Causal enhancement-based energy consumption anomaly diagnosis and adaptive optimization method and system
By employing a causal-enhanced energy consumption anomaly diagnosis and adaptive optimization method, the problem of insufficient accuracy and optimization strategies in energy consumption anomaly detection in highway energy systems has been solved. This method enables accurate identification and adaptive optimization of energy consumption anomalies, thereby improving operation and maintenance efficiency and energy utilization efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG ZHENGCHEN TECH CO LTD
- Filing Date
- 2026-04-30
- Publication Date
- 2026-06-19
AI Technical Summary
Existing highway energy systems suffer from problems such as insufficient detection accuracy, difficulty in explaining the causes of anomalies, and a lack of scenario adaptability in optimization strategies, resulting in high operation and maintenance costs and low efficiency.
A method for energy consumption anomaly diagnosis and adaptive optimization based on causal reinforcement is adopted. Through multi-source data preprocessing, feature construction, causal graph modeling, fuzzy clustering and scene adaptive reinforcement learning, the method can achieve accurate identification, root cause diagnosis and adaptive optimization of energy consumption anomalies.
It improves the accuracy and stability of energy consumption anomaly detection, enables interpretable diagnosis of the causes of energy consumption anomalies, and dynamically adjusts and optimizes strategies under different operating scenarios, thereby enhancing the intelligent and refined management level of the energy system and reducing operation and maintenance costs.
Smart Images
Figure CN122241457A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of highway energy consumption anomaly detection technology, specifically involving a method and system for energy consumption anomaly diagnosis and adaptive optimization based on causal reinforcement. Background Technology
[0002] The statements herein provide only background information in relation to this invention and do not necessarily constitute prior art.
[0003] With the continuous improvement of the intelligence and greenness of highway infrastructure, a large number of energy-consuming devices have been gradually installed in highway service areas, toll stations, tunnels, and monitoring centers. These energy systems are characterized by diverse equipment types, complex operating conditions, and significant energy consumption fluctuations. To reduce operating costs and improve energy efficiency, energy consumption monitoring and management systems have been gradually introduced into existing technologies to collect and analyze energy consumption data.
[0004] Currently, energy consumption anomaly management in highway energy systems mainly employs methods based on threshold judgment, statistical analysis, or simple machine learning models to detect anomalies in energy consumption data. However, these methods typically rely on fixed thresholds or historical statistical characteristics, making it difficult to adapt to dynamic changes in energy consumption levels under different seasons, time periods, and business loads. This can easily lead to false alarms or missed alarms, resulting in insufficient accuracy and stability in anomaly detection.
[0005] Furthermore, existing energy consumption anomaly detection technologies primarily focus on identifying whether an anomaly has occurred, lacking the ability to deeply analyze the causes of the anomaly. When an energy consumption anomaly is detected, the system usually cannot clearly point out the causal relationship between the anomaly and the equipment status, environmental factors, or business load. Maintenance personnel still need to rely on manual experience to troubleshoot, resulting in low fault location efficiency and high maintenance costs.
[0006] On the other hand, the operating scenarios of highway energy systems are highly diverse and time-varying. The appropriate operating strategies for energy-consuming equipment vary significantly depending on the season, time of day, and traffic flow conditions. Existing energy efficiency optimization methods mostly adopt uniform or static control strategies, lacking effective identification and differentiation of operating scenarios. This makes it difficult to dynamically adjust optimization strategies according to changes in scenarios, resulting in limited energy-saving effects.
[0007] Therefore, existing technologies still suffer from problems such as insufficient accuracy in anomaly detection, unexplainable causes of anomalies, and lack of scenario adaptability in optimization strategies in highway energy system energy consumption management. There is an urgent need for a technical solution that can accurately identify energy consumption anomalies, explain the root causes of anomalies, and has the ability to adaptively optimize in different scenarios. Summary of the Invention
[0008] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for energy consumption anomaly diagnosis and adaptive optimization based on causal reinforcement. This method achieves accurate identification, interpretable diagnosis, and scenario-adaptive energy-saving optimization of highway energy systems by constructing a complete technical process of "data acquisition - anomaly screening - causal root cause localization - scenario identification - strategy optimization - closed-loop verification".
[0009] To achieve the above objectives, the present invention is implemented through the following technical solution:
[0010] In a first aspect, the present invention provides a method for energy consumption anomaly diagnosis and adaptive optimization based on causal reinforcement, comprising the following steps:
[0011] S1: Preprocess the multi-source data, construct features, and build a feature sample set;
[0012] S2: Perform initial anomaly screening on the feature sample set to obtain the initial anomaly set;
[0013] S3: Based on causal graph modeling, calculate the average causal effect of each variable on energy consumption anomalies, quantify the root cause contribution and confidence level, and complete the root cause ranking.
[0014] S4: Output scene labels through fuzzy clustering scene recognition;
[0015] S5: Using scene labels and root cause ranking as input, it generates equipment control and load optimization strategies through scene adaptive reinforcement learning.
[0016] As a further technical solution, in step S1, the multi-source data includes energy consumption data, equipment status data, environmental data, and business data; in step S1, the multi-source data is time-aligned and cleaned. Specifically, the multi-source data is unified to the time granularity, missing values and outliers are processed, and the original data is denoised.
[0017] As a further technical solution, in step S1, feature construction includes baseline and deviation feature construction and feature vector construction.
[0018] As a further technical solution, in the construction of baseline and deviation features, a daily energy consumption baseline for the same scenario or type is constructed, and the energy consumption deviation features are calculated; in the construction of feature vectors, a feature vector is constructed for each time window to form a feature sample set, and statistical features are introduced.
[0019] As a further technical solution, in step S2, an isolated forest model is used to perform unsupervised anomaly detection on the feature sample set to obtain anomaly scores, and anomaly thresholds are set to obtain a preliminary anomaly set.
[0020] As a further technical solution, in step S3, causal variables are defined, a directed acyclic graph is constructed, and the average causal effect of each candidate causal variable is calculated. Then, the backdoor adjustment formula is used to remove interference, the independent causal contribution of each variable is separated, the root cause contribution and confidence are obtained, and the root cause ranking is output.
[0021] As a further technical solution, in step S4, a scene feature vector is constructed, a fuzzy C-means clustering algorithm is used to obtain an optimized objective function, and a scene label is defined.
[0022] As a further technical solution, in step S5, the state space of the reinforcement learning model is constructed, the action space is preset, and the operating constraints of the energy system are configured. A scenario-adaptive reward function is constructed, the scenario-adaptive reinforcement learning model is trained, the current state quantity is input into the scenario-adaptive reinforcement learning model, the optimal scheduling action is solved, and the equipment control and load optimization strategies are generated to realize the adaptive optimization operation of the energy system under different operating scenarios.
[0023] As a further technical solution, a closed-loop verification of the optimization effect is performed. When the scene distribution drifts or the anomaly recognition accuracy decreases, the model is triggered to self-learn and update.
[0024] Secondly, the present invention also provides an energy consumption anomaly diagnosis and adaptive optimization system based on causal reinforcement, comprising:
[0025] The first module is used to preprocess multi-source data, construct features, and build a feature sample set;
[0026] The second module is used to perform initial anomaly screening on the feature sample set to obtain an initial anomaly set;
[0027] The third module is used for causal graph modeling, calculating the average causal effect of each variable on energy consumption anomalies, quantifying the root cause contribution and confidence level, and completing the root cause ranking.
[0028] The fourth module is used for scene identification through fuzzy clustering and outputting scene labels;
[0029] The fifth module is used to generate device control and load optimization strategies through scene adaptive reinforcement learning, taking scene labels and root cause ranking as inputs.
[0030] The beneficial effects of the present invention are as follows:
[0031] The energy consumption anomaly diagnosis and adaptive optimization method of the present invention, by introducing a combination of anomaly detection and causal inference mechanisms, achieves accurate identification and root cause diagnosis of energy consumption anomalies in highway energy systems, avoiding misjudgment problems caused by relying solely on empirical thresholds or statistical characteristics; at the same time, by identifying and classifying the operating scenarios of the energy system, an energy efficiency optimization strategy that can be dynamically adjusted according to changes in the operating scenarios is constructed, so that the optimization results are more in line with actual operating needs.
[0032] The energy consumption anomaly diagnosis and adaptive optimization method of the present invention improves the accuracy and stability of energy consumption anomaly detection by introducing multi-source data and performing initial anomaly screening, and determining the root cause ranking based on the root cause contribution. This enables anomaly identification to adapt to energy consumption changes under different seasons, time periods and business load conditions.
[0033] The energy consumption anomaly diagnosis and adaptive optimization method of the present invention achieves interpretable diagnosis of the causes of energy consumption anomalies by root cause ranking and localization, and provides a basis for operation and maintenance decision-making by quantifying the causal contribution of different influencing factors to the abnormal results.
[0034] The energy consumption anomaly diagnosis and adaptive optimization method of the present invention trains a scenario adaptive reinforcement learning model and performs closed-loop verification of the optimization effect, thereby improving the scenario adaptability of the energy efficiency optimization strategy and enabling the energy system to obtain reasonable and effective optimization results under different operating scenarios.
[0035] The energy consumption anomaly diagnosis and adaptive optimization method of the present invention can effectively improve the intelligence and precision of energy consumption management of highway energy systems, reduce operation and maintenance costs, and improve energy utilization efficiency, and has good engineering application value. Attached Figure Description
[0036] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0037] Figure 1 This is a flowchart of the energy consumption anomaly diagnosis and adaptive optimization method based on causal reinforcement according to one or more embodiments of the present invention. Detailed Implementation
[0038] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0039] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless otherwise expressly indicated by the invention, the singular form is intended to include the plural form as well. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0040] As introduced in the background section, existing highway energy system energy consumption management suffers from problems such as insufficient anomaly detection accuracy, difficulty in explaining the causes of anomalies, and lack of scenario adaptability in energy efficiency optimization strategies. In order to solve the above technical problems, this invention proposes an energy consumption anomaly diagnosis and adaptive optimization method and system based on causal reinforcement.
[0041] In a typical embodiment of the present invention, such as Figure 1 As shown, a method for energy consumption anomaly diagnosis and adaptive optimization based on causal reinforcement is proposed, including the following steps:
[0042] S1: Preprocess the multi-source data, construct features, and build a feature sample set;
[0043] S2: Perform initial anomaly screening on the feature sample set to obtain the initial anomaly set;
[0044] S3: Based on causal graph modeling, calculate the average causal effect of each variable on energy consumption anomalies, quantify the root cause contribution and confidence level, and complete the root cause ranking.
[0045] S4: Output scene labels through fuzzy clustering scene recognition;
[0046] S5: Using scene labels and root cause ranking as input, it generates equipment control and load optimization strategies through scene adaptive reinforcement learning.
[0047] In step S1, the multi-source data includes energy consumption data, equipment status data, environmental data, and business data.
[0048] Specifically, energy consumption data includes active power P(t) and electrical energy consumption E(t), where t represents the time index.
[0049] In optional technical solutions, power quality indicators such as power factor and harmonic content can also be added to the energy consumption data.
[0050] Specifically, equipment status data includes equipment start / stop status, operating mode, control settings, fault codes, and equipment efficiency or health indicators.
[0051] The environmental data includes outdoor temperature T. out(t), humidity, irradiance, wind speed, etc.
[0052] The business data includes passenger or vehicle flow intensity F(t), holiday signage, and business load intensity of service areas or buildings.
[0053] In the optional technical solutions, multi-source data can also include electricity price and carbon factor data, which include time-of-use electricity price π(t) and grid emission factor g(t).
[0054] In step S1, time alignment and data cleaning are performed on the multi-source data. Specifically, the multi-source data are unified to a time granularity Δt (e.g., 1 min, 5 min, or 15 min), missing values and outliers are processed, and the original data is denoised.
[0055] Specifically, missing value completion methods include: linear interpolation, spline interpolation, and substitution with the same type of daily average.
[0056] Specifically, the original data is denoised using an exponential smoothing method, the calculation formula of which is: x tilde (t) =α · x(t) + (1 - α) · x tilde (t - Δt).
[0057] Where: x(t) is the original data, x tilde (t) represents the smoothed data, and α is the smoothing coefficient, with a value range of (0,1).
[0058] In step S1, feature construction includes baseline and deviation feature construction and feature vector construction.
[0059] Specifically, in the baseline and deviation feature construction, a daily energy consumption baseline B(t) for the same scenario or type is constructed, and the energy consumption deviation features are calculated:
[0060] d(t) = E(t) - B(t);
[0061] r(t) = (E(t)-B(t)) / (B(t) + ε);
[0062] Where: ε is a very small positive number to prevent the denominator from being zero.
[0063] Specifically, in the feature vector construction, a feature vector x is constructed for each time window w = [t - L, t]. t To form a feature sample set {x t}:
[0064] x t = [E(t), P(t), d(t), r(t), T out(t), F(t), state(t), setpoint(t),…];
[0065] And introduce statistical features:
[0066] μ w = (1 / L) · Σ τ ∈ w E(τ);
[0067] k w = (E(t) - E(t - L)) / L;
[0068] Where: μ w k represents the average energy consumption within the window. w This represents the slope of the energy consumption change.
[0069] In step S2, the Isolation Forest model is used to process the feature sample set {x}. t Unsupervised anomaly detection is performed to obtain anomaly score S. IF (t), setting the abnormal threshold θ IF This yields the initial anomaly set A0. Anomaly set A includes the anomaly type, anomaly confidence level, and scope of impact.
[0070] Specifically, let the average path length of sample x in the forest be E[h(x)], and the number of samples be n, then the standardization constant is:
[0071] c(n) = 2 · H(n -1) - 2 · (n - 1) / n;
[0072] Where: H(k) = Σ i = 1 k (1 / i).
[0073] The abnormality score is defined as:
[0074] S IF (x) = 2 (-E[h(x)] / c(n)) ;
[0075] When S IF When (x) approaches 1, it indicates that the sample is more abnormal;
[0076] When S IF When (x) is close to 0.5, it indicates that the sample is more normal.
[0077] The outlier scores are compared with a set outlier threshold. Samples with outlier scores greater than or equal to the set outlier threshold are selected to form an initial outlier set, A0 = {t | S}. IF (t) ≥θ IF}
[0078] In step S3, the process of causal graph modeling is as follows: define causal variables and construct a directed acyclic graph G = (V,E).
[0079] Specifically, the set of causal variables is defined as follows:
[0080] Y: Anomaly indicators (such as r(t) or anomaly labels);
[0081] X1: Outdoor temperature T out ;
[0082] X2: Passenger or vehicle flow F;
[0083] X3: Equipment efficiency or health status (H);
[0084] X4: Control setpoint U;
[0085] X5: Operating mode M;
[0086] X6: Electricity price π (optional);
[0087] Z: A set of mixed variables (seasons, time periods, holidays, etc.).
[0088] Construct the directed acyclic graph G = (V, E) as follows:
[0089] T out →Y;
[0090] F→Y;
[0091] H→Y;
[0092] U→Y;
[0093] (T out ,F)→U;
[0094] Z→(T out , F, U, Y).
[0095] The process of calculating the average causal effect is as follows:
[0096] For each candidate cause variable X i Calculate its average causal effect:
[0097] ACE i = E[Y | do(X i = x i (1))]-E[Y | do(X i = x i (0))];
[0098] Where: x i(1) Represents the value of a variable under abnormal conditions, x i (0) represents the value of the variable under normal conditions.
[0099] If there exists an adjustment set Z that satisfies the backdoor criterion i The backdoor adjustment formula is:
[0100] E[Y | do(X i = x)] = Σ z E[Y | X i = x, Z i = z] · P(Z i = z);
[0101] Thus, by eliminating interference, the independent causal contribution of each variable can be separated.
[0102] The root cause contribution is defined as follows:
[0103] C i = |ACE i | / (Σ j |ACE j | + ε);
[0104] The overall confidence level is defined as:
[0105] Conf i = σ(λ1 · S IF (t) + λ2 · C i );
[0106] The final root cause ranking is output based on root cause contribution and overall confidence:
[0107] Rank = sort desc {(X i C i Conf i )}.
[0108] Root cause ranking (Rank) includes the causal contribution of each candidate factor.
[0109] In step S4, a scene feature vector is constructed, and a fuzzy C-means clustering algorithm is used to obtain an optimized objective function and define scene labels.
[0110] Specifically, construct the scene feature vector: z t = [T out (t), humidity(t), F(t), hour(t),weekday(t), holiday(t), …].
[0111] The objective function is optimized using the fuzzy C-means clustering algorithm:
[0112] J = Σ t Σ k = 1 K u tk m · ||z t - v k || 2 .
[0113] During the process, the membership degree and cluster centers are updated. The membership degree update formula is as follows:
[0114] u tk = 1 / Σ j = 1 K (||z t - v k || / ||z t - v j ||) (2 / (m - 1)) ;
[0115] The cluster center update formula is: v k = (Σ t u tk m · z t ) / (Σ t u tk m ).
[0116] The scene label is defined as: s(t) = argmax k u tk This is used to describe the current running scenario.
[0117] In step S5, the state space of the reinforcement learning model is constructed, the action space is preset, and the operating constraints of the energy system are configured. A scenario-adaptive reward function is constructed, the scenario-adaptive reinforcement learning model is trained, the current state quantity is input into the scenario-adaptive reinforcement learning model, the optimal scheduling action is solved, and equipment control and load optimization strategies are generated to achieve adaptive optimization operation of the energy system under different operating scenarios.
[0118] Equipment control and load optimization strategies include equipment parameter adjustment, load peak shifting, energy storage charging and discharging, and equipment start-up and shutdown scheduling, etc.
[0119] Specifically, the state space construction process is as follows: based on the feature vector of the feature sample set, the scene feature vector, the scene label, and the primary cause in the root cause ranking, the state is defined as follows:
[0120] S t= [x t , z t , s(t), TopCause(t)].
[0121] Specifically, the action space includes: equipment setpoint adjustment, start-stop control, load shifting strategy, and energy storage charging and discharging power (optional).
[0122] The constraints include:
[0123] T in min ≤ T in (t) ≤ T in max ;
[0124] Q(t) ≥ Q min ;
[0125] a min ≤ a(t) ≤ a max ;
[0126] |a(t) - a(t - Δt)| ≤ Δa max .
[0127] Combining scene labels, an initial set of anomalies, and penalties for violating constraints, the reward function is defined as follows:
[0128] R t = -α(s) · ΔE(t) - β(s) · I[t ∈ A0] - γ · Penalty(t);
[0129] in:
[0130] ΔE(t) = E(t) - B(t) or r(t);
[0131] Penalty(t) = max(0, T in (t) - T in max ) + max(0, T in min - T in (t)) + ….
[0132] In a further technical solution, the optimization effect is verified in a closed loop. When the scene distribution drifts or the anomaly recognition accuracy decreases, the model is triggered to self-learn and update.
[0133] The energy saving rate is defined as:
[0134] η E = (E base - Eopt ) / E base .
[0135] Energy saving rate, comfort or constraint satisfaction rate, and abnormal fallback are used as indicators to evaluate the effectiveness of the strategy. When scene distribution drift or anomaly recognition accuracy is detected, the model parameters are automatically updated, and the "root cause-strategy-effect" triple is written into the knowledge base to achieve cross-scene migration and continuous optimization.
[0136] In another typical embodiment of the present invention, an energy consumption anomaly diagnosis and adaptive optimization system based on causal reinforcement is proposed, comprising:
[0137] The first module is used to preprocess multi-source data, construct features, and build a feature sample set;
[0138] The second module is used to perform initial anomaly screening on the feature sample set to obtain an initial anomaly set;
[0139] The third module is used for causal graph modeling, calculating the average causal effect of each variable on energy consumption anomalies, quantifying the root cause contribution and confidence level, and completing the root cause ranking.
[0140] The fourth module is used for scene identification through fuzzy clustering and outputting scene labels;
[0141] The fifth module is used to generate device control and load optimization strategies through scene adaptive reinforcement learning, taking scene labels and root cause ranking as inputs.
[0142] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for energy consumption anomaly diagnosis and adaptive optimization based on causal reinforcement, characterized in that, Includes the following steps: S1: Preprocess the multi-source data, construct features, and build a feature sample set; S2: Perform initial anomaly screening on the feature sample set to obtain the initial anomaly set; S3: Based on causal graph modeling, calculate the average causal effect of each variable on energy consumption anomalies, quantify the root cause contribution and confidence level, and complete the root cause ranking. S4: Output scene labels through fuzzy clustering scene recognition; S5: Using scene labels and root cause ranking as input, it generates equipment control and load optimization strategies through scene adaptive reinforcement learning.
2. The energy consumption anomaly diagnosis and adaptive optimization method as described in claim 1, characterized in that, In step S1, the multi-source data includes energy consumption data, equipment status data, environmental data, and business data. In step S1, the multi-source data is time-aligned and cleaned. Specifically, the multi-source data is unified to the time granularity, missing values and outliers are processed, and the original data is denoised.
3. The energy consumption anomaly diagnosis and adaptive optimization method as described in claim 1, characterized in that, In step S1, feature construction includes baseline and deviation feature construction and feature vector construction.
4. The energy consumption anomaly diagnosis and adaptive optimization method as described in claim 3, characterized in that, In the baseline and deviation feature construction, a daily energy consumption baseline for the same scenario or type is constructed, and the energy consumption deviation feature is calculated; In the feature vector construction process, a feature vector is constructed for each time window to form a feature sample set, and statistical features are introduced.
5. The energy consumption anomaly diagnosis and adaptive optimization method as described in claim 1, characterized in that, In step S2, the isolated forest model is used to perform unsupervised anomaly detection on the feature sample set to obtain anomaly scores, set anomaly thresholds, and obtain a preliminary anomaly set.
6. The energy consumption anomaly diagnosis and adaptive optimization method as described in claim 1, characterized in that, In step S3, causal variables are defined, a directed acyclic graph is constructed, and the average causal effect of each candidate causal variable is calculated. Then, the backdoor adjustment formula is used to remove interference, the independent causal contribution of each variable is separated, the root cause contribution and confidence are obtained, and the root cause ranking is output.
7. The energy consumption anomaly diagnosis and adaptive optimization method as described in claim 1, characterized in that, In step S4, a scene feature vector is constructed, and a fuzzy C-means clustering algorithm is used to obtain an optimized objective function and define scene labels.
8. The energy consumption anomaly diagnosis and adaptive optimization method as described in claim 1, characterized in that, In step S5, the state space of the reinforcement learning model is constructed, the action space is preset, and the operating constraints of the energy system are configured. The scenario adaptive reward function is constructed, the scenario adaptive reinforcement learning model is trained, the current state quantity is input into the scenario adaptive reinforcement learning model, the optimal scheduling action is solved, and the equipment control and load optimization strategies are generated to realize the adaptive optimization operation of the energy system under different operating scenarios.
9. The energy consumption anomaly diagnosis and adaptive optimization method as described in claim 1, characterized in that, The optimization effect is verified in a closed loop. When the scene distribution drifts or the anomaly recognition accuracy decreases, the model self-learning update is triggered.
10. A system for causal reinforcement-based energy consumption anomaly diagnosis and adaptive optimization, characterized in that, include: The first module is used to preprocess multi-source data, construct features, and build a feature sample set; The second module is used to perform initial anomaly screening on the feature sample set to obtain an initial anomaly set; The third module is used for causal graph modeling, calculating the average causal effect of each variable on energy consumption anomalies, quantifying the root cause contribution and confidence level, and completing the root cause ranking. The fourth module is used for scene identification through fuzzy clustering and outputting scene labels; The fifth module is used to generate device control and load optimization strategies through scene adaptive reinforcement learning, taking scene labels and root cause ranking as inputs.