Equipment dynamic association importance assessment method considering multi-factor fusion

Abstract

The invention discloses an equipment dynamic association importance evaluation method considering multi-factor fusion. The method comprises the steps of problem description, dynamic association model of association indexes, dynamic weight calculation and association importance calculation. The problem description comprises parameter definition and mathematical expression; the dynamic association model of the association indexes comprises a function complementarity association model, an information interactivity association model, a task process association model and a guarantee support association model; the dynamic weight calculation comprises semantic weight calculation, objective weight calculation and combined weight calculation; the correlation importance calculation comprises the step of calculating the correlation importance of the equipment pair; the invention relates to the technical field of equipment importance evaluation. According to the method, the dynamic association importance degree of the equipment pair is evaluated, the rule that the dynamic association importance degree of the equipment changes along with time is shown, the method is suitable for multi-dimensional related scenes with features and labels, components and environments and the like in a complex system, and the problem that multi-dimensional factors are not deeply fused is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of equipment importance evaluation, and in particular to an equipment dynamic correlation importance evaluation method considering multi-factor fusion. BACKGROUND

[0002] The dynamic change of the environment has an important influence on the correlation importance of equipment. The traditional correlation importance evaluation adopts fixed weight and static evaluation, which cannot reflect the dynamic evolution of the cooperative relationship between equipment, resulting in the mismatch between the cooperative deployment and resource allocation of the equipment system and the actual environmental demand. Therefore, it is of great significance to study the equipment dynamic correlation importance evaluation method for the development of equipment maintenance support command decision.

[0003] There are two problems in the traditional equipment correlation importance evaluation: first, the evolution mechanism of relevance and importance in the dynamic scene is not clear. Existing research only designs dynamic feature selection and evaluation algorithm, but does not deeply study the time evolution law of relevance strength and importance weight, resulting in the lack of theoretical support for the adaptability of dynamic algorithm, which can only passively match data changes. Second, the depth of multi-dimensional correlation fusion modeling is not enough. Most researches only focus on single-dimensional correlation, but there are multi-dimensional correlations such as features and labels, components and environment in complex systems. The existing method does not fuse multi-dimensional correlation for importance evaluation, which is easy to lead to one-sidedness in importance evaluation. Therefore, the present application provides an equipment dynamic correlation importance evaluation method considering multi-factor fusion to solve the above problems. SUMMARY

[0004] In view of the deficiencies in the prior art, the present application provides an equipment dynamic correlation importance evaluation method considering multi-factor fusion, which solves the above problems.

[0005] To achieve the above purpose, the present application realizes the following technical scheme: an equipment dynamic correlation importance evaluation method considering multi-factor fusion, including problem description, dynamic correlation model of correlation index, dynamic weight calculation and correlation importance calculation.

[0006] The problem description includes parameter definition and mathematical expression, and the relationship between dynamic influence factors and correlation indexes.

[0007] The dynamic correlation model of the correlation index includes a functional complementarity correlation model, an information interaction correlation model, a task flow correlation model and a support correlation model.

[0008] The dynamic weight calculation includes semantic weight calculation, objective weight calculation and combined weight calculation.

[0009] The correlation importance calculation includes calculating the correlation importance of equipment pairs.

[0010] Preferably, the parameter definition and mathematical expression include:

[0011] The equipment set L = {L1, L2, … L p ,…L q …,L n}, the task cycle is 24 hours, the time sequence node T = {t1, t2, t3, t4, t5}, the first time sequence point interval is 6h, the correlation index C = {C1, C2, C3, C4}, C1 is functional complementarity, C2 is information interaction, C3 is task flow, and C4 is support support;

[0012] A dynamic correlation model X of each index is constructed through the equipment set L and the time sequence T, and the correlation importance of any equipment pair at t i is obtained by fusing the dynamic weight, so as to solve the problem that the static evaluation cannot adapt to the time sequence evolution of the cooperative relationship;

[0013] Preferably, the relationship between the dynamic influence factor and the correlation index includes:

[0014] The equipment correlation importance mainly includes four indexes of functional complementarity, information interaction, task flow, and support support, as shown in Figure 1 , and the relationship between the dynamic influence factor and the correlation index is as follows:

[0015] Functional complementarity: affected by task demand conversion, the necessity of equipment function matching dynamically changes with the task target;

[0016] Information interaction: affected by the environment, the information transmission link is dynamically affected by the electromagnetic environment interference intensity;

[0017] Task flow: affected by the task stage connection, the dependence relationship of equipment in the flow dynamically changes with the stage advancement;

[0018] Support support: affected by the equipment loss rate and the support resource inventory;

[0019] Preferably, the functional complementarity correlation model includes:

[0020] Taking the dynamic matching of function demand-capacity supply as the core, the task stage weight, the function aging coefficient, and the cooperative gain effect are introduced, and the functional complementarity depth of the equipment pair in different scenes is quantified;

[0021] Let the equipment pair be L p ,L q , and the model is constructed in three layers:

[0022] (1) Static function matching degree calculation

[0023] Define the function demand matrix For task t a The demand intensity of function k, m is the number of tasks, n is the function dimension, and the equipment capability matrix A p = [a pk ] 1×n , A q = [a qk ] 1×n , a pk is the supply intensity of L p function k;

[0024] Static complementary base:

[0025]

[0026] In the formula: cos(A p ,A q ) is the cosine similarity of the capability vector, which measures the non-overlapping of function coverage;

[0027] (2) Dynamic adjustment factor

[0028] Task phase weight Based on the task timing, the analytic hierarchy process (AHP) is combined with the entropy weight method to determine; the function aging coefficient α(t) is described by Weibull distribution:

[0029]

[0030] In the formula: β is the decay rate, and γ is the shape parameter, which is fitted by equipment life data;

[0031] Synergistic gain coefficient δ, when L p ,L q function synergy produces additional capacity:

[0032]

[0033] In the formula: ΔA is the newly added capacity of synergy;

[0034] (3) Final function complementarity

[0035]

[0036] In the formula: t is the time dimension, reflecting the dynamic nature; the model output range is [0, 1], and the higher the value, the stronger the complementarity;

[0037] Preferably, the information interaction correlation model comprises:

[0038] Fusion information transmission real-time, reliability, interaction and security, Markov chain is used to describe the interaction state transition, combined with information entropy to quantify uncertainty, reflecting the dynamic interaction process;

[0039] (1) Definition and transition of interactive state

[0040] Define the information interaction state set S = {S0, S1, S2}, where S0, S1, and S2 represent normal, delayed, and interrupted states, respectively. Construct the Markov transition matrix P(t) = [p uv (t)] 3×3 p uv (t) represents the state S at time t. u Transfer to S v The probability is obtained by training with historical interaction data;

[0041] Steady-state interaction reliability:

[0042] R info =π0(t)·(1-ε) (5)

[0043] In the formula: π0(t) is the probability of S0 in steady state at time t, and ε is the packet loss rate, which is measured by the transmission protocol;

[0044] (2) Information interaction volume and uncertainty quantification

[0045] Let L be at time t. p →L q The amount of interactive data is D pq (t), L q →L p D qp (t), total interaction D total (t)=D pq (t)+D qp (t);

[0046] Information entropy is used to measure the uncertainty of interactions. The lower the entropy value, the stronger the certainty of the information, and the higher the quality of the interaction.

[0047]

[0048] (3) Dynamic interaction coefficient and final correlation

[0049] Real-time coefficient τ(t):

[0050]

[0051] In the formula: ΔT(t) is the interaction delay at time t, which is obtained from real-time monitoring data;

[0052] Security factor σ: Based on the encryption level, such as 1.0 for AES-256 and 0.3 for no encryption, and determined by the intrusion detection success rate;

[0053] Final information interaction level:

[0054]

[0055] In the formula: D max To equip the maximum interaction bandwidth, and to avoid the minimum value of entropy of 0, the output range is [0,1].

[0056] Preferably, the task flow association model includes:

[0057] Based on the directed weighted graph of the task flow, the intensity of temporal dependency, resource competition coefficient and critical path weight are introduced to quantify the process association depth of equipment pairs in the task link.

[0058] (1) Task flow topology construction

[0059] Construct a directed graph G = (V, F, W) for the task, where: nodes V = {L1, L2, ..., L...} n} represents the equipment set; edge f pq ∈F: represents L p With L q Task dependencies, prior dependencies / parallel collaboration; edge weights The initial dependency strength is determined by the mission manual in combination with expert scores, such as 0.9 for the dependency strength between radar detection and missile launch, and 0.6 for the parallel coordination of communication and navigation.

[0060] (2) Timing and resource constraint correction

[0061] Temporal dependency coefficient λ ij (t), calculated based on the time difference of task time sequence nodes.

[0062]

[0063] In the formula: t i ,t j For L p ,L q When participating in the mission, T task Total task duration;

[0064] Resource competition coefficient μ pq If L p ,L q Sharing the same resources

[0065]

[0066] In the formula: R shared To share resources, R total The total amount of resources; if there is no sharing, then μ. pq =1;

[0067] Critical path weight η pq : Identify the critical path of a task using CPM (Critical Path Method). If Lp ,L q All are on the critical path, η pq =1.2; Only one on the critical path, η pq =0.8; if none of them are present, then η pq =0.5;

[0068] (3) Calculation of process correlation

[0069] Define the process associated with the contribution value S of the equipment pair pq Introducing the influence of neighboring nodes, i.e., with L p ,L q All depend on equipment L k The correlation propagation, at this time the process correlation contribution value is

[0070]

[0071] Final task flow relevance:

[0072]

[0073] The output range is [0, 1.2]. The critical path equipment pair may exceed 1.0, reflecting the difference in process importance.

[0074] Preferably, the supportive correlation model includes:

[0075] Based on the matching of supply and demand of security resources, integrating the timeliness of security response and resource elasticity, grey relational analysis is used to handle the uncertainty of security data and reflect the dynamic security process.

[0076] (1) Ensure the basic data of supply and demand

[0077] Define a support dimension set {F1, F2, F3, F4,}, representing spare parts, maintenance, energy, and consumables respectively, and construct: demand sequence X i =[x i1 ,x i2 ,x i3 ,x i4 ] represents L p Demand intensity for each dimension of protection; supply sequence Y j =[y j1 ,y j2 ,y j3 ,y j4 ] represents L q L p The supply strength of each dimension of protection provided;

[0078] (2) Grey relational degree

[0079] The steps for calculating the static supply-demand matching degree using grey relational analysis (GRA) are as follows:

[0080] Data standardization:

[0081]

[0082] Absolute difference:

[0083] Δ pk =|x′ pk -y′ qk |;

[0084] Grey relational coefficient:

[0085]

[0086] In the formula: ζ = 0.5 is the resolution coefficient;

[0087] Static support base:

[0088]

[0089] (3) Dynamic protection correction factor

[0090] Response time coefficient θ(t):

[0091]

[0092] In the formula: T resp (t) is L q For L p The guarantee response time is measured by actual guarantee logs.

[0093] Resource elasticity coefficient ρ(t):

[0094]

[0095] In the formula: ΔY q (t) is L q Supply fluctuation, ΔX p (t) is L p Demand fluctuation; ρ(t)≥1 indicates sufficient elasticity, ρ(t)<1 indicates insufficient elasticity;

[0096] Fault impact coefficient κ: If L p L is required after the fault q Only with guarantees can recovery be achieved. For L p Task loss due to failure, L sys This represents the total system loss.

[0097] (4) Ultimate guarantee of support

[0098] C4(p,q,t)=C static-sup ·θ(t)·min(ρ(t),1.5)·κ (16)

[0099] In the formula: min(ρ(t),1.5) is the upper limit constraint of elasticity, and the output range is [0,1.5]. The value is higher for pairs with sufficient elasticity and fast response.

[0100] Preferably, the semantic weight calculation includes:

[0101] Using natural language processing (NLP) technology, semantic features are extracted from text data related to equipment evaluation to quantify the static importance of four related indicators. The key is to use the semantic importance of the keywords related to the indicators in the text as weights.

[0102] (1) Text data collection

[0103] Focus on text directly related to the importance of equipment, ensuring data coverage of indicator importance descriptions. Specific sources include:

[0104] Task requirements document: Describes the intensity of the task's requirements for indicators such as functional complementarity and information interactivity;

[0105] Equipment Technical Manual: Records the capabilities and characteristics of equipment in various indicator dimensions;

[0106] Expert evaluation report: A qualitative description by experts of the importance of indicators in equipment coordination;

[0107] Historical mission log: A record of the actual effects of various indicators in past missions;

[0108] (2) Text preprocessing

[0109] Eliminating noise and standardizing text formatting lays the foundation for semantic extraction. The steps are as follows:

[0110] Word segmentation: The text is segmented using jieba, breaking down semantic units such as "guarantee response efficiency" → "guarantee" and "response efficiency";

[0111] Remove stop words: Access the "stop word list" and a custom stop word list for the equipment field, retaining core keywords;

[0112] Semantic normalization: Through WordNet (English), synonyms and near-synonyms are merged, such as "information transmission quality" and "data transmission reliability" are unified into "information interaction quality";

[0113] Text cleaning: Remove special symbols, numbers, and redundant spaces, retaining valid semantic text.

[0114] (3) Semantic feature extraction

[0115] A BERT+TF-IDF fusion model is used to simultaneously capture the frequency importance and contextual semantic importance of keywords.

[0116] Keyword selection: Based on four related indicators, an indicator-keyword mapping table was constructed (see Table 1) to extract target keywords from the preprocessed text;

[0117] Table 1. Index-Keyword Mapping Table

[0118] Correlation index Core keyword example Functional complementarity Function matching, function coverage, task adaptation Information interaction Data transmission, information dependence, electromagnetic anti-jamming Task flow Timing dependence, process contribution, stage connection Support support Response efficiency, support necessity, resource inventory

[0119] Vector transformation: Using the BERT-base-uncased model, the filtered keywords are converted into 768-dimensional semantic vectors to capture the contextual relationships between words;

[0120] TF-IDF calculation: quantifies the frequency importance of keywords in text, the formula is:

[0121] TF-IDF(w k ,l)=TF(w k ,l)×IDF(w k )

[0122]

[0123] In the formula, TF(w) k ,l) is the keyword w k Frequency of occurrence in the l-th type of text; n word For the keyword w k Frequency of occurrence in the l-th type of text; n text n is the total number of words in the text. total,text Total number of text categories; n word,text The number of text categories containing the keyword w;

[0124] (4) Semantic weight calculation

[0125] By fusing the "TF-IDF frequency score" and the "BERT semantic similarity score," the semantic weights of the four indicators are obtained. The steps are as follows:

[0126] Calculating semantic similarity: with equipment association importance as the core concept (semantic vector V) core The keyword vector V for each indicator is calculated using cosine similarity. w With V core The similarity is calculated using the following formula:

[0127]

[0128] The semantic similarity score S is obtained by averaging the similarity scores of all keywords for each indicator.sim (j)(j=1,2,3,4);

[0129] Fusion score: The deep semantics and surface frequency are balanced by a 60% weighting for semantic similarity score and a 40% weighting for TF-IDF score. The formula is as follows:

[0130] S fusion (j)=0.6×S sim (j)+0.4×TF-IDF(j) (19)

[0131] In the formula, TF-IDF(j) is the average TF-IDF score of all keywords for the j-th indicator;

[0132] Normalization: Convert the fusion score into a weighted form, using the following formula:

[0133]

[0134] Finally, the static semantic weight vector is obtained.

[0135] ω nlp =[ω nlp1 ,ω nlp2 ,ω nlp3 ,ω nlp4 ]

[0136] Preferably, the objective weight calculation includes:

[0137] The CRITIC dynamic weighting method is used to calculate objective weights. The information content of the indicators is calculated through two dimensions: contrast (indicator dispersion) and conflict (indicator correlation). A time decay factor is added to reflect the dynamic nature of the indicators, thus more accurately reflecting their actual contribution over time. The specific steps are as follows:

[0138] Step 1: Time-decrease weighting of time-series data

[0139] Based on the data preprocessing results, highlight the impact of recent data on the weights:

[0140] x′ ijt =x ijt ×λ (T-t) (20)

[0141] In the formula, T represents the current time node, and t represents the historical time node (t <T),x′ ijt The index value after time decay; x ijt For quantitative standardization and qualitative quantification of the indicator value, i = equipment pair number, t = time node, and λ is the introduced time decay factor;

[0142] Step 2: Calculate the indicator contrast

[0143] Using standard deviation σj The dispersion (contrast) of the j-th indicator is measured. The higher the dispersion, the stronger the indicator's ability to distinguish differences in equipment correlation.

[0144]

[0145] In the formula, n is the total number of equipment pairs. Let be the time-decay weighted average of the j-th indicator.

[0146] Step 3: Calculate the conflict of indicators

[0147] The Pearson correlation coefficient is used to measure the redundancy (conflict) between indicators. The lower the correlation, the more unique information the indicator contains.

[0148] Calculate the correlation coefficient between indicators: the Pearson correlation coefficient between the j-th indicator and the k-th indicator (k≠j):

[0149]

[0150] Calculate the average correlation coefficient: the average correlation coefficient between the j-th indicator and the other three indicators:

[0151]

[0152] Conflict Quantification: The larger the value, the stronger the independence of the indicator;

[0153] Step 4: Calculate the information content and objective weight of the indicators.

[0154] Information content of the indicator: The total information content C of the j-th indicator j The product of contrast and conflict:

[0155]

[0156] Objective weight normalization: C j Normalization yields the CRITIC dynamic objective weight ω. obj (t):

[0157]

[0158] Preferably, the combined weight calculation includes:

[0159] By employing the range maximization combination, calculating the combination weights, and optimizing the combination weight coefficient α, the total range of the comprehensive correlation values ​​of equipment pairs is maximized, thereby improving the distinguishability of the weights in prioritizing equipment. The specific steps are as follows:

[0160] Step 1: Define the combined weight form

[0161] Let the combined weight Wj (t) is a linear combination of the NLP subjective weight and the CRITIC objective weight, and α ∈ [0, 1] is the combination coefficient:

[0162] W j (t) = α·ω nlp,j + (1 - α)·ω obj (t) (24)

[0163] In the formula, ω nlp,j is the NLP weight, and ω obj (t) is the CRITIC objective weight;

[0164] Step 2: Calculate the comprehensive correlation value of equipment pairs

[0165] Based on the time-decayed index value x′ ijt and the combined weight W j (t), calculate the comprehensive correlation value H of the i-th equipment pair i :

[0166]

[0167] Step 3: Construct the range maximization objective function

[0168] The objective function is to maximize the sum of the ranges of the comprehensive correlation values of all equipment pairs (i < k) to ensure the optimal weight discrimination:

[0169]

[0170] Step 4: Solve the optimal combined weight coefficient α *

[0171] Use the golden section method to solve α * ∈ [0, 1] (convergence accuracy ò = 10 -4 ), the steps are as follows:

[0172] Initialize the interval [a0 = 0, b0 = 1], and calculate the golden section points:

[0173] c0 = a0 + 0.618×(b0 - a0),

[0174] d0 = b0 - 0.618×(b0 - a0)

[0175] Calculate F(c0) and F(d0):

[0176] If F(c0) > F(d0), then the new interval is [a1 = a0, b1 = d0], d1 = c0, c1 = a1 + 0.618×(b1 - a1)

[0177] If F(c0) < F(d0), then the new interval [a1 = c0, b1 = b0], c1 = d0, d1 = b1 - 0.618 × (b1 - a1)

[0178] Iterate to Pick

[0179] Step 5: Determine the final combined weights

[0180] α * Substituting into the combined weight formula, we obtain the final dynamic combined weights, see... Figure 2

[0181]

[0182] Preferably, the correlation importance calculation includes:

[0183] The normalized correlation value X′ of the four indicators ij (t) and dynamic weights Weighted fusion, j = 1, 2, 3, 4, association importance R pq (t) formula:

[0184] Attached Figure Description

[0185] Figure 1 This invention proposes an equipment correlation importance assessment index.

[0186] Figure 2 This is the flowchart of the dynamic weight calculation proposed in this invention. Detailed Implementation

[0187] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0188] A method for evaluating the dynamic correlation importance of equipment that considers the fusion of multiple factors includes problem description, dynamic correlation model of correlation indicators, dynamic weight calculation and correlation importance calculation;

[0189] The problem description includes parameter definitions and mathematical expressions, as well as the relationship between dynamic influencing factors and related indicators;

[0190] The dynamic correlation models of related indicators include functional complementarity correlation models, information interaction correlation models, task process correlation models, and support correlation models.

[0191] Dynamic weight calculation includes semantic weight calculation, objective weight calculation, and combined weight calculation;

[0192] The correlation importance calculation includes calculating the correlation importance of equipment pairs;

[0193] The parameter definitions and mathematical expressions include:

[0194] Equipment set L = {L1, L2, ... L} p ,…L q …,L n The task cycle is 24 hours, the time sequence nodes T = {t1, t2, t3, t4, t5}, the interval between the first time sequence node is 6 hours, and the related indicators C = {C1, C2, C3, C4}, where C1 is functional complementarity, C2 is information interaction, C3 is task process, and C4 is support and guarantee.

[0195] By using the equipment set L and the time series T, a dynamic correlation model X is constructed for each indicator, and the dynamic weights are fused to obtain t. i The importance of the association between pairs can be arbitrarily assigned at any time, solving the problem that static evaluation cannot adapt to the temporal evolution of collaborative relationships;

[0196] The relationship between dynamic influencing factors and related indicators includes:

[0197] The importance of equipment correlation mainly includes four indicators: functional complementarity, information exchange, mission flow, and support. (See...) Figure 1 The relationship between dynamic influencing factors and related indicators is as follows:

[0198] Functional complementarity: Due to the changing requirements of missions, the necessity of matching equipment functions changes dynamically with mission objectives;

[0199] Information interactivity: Affected by the environment, the information transmission link is dynamically affected by the intensity of electromagnetic interference.

[0200] Task process: Due to the influence of the connection between task phases, the dependencies of equipment in the process change dynamically as the phases progress;

[0201] Support capability: Affected by equipment wear and tear rate and dynamic inventory of support resources;

[0202] Functional complementarity association models include:

[0203] With the dynamic matching of functional requirements and capability supply as the core, the system introduces task phase weight, functional aging coefficient and synergistic gain effect to quantify the functional complementarity depth of equipment pairs in different scenarios.

[0204] Let the equipment pair be L p ,L q The model is constructed in three layers:

[0205] (1) Calculation of static functional matching degree

[0206] Define the functional requirements matrix For task t a The intensity of demand for function k, m is the number of tasks, n is the function dimension, and the equipment capability matrix A. p =[a pk ] 1×n A q =[a qk ] 1×n a pk For L p The intensity of supply to function k;

[0207] Static complementary cardinality:

[0208]

[0209] In the formula: cos(A) p A q The cosine similarity of capability vectors measures the non-overlapping nature of capability coverage.

[0210] (2) Dynamic adjustment factor

[0211] Task phase weights The functional aging coefficient α(t) is determined based on the task time sequence using the Analytic Hierarchy Process (AHP) combined with the entropy weight method; the Weibull distribution is used to describe the functional decay.

[0212]

[0213] In the formula: β is the decay rate, and γ is the shape parameter, which is fitted from the equipment life data;

[0214] Synergistic gain coefficient δ, when L p ,L q When functional synergy generates additional capabilities:

[0215]

[0216] In the formula: ΔA represents the newly added collaborative capability;

[0217] (3) Ultimate functional complementarity

[0218]

[0219] In the formula: t represents the time dimension, reflecting the dynamic nature; the model output range is [0,1], and the higher the value, the stronger the complementarity.

[0220] Information interaction association models include:

[0221] It integrates the real-time performance, reliability, interaction volume, and security of information transmission, uses Markov chains to describe the interaction state transitions, and combines information entropy to quantify uncertainty, thus reflecting the dynamic interaction process.

[0222] (1) Definition and transition of interactive state

[0223] Define the information interaction state set S = {S0, S1, S2}, where S0, S1, and S2 represent normal, delayed, and interrupted states, respectively. Construct the Markov transition matrix P(t) = [p uv (t)] 3×3 p uv (t) represents the state S at time t. u Transfer to S v The probability is obtained by training with historical interaction data;

[0224] Steady-state interaction reliability:

[0225] R info =π0(t)·(1-ε) (5)

[0226] In the formula: π0(t) is the probability of S0 in steady state at time t, and ε is the packet loss rate, which is measured by the transmission protocol;

[0227] (2) Information interaction volume and uncertainty quantification

[0228] Let L be at time t. p →L q The amount of interactive data is D pq (t), L q →L p D qp (t), total interaction D total (t)=D pq (t)+D qp (t);

[0229] Information entropy is used to measure the uncertainty of interactions. The lower the entropy value, the stronger the certainty of the information, and the higher the quality of the interaction.

[0230]

[0231] (3) Dynamic interaction coefficient and final correlation

[0232] Real-time coefficient τ(t):

[0233]

[0234] In the formula: ΔT(t) is the interaction delay at time t, which is obtained from real-time monitoring data;

[0235] Security factor σ: Based on the encryption level, such as 1.0 for AES-256 and 0.3 for no encryption, and determined by the intrusion detection success rate;

[0236] Final information interaction level:

[0237]

[0238] In the formula: D max To equip the maximum interactive bandwidth, To avoid the minimum value of entropy 0, the output range is [0,1].

[0239] The task process association model includes:

[0240] Based on the directed weighted graph of the task flow, the intensity of temporal dependency, resource competition coefficient and critical path weight are introduced to quantify the process association depth of equipment pairs in the task link.

[0241] (1) Task flow topology construction

[0242] Construct a directed graph G = (V, F, W) for the task, where: nodes V = {L1, L2, ..., L...} n} represents the equipment set; edge f pq ∈F: represents L p With L q Task dependencies, prior dependencies / parallel collaboration; edge weights The initial dependency strength is determined by the mission manual in combination with expert scores, such as 0.9 for the dependency strength between radar detection and missile launch, and 0.6 for the parallel coordination of communication and navigation.

[0243] (2) Timing and resource constraint correction

[0244] Temporal dependency coefficient λ ij (t), calculated based on the time difference of task time sequence nodes.

[0245]

[0246] In the formula: t i ,t j For L p ,L q When participating in the mission, T task Total task duration;

[0247] Resource competition coefficient μ pq If L p ,L q Sharing the same resources

[0248]

[0249] In the formula: Rshared To share resources, R total The total amount of resources; if there is no sharing, then μ. pq =1;

[0250] Critical path weight η pq : Identify the critical path of a task using CPM (Critical Path Method). If L p ,L q All are on the critical path, η pq =1.2; Only one on the critical path, η pq =0.8; if none of them are present, then η pq =0.5;

[0251] (3) Calculation of process correlation

[0252] Define the process associated with the contribution value S of the equipment pair pq Introducing the influence of neighboring nodes, i.e., with L p ,L q All depend on equipment L k The correlation propagation, at this time the process correlation contribution value is

[0253]

[0254] Final task flow relevance:

[0255]

[0256] The output range is [0, 1.2]. The critical path equipment pair may exceed 1.0, reflecting the difference in process importance.

[0257] The supportive correlation model includes:

[0258] Based on the matching of supply and demand of security resources, integrating the timeliness of security response and resource elasticity, grey relational analysis is used to handle the uncertainty of security data and reflect the dynamic security process.

[0259] (1) Ensure the basic data of supply and demand

[0260] Define a support dimension set {F1, F2, F3, F4,}, representing spare parts, maintenance, energy, and consumables respectively, and construct: demand sequence X i =[x i1 ,x i2 ,x i3 ,x i4 ] represents L p Demand intensity for each dimension of protection; supply sequence Y j =[y j1 ,y j2 ,y j3 ,yj4 ] represents L q L p The supply strength of each dimension of protection provided;

[0261] (2) Grey relational degree

[0262] The steps for calculating the static supply-demand matching degree using grey relational analysis (GRA) are as follows:

[0263] Data standardization:

[0264]

[0265] Absolute difference:

[0266] Δ pk =|x′ pk -y′ qk |;

[0267] Grey relational coefficient:

[0268]

[0269] In the formula: ζ = 0.5 is the resolution coefficient;

[0270] Static support base:

[0271]

[0272] (3) Dynamic protection correction factor

[0273] Response time coefficient θ(t):

[0274]

[0275] In the formula: T resp (t) is L q For L p The guarantee response time is measured by actual guarantee logs.

[0276] Resource elasticity coefficient ρ(t):

[0277]

[0278] In the formula: ΔY q (t) is L q Supply fluctuation, ΔX p (t) is L p Demand fluctuation; ρ(t)≥1 indicates sufficient elasticity, ρ(t)<1 indicates insufficient elasticity;

[0279] Fault impact coefficient κ: If L p L is required after the fault q Only with guarantees can recovery be achieved. For L p Task loss due to failure, L sys This represents the total system loss.

[0280] (4) Ultimate guarantee of support

[0281] C4(p,q,t)=C static-sup ·θ(t)·min(ρ(t),1.5)·κ (16)

[0282] In the formula: min(ρ(t),1.5) is the upper limit constraint of elasticity, and the output range is [0,1.5]. The value is higher for pairs with sufficient elasticity and fast response.

[0283] Semantic weight calculation includes:

[0284] Using natural language processing (NLP) technology, semantic features are extracted from text data related to equipment evaluation to quantify the static importance of four related indicators. The key is to use the semantic importance of the keywords related to the indicators in the text as weights.

[0285] (1) Text data collection

[0286] Focus on text directly related to the importance of equipment, ensuring data coverage of indicator importance descriptions. Specific sources include:

[0287] Task requirements document: Describes the intensity of the task's requirements for indicators such as functional complementarity and information interactivity;

[0288] Equipment Technical Manual: Records the capabilities and characteristics of equipment in various indicator dimensions;

[0289] Expert evaluation report: A qualitative description by experts of the importance of indicators in equipment coordination;

[0290] Historical mission log: A record of the actual effects of various indicators in past missions;

[0291] (2) Text preprocessing

[0292] Eliminating noise and standardizing text formatting lays the foundation for semantic extraction. The steps are as follows:

[0293] Word segmentation: The text is segmented using jieba, breaking down semantic units such as "guarantee response efficiency" → "guarantee" and "response efficiency";

[0294] Remove stop words: Access the "stop word list" and a custom stop word list for the equipment field, retaining core keywords;

[0295] Semantic normalization: Through WordNet (English), synonyms and near-synonyms are merged, such as "information transmission quality" and "data transmission reliability" are unified into "information interaction quality";

[0296] Text cleaning: Remove special symbols, numbers, and redundant spaces, retaining valid semantic text.

[0297] (3) Semantic feature extraction

[0298] A BERT+TF-IDF fusion model is used to simultaneously capture the frequency importance and contextual semantic importance of keywords.

[0299] Keyword selection: Based on four related indicators, an indicator-keyword mapping table was constructed (see Table 1) to extract target keywords from the preprocessed text;

[0300] Table 1. Index-Keyword Mapping Table

[0301] Correlation index Core keyword example Functional complementarity Function matching, function coverage, task adaptation Information interaction Data transmission, information dependence, electromagnetic anti-jamming Task flow Timing dependence, process contribution, stage connection Support support Response efficiency, support necessity, resource inventory

[0302] Vector transformation: Using the BERT-base-uncased model, the filtered keywords are converted into 768-dimensional semantic vectors to capture the contextual relationships between words;

[0303] TF-IDF calculation: quantifies the frequency importance of keywords in text, the formula is:

[0304] TF-IDF(w k ,l)=TF(w k ,l)×IDF(w k )

[0305]

[0306] In the formula, TF(w) k ,l) is the keyword w k Frequency of occurrence in the l-th type of text; n word For the keyword w k Frequency of occurrence in the l-th type of text; n text n is the total number of words in the text. total,text Total number of text categories; n word,text The number of text categories containing the keyword w;

[0307] (4) Semantic weight calculation

[0308] By fusing the "TF-IDF frequency score" and the "BERT semantic similarity score," the semantic weights of the four indicators are obtained. The steps are as follows:

[0309] Calculating semantic similarity: with equipment association importance as the core concept (semantic vector V) coreThe keyword vector V for each indicator is calculated using cosine similarity. w With V core The similarity is calculated using the following formula:

[0310]

[0311] The semantic similarity score S is obtained by averaging the similarity scores of all keywords for each indicator. sim (j)(j=1,2,3,4);

[0312] Fusion score: The deep semantics and surface frequency are balanced by a 60% weighting for semantic similarity score and a 40% weighting for TF-IDF score. The formula is as follows:

[0313] S fusion (j)=0.6×S sim (j)+0.4×TF-IDF(j) (19)

[0314] In the formula, TF-IDF(j) is the average TF-IDF score of all keywords for the j-th indicator;

[0315] Normalization: Convert the fusion score into a weighted form, using the following formula:

[0316]

[0317] Finally, the static semantic weight vector is obtained.

[0318] ω nlp =[ω nlp1 ,ω nlp2 ,ω nlp3 ,ω nlp4 ]

[0319] Objective weight calculation includes:

[0320] The CRITIC dynamic weighting method is used to calculate objective weights. The information content of the indicators is calculated through two dimensions: contrast (indicator dispersion) and conflict (indicator correlation). A time decay factor is added to reflect the dynamic nature of the indicators, thus more accurately reflecting their actual contribution over time. The specific steps are as follows:

[0321] Step 1: Time-decrease weighting of time-series data

[0322] Based on the data preprocessing results, highlight the impact of recent data on the weights:

[0323] x′ ijt =x ijt ×λ (T-t) (20)

[0324] In the formula, T represents the current time node, and t represents the historical time node (t <T),x′ ijt The index value after time decay; x ijt For quantitative standardization and qualitative quantification of the indicator value, i = equipment pair number, t = time node, and λ is the introduced time decay factor;

[0325] Step 2: Calculate the indicator contrast

[0326] Using standard deviation σ j The dispersion (contrast) of the j-th indicator is measured. The higher the dispersion, the stronger the indicator's ability to distinguish differences in equipment correlation.

[0327]

[0328] In the formula, n is the total number of equipment pairs. Let be the time-decay weighted average of the j-th indicator.

[0329] Step 3: Calculate the conflict of indicators

[0330] The Pearson correlation coefficient is used to measure the redundancy (conflict) between indicators. The lower the correlation, the more unique information the indicator contains.

[0331] Calculate the correlation coefficient between indicators: the Pearson correlation coefficient between the j-th indicator and the k-th indicator (k≠j):

[0332]

[0333] Calculate the average correlation coefficient: the average correlation coefficient between the j-th indicator and the other three indicators:

[0334]

[0335] Conflict Quantification: The larger the value, the stronger the independence of the indicator;

[0336] Step 4: Calculate the information content and objective weight of the indicators.

[0337] Information content of the indicator: The total information content C of the j-th indicator j The product of contrast and conflict:

[0338]

[0339] Objective weight normalization: C j Normalization yields the CRITIC dynamic objective weight ω. obj (t):

[0340]

[0341] The combined weight calculation includes:

[0342] Using the range maximization combination, calculate the combined weight. By optimizing the combined weight coefficient α, maximize the sum of the ranges of the comprehensive correlation values of the equipment pairs, and improve the discrimination of the weight for the priority ranking of the equipment. The specific steps are as follows:

[0343] Step 1: Define the combined weight form

[0344] Let the combined weight W j (t) be the linear combination of the NLP subjective weight and the CRITIC objective weight, and α ∈ [0,1] be the combination coefficient:

[0345] W j (t) = α·ω nlp,j +(1 - α)·ω obj (t) (24)

[0346] In the formula, ω nlp,j is the NLP weight, and ω obj (t) is the CRITIC objective weight;

[0347] Step 2: Calculate the comprehensive correlation value of the equipment pairs

[0348] Based on the time-decayed index value x′ ijt and the combined weight W j (t), calculate the comprehensive correlation value H i of the i-th equipment pair:

[0349]

[0350] Step 3: Construct the range maximization objective function

[0351] The objective function is to maximize the sum of the ranges of the comprehensive correlation values of all equipment pairs (i < k) to ensure the optimal weight discrimination:

[0352]

[0353] Step 4: Solve the optimal combined weight coefficient α *

[0354] Use the golden section method to solve α * ∈ [0,1] (convergence accuracy ), and the steps are as follows:

[0355] Initialize the interval [a0 = 0, b0 = 1], and calculate the golden section point:

[0356] c0 = a0 + 0.618×(b0 - a0),

[0357] d0 = b0 - 0.618 × (b0 - a0)

[0358] Calculate F(c0) and F(d0):

[0359] If F(c0) > F(d0), then the new interval [a1 = a0, b1 = d0], d1 = c0, c1 = a1 + 0.618 × (b1 - a1)

[0360] If F(c0) < F(d0), then the new interval [a1 = c0, b1 = b0], c1 = d0, d1 = b1 - 0.618 × (b1 - a1)

[0361] Iterate to Pick

[0362] Step 5: Determine the final combined weights

[0363] α * Substituting into the combined weight formula, we obtain the final dynamic combined weights, see... Figure 2

[0364]

[0365] Association importance calculation includes:

[0366] The normalized correlation value X′ of the four indicators ij (t) and dynamic weights Weighted fusion, j = 1, 2, 3, 4, association importance R pq (t) formula:

[0367]

[0368] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0369] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0370] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the importance of equipment dynamic correlation considering multi-factor fusion, characterized in that: The problem description comprises parameter definition and mathematical expression and the relationship between dynamic influencing factors and correlation indicators. The problem description comprises parameter definition and mathematical expression and the relationship between dynamic influencing factors and correlation indicators. The dynamic correlation model of the correlation indicators comprises a functional complementarity correlation model, an information interaction correlation model, a task flow correlation model and a support correlation model. The dynamic weight calculation comprises semantic weight calculation, objective weight calculation and combined weight calculation. The correlation importance calculation comprises calculating the correlation importance of equipment pairs.

2. The method of claim 1, wherein the method further comprises: The problem description comprises: The parameter definition and mathematical expression comprise: L = {L1, L2, … L p ,…L q …,L n}, task cycle 24 hours, timing node T = {t1, t2, t3, t4, t5}, the first timing point interval 6h, the correlation index C = {C1, C2, C3, C4}, C1 is functional complementarity, C2 is information interaction, C3 is task flow, and C4 is support. The dynamic correlation model X of each index is constructed through the equipment set L and the time sequence T, and the dynamic weight is fused to obtain t i The correlation importance of any equipment pair at any time is solved, and the problem that the static evaluation cannot adapt to the time sequence evolution of the cooperative relationship is solved. The relationship between dynamic influencing factors and correlation indicators comprises: The equipment correlation importance mainly comprises four indicators, namely, functional complementarity, information interaction, task flow and support, as shown in FIG. 1, and the relationship between dynamic influencing factors and correlation indicators is as follows: Functional complementarity: influenced by task demand conversion, the necessity of equipment function matching dynamically changes with the task target; Information interaction: influenced by the environment, the information transmission link is dynamically influenced by the electromagnetic environment interference intensity; Task flow: influenced by task stage connection, the dependence relationship of equipment in the flow dynamically changes with the stage advancement; Support: influenced by equipment loss rate and support resource inventory.

3. The method of claim 1, wherein the method is a method of evaluating the importance of equipment in a dynamic correlation considering a plurality of factors. The dynamic correlation model of the correlation indicators comprises: The functional complementarity correlation model comprises: Taking dynamic matching of function demand-capacity supply as the core, introducing task stage weight, function aging coefficient and synergistic gain effect, the functional complementarity depth of equipment pairs in different scenarios is quantified; The equipment pair is L p , L q , The model is constructed in three layers: (1) Static function matching degree calculation Definition of functional requirement matrix For task t a The requirement intensity of function k, m is the number of tasks, n is the function dimension, equipment capability matrix A p = [a pk ] 1×n , A q = [a qk ] 1×n , a pk is the supply intensity of equipment L p to function k; a qk is the supply intensity of equipment L q to function k; Static complementary base: where: cos(A p ,A q ) is the cosine similarity of capability vectors, measuring the non-overlapping of the functional coverage. (2) Dynamic adjustment factor Task phase weight Determined based on task timing using analytic hierarchy process (AHP) combined with entropy weight method; function aging coefficient α(t), described using Weibull distribution In the formula, β is the decay rate, and γ is the shape parameter, which is fitted from equipment life data; Synergy gain factor δ, when L p ,L q When functions synergize to produce additional capabilities: In the formula, ΔA is the synergistic new capacity; (3) Final function complementarity degree In the formula, t is the time dimension, which reflects the dynamic nature; the model output range is [0, 1], and the higher the value, the stronger the complementarity; The information interaction correlation model comprises: Fusing the real-time, reliability, interaction amount and security of information transmission, the interactive state transition is described by Markov chain, the uncertainty is quantified by combining information entropy, and the dynamic interaction process is embodied; (1) Interactive state definition and transition Define the information interaction state set S = {S0, S1, S2}, S0, S1, S2 represent normal, delay, interruption respectively, and construct Markov transition matrix P(t) = [p uv (t)] 3×3 , p uv (t) is the probability of transition from state S u to S v at time t, which is trained by historical interaction data; Steady-state interaction reliability: R info = π0(t) · (1 - ε) (5) In the formula, π0(t) is the probability of S0 at t time in the steady state, and ε is the information packet loss rate, which is measured from the transmission protocol; (2) Information interaction amount and uncertainty quantification Let t be the time when L p → L q The amount of interaction data is D pq (t), L q → L p D qp (t), the total amount of interaction D total (t) = D pq (t) + D qp (t); The interaction uncertainty is measured by information entropy, the lower the entropy value, the stronger the information certainty, and the higher the interaction quality: (3) Dynamic interaction coefficient and final correlation degree Real-time coefficient τ(t): In the formula, ΔT(t) is the interaction time delay at t time, which is obtained from real-time monitoring data; Security coefficient σ: based on encryption level, such as AES-256, 1.0 for encryption, and 0.3 for no encryption, which is determined with the intrusion detection success rate; Final information interaction degree: where D max to equip the maximum interaction bandwidth, to avoid the minimum value of the entropy value is 0, the output range [0,1]; The task flow correlation model comprises: Taking the directed weighted graph of the task flow as the basis, introducing time sequence dependence intensity, resource competition coefficient and critical path weight, the flow correlation depth of equipment pairs in the task link is quantified; (1) Task flow topology construction The construction task directed graph G=(V, F, W) is constructed, wherein: nodes V={L1, L2, …, L n} are equipment sets; edges f pq ∈F represent the task dependency relationship of L p and L q , the pre-dependence / parallel cooperation; edge weight The initial dependency strength is determined by the task manual combined with expert scoring, such as 0.9 for the radar detection and missile launch dependency strength, and 0.6 for the communication-navigation parallel cooperation. (2) Time sequence and resource constraint correction Timing dependency coefficient λ ij (t), based on a time difference computation of the task timing nodes, wherein: t i t j L p L q T task total duration of the task; Resource contention coefficient μ pq : If L p , L q share the same resource, where: R shared is the amount of shared resources, R total is the total amount of resources; without sharing μ pq = 1; Critical path weight η pq : Identify critical path of tasks by CPM (Critical Path Method), if L p , L q are both on the critical path, η pq = 1.2; only one is on the critical path, η pq = 0.8; none is on the critical path, η pq = 0.5; (3) Flow correlation degree calculation Defining the process correlation contribution value S of a pair of equipment pq , the neighbor node influence, i.e. the L p , q The associated transmission of equipment L k that both have dependencies, at this time the process correlation contribution value is Final task flow correlation degree: Output range [0, 1.2], key path equipment pair may break 1.0, reflecting the importance difference of the process; Supporting correlation model includes: Based on the matching of supply and demand of support resources, the dynamic support process is reflected by combining the support response timeliness and resource elasticity and using gray correlation analysis to process the uncertainty of support data; (1) Support supply and demand basic data Define a support dimension set {F1, F2, F3, F4,}, representing spare parts, maintenance, energy, and consumables respectively, and construct: demand sequence X i =[x i1 ,x i2 ,x i3 ,x i4 ] represents L p Demand intensity for each dimension of protection; supply sequence Y j =[y j1 ,y j2 ,y j3 ,y j4 ] represents L q L p The supply strength of each dimension of protection provided; (2) Gray correlation degree Use gray correlation analysis (GRA) to calculate the static supply and demand matching degree, the steps are as follows: Data standardization: Absolute difference: Δ pk = |x′ pk -y′ qk |; wherein: Δ pk is the absolute difference; Gray correlation coefficient: where: ξ pk is the grey correlation coefficient; ζ = 0.5 is the resolution coefficient; max(Δ pk ) is the maximum absolute difference; min(Δ pk ) is the minimum absolute difference; Static support base: In the formula: C static-sup is a static support base; (3) Dynamic support correction factor Response timeliness coefficient θ(t): In the formula, T resp (t) is L q The guarantee response time of L p is measured by the guarantee log. Resource elasticity coefficient ρ(t): where: ΔY q (t) is L q Supply fluctuation; Y q (t) is L q Supply; ΔX p (t) is L p Demand fluctuation; X p (t) is L p Demand; p(t) > 1 indicates elastic sufficiency, p(t) < 1 indicates insufficient elasticity; Failure impact coefficient K: if L p Failure impact coefficient K: if L q Guarantee to restore, For L p Task loss caused by failure, L sys Total loss of the system; (4) Final support support degree C4(p, q, t) = C static-sup • θ(t) • min(p(t), 1.5) • K (16) Where: min(ρ(t), 1.5) is the upper limit constraint of elasticity, the output range is [0, 1.5], and the value is higher for the pair with sufficient elasticity and fast response.

4. The method of claim 1, wherein the method is a method of evaluating the importance of equipment in a dynamic correlation considering a plurality of factors. The dynamic weight calculation includes: Semantic weight calculation includes: Through natural language processing technology (NLP), semantic features are extracted from text data related to equipment evaluation, and the static importance of the four correlation indicators is quantified, focusing on using the semantic importance of index-related keywords in the text as weights; (1) Text data collection Focus on the text related to the importance of equipment correlation, ensure that the data cover the description of the importance of the indicators, and the specific sources include: Task demand document: describes the demand intensity of the equipment in the function complementarity, information interaction and other indicators; Equipment technical manual: records the ability characteristics of the equipment in each indicator dimension; Expert evaluation report: expert's qualitative description of the importance of indicators in equipment cooperation; Historical task log: record of the actual role of each indicator in the past task; (2) Text preprocessing Eliminate noise and unify the text format to lay the foundation for semantic extraction, the steps are as follows: Word segmentation: the text uses jieba word segmentation to split semantic units such as "guarantee response efficiency" → "guarantee" "response efficiency"; Stop words: call "stop word list", equipment field custom stop word list, keep core keywords; Semantic normalization: through WordNet (English), merge synonyms and near synonyms, such as "information transmission quality" "data transmission reliability" unified as "information interaction quality"; Text cleaning: remove special symbols, numbers and redundant spaces, and keep valid semantic text; (3) Semantic feature extraction Use BERT+TF-IDF fusion model to capture the frequency importance and context semantic importance of keywords: Keyword selection: based on the four correlation indicators, construct an index-keyword mapping table, and extract target keywords from preprocessed text; Vector conversion: convert the selected keywords into 768-dimensional semantic vectors through the BERT-base-uncased model to capture the context correlation between words; TF-IDF calculation: quantize the frequency importance of keywords in the text, the formula is: TF-IDF(w k ,l) = TF(w k ,l) x IDF(w k ) wherein TF(w, l) is the frequency of keyword w in the lth text category; n k TF(w, l) is the frequency of keyword w in the lth text category; n k TF(w, l) is the frequency of keyword w in the lth text category; n word TF(w, l) is the frequency of keyword w in the lth text category; n k TF(w, l) is the frequency of keyword w in the lth text category; n text TF(w, l) is the frequency of keyword w in the lth text category; n total,text TF(w, l) is the frequency of keyword w in the lth text category; n word,text TF(w, l) is the frequency of keyword w in the lth text category; n (4) Semantic weight calculation Fuse "TF-IDF frequency score" and "BERT semantic similarity score" to get the semantic weight of the four indicators, the steps are as follows: Calculate semantic similarity: take the equipment correlation importance as the core concept (semantic vector V core ), calculate the similarity of each index keyword vector V w and V core , the formula is: The semantic similarity score S of each indicator is obtained by averaging the similarity of all keywords of the indicator sim (j) (j = 1, 2, 3, 4); Fusion score: balance deep semantics and surface frequency with semantic similarity score accounting for 60% and TF-IDF score accounting for 40%, formula is: S fusion (j) = 0.6 x S sim (j) + 0.4 x TF-IDF(j) (19) In the formula: S fusion (j) is the fusion score of the jth indicator; TF-IDF(j) is the mean of the TF-IDF scores of all keywords of the jth indicator; Normalization: convert fusion score into weight form, formula is: Finally get static semantic weight vector ω nlp = [ω nlp1 , ω nlp2 , ω nlp3 , ω nlp4 ] Objective weight calculation includes: Use CRITIC dynamic weight method to calculate objective weight, calculate index information quantity through contrast (index dispersion) and conflict (index correlation), combine time decay factor to reflect dynamics, more accurately reflect the actual contribution of index in time sequence, specific steps are as follows: Step 1: Time decay weighting of time series data Based on the data preprocessing results, highlight the influence of recent data on weight: x' ijt = x ijt x λ (T-t) (20) In the formula, T is a current time node, t is a historical time node (t ijt is an index value after time decay; x ijt is an index value after quantitative standardization and qualitative quantization, i = equipment pair subsequence number; t = time node; λ is an introduced time decay factor; Step 2: Calculate index contrast with the standard deviation σ j measures the dispersion (contrast) of the jth indicator, the higher the dispersion, the stronger the indicator's ability to distinguish between differences in equipment association: In the formula, n is the total number of equipment pairs, is the time-decay weighted mean of the jth indicator. Step 3: Calculate index conflict Use Pearson correlation coefficient to measure the redundancy (conflict) between indexes, the lower the correlation, the more unique information the index has: Calculate the correlation coefficient between indexes: the Pearson correlation coefficient of the jth index and the kth index (k≠j): Calculate the average correlation coefficient: the average correlation coefficient of the jth index and all other three indexes: Conflictive quantification: The greater the value, the stronger the independence of the indicator. Step 4: Calculate index information quantity and objective weight Indicator information content: total information content C of the jth indicator j Product of contrast and conflict: Objective weight normalization: C j is normalized to obtain the CRITIC dynamic objective weight ω obj (t): Combination weight calculation includes: Use range maximization combination to calculate combination weight, maximize the sum of the range of equipment pair comprehensive correlation value by optimizing combination weight coefficient α, improve the discrimination of weight on equipment priority ranking, specific steps are as follows: Step 1: Define combination weight form Let the combined weight W j (t) is a linear combination of the NLP subjective weight and the CRITIC objective weight, and a ∈ [0, 1] is the combination coefficient: W j (t) = a - ω nlp,j + (1 - a) - ω obj (t) (24) where ω nlp,j is the NLP weight, ω obj (t) is the CRITIC objective weight; Step 2: Calculate equipment pair comprehensive correlation value based on the time-decayed index value x' ijt with the combination weight W j (t), the integrated correlation value H of the ith equipment pair is calculated i : Step 3: Build range maximization objective function Objective function is to maximize the sum of the range of all equipment pair comprehensive correlation value, ensure the optimal discrimination of weight: Step 4: Solving for the optimal combination weight coefficients a * The golden section method is used to solve α * ∈[0,1](convergence accuracy ò=10 -4 ), and the steps are as follows: Initialize interval [a0=0, b0=1], calculate golden section point: c0=a0+0.618×(b0-a0), d0=b0-0.618×(b0-a0) Calculate F(c0) and F(d0): If F(c0)>F(d0), then new interval [a1=a0, b1=d0], d1=c0, c1=a1+0.618×(b1-a1) If F(c0)<F(d0), then new interval [a1=c0, b1=b0], c1=d0, d1=b1-0.618×(b1-a1) iterating to b k -a k <ò, take Step 5: Determine the final combination weight Substitute the α * into the combination weight formula to obtain the final dynamic combination weight, see Figure 2 5. The method of claim 1, wherein the method further comprises: determining a correlation importance of each equipment based on the multi-factor fusion. The said correlation importance calculation includes: The normalized correlation values X' of the 4 indicators ij (t) with dynamic weights Weighted fusion, j = 1, 2, 3, 4, correlation importance R pq (t) Formula:

Images