore, in the process of intelligent decision-making, in-depth research on data fusion algorithms is of great significance.

[0003] In a big data integration system, information fusion with multiple data sources and complex structures is a problem to be solved. Problems such as data registration, heterogeneous data, uncertainty, incompleteness and inconsistency of sensor data, false data, data association and situation database will seriously affect the system decision-making process. In order to achieve collaborative work between multiple subsystems, in current information fusion technology, data-level fusion processing measures the data itself and then makes a decision on the data space.

[0004] However, in the prior art, measuring the data itself lacks the support of context, the current state description of the thing attributes is incomplete, and accurate decisions cannot be made based on the current state of the thing attributes. Summary of the Invention

[0005] In view of this, it is necessary to provide a decision-making method, device, equipment and storage medium for multi-data fusion to solve the problem that in the prior art, the current state description of the thing attributes is incomplete and accurate decisions cannot be made based on the current state of the thing attributes.

[0006] To achieve the above technical purpose, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a decision-making method for multi-data fusion, including:

[0008] Collect multi-dimensional target data, where the multi-dimensional target data includes normal data and / or abnormal data;

[0009] Detect abnormal data, and based on the RBF neural network model of the cuckoo search algorithm, correct the abnormal data to obtain corrected data;

[0010] Locally fuse the corrected data and the normal data in the same dimension to obtain feature data in the same dimension;

[0011] Based on the improved D-S evidence theory, the feature data is globally fused to determine the target decision-making scheme.

[0012] Preferably, based on the cuckoo search optimization algorithm-based RBF neural network model, the abnormal data is corrected to obtain the corrected data, including:

[0013] According to the cuckoo search optimization algorithm, calculate the target parameters of the RBF neural network model;

[0014] According to the abnormal data and the target parameters, calculate the output of the Gaussian radial basis function of the RBF neural network model;

[0015] According to the target parameters and the output of the Gaussian radial basis function, calculate the corrected data.

[0016] Preferably, according to the cuckoo search optimization algorithm, calculating the target parameters of the RBF neural network model includes:

[0017] Obtain the training set sample data, initialize the RBF neural network model, and input the training set sample data;

[0018] Initialize the parameters of the cuckoo search optimization algorithm, and calculate the optimal value of the target parameters of the current cuckoo search optimization algorithm parameters;

[0019] Repeat to update the parameters of the cuckoo search optimization algorithm, calculate and update the optimal value of the target parameters until the preset termination condition is reached to obtain the target parameters.

[0020] Preferably, the corrected data and the normal data in the same dimension are locally fused to obtain the feature data in the same dimension, including:

[0021] According to the preset prediction method, predict the real data in the same dimension to obtain the predicted real data in the same dimension;

[0022] According to the predicted real data, determine the variance in the same dimension;

[0023] According to the variance in the same dimension, determine the weights of the corrected data and the normal data in the same dimension;

[0024] According to the weights, determine the feature data in the same dimension.

[0025] Preferably, based on the improved D-S evidence theory, globally fusing the feature data to determine the target decision-making scheme includes:

[0026] According to the multi-dimensional target data, determine all decision-making schemes;

[0027] Based on the fuzzy theory, determine the support probability of all feature data for each decision-making scheme;

[0028] Based on the cosine angle method, according to the support probability, determine whether the feature data has conflicts;

[0029] When the feature data conflicts, calculate the corrected probability of the support probability, and determine the target decision-making scheme according to the corrected probability.

[0030] Preferably, based on the cosine angle method, according to the support probability, determine whether there are conflicts between the feature data, including:

[0031] According to the support probability, calculate the cosine similarity between every two of all the feature data;

[0032] According to the cosine similarity, determine whether there are conflicts between the feature data.

[0033] Preferably, when the feature data conflicts, calculate the corrected probability of the support probability, including:

[0034] Calculate the average probability of the support probability of each feature data for the same decision-making scheme;

[0035] According to the average probability, determine the distance between each feature data and the average probability of the same decision-making scheme;

[0036] According to the distance of the average probability, determine the weight of the feature data;

[0037] According to the weight of the feature data, calculate the corrected probability of the support probability.

[0038] In a second aspect, the present invention also provides a decision-making device for multi-data fusion, including:

[0039] An acquisition module, configured to acquire multi-dimensional target data, where the multi-dimensional target data includes normal data and / or abnormal data;

[0040] A correction module, configured to detect abnormal data, and based on an RBF neural network model of the cuckoo optimization algorithm, correct the abnormal data to obtain corrected data;

[0041] A local fusion module, configured to locally fuse the corrected data and the normal data of the same dimension to obtain feature data of the same dimension;

[0042] A global fusion module, configured to globally fuse the feature data based on the improved D-S evidence theory to determine the target decision-making scheme.

[0043] In a third aspect, the present invention also provides an electronic device, including a memory and a processor, wherein,

[0044] The memory is used to store programs;

[0045] A processor, coupled to a memory, is configured to execute a program stored in the memory to implement the steps in the multi-data fusion decision method in any of the above implementation manners.

[0046] In a fourth aspect, the present invention further provides a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, the steps in the multi-data fusion decision method in any of the above implementation manners can be implemented.

[0047] The beneficial effects of adopting the above embodiments are as follows: A multi-data fusion decision method, device, equipment, and storage medium provided by the present invention detect whether the collected data is abnormal, correct the abnormal data through an RBF neural network model of the cuckoo search algorithm, improve the accuracy of decision-making, perform local fusion on the corrected data and normal data to obtain feature data of the same dimension, complete the description of the current state of the attributes of things, and then perform global fusion through an improved D-S evidence theory to determine the target decision-making scheme, further improving the accuracy of the decision-making method. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic flowchart of an embodiment of the multi-data fusion decision method provided by the present invention;

[0049] Figure 2 It is a schematic flowchart of an embodiment of abnormal data correction provided by the present invention;

[0050] Figure 3 It is a schematic flowchart of an embodiment of local fusion provided by the present invention;

[0051] Figure 4 It is a schematic flowchart of an embodiment of global fusion provided by the present invention;

[0052] Figure 5 It is a schematic structural diagram of an embodiment of the multi-data fusion decision device provided by the present invention;

[0053] Figure 6 It is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] The following will specifically describe the preferred embodiments of the present invention in conjunction with the accompanying drawings. The accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention, rather than to limit the scope of the present invention.

[0055] In the description of this application, "a plurality of" means two or more, unless otherwise specifically defined.

[0056] References to "embodiments" in this specification mean that specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of the invention. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0057] The present invention provides a decision-making method, device, equipment, and storage medium for multi-data fusion, which will be described separately below.

[0058] Please refer to Figure 1 , Figure 1 , which is a schematic flowchart of an embodiment of the decision-making method for multi-data fusion provided by the present invention. A specific embodiment of the present invention discloses a decision-making method for multi-data fusion, including:

[0059] S101. Collect multi-dimensional target data, where the multi-dimensional target data includes normal data and / or abnormal data;

[0060] S102. Detect abnormal data, and based on the RBF neural network model of the cuckoo search algorithm, correct the abnormal data to obtain corrected data;

[0061] S103. Locally fuse the corrected data and normal data of the same dimension to obtain feature data of the same dimension;

[0062] S104. Based on the improved D-S evidence theory, globally fuse the feature data to determine the target decision-making scheme.

[0063] In a specific embodiment of the present invention, step S101 collects multi-dimensional target data of the current environment. The target data is data that has a significant impact on the target decision-making scheme, and the collection method can be specifically set according to the actual situation. To improve the accuracy of the target decision-making scheme, data should be collected at multiple points. A specific embodiment provided by the present invention collects multiple-dimensional target parameters in the current environment through multiple sensors. It can be understood that the present invention does not further limit the means of data collection, as long as it can accurately collect the accurate values of the target data in the current environment.

[0064] In a specific embodiment of the present invention, in step S102, box plots are used to detect outliers in the original data to determine the abnormal data, and then the abnormal data is corrected. Abnormal data will affect the accuracy of the final target decision-making scheme and must be processed to avoid errors in the final decision-making scheme. The RBF neural network model of the cuckoo search algorithm can predict the current outlier and replace it by leveraging the correlation between data in adjacent time windows, eliminating the impact of abnormal data on the decision-making scheme.

[0065] In a specific embodiment of the present invention, in step S103, since there may be differences in the measurement of the same data at different points in the current environment, the data at all measurement points in the same dimension are locally fused to obtain the feature data of all the same dimensions. Local fusion can complete the description of the current state of the attributes of things, thereby further improving the accuracy of the target decision-making scheme.

[0066] In a specific embodiment of the present invention, in step S104, the feature data of different dimensions have different impacts on the decision-making scheme. The feature data of all the same dimensions obtained by local fusion are globally fused to determine the impact of each feature data on the final decision and further determine the best decision-making scheme, that is, the target decision-making scheme.

[0067] Compared with the prior art, a decision-making method for multi-data fusion provided in this embodiment detects whether the collected data is abnormal, corrects the abnormal data through the RBF neural network model of the cuckoo search algorithm, improves the accuracy of the decision-making, locally fuses the corrected data and the normal data to obtain the feature data of the same dimension, completes the description of the current state of the attributes of things, and then performs global fusion through the improved D-S evidence theory to determine the target decision-making scheme, further improving the accuracy of the decision-making method.

[0068] Please refer to Figure 2 , Figure 2 which is a schematic flowchart of an embodiment of abnormal data correction provided by the present invention. In some embodiments of the present invention, based on the RBF neural network model of the cuckoo search algorithm, the abnormal data is corrected to obtain the corrected data, including:

[0069] S201. Calculate the target parameters of the RBF neural network model according to the cuckoo search algorithm;

[0070] S202. Calculate the output of the Gaussian radial basis function of the RBF neural network model according to the abnormal data and the target parameters;

[0071] S203. Calculate the corrected data according to the target parameters and the output of the Gaussian radial basis function.

[0072] In a specific embodiment of the present invention, the target parameter in step S201 is the connection weight w j The center c of the basis function j and the variance σ. The target parameter will directly affect the correction result of the RBF neural network model, and the cuckoo optimization can improve the calculation speed and accuracy.

[0073] In a specific embodiment of the present invention, in step S202, the abnormal data and the target parameter are used as inputs and input into the RBF neural network model to obtain the output of the Gaussian radial basis function. The calculation formula of the Gaussian radial basis function is as follows:

[0074]

[0075] where x ij is the j-th data in the i-th group of data, c i is the center of the i-th radial basis function, ||x ij -c i || is the distance between x ij and c i , σ is the field width of the hidden node, and the smaller σ is, the smaller the width of the radial basis function is. It should be noted that x ij here only represents the abnormal data in the original data, that is, it means that the j-th data in the i-th group of data is abnormal.

[0076] In a specific embodiment of the present invention, in step S203, the correction data is further calculated through the target parameter and the output of the Gaussian radial basis function calculated in step S202, and its specific formula is as follows:

[0077]

[0078] where w0 is the offset of the output neuron, and w i is the connection weight between the output result of the i-th hidden layer neuron and the output layer node.

[0079] In some embodiments of the present invention, according to the cuckoo optimization algorithm, the target parameters of the RBF neural network model are calculated, including:

[0080] Obtain the training set sample data, initialize the RBF neural network model, and input the training set sample data;

[0081] Initialize the parameters of the cuckoo optimization algorithm, and calculate the optimal value of the target parameter of the current cuckoo optimization algorithm parameters;

[0082] Repeat to update the parameters of the cuckoo optimization algorithm, calculate and update the optimal value of the target parameter until the preset termination condition is reached, and obtain the target parameter.

[0083] In the above embodiments, the RBF neural network is first initialized, and the acquired training set sample data is input. Then the Cuckoo Search (CS) algorithm is initialized: that is, parameters such as the number of nests n and the discovery probability Pa are set. Then the CS algorithm is executed to randomly generate the initial positions of each nest, and the combined data of the three target parameters [c1, σ1, w1, c2, σ2, w2... c n , σ n , w n are used as the position values of each nest in the CS algorithm, substituted into the fitness function to calculate and evaluate the quality of the solution, and the current optimal nest position solution and the fitness function value are obtained. Continue with the global search and position update, calculate the fitness value of the new position and evaluate the quality of the fitness value to obtain a new optimal solution, and retain the value of the better optimal solution. Determine whether the error meets the threshold requirement. If it meets, output the optimal result to obtain the optimal parameter values of the RBF neural network; otherwise, continue to calculate the value of the optimal solution. That is, the optimal target parameters are obtained.

[0084] Please refer to Figure 3 , Figure 3 , which is a schematic flowchart of an embodiment of local fusion provided by the present invention. In some embodiments of the present invention, the corrected data and the normal data in the same dimension are locally fused to obtain the feature data in the same dimension, including:

[0085] S301. According to a preset prediction method, predict the real data in the same dimension to obtain the predicted real data in the same dimension;

[0086] S302. Determine the variance in the same dimension according to the predicted real data;

[0087] S303. Determine the weights of the corrected data and the normal data in the same dimension according to the variance in the same dimension;

[0088] S304. Determine the feature data in the same dimension according to the weights.

[0089] In a specific embodiment of the present invention, the preset prediction method in step S301 is the distribution map method, and the predicted real data A is obtained through the distribution map method. It should be noted that the predicted real data A obtained through the distribution map method may have a certain error, but here it is only for calculating the weights of each sensor, and the error therein is acceptable.

[0090] In a specific embodiment of the present invention, in step S302, the data variance in the same dimension is calculated through the predicted real data A, the corrected data, and the normal data. The variance reflects the deviation between the reliable data (i.e., the corrected data and the normal data) and the predicted real data A.

[0091] In a specific embodiment of the present invention, in step S303, according to the deviation between the trusted data and the predicted true data A, the weight of the trusted data is further determined, and the weight reflects the credibility between the trusted data in the same dimension.

[0092] In a specific embodiment of the present invention, in step S304, through the weights of the correction data and the normal data in the same dimension, local fusion is performed, and the trusted data in the same dimension is weighted and fused, and finally the characteristic data that can reflect the most real situation of this dimension is obtained.

[0093] In the above embodiment, by calculating the variance and the weight, the accuracy of the characteristic data in the same dimension finally determined is improved, which can truly reflect the actual situation, and also completes the description of the current state of the attributes of things, thereby improving the accuracy of the final decision-making scheme.

[0094] Please refer to Figure 4 , Figure 4 which is a schematic flowchart of an embodiment of the global fusion provided by the present invention. In some embodiments of the present invention, based on the improved D-S evidence theory, the characteristic data is globally fused to determine the target decision-making scheme, including:

[0095] S401. Determine all decision-making schemes according to the multi-dimensional target data;

[0096] S402. Based on the fuzzy theory, determine the support probability of all characteristic data for each decision-making scheme;

[0097] S403. Based on the cosine of the included angle method, judge whether the characteristic data has conflicts according to the support probability;

[0098] S404. When the characteristic data conflicts, calculate the corrected probability of the support probability, and determine the target decision-making scheme according to the corrected probability.

[0099] In a specific embodiment of the present invention, in step S401, according to the collected multi-dimensional target data, the conditions of the target data are set to determine all feasible decision-making schemes, and the decision-making schemes corresponding to different ranges of the target data are different. It should be noted that the decision-making scheme also includes the conditions for executing the decision-making scheme.

[0100] In a specific embodiment of the present invention, in step S402, the support probability is represented by the membership function of the fuzzy theory. Each characteristic data has a unique support probability for each decision-making scheme. According to the characteristic data after local fusion, the support probability of each characteristic data for each decision-making scheme is calculated respectively.

[0101] In a specific embodiment of the present invention, step S403 uses the cosine angle method to determine whether the feature data conflicts. By using the cosine angle method, the cosine values of the angles between all pairs of feature data are calculated. The cosine value reflects the conflict situation between the feature data. Therefore, it is possible to determine whether there is a conflict between the feature data based on the cosine value.

[0102] In a specific embodiment of the present invention, in step S404, the probabilities of the conflicting feature data are re-corrected so that the corrected probabilities can accurately reflect the actual state of things, and then the target decision-making scheme is determined according to the corrected probabilities.

[0103] In some embodiments of the present invention, based on the cosine angle method, it is determined whether there is a conflict between the feature data according to the support probability, including:

[0104] Calculate the cosine similarity between all pairs of feature data according to the support probability;

[0105] Determine whether there is a conflict between the feature data according to the cosine similarity.

[0106] In the above embodiment, the cosine similarity calculation formula for the support probability m1 of the feature data and the support probability m2 of the feature data is as follows:

[0107]

[0108] where s ij is the cosine similarity.

[0109] It is necessary to determine whether there is a conflict between the feature data according to the calculation result. It should be noted that when s ij ∈ {1, 0, -1} indicates that the two feature data are {exactly the same, completely unrelated, completely conflicting}.

[0110] In some embodiments of the present invention, when the feature data conflicts, calculate the corrected probability of the support probability, including:

[0111] Calculate the average probability of the support probability of each feature data for the same decision-making scheme;

[0112] Determine the distance between each feature data and the average probability of the same decision-making scheme according to the average probability;

[0113] Determine the weight of the feature data according to the distance of the average probability;

[0114] Calculate the corrected probability of the support probability according to the weight of the feature data.

[0115] In the above embodiment, first calculate the average value of the support probability of each feature data for the k-th decision-making scheme:

[0116]

[0117] Among them, n is the total number of support probabilities of characteristic data for the decision-making scheme.

[0118] Then, according to the average probability, calculate the distance between the characteristic data and the average probability of the decision-making scheme:

[0119]

[0120] where m ij represents the basic probability function value of the support probability of the i-th characteristic data for the j-th decision-making scheme.

[0121] As can be seen from the above, the higher the similarity between the two evidences, the smaller the value of d i is.

[0122] Then, according to the distance of the average probability, determine the weight of the characteristic data:

[0123]

[0124] This weight is recalculated through the distance of the average probability, which can correct the support probability of conflicting characteristic data to obtain an accurate corrected probability.

[0125] The present invention also provides a specific embodiment for judging the driving state decision for illustration:

[0126] The judgment of the driving state is jointly determined by multiple parameters. The present invention selects parameters in three dimensions of the cab temperature, smoke concentration, and alcohol concentration to judge the driving state.

[0127] First, detect the parameters of the cab temperature, smoke concentration, and alcohol concentration at multiple points through means such as sensors. These parameters can be multi-time or multi-space parameters. After obtaining multiple groups of multi-dimensional data, first detect abnormal data through a box plot. Arrange the data in ascending order and represent it as x1, x2, x3...x n , and then draw a box plot according to the distribution of the data. The positions of Q1, Q2, and Q3 can be obtained according to the data distribution:

[0128]

[0129]

[0130]

[0131] Positions of the upper and lower limits of the outlier:

[0132]

[0133]

[0134] I QR = Q3 - Q2;

[0135] Wherein, I QR represents the distance between Q1 and Q3, F U represents the upper limit, F I represents the lower limit. n represents the number of data (taking 6 in the formula), n1, n2, and n3 respectively represent the positions of the first quartile, the second quartile, and the third quartile of a set of data, and β represents the interval coefficient used to delimit the abnormal interval, which is usually taken as 3 in this embodiment.

[0136] n1, n2, and n3 are used to determine the quartiles. The abnormal data is greater than Q1 + 1.5I QR or less than Q1 - 1.5I QR That is, the data outside the upper and lower limits. The box plot can truly and intuitively identify the outliers.

[0137] After determining the positions of the abnormal data through the box plot, the abnormal data is corrected by the RBF neural network of the cuckoo search algorithm. The specific process and method of abnormal data correction have been described in detail above and will not be elaborated here.

[0138] After correcting the abnormal data, the corrected data replaces the original abnormal data. The corrected data and the normal data together form the credible data. In this embodiment, the credible data is locally fused, that is, the parameters of the three dimensions of the cab temperature, the smoke concentration, and the alcohol concentration are fused respectively. In this embodiment, the adaptive weighted average method is used to dynamically allocate weights to the data in each time period for fusion, and three characteristic data of the cab temperature, the smoke concentration, and the alcohol concentration are obtained.

[0139] Then, through global fusion, the driving state is judged according to these three characteristic data.

[0140] First, the recognition framework is established by dividing the current state into three levels:

[0141] A = {Normal driving, cab temperature < 28, smoke concentration < 3% OBS / M, alcohol concentration is 0 - 20 mg / ml};

[0142] B = {Abnormal driving, cab temperature > 28, smoke concentration > 3% OBS / M, alcohol concentration is 0 - 20 mg / ml};

[0143] C = {Dangerous driving, cab temperature < 28, smoke concentration < 3% OBS / M, alcohol concentration > 20 mg / ml};

[0144] For normal driving of A, m1(μA ) represents the support probability of the cab temperature for grade A, m2(μ A ) represents the support probability of the smoke concentration for grade A, m3(μ A ) represents the support probability of the alcohol concentration for grade A. Similarly, for grade B, there are m1(μ B )、m2(μ B )、m3(μ B );for grade C, there are m1(μ C )、m2(μ C )、m3(μ C ).

[0145] Taking temperature as an example, in this embodiment, the cab temperature obtained through local fusion is 20°C. By passing through 20°C, m1(μ A )、m1(μ B )、m1(μ C ) are calculated. The three of them respectively represent how much the support degree of 20°C is for grade A, how much the support degree is for grade B, and how much the support degree is for grade C. Their sum is equal to 1. The process of calculating the support degree uses the trapezoidal function in fuzzy theory. It assigns a probability to the temperature in each interval. Substituting x = 20 into the following formula, m1(μ A )、m1(μ B )、m1(μ C ) can be obtained.

[0146]

[0147]

[0148]

[0149] Similarly, for the smoke concentration: m2(μ A )、m2(μ B )、m2(μ C );for the alcohol concentration: m3(μ A )、m3(μ B )、m3(μ C ) are all obtained in this way. They are all probability values, and the probability values in Table 1 are obtained through calculation.

[0150] Table 1

[0151]

[0152] Then, the cosine of the angle method is used to determine whether the characteristic data has conflicts. When there are conflicts between the characteristic data, the support probability of the characteristic data is corrected, and then its corrected probability is obtained. The specific process and method have been described in detail above, and will not be elaborated in this embodiment.

[0153] After the present invention obtains the corrected probability, it recalculates the membership function values of the BPA of the conflicting characteristic data and constructs a confidence matrix M.

[0154]

[0155] Transpose the i-th row of the matrix M and multiply it by the j-th row. Obtain a new matrix A:

[0156]

[0157] Therefore, the product of the elements on the diagonal represents the fusion result of the characteristic data i and the characteristic data j, and the sum of all other elements in the matrix is the uncertainty coefficient of the fusion result:

[0158]

[0159] Finally, fuse multiple fusion decision factor information:

[0160]

[0161] The final fusion result of the present invention provides evidence support for system decision-making, reduces the uncertainty of decision-making between heterogeneous data, and at the same time solves the problem of paradox caused by conflicts of characteristic data in the data fusion process, improving the stability of the decision-making system.

[0162] To better implement the decision-making method for multi-data fusion in the embodiments of the present invention, based on the decision-making method for multi-data fusion, correspondingly, please refer to Figure 5 , Figure 5 is a schematic structural diagram of an embodiment of a decision-making device for multi-data fusion provided by the present invention. The embodiments of the present invention provide a decision-making device 500 for multi-data fusion, including:

[0163] An acquisition module 501, configured to acquire multi-dimensional target data, and the multi-dimensional target data includes normal data and / or abnormal data;

[0164] A correction module 502, configured to detect abnormal data and correct the abnormal data based on an RBF neural network model of the cuckoo search algorithm to obtain corrected data;

[0165] A local fusion module 503, configured to locally fuse the corrected data and the normal data in the same dimension to obtain characteristic data in the same dimension;

[0166] A global fusion module 504, which is used to globally fuse feature data based on the improved D-S evidence theory to determine a target decision-making scheme.

[0167] It should be noted here that: The device 500 provided in the above embodiments can implement the technical solutions described in the above method embodiments. For the specific implementation principles of the above modules or units, reference can be made to the corresponding content in the above method embodiments, which will not be elaborated here.

[0168] Please refer to Figure 6 , Figure 6 which is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. Based on the above decision-making method for multi-data fusion, the present invention also correspondingly provides a decision-making device for multi-data fusion. The decision-making device for multi-data fusion can be a computing device such as a mobile terminal, a desktop computer, a notebook, a palm computer, and a server. The decision-making device for multi-data fusion includes a processor 610, a memory 620, and a display 630. Figure 6 Only some components of the electronic device are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0169] The memory 620 can be an internal storage unit of the decision-making device for multi-data fusion in some embodiments, such as the hard disk or memory of the decision-making device for multi-data fusion. The memory 620 can also be an external storage device of the decision-making device for multi-data fusion in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the decision-making device for multi-data fusion. Further, the memory 620 can also include both the internal storage unit and the external storage device of the decision-making device for multi-data fusion. The memory 620 is used to store the application software installed in the decision-making device for multi-data fusion and various types of data, such as the program code installed in the decision-making device for multi-data fusion. The memory 620 can also be used to temporarily store the data that has been output or will be output. In one embodiment, a decision-making program 640 for multi-data fusion is stored on the memory 620, and the decision-making program 640 for multi-data fusion can be executed by the processor 610, so as to implement the decision-making method for multi-data fusion in various embodiments of the present application.

[0170] The processor 610 can be a central processing unit (CPU), a microprocessor, or other data processing chips in some embodiments, and is used to run the program code stored in the memory 620 or process data, such as executing the decision-making method for multi-data fusion, etc.

[0171] The display 630 can be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. in some embodiments. The display 630 is used to display the information of the multi-data fusion decision-making device and to display a visual user interface. The components 610-630 of the multi-data fusion decision-making device communicate with each other through the system bus.

[0172] In one embodiment, when the processor 610 executes the multi-data fusion decision-making program 640 in the memory 620, the steps in the above multi-data fusion decision-making method are implemented.

[0173] This embodiment also provides a computer-readable storage medium, on which a multi-data fusion decision-making program is stored. When the multi-data fusion decision-making program is executed by a processor, the following steps are implemented:

[0174] Collect multi-dimensional target data, where the multi-dimensional target data includes normal data and / or abnormal data;

[0175] Detect abnormal data, and based on the RBF neural network model of the cuckoo search algorithm, correct the abnormal data to obtain corrected data;

[0176] Locally fuse the corrected data and the normal data of the same dimension to obtain the feature data of the same dimension;

[0177] Based on the improved D-S evidence theory, globally fuse the feature data to determine the target decision-making scheme.

[0178] In summary, a multi-data fusion decision-making method, device, equipment and storage medium provided in this embodiment detect whether the collected data is abnormal, correct the abnormal data through the RBF neural network model of the cuckoo search algorithm, improve the accuracy of decision-making, locally fuse the corrected data and the normal data to obtain the feature data of the same dimension, complete the description of the current state of the attributes of things, and then through the improved D-S evidence theory, perform global fusion to determine the target decision-making scheme, further improving the accuracy of the decision-making method.

[0179] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A decision-making method for multi-data fusion, characterized in that Including: Collecting multi-dimensional target data, where the multi-dimensional target data includes normal data and / or abnormal data; Detecting the abnormal data, and based on the RBF neural network model of the cuckoo search algorithm, correcting the abnormal data to obtain corrected data; Locally fusing the corrected data and the normal data of the same dimension to obtain feature data of the same dimension; Based on the improved D-S evidence theory, globally fusing the feature data to determine the target decision-making scheme, including: Determining all decision-making schemes according to the multi-dimensional target data; Based on the fuzzy theory, determining the support probability of all the feature data for each decision-making scheme; Calculating the cosine similarity between all pairs of the feature data according to the support probability; Judging whether there is a conflict between the feature data according to the cosine similarity; Calculating the average probability of the support probability of each feature data for the same decision-making scheme; Determining the distance between each feature data and the average probability of the same decision-making scheme according to the average probability; Determining the weight of the feature data according to the distance of the average probability; Calculating the corrected probability of the support probability according to the weight of the feature data; Wherein, the distance between each feature data and the average probability of the same decision-making scheme is: ； The weight of the said characteristic data is: ; Among them, represents the basic probability function value of the -th characteristic data for the j -th decision-making scheme.

2. The decision-making method for multi-data fusion according to claim 1, wherein The RBF neural network model based on the cuckoo search algorithm corrects the abnormal data to obtain corrected data, including: Calculating the target parameters of the RBF neural network model according to the cuckoo search algorithm; Calculating the Gaussian radial basis function output of the RBF neural network model according to the abnormal data and the target parameters; Calculating the corrected data according to the target parameters and the Gaussian radial basis function output.

3. The decision-making method for multi-data fusion according to claim 2, characterized in that, The calculating the target parameters of the RBF neural network model according to the cuckoo search algorithm includes: Obtaining training set sample data, initializing the RBF neural network model, and inputting the training set sample data; Initializing the parameters of the cuckoo search algorithm, and calculating the optimal value of the target parameters of the current cuckoo search algorithm parameters; Repeatedly updating the parameters of the cuckoo search algorithm, calculating and updating the optimal value of the target parameters until a preset termination condition is reached to obtain the target parameters.

4. The decision-making method for multi-data fusion according to claim 2, wherein The locally fusing the corrected data and the normal data of the same dimension to obtain feature data of the same dimension includes: Predicting the real data of the same dimension according to a preset prediction method to obtain the predicted real data of the same dimension; Determining the variance of the same dimension according to the predicted real data; Determining the weights of the corrected data and the normal data of the same dimension according to the variance of the same dimension; Determining the feature data of the same dimension according to the weights.

5. A decision-making device for multi-data fusion, characterized in that, Including: A collection module for collecting multi-dimensional target data, where the multi-dimensional target data includes normal data and / or abnormal data; A correction module for detecting the abnormal data and correcting the abnormal data based on the RBF neural network model of the cuckoo search algorithm to obtain corrected data; A local fusion module for locally fusing the corrected data and the normal data of the same dimension to obtain feature data of the same dimension; A global fusion module for globally fusing the feature data based on the improved D-S evidence theory to determine a target decision scheme, including: Determining all decision schemes according to the multi-dimensional target data; Based on the fuzzy theory, determining the support probability of all the feature data for each decision scheme; Calculating the cosine similarity between every two of all the feature data according to the support probability; Judging whether there is a conflict between the feature data according to the cosine similarity; Calculating the average probability of the support probability of each feature data for the same decision scheme; Determining the distance between each feature data and the average probability of the same decision scheme according to the average probability; Determining the weight of the feature data according to the distance of the average probability; Calculating the corrected probability of the support probability according to the weight of the feature data; Wherein, the distance between each feature data and the average probability of the same decision scheme is: ； The weight of the said characteristic data is: ; Among them, represents the basic probability function value of the -th decision-making scheme with respect to the support probability of the j -th characteristic data.

6. An electronic device, characterized in that, Including a memory and a processor, wherein, The memory is used for storing programs; The processor is coupled to the memory and is used for executing the program stored in the memory to implement the steps in the decision method for multi-data fusion according to any one of claims 1 to 4 above.

7. A computer-readable storage medium, characterized in that, For storing computer-readable programs or instructions, when the programs or instructions are executed by a processor, the steps in the decision method for multi-data fusion according to any one of claims 1 to 4 above can be implemented.

Images