Energy consumption monitoring method and device suitable for expressway operation

Abstract

The invention discloses an energy consumption monitoring method and device suitable for expressway operation. The method comprises the following steps: acquiring features for intelligent agent input from traffic flow related features, environment related features, operation related features, equipment load features and stability features according to current scene information; acquiring a trained reinforcement learning agent; inputting the features for agent input into a trained reinforcement learning agent so as to obtain an optimal collection parameter combination and a frequency decision; obtaining a scenarized optimization data set according to the optimal collection parameter combination and the frequency decision; acquiring scene basic data; and performing energy consumption monitoring according to the scenarized optimization data set and the scene basic data. According to the energy consumption monitoring method suitable for expressway operation, working condition perception and strategy dynamic adjustment closed loop are realized based on reinforcement learning agent and working condition feature quantification.

Description

Technical Field

[0001] This application relates to the field of energy consumption monitoring technology for highway operation, specifically to an energy consumption monitoring method and an energy consumption monitoring device suitable for highway operation. Background Technology

[0002] With the continuous expansion of the expressway network and the improvement of its intelligent operation, the total energy consumption of toll stations, service areas, tunnels, and roadside areas has been increasing year by year. Energy consumption monitoring has become a core link in the green operation and cost control of expressways.

[0003] Highway energy consumption scenarios are characterized by significant complexity and dynamism: on the one hand, energy consumption entities are diverse and dispersed, covering dozens of types of equipment such as ventilation fans, lighting systems, charging piles, and air conditioning. The energy consumption coupling relationship varies significantly in different scenarios (such as ventilation + lighting in tunnels, charging piles + catering + air conditioning in service areas). On the other hand, energy consumption is highly affected by operating conditions. Factors such as traffic flow tidal changes, sudden environmental changes (heavy rain, extreme temperatures), and adjustments to operating modes (maintenance construction, holiday operation) can all lead to drastic fluctuations in energy consumption, placing extremely high demands on the adaptability and accuracy of monitoring methods.

[0004] Existing highway energy consumption monitoring methods typically employ fixed data collection strategies, which cannot adapt to dynamic operating conditions. For example, current technologies often use a fixed parameter + fixed frequency collection mode, such as collecting total energy consumption and power consumption of a few devices every 15 minutes across all scenarios, without considering dynamic changes in operating conditions. Under complex conditions such as peak traffic flow and heavy rain, the low collection frequency and incomplete parameter coverage lead to missed data collection of key energy consumption fluctuations (such as charging pile overload and tunnel ventilation fan full-load operation). In low-value scenarios such as off-peak hours in the early morning and stable equipment operation, excessive collection of redundant data increases storage and analysis costs, and the collected data lacks specificity and cannot support subsequent in-depth analysis. Summary of the Invention

[0005] The purpose of this invention is to provide an energy consumption monitoring method suitable for highway operation to at least solve one of the above-mentioned technical problems.

[0006] One aspect of the present invention provides an energy consumption monitoring method suitable for highway operation, the energy consumption monitoring method suitable for highway operation comprising:

[0007] Acquire traffic flow-related features, environmental features, operational features, equipment load features, and stability features;

[0008] Obtain current scene information;

[0009] Based on the current scenario information, the input features for the intelligent agent are obtained from the traffic flow-related features, environment-related features, operation-related features, equipment load features, and stability features;

[0010] Acquire a trained reinforcement learning agent;

[0011] The input of the agent is fed into a trained reinforcement learning agent with features to obtain the optimal combination of acquisition parameters and frequency decision.

[0012] Obtain a scenario-optimized dataset based on the optimal combination of acquisition parameters and frequency decisions;

[0013] Obtain basic scene data;

[0014] Energy consumption monitoring is performed based on scenario-based optimization datasets and scenario-based basic data.

[0015] Optionally, the energy consumption monitoring based on the scenario-based optimized dataset includes:

[0016] Obtain a scene database, which includes at least one preset scene-based dynamic physical mechanism constraint equation and current scene information associated with each preset scene-based dynamic physical mechanism constraint equation;

[0017] Construct a coupled input matrix based on the scenario-optimized dataset and the scenario-based basic data;

[0018] Based on the current scene information, the preset scene-based dynamic physical mechanism constraint equations associated with the current scene information are obtained from the scene database;

[0019] The whitened data is obtained based on the coupling input matrix;

[0020] Obtain the separation matrix based on the whitened data;

[0021] The optimal separation matrix is ​​obtained by solving the separation matrix using gradient descent.

[0022] The independent energy consumption matrix is ​​obtained based on the optimal separation matrix and the whitened data.

[0023] Obtain a deep feature dataset based on the independent energy consumption matrix;

[0024] Generate a dynamic benchmark dataset based on the deep feature dataset;

[0025] Anomaly tracing reports are obtained based on deep feature datasets, dynamic benchmark datasets, and implementation detection data.

[0026] Optionally, the whitened data obtained from the coupling input matrix is ​​acquired in the following manner:

[0027] The covariance matrix of e can be obtained using the following formula:

[0028]

[0029] Where ∑ is the covariance matrix of e; e is the coupled energy consumption data. Coupled input matrix

[0030] Perform eigenvalue decomposition on the covariance matrix ∑ of e:

[0031] ∑=UΛV T ;

[0032] Where U is the orthogonal eigenvector matrix and Λ is the diagonal eigenvalue matrix;

[0033] The whitening conversion is performed using the following formula:

[0034]

[0035] Where z represents the whitened data; T represents the matrix transpose operation.

[0036] Optionally, obtaining the separation matrix based on the whitened data includes:

[0037] Obtain the target function;

[0038] The separation matrix that minimizes the objective function is found through iterative optimization using the gradient descent method.

[0039] Optionally, the objective function is as follows:

[0040]

[0041] Where J(W) is the objective function value, which needs to be minimized; Let W be the separation matrix to be solved; E[·] is the mathematical expectation operation; det W is the determinant of the separation matrix W; p s (·) represents the probability density function of the independent components; Wz is the product of the separation matrix and the whitened data; λ is the physical constraint weight. The physical constraint matrix; is the expected vector of independent components; T is the transpose of the matrix.

[0042] Optionally, the iterative optimization using gradient descent to find the separation matrix that minimizes the objective function includes:

[0043] Perform initialization, randomly setting the initial value of W;

[0044] The gradient of the objective function is calculated using the following formula:

[0045] in, The gradient of the objective function with respect to the separation matrix W; W -T φ(Wz)z is the transpose of the inverse of the separation matrix. T is the product of the derivative and the transpose of the whitened data; C is the physical constraint matrix; Wz is the product of the separation matrix and the whitened data;

[0046] Iteratively update W until the convergence condition is met, thereby obtaining the separation matrix that minimizes the objective function; Let be the expected vector of the independent components.

[0047] Optionally, obtaining the independent energy consumption matrix based on the optimal separation matrix and the whitened data includes:

[0048] The whitened data is processed using the optimal separation matrix to obtain the independent energy consumption components, which are then concatenated into a matrix to obtain the independent energy consumption matrix.

[0049] Optionally, obtaining the deep feature dataset based on the independent energy consumption matrix includes:

[0050] Obtain the operating condition parameter matrix;

[0051] A high-dimensional input feature matrix is ​​constructed based on the operating condition parameter matrix, the scenario-based optimization dataset, and the independent energy consumption matrix.

[0052] Obtain the trained feature extraction network;

[0053] The high-dimensional input feature matrix is ​​input into a trained feature extraction network to obtain a deep feature dataset.

[0054] Optionally, generating a dynamic benchmark dataset based on the deep feature dataset includes:

[0055] Extract energy consumption modeling data, operating condition modeling data, and metadata from the deep feature dataset;

[0056] Energy consumption modeling data and operating condition modeling data are time-series normalized and modeling matrix is ​​constructed using metadata to obtain a standardized modeling dataset;

[0057] Obtain a historical operating condition sample database, which includes multiple historical samples;

[0058] The similarity between the current working condition and each historical sample is calculated based on the standardized modeling dataset and historical samples, thereby obtaining the historical sample with the highest similarity as the sample pool for the same working condition.

[0059] An initial dynamic baseline is generated based on a sample pool with the same operating conditions.

[0060] Generate a dynamic benchmark dataset based on the initial dynamic benchmark.

[0061] This application also provides an energy consumption monitoring device suitable for highway operation, the energy consumption monitoring device suitable for highway operation comprising:

[0062] The feature acquisition module is used to acquire traffic flow-related features, environment-related features, operation-related features, equipment load features, and stability features.

[0063] Current scene information acquisition module, the current scene information acquisition module is used to acquire current scene information;

[0064] The intelligent agent input feature acquisition module is used to acquire intelligent agent input features from traffic flow-related features, environment-related features, operation-related features, equipment load features, and stability features based on current scene information.

[0065] A reinforcement learning agent acquisition module is used to acquire a trained reinforcement learning agent.

[0066] The optimal acquisition parameter combination and frequency decision acquisition module is used to input the agent input with features into the trained reinforcement learning agent, thereby obtaining the optimal acquisition parameter combination and frequency decision.

[0067] A scenario-based optimized dataset acquisition module is used to acquire a scenario-based optimized dataset based on the optimal combination of acquisition parameters and frequency decision.

[0068] A scene basic data acquisition module, which is used to acquire scene basic data;

[0069] An energy consumption monitoring module is used to monitor energy consumption based on a scenario-optimized dataset and scenario-based basic data.

[0070] The energy consumption monitoring method applicable to highway operation proposed in this application is based on reinforcement learning agents and quantification of operating condition characteristics, realizing a closed loop of operating condition perception and dynamic strategy adjustment. For complex operating conditions (peak hours, heavy rain), the method automatically increases the frequency of core energy consumption parameter collection and supplements related parameters to avoid missing key data. For low-value operating conditions (off-peak hours, early morning), the method automatically simplifies parameters and reduces the frequency to reduce redundant data, balancing monitoring accuracy with storage / analysis costs, and adapting to the dynamic operating characteristics of highways. Attached Figure Description

[0071] Figure 1 This is a schematic flowchart of an embodiment of the energy consumption monitoring method applicable to highway operation according to this application. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some, but not all, embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0073] like Figure 1 The energy consumption monitoring methods applicable to highway operation shown include:

[0074] Acquire traffic flow-related features, environmental features, operational features, equipment load features, and stability features;

[0075] In this embodiment, traffic flow-related features include:

[0076] (1) Traffic flow level, for example, the number of vehicles traveling in the same lane at the entrance / exit of a toll station within 15 minutes; the number of vehicles entering a tunnel within 15 minutes; the number of vehicles entering a service area within 15 minutes, etc. Various quantitative rules can also be set, for example, level 0: ≤50 vehicles / 15 minutes (extremely low traffic flow); level 1: 51-150 vehicles / 15 minutes (low traffic flow); level 2: 151-300 vehicles / 15 minutes (off-peak traffic flow); level 3: 301-500 vehicles / 15 minutes (medium-peak traffic flow); level 4: 501-800 vehicles / 15 minutes (peak traffic flow); level 5: >800 vehicles / 15 minutes (extreme peak, such as holidays).

[0077] (2) Traffic flow fluctuation coefficient. In this embodiment, the traffic flow fluctuation coefficient = (standard deviation of traffic flow over 3 consecutive periods) / (mean of traffic flow over 3 consecutive periods × 100%). Quantification rules (levels 0-5, the higher the value, the greater the fluctuation): Level 0: ≤5% (stable traffic flow); Level 1: 6%-10% (slight fluctuation); Level 2: 11%-15% (moderate fluctuation); Level 3: 16%-20% (significant fluctuation); Level 4: 21%-30% (violent fluctuation); Level 5: >30% (extreme fluctuation, such as sudden traffic control).

[0078] Environmental characteristics include:

[0079] Temperature levels, such as temperature information at toll booths, tunnels, and along roadsides. Quantification rules (0-5 levels, combined with equipment energy consumption sensitivity ranges):

[0080] Level 0: ≤-10℃ (extremely low temperature, requires high heating energy consumption); Level 1: -9℃~0℃ (low temperature, requires heating); Level 2: 1℃~15℃ (comfortable temperature range, low energy consumption); Level 3: 16℃~28℃ (optimal temperature range, lowest energy consumption); Level 4: 29℃~35℃ (high temperature, requires cooling); Level 5: >35℃ (extreme high temperature, high cooling energy consumption).

[0081] Precipitation levels, quantification rules (0-5 levels, referencing meteorological standards):

[0082] Level 0: No precipitation (no additional energy consumption adjustment required); Level 1: Light rain (daily precipitation ≤ 10mm, slight increase in lighting brightness required); Level 2: Moderate rain (10mm < daily precipitation ≤ 25mm, increase in lighting brightness required); Level 3: Heavy rain (25mm < daily precipitation ≤ 50mm, significantly increase in lighting + check drainage equipment required); Level 4: Rainstorm (50mm < daily precipitation ≤ 100mm, maximum lighting required + activate emergency drainage); Level 5: Extreme rainstorm (daily precipitation > 100mm, extreme lighting required + full-load drainage required).

[0083] Visibility rating, quantification rules (0-5 levels, related to lighting power and ventilation frequency): Level 0: >2000 meters (excellent, minimum lighting); Level 1: 1001-2000 meters (good, low lighting); Level 2: 501-1000 meters (fair, medium lighting); Level 3: 201-500 meters (poor, high lighting + ventilation); Level 4: 101-200 meters (poor, maximum lighting + high-frequency ventilation); Level 5: ≤100 meters (very poor, emergency lighting + full-load ventilation).

[0084] Operational characteristics:

[0085] Operational levels, i.e., the operational status of each scenario, are quantified by rules (levels 0-5, associated with energy consumption fluctuation risks): Level 0: Normal operation (no intervention, stable energy consumption); Level 1: Light maintenance (single equipment / single area maintenance, such as the closure of one lane at a toll station); Level 2: Moderate maintenance (multiple equipment / multiple areas maintenance, such as the overhaul of a ventilation fan on one side of a tunnel); Level 3: Heavy maintenance (partial functional limitations across all scenarios, such as half of the charging piles in a service area being out of service); Level 4: Holiday operation (surge in traffic / pedestrian flow, such as during the Spring Festival and National Day); Level 5: Emergency status (accidents / extreme weather, such as vehicle breakdowns in tunnels or road closures due to blizzards).

[0086] Equipment load characteristics:

[0087] Ventilation fan load rating (tunnel scenario only)

[0088] Main body and data source: Tunnel axial flow / jet fan (ratio of actual power to rated power of a single unit).

[0089] Calculation logic: Load rate = (Actual operating power ÷ Rated power) × 100%.

[0090] Quantification rules (levels 0-5, reflecting energy consumption intensity): Level 0: 0% (shutdown); Level 1: 1%-30% (low load, energy-saving mode); Level 2: 31%-50% (low to medium load, normal ventilation); Level 3: 51%-70% (medium to high load, enhanced ventilation); Level 4: 71%-90% (high load, strong ventilation); Level 5: >90% (full load, emergency ventilation).

[0091] Air conditioning load level (toll station / service area)

[0092] Subject and data source:

[0093] Toll station: air conditioning in toll booths, central air conditioning in office area (actual power to rated power ratio);

[0094] Service Area: Food and Beverage Area / Guest Room Air Conditioning (Actual Power Ratio to Rated Power).

[0095] Calculation logic: Load rate = (Actual operating power ÷ Rated power) × 100%.

[0096] Quantification rules (levels 0-5, associated with cooling / heating intensity): Level 0: 0% (off); Level 1: 1%-20% (standby / low fan mode, low energy consumption); Level 2: 21%-40% (weak mode, mild temperature adjustment); Level 3: 41%-60% (medium mode, normal temperature adjustment); Level 4: 61%-80% (strong mode, deep temperature adjustment); Level 5: >80% (full load, extreme temperature adjustment).

[0097] Charging station utilization rate (service area only)

[0098] Main data and source: Service area charging piles (ratio of occupied charging piles to total number of charging piles within 15 minutes).

[0099] Calculation logic: Utilization rate = (Number of occupied piles ÷ Total number of piles) × 100%.

[0100] Quantification rules (levels 0-5, reflecting charging energy consumption pressure): Level 0: 0% (no vehicles charging); Level 1: 1%-20% (low utilization, low energy consumption); Level 2: 21%-40% (low to medium utilization, normal energy consumption); Level 3: 41%-60% (utilization, medium energy consumption); Level 4: 61%-80% (high utilization, high energy consumption); Level 5: >80% (full load, requires capacity expansion or peak shifting).

[0101] Lighting operation modes (all scenes)

[0102] Subject and data source:

[0103] Toll booth: Plaza lighting;

[0104] Service area: Public area / parking lot lighting;

[0105] Tunnel: Main lighting inside the tunnel;

[0106] LED streetlights along the road.

[0107] Quantification rules (levels 0-5, based on power percentage): Level 0: Off (0 power); Level 1: Low brightness (≤30% of rated power); Level 2: Medium brightness (31%-60% of rated power); Level 3: High brightness (61%-90% of rated power); Level 4: Ultra-high brightness (>90% of rated power); Level 5: Emergency mode (backup power supply, power is variable but must be monitored first).

[0108] Stability characteristics (main body: power grid and equipment operation):

[0109] Power grid stability (all scenarios)

[0110] Main body and data source: power supply lines in various scenarios (voltage fluctuation range within 15 minutes).

[0111] Calculation logic: Fluctuation amplitude = (maximum voltage - minimum voltage) ÷ rated voltage × 100%.

[0112] Quantification rules (levels 0-5, reflecting the impact of voltage stability on energy consumption): Level 0: ≤1% (extremely stable, no deviation in energy consumption calculation); Level 1: 1.1%-2% (stable, slight impact); Level 2: 2.1%-3% (generally stable, moderate impact); Level 3: 3.1%-5% (unstable, significant impact); Level 4: 5.1%-8% (relatively unstable, need to be alert for equipment failure); Level 5: >8% (extremely unstable, high probability of abnormal energy consumption).

[0113] Device start / stop frequency (core devices across all scenarios):

[0114] Subject and data source:

[0115] Toll booths: barriers, toll booths, and air conditioning;

[0116] Service area: catering equipment, air conditioning;

[0117] Tunnel: ventilation fans, emergency power supply.

[0118] Calculation logic: The number of times the equipment switches from shutdown to operation within 1 hour.

[0119] Quantitative rules (levels 0-5, reflecting equipment aging / control rationality): Level 0: 0 times (continuous operation or shutdown, optimal state); Level 1: 1-2 times (normal start-up and shutdown); Level 2: 3-5 times (relatively frequent, slight energy loss); Level 3: 6-10 times (frequent, significant energy loss); Level 4: 11-20 times (extremely frequent, high energy loss); Level 5: >20 times (abnormal start-up and shutdown, possible equipment failure).

[0120] Obtain current scene information;

[0121] Based on the current scenario information, the input features for the intelligent agent are obtained from the traffic flow-related features, environment-related features, operation-related features, equipment load features, and stability features;

[0122] Specifically, if any of the features mentioned above are specific to certain scenarios and are not currently used to calculate the energy consumption of that scenario, they should be removed and replaced with other features or marked with invalid values.

[0123] For example, in a tollbooth scenario (removing service area-specific features and adding tollbooth-specific features):

[0124] Retain common features: traffic flow level, traffic flow fluctuation coefficient, temperature level, precipitation level, operation level, power grid stability, and equipment start-up and shutdown frequency (number of times the air conditioners of the barrier gate and toll booth are started and stopped);

[0125] Remove service area-specific features: air conditioning load level (specifically referring to service area catering / guest room air conditioning), charging pile utilization rate, water consumption characteristics of catering area, energy consumption characteristics of guest room, and load level of sewage treatment equipment;

[0126] Additional features specific to toll stations (5 items):

[0127] Load rating of the barrier gate equipment (ratio of operating power to rated power);

[0128] Air conditioning load level of toll booths (power ratio of air conditioning in a single booth);

[0129] Plaza lighting load rating (power percentage of lighting circuits);

[0130] ETC gantry equipment operating power rating;

[0131] Total energy consumption fluctuation coefficient of office area;

[0132] The final feature vector dimension is 12 (7 general + 5 specific).

[0133] Tunnel scenario (removing service area-specific features and adding tunnel-specific features):

[0134] Retain common features: traffic flow level (traffic flow at tunnel entrance), traffic flow fluctuation coefficient, temperature level (temperature inside the tunnel), precipitation level, operation level, power grid stability, and equipment start-up and shutdown frequency (number of times fans and emergency power supplies are started and stopped).

[0135] Eliminate service area-specific features: charging pile utilization rate, catering / guest room related features, and sewage treatment equipment load level;

[0136] Additional tunnel-specific features (5 items):

[0137] Ventilation fan load rating (power ratio of axial flow / jet flow fans);

[0138] Tunnel lighting operation mode (main lighting power level inside the tunnel);

[0139] Visibility level inside the cave;

[0140] CO concentration level inside the cave (related to ventilation energy consumption);

[0141] Emergency lighting load rating;

[0142] Final feature vector dimensions: 12 dimensions (7 general + 5 specific), adapted to core energy consumption scenarios of tunnel ventilation and lighting.

[0143] Acquire a trained reinforcement learning agent;

[0144] In this embodiment, training is performed first to construct a trained reinforcement learning agent:

[0145] Offline training (building the initial decision model):

[0146] Dataset: Historical data for one or more years (including 1 million samples of working condition features S-collection strategy A-actual reward R).

[0147] Training algorithm: TD(λ) algorithm, which iteratively updates the action value function Q(S,A) (representing the expected reward for performing action A in state S): Q_new(S,A)=Q_old(S,A)+α×[R+γ×maxQ(S',A')-Q_old(S,A)]; where α=0.1 (learning rate), γ=0.9 (discount factor, emphasizing future rewards), and λ=0.7 (eligibility trace parameter, accelerating convergence).

[0148] Convergence condition: The change in Q value is less than 0.001 in 1000 consecutive iterations. After training, an initial Q table is generated (to store the optimal Q value of each SA).

[0149] Online Iteration (Dynamic Optimization Decision Model):

[0150] The Q-table is fine-tuned daily using newly generated real-time data (S, A, R), with the learning rate α decaying to 0.01, ensuring that the agent adapts to changes in operating modes (such as the addition of new equipment or seasonal changes).

[0151] The exploration rate ε = 0.05 (95% probability of selecting the current optimal action, 5% probability of randomly exploring new actions) avoids strategy rigidity.

[0152] In this embodiment, the process of inputting the agent's input with features into a trained reinforcement learning agent to obtain the optimal combination of acquisition parameters and frequency decision includes:

[0153] State input: Input the agent with feature S every 5 minutes; (2) Action selection: The agent queries the Q table and selects the action A = (A1_opt, A2_opt) that maximizes Q(S,A); (3) Policy output: Convert the optimal parameter combination A1_opt and frequency A2_opt into executable instructions.

[0154] The input of the agent is fed into a trained reinforcement learning agent with features to obtain the optimal combination of acquisition parameters and frequency decision.

[0155] The above-mentioned intelligent agent part will be further elaborated below by way of example. It should be understood that the example does not constitute any limitation on this application.

[0156] Select the tunnel scenario (non-service area, verify dynamic adaptation of features). Specific working conditions: a long tunnel (3km in length) on a highway, from 10:00 to 10:15 am on a weekday, with off-peak traffic and moderate rain. The equipment inside the tunnel is operating stably, with no maintenance / emergency situations.

[0157] The obtained data is shown in Table 1 below:

[0158] Table 1: Standardized Valid Data for Tunnels

[0159]

[0160] The above data is then quantified based on operating condition characteristics, as follows:

[0161] Based on a subset of tunnel scene features (7 general + 5 specific), the raw data is quantized into a 0-5 level value:

[0162] 1. General Feature Quantization

[0163] (1) Traffic flow level (s1)

[0164] Original data: 220 vehicles / 15 minutes;

[0165] Quantification rules: Level 0 ≤ 50, Level 1 51-150, Level 2 151-300, Level 3 301-500, Level 4 501-800, Level 5 > 800;

[0166] Calculation result: 220∈[151,300]→s1=2 levels;

[0167] (2) Traffic flow fluctuation coefficient (s2)

[0168] Original data: Traffic flow over three consecutive periods (210, 220, 205)

[0169] Volatility coefficient = (standard deviation ÷ mean) × 100%

[0170] Volatility coefficient = (7.64 ÷ 211.67) × 100% ≈ 3.61%

[0171] Quantification rules: Level 0 ≤ 5%, Level 1 6%-10%, ..., Level 5 > 30%

[0172] Calculation result: 3.61% ≤ 5% → s2 = Level 0

[0173] (3) Temperature rating (s3):

[0174] Original data: Average temperature inside the cave: 23℃

[0175] Quantification rules: Level 0 ≤ -10, Level 1 -9 to 0, Level 2 1 to 15, Level 3 16 to 28, Level 4 29 to 35, Level 5 > 35;

[0176] Calculation result: 23∈[16,28]→s3=Level 3 (Optimal energy consumption zone)

[0177] (4) Precipitation level (s4)

[0178] Original data: Moderate rain (35mm / day);

[0179] Quantification rules: Level 0: No precipitation; Level 1: Light rain ≤10; Level 2: Moderate rain 10-25; Level 3: Heavy rain 25-50; Level 4: Rainstorm 50-100; Level 5: Extremely heavy rain >100.

[0180] Calculation result: 35∈[25,50]→s4=Level 3 (heavy rain);

[0181] (5) Operational Level (S5):

[0182] Original state: Normal operation (no maintenance / emergency);

[0183] Quantification rules: Level 0 is normal, Level 1 is light maintenance, ..., Level 5 is an emergency;

[0184] Calculation result: s5 = level 0

[0185] (6) Power grid stability (s6):

[0186] Original value: Voltage fluctuation range 1.2%;

[0187] Quantification rules: Level 0 ≤ 1%, Level 1 1.1%-2%, Level 2 2.1%-3%, ..., Level 5 > 8%;

[0188] Calculation result: 1.2% ∈ [1.1,2] → s6 = Level 1 (stable);

[0189] (7) Equipment start-up and shutdown frequency (s7):

[0190] Original value: Start and stop once per hour;

[0191] Quantification rules: Level 0: 0 times; Level 1: 1-2 times; Level 2: 3-5 times; ...; Level 5: >20 times.

[0192] Calculation result: 1st time ∈ [1,2] → s7 = Level 1 (normal start-stop);

[0193] 2. Quantification of tunnel-specific features (5 items)

[0194] (8) Fan load rating (s8);

[0195] Original values: Actual power of the fan is 25kW, rated power is 50kW;

[0196] Calculation logic: Load rate = (actual power ÷ rated power) × 100% = (25 ÷ 50) × 100% = 50%;

[0197] Quantification rules: Level 0: 0%, Level 1: 1%-30%, Level 2: 31%-50%, Level 3: 51%-70%, Level 4: 71%-90%, Level 5: >90%;

[0198] Calculation result: 50%∈[31,50]→s8=Level 2 (medium to low load);

[0199] (9) Tunnel lighting operation mode (s9);

[0200] Original values: Actual lighting power 6kW, rated power 10kW, power ratio 60%;

[0201] Quantification rules: Level 0: Closed; Level 1: ≤30%; Level 2: 31%-60%; Level 3: 61%-90%; Level 4: >90%; Level 5: Emergency.

[0202] Calculation result: 60%∈[31,60]→s9=Level 2 (medium brightness);

[0203] (10) Visibility level inside the cave (s10);

[0204] Original data: Visibility inside the cave: 650 meters

[0205] Quantification rules: Level 0 > 2000, Level 1 1001-2000, Level 2 501-1000, Level 3 201-500, Level 4 101-200, Level 5 ≤ 100;

[0206] Calculation result: 650∈[501,1000]→s10=Level 2 (General Visibility)

[0207] (11) CO concentration level inside the cave (s11)

[0208] Raw data: CO concentration 0.005% (normal range)

[0209] Quantification rules: Level 0 ≤ 0.005% (normal), Level 1 0.005%-0.01%, Level 2 0.01%-0.02%, Level 3 0.02%-0.03%, Level 4 0.03%-0.05%, Level 5 > 0.05%;

[0210] Calculation result: 0.005% ∈ [0, 0.005] → s11 = Level 0

[0211] (12) Emergency lighting load level (s12)

[0212] Initial state: Emergency lighting off (main lighting operating normally)

[0213] Quantification rules: Level 0: Off; Level 1: ≤20% load; Level 2: 21%-40% load; ...; Level 5: Full load.

[0214] Calculation result: s12 = Level 0

[0215] 3. Final operating condition feature vector (12-dimensional with numerical values)

[0216] Combining the above 12 level values, we obtain the input features for the intelligent agent: S = [s1 = 2, s2 = 0, s3 = 3, s4 = 3, s5 = 0, s6 = 1, s7 = 1, s8 = 2, s9 = 2, s10 = 2, s11 = 0, s12 = 0].

[0217] The above-mentioned agent inputs are fed into a trained reinforcement learning agent using feature inputs, thereby obtaining the optimal acquisition strategy for the agent's output, for example:

[0218] The following parameters were collected: fan power (kW), lighting power (kW), cumulative energy consumption (kWh), tunnel temperature (°C), tunnel visibility (m), and CO concentration (%).

[0219] Data collection frequency: 300 seconds / time (i.e., 5 minutes / time).

[0220] Obtain a scenario-optimized dataset based on the optimal combination of acquisition parameters and frequency decisions;

[0221] Taking the above example as an example, the collection of 6 target parameters will be enabled (fan power, lighting power, cumulative energy consumption, tunnel temperature, tunnel visibility, CO concentration), while other irrelevant parameters will be turned off (such as service area charging pile related parameters, which are not available in the tunnel scenario).

[0222] The data collection period is set to 300 seconds. After each collection, the data is uploaded via the MQTT protocol to obtain a scenario-optimized dataset.

[0223] Obtain basic scene data;

[0224] In this embodiment, the basic scene data may include the following data:

[0225] Obtain basic scene identifiers (e.g., tunnel length 3km, service area 5000㎡); basic energy consumption data (total scene energy consumption, total incoming line electricity meter data); core equipment list, such as basic parameters of all energy-consuming equipment in the scene (model, rated power, installation location); basic environmental data, such as ambient temperature and humidity (°C / %), atmospheric pressure (kPa), etc.

[0226] Energy consumption monitoring is performed based on scenario-based optimization datasets and scenario-based basic data.

[0227] In this embodiment, the energy consumption monitoring based on the scenario-based optimization dataset includes:

[0228] Obtain a scene database, which includes at least one preset scene-based dynamic physical mechanism constraint equation and current scene information associated with each preset scene-based dynamic physical mechanism constraint equation;

[0229] Taking a tunnel scenario as an example, the total energy consumption in a tunnel scenario is as follows:

[0230] E total (t)=E fan (t)+E light (t)+E traffic (t)+E env (t)+δ(t)

[0231] Among them, E fan (t) represents the wind turbine energy consumption, El ight (t) represents lighting energy consumption, E traffic (t) represents energy consumption related to traffic flow, E env (t) represents the environmentally related energy consumption; δ(t) represents the coupling error at time t, satisfying |δ(t) / E total (t)|≤2%).

[0232] In this embodiment, the wind turbine energy consumption is obtained using the following formula:

[0233] ; where η fan (t)=η0-α·(P fan (t) / P fan，rated ) 2 η fan (t) represents the dynamic efficiency, η0 = 0.85 represents the rated efficiency, and α = 0.12 represents the efficiency decay coefficient; k fan，vis The dynamic correlation coefficient for visibility (fitted in real time using gradient descent); k fan，CO Δt is the dynamic correlation coefficient of CO concentration (fitted in real time using gradient descent); P is the acquisition interval; Δt is the dynamic correlation coefficient of CO concentration. fan (t) represents the actual operating power of the wind turbine at time t; C CO (t) represents the CO concentration inside the cave at time t; P fan，rated This refers to the rated power of the fan.

[0234] In this embodiment, the lighting energy consumption formula is as follows:

[0235] E light (t)=P light (t)·Δt·[1+k light，vis ·exp(-V vis (t) / V0)]

[0236] Where V0 is 1000m, representing the baseline visibility; k light，vis The value is 0.75, which is the visibility correlation coefficient. (P) light (t) represents the actual operating power of the lighting at time t;

[0237] In this embodiment, E traffic The formula for (t) is as follows:

[0238] E traffic (t)=k traffic ·N traffic (t)·[1+β·ρ traffic (t)];

[0239] Where, k traffic It is 0.03 kWh / vehicle, k traffic N is the basic correlation coefficient between traffic flow and energy consumption. traffic (t) represents the traffic flow within 15 minutes at time t; β is 0.002m. -1 ρ is the traffic density correction factor; tra.fic (t) represents the traffic density at time t.

[0240] In this embodiment, Among them, Ltunnel This represents the tunnel length.

[0241] In this embodiment, E env (t) is obtained through the following formula:

[0242]

[0243] Where, k env，T 0.05 kWh / °, representing the correlation coefficient between temperature and energy consumption; T(t) is the average temperature inside the cave at time t; T0 is 25°; P atm (t) represents the real-time atmospheric pressure at time t; P0 represents the standard atmospheric pressure.

[0244] For example, the total energy consumption in a service area scenario is as follows:

[0245] E total，服务区 =E 空调 +E 充电桩 +E 餐饮设备 +E 公共照明 +E 办公设备 ;

[0246] Among them, E 空调 Energy consumption for guest room air conditioning + lobby central air conditioning + kitchen air conditioning; E 充电桩 Energy consumption (including standby power loss) for all fast / slow charging stations; E 餐饮设备 Energy consumption of kitchen refrigerators, stoves, and microwave ovens; E 公共照明 Energy consumption for plaza streetlights + corridor lighting + parking lot lighting; E 办公设备 Energy consumption for front desk computers, printers, and monitoring systems.

[0247] In this embodiment, all of the above-mentioned data can undergo various existing preprocessing methods, such as noise reduction and standardization.

[0248] Construct a coupled input matrix based on the scenario-optimized dataset and the scenario-based basic data;

[0249] In this embodiment, the coupling input matrix is Where N is the time series length, and M = M1 + M2 (M1 is the dimension of energy consumption parameters, and M2 is the dimension of drive parameters);

[0250] Matrix structure: X = [E T D T ] T ,in Standardized total energy consumption, equipment power, and other information. (Standardized traffic flow, environmental parameters, power grid parameters, etc.)

[0251] Based on the current scene information, a preset scenario-based dynamic physical mechanism constraint equation associated with the current scene information is obtained from the scene database; for example, if the current scene is a service area scene, the total energy consumption is as follows:

[0252] E total，服务区 =E 空调 +E 充电桩 +E 餐饮设备 +E 公共照明 +E 办公设备 ;

[0253] The whitened data is obtained based on the coupling input matrix;

[0254] In this embodiment, the whitened data is obtained based on the coupling input matrix in the following manner:

[0255] The covariance matrix of e can be obtained using the following formula:

[0256]

[0257] Where ∑ is the covariance matrix of e; e is the coupled energy consumption data. Coupled input matrix

[0258] Perform eigenvalue decomposition on the covariance matrix ∑ of e:

[0259] ∑=UΛU T ;

[0260] Where U is the orthogonal eigenvector matrix and Λ is the diagonal eigenvalue matrix;

[0261] The whitening conversion is performed using the following formula:

[0262]

[0263] Where z represents the whitened data; T represents the matrix transpose operation.

[0264] Obtain the separation matrix based on the whitened data;

[0265] In this embodiment, obtaining the separation matrix based on the whitened data includes:

[0266] Obtain the target function;

[0267] The separation matrix that minimizes the objective function is found through iterative optimization using the gradient descent method.

[0268] In this embodiment, the objective function is as follows:

[0269]

[0270] Where J(W) is the objective function value, which needs to be minimized; The separation matrix to be solved is given by K (where K is the number of independent energy consumption components, determined by scene constraints, such as K=4 in a tunnel scene); E[·] is the mathematical expectation operation; det W is the determinant of the separation matrix W; p s (·) represents the probability density function of the independent components (using a kurtosis-optimized Gaussian mixture distribution); Wz is the product of the separation matrix and the whitened data; λ is the physical constraint weight, which is 0.15 in this application; The physical constraint matrix consists of the coefficients of the dynamic constraint equations, such as C containing η in a tunnel scenario. fan (t), k fan，vis Equal dynamic coefficient; Let E be the expected vector of the independent components. source =[E[E fan ], E[E light ], E[E traffic ], E[E env ]] T , E[E fan ] represents the theoretical expectation of wind turbine energy consumption, derived from the calculated values ​​of operating parameters and constraint equations; T represents the transpose of the matrix.

[0271] In this embodiment, the gradient descent method is used for iterative optimization to find the separation matrix W that minimizes the objective function. * include:

[0272] Perform initialization by randomly setting the initial value of W (element range [-0.5, 0.5], ensuring that the determinant ≠ 0, and avoiding singular matrices);

[0273] The gradient of the objective function is calculated using the following formula:

[0274] in, The gradient of the objective function with respect to the separation matrix W; W -T φ(Wz)z is the transpose of the inverse of the separation matrix. T is the product of the derivative and the transpose of the whitened data; C is the physical constraint matrix; Wz is the product of the separation matrix and the whitened data;

[0275] Iterative update of W (adding a momentum term to accelerate convergence):

[0276] Where μ is 0.012, and is the learning rate (which decreases with iteration, (μ...) t+1 =μ t 0.98 t v = 0.08, which is the momentum coefficient (to reduce iterative oscillations);

[0277] Continue until the convergence condition is met, thereby obtaining the separation matrix that minimizes the objective function;

[0278]

[0279] Let be the expected vector of the independent components.

[0280] In this embodiment, convergence is determined when the change in W over 100 consecutive iterations satisfies ||W (t+1) -W (t) || F <10 -7 (Frobenius norm), stop iteration, and obtain the optimal separation matrix W. * ;

[0281] The independent energy consumption matrix is ​​obtained based on the optimal separation matrix and the whitened data.

[0282] In this embodiment, obtaining the independent energy consumption matrix based on the optimal separation matrix and the whitened data includes:

[0283] The whitened data is processed using the optimal separation matrix to obtain the independent energy consumption components, which are then concatenated into a matrix to obtain the independent energy consumption matrix.

[0284] Specifically, the formula for the independent energy consumption matrix is ​​as follows:

[0285] S = W * ·z; where, (N is the time series length, K is the number of independent components); each column corresponds to an independent energy consumption component (e.g., in a tunnel scenario, S = [E...]). fan E light E traffic E env ] T ).

[0286] Inverse standardization restoration: The standardized data of S are restored using x = x * ·σ x +μ x It is converted back to the actual energy consumption unit (kWh) to ensure that the physical meaning can be directly interpreted.

[0287] Obtain a deep feature dataset based on the independent energy consumption matrix;

[0288] In this embodiment, obtaining the deep feature dataset based on the independent energy consumption matrix includes:

[0289] Obtain the operating condition parameter matrix;

[0290] A high-dimensional input feature matrix is ​​constructed based on the operating condition parameter matrix, the scenario-based optimization dataset, and the independent energy consumption matrix.

[0291] In this embodiment, the high-dimensional input feature matrix F has a dimension of The matrix structure is as follows:

[0292] F = [S T D T G T ] T

[0293] Wherein, S is the independent energy consumption matrix; D is the standardized operating condition parameter matrix (M2 in the above text is the operating condition parameter dimension, such as traffic flow, temperature, etc.); G is the feature vector matrix related to the current scenario selected from the traffic flow related features, environment related features, operation related features, equipment load features, and stability features. For example, the final feature vector dimension of the toll station scenario in the above text is 12 dimensions (7 general + 5 specific).

[0294] Using a temporal sliding window technique (sliding window length W = 6, meaning 6 consecutive time points constitute a sample), F is converted into a three-dimensional input tensor. (Adapted to CNN-LSTM network input format).

[0295] To obtain the trained feature extraction network, this application uses a CNN-LSTM network, including a CNN, an attention mechanism layer, an LSTM layer, and a fully connected layer;

[0296] The high-dimensional input feature matrix is ​​input into a trained feature extraction network to obtain a deep feature dataset.

[0297] In this embodiment, an 18-dimensional latent feature V is extracted using a feature extraction network, as detailed below:

[0298]

[0299]

[0300] Generate a dynamic benchmark dataset based on the deep feature dataset;

[0301] In this embodiment, generating a dynamic benchmark dataset based on a deep feature dataset includes:

[0302] Energy consumption modeling data, operating condition modeling data, and metadata are extracted from the deep feature dataset; specifically, the energy consumption modeling data includes an independent energy consumption matrix S and total scene energy consumption data E. total The working condition modeling data includes the working condition feature vector G and the 18-dimensional latent feature V.

[0303] Metadata deduction of time sequence timestamps and scene identifiers.

[0304] Energy consumption modeling data and operating condition modeling data are time-series normalized and modeling matrix is ​​constructed using metadata to obtain a standardized modeling dataset;

[0305] Specifically, the data is sorted in ascending order by timestamp to ensure that energy consumption data corresponds one-to-one with operating condition data (time synchronization error ≤10ms).

[0306] Modeling matrix M construction: dimension is The matrix structure is as follows:

[0307] M = [S T G T V T ] T .

[0308] Obtain a historical operating condition sample database, which includes multiple historical samples;

[0309] In this embodiment, the historical sample library is sourced from standardized modeling datasets from the past year or more (stored according to scenario categories, such as tunnel scenario sample library and service area scenario sample library);

[0310] Each historical sample is:

[0311]

[0312] Where i = 1, 2, ..., M, M is the total number of historical samples; and the corresponding independent energy consumption data S is associated with it. hist,i Total energy consumption data E total,hist,i ;

[0313] The similarity between the current working condition and each historical sample is calculated based on the standardized modeling dataset and historical samples, thereby obtaining the historical sample with the highest similarity as the sample pool for the same working condition.

[0314] In this embodiment, calculating the similarity between the current working condition and each historical sample based on the standardized modeling dataset and historical samples, and then obtaining the historical sample with the highest similarity as the sample pool for the same working condition, includes:

[0315] Define the current operating condition feature vector G curr :

[0316]

[0317] G0 is the current working condition feature vector G; V0 is the current 18-dimensional latent feature vector.

[0318] The similarity between the current working condition and historical samples is calculated using weighted Euclidean distance (the smaller the distance, the higher the similarity), as shown in the following formula:

[0319]

[0320] Where, d i w represents the distance between the current working condition and the i-th historical sample. j The feature weights can be set as needed, for example, traffic flow level (0.15), temperature level (0.12), precipitation level (0.10), equipment load level (0.10), visibility level (0.08), power grid stability (0.07), and the total weight of the remaining features is 0.38; 18-dimensional implicit feature weights: equipment dynamic efficiency decay rate (0.12), energy consumption-operating condition nonlinear correlation coefficient (0.10), equipment health correlation feature (0.09), and the total weight of the remaining features is 0.69; The calculation is performed item by item for the range j = 1 to j = 12, corresponding to the 12-dimensional working condition feature vector mentioned above (if there are changes, the changed ones shall prevail; for example, if it is 9-dimensional, then j = 9). For the range j = 13 to j = 30, this corresponds to the term-by-term calculation of the 18-dimensional latent eigenvectors mentioned above;

[0321] G curr，j This is the quantized value of the j-th 12-dimensional feature vector of the current working condition, which has been standardized to a level value of 0-5 (such as traffic flow level, temperature level, visibility level, etc.);

[0322] G hist，i，j The quantized value of the j-th 12-dimensional feature vector of the i-th historical sample (j = 1 to 12), and G curr，j For features of the same type, ensure consistency in calculation;

[0323] V curr，j-12 This refers to the value of the (j-12)th 18-dimensional latent feature vector under the current operating condition (j = 13 to 30, corresponding to latent feature numbers 1 to 18), such as the equipment dynamic efficiency decay rate, energy consumption-operating condition nonlinear correlation coefficient, etc.

[0324] V hist，i，j-12 Let V be the value of the (j-12)th 18-dimensional latent feature vector of the i-th historical sample (j = 13 to 30), and V curr，j-12 These are latent features of the same type.

[0325] Historical working conditions with similarity exceeding a preset threshold are selected as the sample pool P of the same working conditions. If multiple conditions exceed the preset threshold, the one with the highest similarity is selected. If no conditions exceed the preset threshold, the one with the highest similarity is also selected.

[0326] An initial dynamic baseline is generated based on a sample pool with the same operating conditions.

[0327] In this embodiment, generating an initial dynamic baseline based on a sample pool under the same operating conditions includes:

[0328] Based on the energy consumption distribution of the sample pool under the same operating conditions, the kernel density estimation method is used to calculate the normal range of each independent energy consumption component and the total energy consumption, ensuring that the benchmark range can accurately describe the energy consumption distribution pattern under the current operating conditions (even if the energy consumption distribution is not normal).

[0329] Because the distribution patterns of different independent energy consumption components differ significantly (e.g., wind turbine energy consumption is skewed and lighting energy consumption is approximately normally distributed), it is necessary to calculate the baseline separately for each dimension and then superimpose them to obtain the total energy consumption baseline.

[0330] Specifically as follows:

[0331] Extract sample data of the k-th independent energy consumption component from the sample pool P under the same operating conditions.

[0332] S k，pool ={S hist，i，k |(S hist，i E total,hist，i )∈P}S hist，i,k (This refers to the k-th type of independent energy consumption data for the i-th historical sample);

[0333] Construct sample S k，pool probability density function f k (x), formula:

[0334] Where n is the number of samples in sample pool P; h k K is the bandwidth; K(·) is the kernel function, using a Gaussian kernel;

[0335] Baseline interval determination: based on probability density function f k Calculate the cumulative distribution function (x). Take the 99% confidence interval as the baseline interval for the k-th type of independent energy consumption component:

[0336] [L k U k ]={x|F k (x) = 0.005, F k (x) = 0.995};

[0337] The total energy consumption benchmark range is not calculated directly from historical total energy consumption samples, but is obtained by superimposing independent energy consumption benchmarks to ensure consistency with independent energy consumption benchmarks. The formula is as follows:

[0338]

[0339] L total This is the lower limit of the total energy consumption benchmark range, i.e., the sum of the lower limits of all independent energy consumption components; U total The upper limit of the total energy consumption benchmark range, which is the result of the superposition of the benchmark upper limits of all independent energy consumption components;

[0340] Generate a dynamic benchmark dataset based on the initial dynamic benchmark.

[0341] In this embodiment, generating a dynamic benchmark dataset based on an initial dynamic benchmark includes:

[0342] Acquire real-time monitoring data (independent energy consumption, operating condition characteristics, and latent characteristics);

[0343] The proportion of real-time data falling within the baseline range is counted (fit score). If the fit is normal (score ≥ 90%), the real-time data is included in the historical sample database, the sample pool is updated, and the baseline range is fine-tuned (adjustment range ≤ 5%).

[0344] For slight mismatch (score 80%-90%), the baseline range is slightly expanded to accommodate short-term operating condition fluctuations;

[0345] Severe mismatch (score < 80%) indicates a change in operating model, and a new sample library is constructed based on data from the past 3 months, with the baseline recalculated.

[0346] Thus, the iterative independent energy consumption component benchmark (L) is obtained. k U k Total energy consumption benchmark (L) total U total The self-evolutionary log (iteration time, adjustment range) serves as a dynamic benchmark dataset.

[0347] In other words, the dynamic benchmark dataset of this application includes:

[0348] Independent energy consumption component benchmarks: Lower limit of the benchmark range L for each component k upper limit of the benchmark interval U k ;

[0349] Total energy consumption benchmark: Lower limit L of the total energy consumption benchmark range total Total energy consumption benchmark range upper limit U total ;

[0350] Auxiliary data (self-evolution log, benchmark validity report, scene identifier, timestamp).

[0351] Anomaly tracing reports are obtained based on deep feature datasets, dynamic benchmark datasets, and implementation detection data.

[0352] In this embodiment, obtaining an anomaly tracing report based on the deep feature dataset, dynamic benchmark dataset, and implementation detection data includes:

[0353] Compare real-time energy consumption with a dynamic baseline to determine if there is an anomaly. Specifically, anomaly determination for independent energy consumption components: if the real-time value of a certain type of independent energy consumption component exceeds its baseline range (the real-time independent energy consumption value is less than the lower limit of the baseline range minus 5% of the width of the baseline range (L...)). k -0.05×(U k -L k Only after this is it preliminarily determined to be an anomaly (low energy consumption anomaly) or a real-time independent energy consumption value greater than U. total +0.05×(U total -L total If the component fails to recover after two consecutive collection cycles, it is marked as abnormal.

[0354] Total energy consumption anomaly determination: If the real-time value of total energy consumption exceeds the total energy consumption baseline range (for example, the real-time value of total energy consumption is less than L). total -0.05×(U total -L total ) or greater than U total +0.05×(U total -L total This is marked as an anomaly in total energy consumption.

[0355] The energy consumption monitoring method applicable to highway operation proposed in this application has the following advantages:

[0356] The energy consumption monitoring method for highway operation provided in this application acquires traffic flow-related features, environmental features, operational features, equipment load features, stability features, and current scenario information. From these features, it selects input features for an intelligent agent. These features are then input into a trained reinforcement learning agent to obtain the optimal combination of acquisition parameters and frequency decisions. Based on these decisions, a scenario-optimized dataset is acquired, and energy consumption monitoring is conducted in conjunction with scenario-based basic data. This achieves dynamic adaptation of energy consumption acquisition strategies, solving the technical problems of traditional fixed acquisition modes, such as missed key data under complex operating conditions, excessive redundant data under low-value operating conditions, and the difficulty in balancing monitoring accuracy and storage / analysis costs.

[0357] In the energy consumption monitoring process, a scenario database containing preset scenario-based dynamic physical mechanism constraint equations is acquired. A coupled input matrix is ​​constructed by combining the scenario-based optimization dataset with the scenario's basic data. The coupled input matrix is ​​then whitened, and the optimal separation matrix is ​​solved based on the objective function and gradient descent method to obtain the independent energy consumption matrix. A feature extraction network is then used to mine deep feature datasets from the independent energy consumption matrix, and combined with historical operating condition samples to generate a dynamic benchmark dataset. Finally, an anomaly tracing report is generated based on the deep feature dataset, the dynamic benchmark dataset, and real-time detection data. This achieves refined monitoring throughout the entire process, from energy consumption data decoupling and deep feature mining to accurate anomaly tracing, solving the technical problems of traditional energy consumption monitoring that only focuses on surface data, lacks physical meaning in decoupling results, and is prone to misjudgment and missed anomaly detection.

[0358] This application employs a technical approach that integrates reinforcement learning and physical mechanisms. By dynamically sensing highway operating conditions, it enables energy consumption data collection strategies and monitoring and analysis processes to adapt to dynamic scenarios such as traffic flow fluctuations, sudden environmental changes, and equipment load fluctuations, rather than using a fixed monitoring model. Compared to traditional fixed data collection and analysis schemes, this approach avoids monitoring blind spots caused by the lack of key energy consumption information under complex operating conditions, reduces the storage and analysis costs of redundant data under low-value operating conditions, and improves the accuracy of energy consumption decoupling and anomaly tracing, providing more precise technical support for green highway operation and cost control.

[0359] This application also provides an energy consumption monitoring device suitable for highway operation, the energy consumption monitoring device suitable for highway operation comprising:

[0360] The feature acquisition module is used to acquire traffic flow-related features, environment-related features, operation-related features, equipment load features, and stability features.

[0361] Current scene information acquisition module, the current scene information acquisition module is used to acquire current scene information;

[0362] The intelligent agent input feature acquisition module is used to acquire intelligent agent input features from traffic flow-related features, environment-related features, operation-related features, equipment load features, and stability features based on current scene information.

[0363] A reinforcement learning agent acquisition module is used to acquire a trained reinforcement learning agent.

[0364] The optimal acquisition parameter combination and frequency decision acquisition module is used to input the agent input with features into the trained reinforcement learning agent, thereby obtaining the optimal acquisition parameter combination and frequency decision.

[0365] A scenario-based optimized dataset acquisition module is used to acquire a scenario-based optimized dataset based on the optimal combination of acquisition parameters and frequency decision.

[0366] A scene basic data acquisition module, which is used to acquire scene basic data;

[0367] An energy consumption monitoring module is used to monitor energy consumption based on a scenario-optimized dataset and scenario-based basic data.

[0368] It should be noted that the above parameters are designed for verification experiments and are adjustable to a certain extent. The size, shape, position, and program of all devices in this embodiment can be set as needed. The requirements and detailed descriptions in the above embodiments should not be construed as limiting the scope of this invention. Where there is no conflict, all combinations of different technical features should be considered within the scope of this specification. It should be noted that even if modifications or improvements are made by those skilled in the art, as long as these modifications or improvements do not constitute substantial innovation or transformative improvements, they still fall within the original concept of this invention and should be considered within the scope of protection of this invention. The specific scope of protection should be based on the appended claims.

[0369] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for monitoring energy consumption in highway operation, characterized in that, The energy consumption monitoring method applicable to highway operation includes: Acquire traffic flow-related features, environmental features, operational features, equipment load features, and stability features; Obtain current scene information; Based on the current scenario information, the input features for the intelligent agent are obtained from the traffic flow-related features, environment-related features, operation-related features, equipment load features, and stability features; Acquire a trained reinforcement learning agent; The input of the agent is fed into a trained reinforcement learning agent with features to obtain the optimal combination of acquisition parameters and frequency decision. Obtain a scenario-optimized dataset based on the optimal combination of acquisition parameters and frequency decisions; Obtain basic scene data; Energy consumption monitoring is performed based on scenario-based optimization datasets and scenario-based basic data.

2. The energy consumption monitoring method for highway operation as described in claim 1, characterized in that, The energy consumption monitoring based on the scenario-optimized dataset includes: Obtain a scene database, which includes at least one preset scene-based dynamic physical mechanism constraint equation and current scene information associated with each preset scene-based dynamic physical mechanism constraint equation; Construct a coupled input matrix based on the scenario-optimized dataset and the scenario-based basic data; Based on the current scene information, the preset scene-based dynamic physical mechanism constraint equations associated with the current scene information are obtained from the scene database; The whitened data is obtained based on the coupling input matrix; Obtain the separation matrix based on the whitened data; The optimal separation matrix is ​​obtained by solving the separation matrix using gradient descent. The independent energy consumption matrix is ​​obtained based on the optimal separation matrix and the whitened data. Obtain a deep feature dataset based on the independent energy consumption matrix; Generate a dynamic benchmark dataset based on the deep feature dataset; Anomaly tracing reports are obtained based on deep feature datasets, dynamic benchmark datasets, and implementation detection data.

3. The energy consumption monitoring method for highway operation as described in claim 2, characterized in that, The whitened data based on the coupling input matrix is ​​obtained in the following manner: The covariance matrix of e can be obtained using the following formula: Where ∑ is the covariance matrix of e; e is the coupled energy consumption data. Coupled input matrix Perform eigenvalue decomposition on the covariance matrix ∑ of e: ∑=UΛUT T ; Where U is the orthogonal eigenvector matrix and Λ is the diagonal eigenvalue matrix; The whitening conversion is performed using the following formula: Where z represents the whitened data; T represents the matrix transpose operation.

4. The energy consumption monitoring method for highway operation as described in claim 3, characterized in that, The step of obtaining the separation matrix based on the whitened data includes: Obtain the target function; The separation matrix that minimizes the objective function is found through iterative optimization using the gradient descent method.

5. The energy consumption monitoring method for highway operation as described in claim 4, characterized in that, The objective function is as follows: Where J(W) is the objective function value, which needs to be minimized; Let W be the separation matrix to be solved; E[·] is the mathematical expectation operation; det W is the determinant of the separation matrix W; p s (·) represents the probability density function of the independent components; Wz is the product of the separation matrix and the whitened data; λ is the physical constraint weight. The physical constraint matrix; is the expected vector of independent components; T is the transpose of the matrix.

6. The energy consumption monitoring method for highway operation as described in claim 5, characterized in that, The step of iteratively optimizing using the gradient descent method to find the separation matrix that minimizes the objective function includes: Perform initialization, randomly setting the initial value of W; The gradient of the objective function is calculated using the following formula: ； in, The gradient of the objective function with respect to the separation matrix W; W -T φ(Wz)z is the transpose of the inverse of the separation matrix. T is the product of the derivative and the transpose of the whitened data; C is the physical constraint matrix; Wz is the product of the separation matrix and the whitened data; Iteratively update W until the convergence condition is met, thereby obtaining the separation matrix that minimizes the objective function; Let be the expected vector of the independent components.

7. The energy consumption monitoring method for highway operation as described in claim 6, characterized in that, The process of obtaining the independent energy consumption matrix based on the optimal separation matrix and the whitened data includes: The whitened data is processed using the optimal separation matrix to obtain the independent energy consumption components, which are then concatenated into a matrix to obtain the independent energy consumption matrix.

8. The energy consumption monitoring method for highway operation as described in claim 7, characterized in that, The process of obtaining the deep feature dataset based on the independent energy consumption matrix includes: Obtain the operating condition parameter matrix; A high-dimensional input feature matrix is ​​constructed based on the operating condition parameter matrix, the scenario-based optimization dataset, and the independent energy consumption matrix. Obtain the trained feature extraction network; The high-dimensional input feature matrix is ​​input into a trained feature extraction network to obtain a deep feature dataset.

9. The energy consumption monitoring method for highway operation as described in claim 8, characterized in that, The process of generating a dynamic benchmark dataset based on a deep feature dataset includes: Extract energy consumption modeling data, operating condition modeling data, and metadata from the deep feature dataset; Energy consumption modeling data and operating condition modeling data are time-series normalized and modeling matrix is ​​constructed using metadata to obtain a standardized modeling dataset; Obtain a historical operating condition sample database, which includes multiple historical samples; The similarity between the current working condition and each historical sample is calculated based on the standardized modeling dataset and historical samples, thereby obtaining the historical sample with the highest similarity as the sample pool for the same working condition. An initial dynamic baseline is generated based on a sample pool with the same operating conditions. Generate a dynamic benchmark dataset based on the initial dynamic benchmark.

10. An energy consumption monitoring device suitable for highway operation, characterized in that, The energy consumption monitoring device suitable for highway operation includes: The feature acquisition module is used to acquire traffic flow-related features, environment-related features, operation-related features, equipment load features, and stability features. Current scene information acquisition module, the current scene information acquisition module is used to acquire current scene information; The intelligent agent input feature acquisition module is used to acquire intelligent agent input features from traffic flow-related features, environment-related features, operation-related features, equipment load features, and stability features based on current scene information. A reinforcement learning agent acquisition module is used to acquire a trained reinforcement learning agent. The optimal acquisition parameter combination and frequency decision acquisition module is used to input the agent input with features into the trained reinforcement learning agent, thereby obtaining the optimal acquisition parameter combination and frequency decision. A scenario-based optimized dataset acquisition module is used to acquire a scenario-based optimized dataset based on the optimal combination of acquisition parameters and frequency decision. A scene basic data acquisition module, which is used to acquire scene basic data; An energy consumption monitoring module is used to monitor energy consumption based on a scenario-optimized dataset and scenario-based basic data.

Images