A multi-scenario building lighting full life cycle energy consumption optimization scheduling method and system

Abstract

This invention discloses a method and system for optimizing and scheduling energy consumption throughout the entire lifecycle of building lighting in multiple scenarios. It relates to the field of intelligent control and energy consumption optimization for building lighting, and addresses shortcomings in unified scenario adaptation, closed-loop optimization throughout the entire lifecycle, cross-scenario knowledge transfer, and collaborative scheduling of multiple subsystems. The system achieves full-chain optimization through a six-layer asynchronous collaborative closed loop: scheduling priorities are allocated using a combined urgency-weighted index U(P); automatic cross-scenario knowledge transfer is achieved using a recursive scenario access tree; global optimal scheduling of the wavefield is achieved by using Fuch chaotic initialization combined with sine and cosine surround search and Tent-Cauchy hybrid mutation to escape local extrema; residual energy consumption information entropy H is adaptively eroded and deposited during start-stop operations; the opportunity to preemptively maintain the deferral margin τ_margin and the comprehensive deferral benefit Benefit_delay is quantified; and switching overhead C_switch admission control prevents scheduling costs from negatively impacting energy-saving benefits.

Description

Technical Field

[0001] This invention relates to the field of intelligent control and energy consumption optimization technology for building lighting, specifically to a method and system for optimizing and scheduling energy consumption throughout the entire life cycle of building lighting in multiple scenarios. Background Technology

[0002] Architectural lighting accounts for 15% to 20% of global building energy consumption. The introduction of LED light sources has greatly enhanced the potential for energy saving in lighting, but the energy consumption patterns vary greatly among different building scenarios (office buildings, hospitals, shopping malls, schools, and industrial plants each have different usage periods, illuminance standards, and personnel density curves), making a single scheduling model unsuitable for all situations.

[0003] Existing technologies CN120264536A, CN119907170A, and CN120475592A employ methods such as simulated annealing and elite differential mutation search for lighting scheduling optimization, which have the following limitations: Limitation 1: Static priority ranking. Existing solutions determine the scheduling priority of each scenario with fixed weights, without considering the dynamic trade-off between energy consumption urgency and device health urgency when the total system load changes. Energy consumption is prioritized under high load, and health is prioritized under low load.

[0004] Limitation 2: Cross-scene alignment is a one-to-one planar approach. Existing cross-architectural knowledge transfer uses a pairwise matching method (DTW+MMD), which has a computational complexity of O(N²), and each new scene needs to be re-aligned with all existing scenes.

[0005] Limitation 3: Limited search strategy. Existing solutions employ fixed search strategies (pure gradient descent or single random perturbation) and do not adaptively switch search modes based on the local terrain features of the wavefield potential surface. While large-scale circular searches are suitable for flat areas, hybrid variation escape is required for rugged areas.

[0006] Limitation 4: Lack of a multi-scale perspective on erosion and deposition. Existing energy consumption peak shaving and valley filling are performed on a single time scale, failing to distinguish between the need for rapid correction of minute-level fluctuations and the need for slow structural adjustment of hour-level trends.

[0007] Limitation 5: Maintenance decisions ignore opportunity costs. Existing solutions only consider energy-saving benefits and failure risks in the benefit assessment of postponing maintenance, without including the opportunity cost of "missing a better scheduling window due to postponed maintenance".

[0008] Limitation 6: Lack of self-learning capability in scheduling handover overhead assessment. The existing solution determines whether to allow policy handover based on a fixed percentage, without adaptively adjusting the rejection threshold based on actual feedback from historical handover benefits. Summary of the Invention

[0009] The technical problem to be solved by this invention is how to construct a multi-scenario building lighting full life cycle energy consumption optimization scheduling method and system that can be uniformly adapted across multiple building scenarios, cover the entire chain optimization of lighting systems from initial configuration to full life cycle retirement, realize cross-scenario scheduling knowledge automatic transfer, have global optimal escape capability and maintain quantitative margin decision-making, and control the scheduling system's own computing and communication overhead to prevent it from eroding the energy-saving benefits of scheduling.

[0010] To solve the above technical problems, the following technical solution is adopted: A method for optimizing and scheduling energy consumption throughout the entire lifecycle of multi-scenario building lighting includes the following steps: S1) Dynamic urgency priority sorting steps: Calculate the energy consumption urgency U_E, equipment health urgency U_H, and neighborhood propagation urgency U_P for each building scenario, and arrange the scheduling task queue in descending order according to the weighted combination U(P_i)=α(t)·U_E+β(t)·U_H+γ·U_P; Where α(t) and β(t) are weighting coefficients that are adaptively adjusted with the total system load L_total(t). The higher the load, the higher α(t) becomes, which increases the priority of energy consumption. The lower the load, the lower α(t) becomes, which increases the priority of health. U_P is the urgency component transmitted from adjacent scenarios that share equipment and facilities or overlap with the usage period of this scenario. S2) Adaptive incremental scene tree alignment step: Using the scene wavefield feature similarity matrix in historical running data as input, a recursive scene access tree T is constructed through hierarchical clustering. The leaf nodes carry the scene feature kernel e_i, and the internal nodes represent the shared wavefield feature basis space of the sub-scene groups. When a new scene is accessed, the matching degree of the wavefield interferometer detector κ is calculated layer by layer from top to bottom along the tree structure. After determining the belonging path, only the alignment parameters of the nodes on that path are updated. Adaptive pruning is performed on low-activity subtrees that have not received new data for a long time to reduce the tree structure maintenance overhead. S3) Terrain-aware hybrid mutation wavefield scheduling steps: Initialize the pulse source intensity distribution of each subsystem of lighting, shading, and air conditioning with Fuch chaotic mapping; In each iteration, calculate the local roughness ρ_rough of the wavefield potential energy surface in the neighborhood of the current search point, and adaptively select the search strategy according to ρ_rough. When ρ_rough is lower than the flatness threshold, perform a large step sine and cosine surround search. When ρ_rough is higher than the ruggedness threshold, start the hybrid mutation escape. The mutation operator is selected from the parameterized mutation family {M_θ} according to the current search stage and stagnation count. After mutation, the new solution is accepted with the improved potential energy surface conditions. S4) Adaptive erosion deposition steps driven by multi-scale information entropy: Calculate the distribution information entropy H_fast and H_slow of the residual wavefield ψ_residual in two time windows, one fast and one slow, respectively; Erosion initiation is controlled by comparing H_fast with the initiation threshold H_start; erosion termination is controlled by comparing H_slow with the stopping threshold H_stop; the deposition direction d_deposit is determined by the energy consumption gradient. G, Information Entropy Gradient The weighted combination of H, the historical depositional direction inertia term d_inertia, and the other three factors determines the determination. S5) Full-lifecycle compensation steps for multi-factor preemption margin postponement: Calculate and maintain the deferable margin based on the inverse function of the wave field amplitude decay model ψ(t)=ψ_max·exp(-(t / τ)^β). τ_margin=τ·[ln(ψ_max / ψ_min)^(1 / β) ln(ψ_max / ψ(t))^(1 / β)] T_maint; Calculate the overall benefits of delaying maintenance: Benefit_delay=C_energy_saving C_risk C_opportunity; Where C_opportunity is the opportunity cost of missing a better scheduling window due to delayed maintenance. Maintenance is delayed only if Benefit_delay>0 and τ_margin>0; otherwise, it is executed immediately. S6) Predictive scheduling handover overhead control steps: Based on the statistical characteristics of historical scheduling handover records, learn the overhead parameters of various types of scheduling handover online; Calculate the predicted switching overhead value C_switch_pred for this strategy change. When C_switch_pred exceeds the adaptive rejection ratio θ_switch(t) of the expected energy saving benefit ΔE_saving for this switch, the switch is rejected. θ_switch(t) is dynamically adjusted based on the recent switch benefit realization rate. When the actual benefit of multiple consecutive switches is lower than expected, the rejection threshold is automatically tightened.

[0011] The working principle of this invention: This invention uses a six-layer asynchronous collaborative closed loop to connect all optimization stages of building lighting from "new scene access" to "lifecycle retirement" into a unified chain. S1 (second-level sorting), S2 (scene access / weekly alignment), S3 (minute-level scheduling), S4 (hourly fine-tuning), S5 (weekly / monthly maintenance decision-making), and S6 (admission control before each switchover). These six layers operate asynchronously at different time granularities, without blocking each other.

[0012] This invention integrates "the degree to which scene energy consumption deviates from the target × the rate of deterioration" and "the degree to which equipment light decay approaches the maintenance threshold" into a single ranking index U(P). Computational resources are prioritized for the most urgent scenes, avoiding ineffective scheduling of scenes already in good condition. N scenes are organized into a recursive scene access tree. When a new scene is accessed, it is recursively aligned only along the "root, target leaf node" path, with a complexity of O(logN) instead of O(N²). When accessing office buildings, the scheduling experience of schools is automatically reused, and reasonable scheduling parameters can be obtained on the first day. Fuch chaotic mapping initialization ensures no blind spots in search coverage. Sine and cosine loop search replaces pure gradient descent to avoid gradient zero stagnation. Tent-Cauchy hybrid mutation (Tent fine-grained local search + Cauchy long-tailed long-jump escape) breaks through local extrema, and elite retention ensures that only improvements are accepted. The information entropy H of the residual energy consumption distribution automatically determines whether peak shaving and valley filling are needed (H is higher than the start threshold) or whether it is sufficiently uniform (H is lower than the stop threshold). The dual thresholds form a hysteresis control band to avoid frequent oscillations. The deposition direction is based on the energy consumption gradient. G (local optimum) and information entropy gradient The weighted combination of H (global uniformity) is determined. The maintenance delay margin τ_margin is derived from the wavefield attenuation model, which is "the maximum safe delay without violating the safe illuminance constraint," and then Benefit_delay is used to save electricity costs. An economic rationality test of the cost of failure risk is conducted, and delays are only implemented when the delay is a net benefit. Before each policy switch, the physical switching overhead C_switch (LED dimming energy consumption + sunshade motor power consumption + air conditioner startup impact) is calculated, and switching is only allowed if C_switch is less than 15%~25% of the expected energy-saving benefits. Simultaneously, traditional periodic full polling and duplicate reporting within the same area are eliminated, saving 65%~72% of network traffic. By employing a six-layer closed-loop collaborative overall technical solution, this invention combines urgency-priority sorting for computing resource allocation, recursive scenario access trees for O(logN) cross-scenario knowledge transfer, chaotic hybrid mutation wave field scheduling for global optimal search, information entropy-driven adaptive erosion deposition for fully automatic peak shaving and valley filling, preemption margin deferral for full-lifecycle compensation for quantitative maintenance decisions, and scheduling switching overhead control to prevent computational consumption backlash. This achieves unified optimization across the entire chain, from cross-scenario access to full-lifecycle retirement. Compared to existing solutions with N separate deployments for each scenario, this invention covers all scenarios with a single six-layer architecture, significantly reducing deployment and maintenance costs. Compared to solutions that only optimize the operational phase, this invention also incorporates maintenance and replacement into the optimization loop, further reducing overall energy consumption throughout the entire lifecycle.

[0013] In the above scheme, for optimization, further, in step S1: The urgency of energy consumption: U_E(scene_i)=|ψ_actual(i) ψ_target(i)| / ψ_target(i)·dψ / dt; The urgency of the device's health: U_H(scene_i)=α(i) / θ_main; Where α(i) is the wave field amplitude attenuation rate of the i-th scene, and θ_main is the preset maintenance threshold; The urgency of the neighborhood propagation: U_P(scene_i)=Σ_{j∈N(i)}σ_share(i,j)·U(j) / |N(i)|; Wherein, N(i) is the set of adjacent scenes that share devices or have overlapping usage periods with scene i, and σ_share(i,j)∈[0,1] is the sharing coupling strength; The weighting coefficients are α(t) = α_0·(1+k_α·L_total(t)) and β(t) = β_0·(1+k_α·L_total(t)). k_β·L_total(t)); Where L_total(t) is the current total system load rate, and k_α and k_β are sensitivity coefficients.

[0014] The design of U_E uses the product of deviation degree × deterioration rate, rather than simply using deviation or rate. The principle is that scenarios with large deviations but improving (dψ / dt is positive, the wave field is returning to the target) should have lower urgency; scenarios with small deviations but deteriorating (dψ / dt is negative and has a large absolute value, the wave field is rapidly deviating from the target) should have higher urgency. The product of these two factors captures both the "current position" and "direction of motion," providing an accurate measure of urgency. U_H = α / θ_main uses the ratio of the decay rate α to the maintenance threshold θ_main as the normalized urgency of the equipment health dimension; the closer the decay is to the threshold, the higher the urgency. The weighting coefficients α(t) and β(t) are dynamically adjusted with the total system load—at high loads, an increase in α(t) increases the weight of energy consumption urgency, while a decrease in β(t) reduces the weight of health urgency; conversely, at low loads, the opposite occurs. This, combined with the neighborhood propagation term γ·U_P, achieves adaptive three-factor priority ranking based on system load. Beneficial effects: By employing a weighted combination U(P) of U_E and U_H as the scheduling priority for multiple scenarios, and considering both the degree of deviation and the rate of deterioration in U_E, this invention ensures that limited computing resources always prioritize the scenarios with the largest deviation and continuous deterioration, avoiding ineffective computation of scenarios that are already in an optimized state, compared to existing schemes that only sort by energy consumption deviation or use round-robin scheduling. In parallel scheduling of multiple scenarios, resource allocation based on U(P) significantly improves the overall scheduling quality compared to equal allocation schemes.

[0015] Furthermore, the adaptive incremental scene tree alignment described in step S2 includes: A similarity matrix is ​​constructed using wave field waveform similarity σ(ψ_i,ψ_j), and a recursive scene access tree T is built from bottom to top using average chain hierarchical clustering. Scenes with higher σ are merged first. The wavefield interferometer detector κ calculates the interferometric matching degree between the current scene wavefield ψ_cur and the representative wavefield ψ_rep of each subspace in each tree layer: μ=|ψ_cur ψ_rep| / (‖ψ_cur‖·‖ψ_rep‖); Choose the subspace with the highest μ as the downlink direction; For subtree nodes that have no new wavefield data injection for T consecutive inactive periods, remove them from the active alignment path to reduce maintenance overhead by pruning the tree structure.

[0016] The tree structure in the preferred scheme reflects hierarchical scene clustering: The scene feature kernel e_i of the leaf node includes four dimensions: illumination standard, usage time distribution, personnel density curve, and safety redundancy coefficient. After Z-score normalization, the kernel is concatenated and then encrypted and stored using SHA-256 hash (taking the first 16 bytes). This ensures that the similarity between different scenes can be compared without revealing specific parameters.

[0017] Internal nodes are automatically clustered based on wave field waveform similarity σ, and scenes with higher σ are merged into the same parent node earlier.

[0018] The tree alignment propagation function is recursively applied hierarchically along the tree path: leaf nodes are mapped individually by the wavefront potential function Φ, and internal nodes are fused by the wavefield interferometer detector κ. This hierarchical division of labor avoids the unreasonable situation in planar alignment where "the most similar and the second most similar scenes participate in alignment with the same weight".

[0019] By employing tree-structured hierarchical recursive alignment instead of traditional pairwise alignment, when a new scene is added, only the node parameters on the path from the root to the leaf node are updated, while the remaining subtrees remain unaffected, reducing the integration complexity from O(N²) to O(logN). The hierarchical fusion mechanism of internal nodes within the tree structure grants higher alignment weights to the most similar scenes, preventing accuracy dilution caused by second-most similar scenes participating in alignment with the same weight. As the scene library expands, the branching capability of the tree structure simultaneously improves alignment accuracy and efficiency—a scalability advantage that planar alignment lacks.

[0020] Furthermore, in step S3: The local roughness ρ_rough=(1 / |Ω|)·Σ_{x∈Ω}|Ω(x) Ω_mean / Ω_range; where Ω is the set of potential energy values of M sampling points in the neighborhood of the current search point, and Ω_mean and Ω_range are the mean and range thereof respectively; When ρ_rough<θ_flat, perform sine-cosine surrounding search, and perform oscillatory perturbation on the pulse source intensity in a direction selected randomly with equal probability by a sine or cosine function; When ρ_rough≥θ_rough, select an operator from the variation family {M_θ} according to the current stagnation count c_stag: When c_stag<N_low, select directional variation with decreasing amplitude; When N_low≤c_stag<N_high, select mixed variation with medium amplitude; When c_stag≥N_high, select large-amplitude long-jump variation.

[0021] Due to the adoption of Fuch chaotic initialization, which has more ergodicity and lower iterative convergence requirements than random initialization, Tent mapping and Logistic mapping, the blind areas of the initial pulse source distributed in the wave field space are significantly reduced. The sine-cosine position update operator replaces pure gradient descent with periodic surrounding search, avoiding stagnation at points with zero gradient. The Tent-Cauchy mixed variation, especially the long-distance jump capability provided by the long-tail distribution of Cauchy variation, endows the search with a mechanism to jump out of "deep and narrow" local extrema. The superposition of the three mechanisms makes the global search capability of the present invention significantly better than that of a single gradient descent or a single meta-heuristic algorithm.

[0022] Further, in step S4: The fast-scale information entropy H_fast is calculated within the sliding time window W_fast, and the slow-scale information entropy H_slow is calculated within the sliding time window W_slow, where W_slow>W_fast; The historical deposition direction inertia term: d_inertia(t)=(1 η)·d_deposit(t 1)+η·Δd; where η is the inertia attenuation coefficient, and Δd is the direction variation of the latest effective deposition; Deposition direction: d_deposit=λ_G· G+λ_H· H+λ_I·d_inertia; where λ_G+λ_H+λ_I=1.

[0023] By using information entropy H as the start / stop criterion for erosion-deposition instead of the fixed-iteration proportional erosion-deposition, "over-smoothing" and invalid computation caused by continuing erosion-deposition after the energy consumption distribution has been fully balanced are avoided. The dual-threshold design forms a hysteresis control band, avoiding frequent start / stop oscillations under a single-threshold criterion. Information entropy gradient The correction of the deposition direction by H changes the valley filling target from "the deepest local valley" to "the location that can improve global uniformity the fastest". In multi-hotspot scenarios, this correction ensures the global efficiency of valley filling.

[0024] Furthermore, in step S5: The maintenance can be postponed with a margin: τ_margin(t)=τ·[ln(ψ_max / ψ_min)^(1 / β) ln(ψ_max / ψ(t))^(1 / β)] T_maint; Where ψ_min is the minimum wave field intensity corresponding to the scene's safe illumination, and T_maint is the time required for maintenance operations; The overall benefits of the postponement: Benefit_delay=C_energy_saving C_risk C_opportunity; Where C_energy_saving is the operating electricity cost saved by delaying the operation to off-peak electricity price periods, C_risk = p_failure(τ_margin)·C_consequence is the product of the probability and consequences of sudden failures during the delay period, and C_opportunity = max(0, C_optimal) C_current represents the opportunity cost of missing a better scheduling window due to delayed maintenance.

[0025] By replacing the binary threshold with a maintenance deferral margin τ_margin derived from the inverse solution of the wave field decay model, the maintenance decision-making process is upgraded from "whether maintenance is needed" to "how long can it be safely postponed at most." Furthermore, the economic rationality test of Benefit_delay ensures that the postponement is a net benefit. Maintenance is no longer limited to a passive response mode of "acting immediately when degradation reaches the threshold," but can be proactively scheduled during off-peak electricity price periods to reduce overall maintenance costs. This mechanism has significant economic benefits in the electricity market.

[0026] Furthermore, in step S6: The predicted switching overhead value C_switch_pred=Σ_kωk·C_k, where C_k includes the state transition energy consumption of each actuator; The adaptive rejection ratio θ_switch(t) = θ_0·(1+ε·f_under(t)); Where θ_0 is the basic rejection threshold, ε is the adjustment step size, and f_under(t)=N_under(t) / N_total(t) is the frequency at which the actual benefit of recent switching is lower than expected. The higher f_under(t) is, the more automatically θ_switch(t) increases, and the stricter the standard for rejecting policy switching. When C_switch_pred > θ_switch(t)·ΔE_saving, the current switch is rejected and the current scheduling configuration is maintained.

[0027] By adopting the handover admission condition of C_switch ≤ θ_switch · ΔE_saving, the system first determines whether the physical overhead of the handover itself will consume the energy-saving benefits before each scheduling policy change, avoiding the paradox that "the scheduling algorithm theoretically calculates an energy-saving solution, but in physical execution, the handover overhead increases the total energy consumption." Removing redundant periodic polling and duplicate reporting within the same region further reduces communication overhead and the processing load on edge nodes. The energy efficiency of the scheduling system itself is guaranteed.

[0028] Furthermore, the six steps operate in a heterogeneous, collaborative manner: S1 continuously updates the priority queue at the second level; S2 is executed and adaptively pruned when a scene is accessed; S3 is executed at the minute level in descending order of the priority output by S1; S4 runs asynchronously in both minute and hourly windows; S5 evaluates τ_margin at the weekly / monthly level; and S6 performs predictive go / no-go judgments before each scheduling strategy change. The six-step closed loop covers all dimensions of optimization, from priority sorting to cross-scene alignment to scheduling optimization to secondary correction to full lifecycle management to overhead control.

[0029] The key design principles between the six layers are: second-level sorting does not affect minute-level scheduling, hour-level fine-tuning does not block weekly maintenance assessment, and full-dimensional coverage is achieved from micro (second-level urgency perception) to macro (year-round life cycle).

[0030] This invention also provides a multi-scenario building lighting full life cycle energy consumption optimization and scheduling system, including: The dynamic urgency priority sorting module is used to calculate the energy consumption urgency U_E, equipment health urgency U_H, and neighborhood propagation urgency U_P of each scenario, and to weight them with adaptive weights α(t) and β(t) and arrange the scheduling task queues in descending order. The adaptive incremental scene tree alignment module is used to construct a recursive scene access tree T through hierarchical clustering. When a new scene is accessed, the matching degree of the wave field interferometer detector κ is calculated layer by layer along the tree structure to determine the belonging path. Only the alignment parameters of the nodes on the path are updated, and adaptive pruning is performed on the inactive subtrees. The terrain-aware hybrid variation wavefield scheduling module is used to initialize the pulse source intensity distribution with Fuch chaotic mapping and adaptively select sine and cosine surround search or hybrid variation escape strategy according to the local roughness ρ_rough of the wavefield potential energy surface. The multi-scale information entropy-driven adaptive erosion deposition module is used to calculate the residual wavefield distribution information entropy H_fast and H_slow in fast and slow dual time windows respectively, and enhance the stability of the deposition direction with the historical deposition direction inertia term. The full-lifecycle compensation module for multi-factor preemption margin postponement is used to calculate the maintenance deferable margin τ_margin based on the inverse function of the wave field amplitude decay model. The decision on whether to postpone maintenance is made through a comprehensive benefit assessment of three factors including opportunity cost. The predictive scheduling switching overhead control module is used to learn the overhead parameters of various types of scheduling switching online, and to determine whether to allow policy changes based on the adaptive rejection ratio θ_switch(t).

[0031] Because it adopts a modular, loosely coupled six-module system architecture, the modules communicate asynchronously via a data bus and message queue, ensuring that the failure of one module will not block the operation of other modules. This architecture provides inherent fault tolerance and independent upgrade capabilities in engineering deployments.

[0032] Furthermore, the terrain-aware hybrid variation wavefield scheduling module also includes: a local roughness estimation submodule, an adaptive strategy selection submodule, and a parameterized variation family management submodule; the multi-scale information entropy-driven adaptive erosion deposition module also includes: a fast and slow dual-window entropy calculation submodule, a historical deposition direction inertia preservation submodule, and a deposition direction three-factor weighted synthesis submodule.

[0033] The internal functions of the S3 chaotic hybrid mutation wavefield scheduling module are subdivided into 5 sub-modules: Fuch chaotic initialization sub-module (once at startup), sine and cosine position update sub-module (every scheduling cycle), stagnation detection sub-module (every iteration), Tent-Cauchy hybrid mutation sub-module (after stagnation triggering), and elite retention selection sub-module (after mutation). The internal functions of the information entropy module are also divided according to the sub-functions of step S4. Attached Figure Description

[0034] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0035] Figure 1 This is a schematic diagram of the overall system architecture according to an embodiment of the present invention.

[0036] Figure 2 This is a schematic diagram of the access tree structure for a recursive scenario.

[0037] Figure 3 This is a schematic diagram of the scheduling process for chaotic mixed variant wavefields.

[0038] Figure 4 This is a schematic diagram of the process for information entropy-driven adaptive erosion deposition.

[0039] Figure 5 A timeline diagram illustrating the postponement of full-life compensation in order to gain leeway. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0041] Example 1 This embodiment provides a multi-scenario building lighting full life cycle energy consumption optimization scheduling method, including a combined urgency priority sorting step S1, a recursive scene access tree alignment step S2, a chaotic hybrid variation wave field scheduling step S3, an information entropy driven adaptive erosion deposition step S4, a preemption margin postponement full life cycle compensation step S5, and a scheduling switching overhead control step S6.

[0042] System deployment environment. See also Figure 1 In this embodiment, the six-layer architecture is deployed on an edge computing server. This server uses an Intel Xeon E-2288G processor (8 cores / 16 threads, 3.7GHz), configured with 32GB DDR4 ECC memory and 2TB NVMe SSD storage. The server connects to the DALI-2 lighting control system, Modbus-TCP shading controller, and BACnet / IP air conditioning control system of each building via a gigabit Ethernet switch. The data acquisition layer deploys illuminance sensors (using BH1750FVI digital illuminance sensors, accuracy ±1 lux, transmitted via RS-485 bus), human infrared sensors (using AM312 miniature human body sensing modules), and smart meters (using commercially available three-phase multi-function power monitoring instruments, supporting the Modbus-RTU protocol, used to collect real-time power data from each subsystem). The server operating system is Ubuntu Server 22.04 LTS, the algorithm modules are implemented in Python 3.10, the core computing parts are accelerated using NumPy and PyTorch, and the message queue uses RedisStreams to implement asynchronous communication between modules.

[0043] Step S1: Combine urgency priority sorting.

[0044] The system performs urgency calculations for each connected building scenario i (this embodiment covers 6 scenarios: office building A, commercial complex B, industrial plant C, hospital D, school E, and underground parking lot F) with a cycle of 1 second.

[0045] First, calculate the energy urgency of each scenario. U_E(i) = |ψ_actual(i) - ψ_target(i)| / ψ_target(i) · dψ / dt, where ψ_actual(i) is the current wave field intensity of the i-th scenario (calculated by combining the real-time total power of all lighting circuits in the scenario with the current natural lighting compensation, in kW), ψ_target(i) is the target wave field intensity determined according to the illuminance standard and personnel density described in Section 6.2 of the scenario, and dψ / dt is estimated by the slope of the linear regression of the most recent 5 sampling points. Then, the equipment health urgency U_H(i) = α(i) / θ_main is calculated, where α(i) is the real-time value of the wave field amplitude attenuation rate of the i-th scenario, θ_main is taken as 0.15, and a maintenance assessment should be triggered when the light attenuation is about 15%.

[0046] The system first calculates the energy urgency of each scenario: U_E(i)=|ψ_actual(i) ψ_target(i)| / ψ_target(i)·dψ / dt; Calculate the device health urgency U_H(i) = α(i) / θ_main; Where ψ_actual(i) is the current wave field intensity of the i-th scene, ψ_target(i) is the target wave field intensity, and dψ / dt is estimated by the linear regression slope of the most recent 5 sampling points.

[0047] Where α(i) is the real-time value of the wave field amplitude attenuation rate.

[0048] Simultaneously calculate the neighborhood propagation urgency U_P(i)=Σ_{j∈N(i)}σ_share(i,j)·U(j) / |N(i)| (where N(i) is the set of neighboring scenes that share the same device or overlap with the usage time of scene i).

[0049] The current total system load L_total(t) = 0.72 (72% load rate). α(t)=0.5·(1+0.4×0.72)=0.644, β(t)=0.3·(1 (0.4 × 0.72) = 0.214, γ=0.142.

[0050] U(P_i) = 0.644·U_E(i) + 0.214·U_H(i) + 0.142·U_P(i), and all scenarios are sorted in descending order of U(P_i) and stored in the scheduling task queue Q.

[0051] For example, at 09:30 on a certain day, office building A has just entered a high-frequency usage period (ψ_actual deviates from ψ_target by about 18% and is still rising) and hospital D is in the peak of surgery period (deviation of 10% but stable). The U(P) of office building A may be 0.72, which is higher than that of hospital D (0.45). Therefore, office building A is at the front of the queue and has priority to obtain the scheduling computing resources of step S3.

[0052] Step S2: Align the recursive scenario access tree.

[0053] This is executed once when a new scenario is introduced, and the tree structure is subsequently updated weekly. See [link / documentation]. Figure 2 .

[0054] Taking office building A as an example, this embodiment assumes that there are five existing scenarios: commercial complex B, industrial plant C, hospital D, school E, and parking lot F.

[0055] The system first calculates the waveform similarity σ between office building A and five existing scenarios (the normalization distance between the typical energy consumption time series of each scenario in the most recent week is calculated by the Dynamic Time Warping (DTW) algorithm; the smaller the normalization distance, the more similar the waveforms).

[0056] The typical lighting load curve of office building A is "peak from 09:00 to 12:00, trough from 12:00 to 14:00 (lunch break), and secondary peak from 14:00 to 18:00". It is most similar to the waveform of school E, which is "peak from 08:00 to 12:00, trough from 12:00 to 14:00, and secondary peak from 14:00 to 17:00" (σ value is 0.12, the highest). It differs the most from the waveform of "balanced load throughout the day" of industrial plant C (σ value is 0.68, the lowest).

[0057] The system automatically clusters office building A and school E into a sub-node of the same internal node "Education and Office".

[0058] When a new scene is accessed, the matching degree of the wave field interferometer detector κ is calculated layer by layer from top to bottom starting from the root node. At the root node level, it is determined whether office building A should enter the "education office" subtree or the "public facilities" subtree. Based on the feature distance calculated by the wave field interferometer detector κ, it is determined whether to enter the "education office" subtree. Within the internal node layer of the "Education and Office" category, the scene feature kernel e_A=[300lux,M-F_09-18,0.15 people / m²,safety factor_0.6] of office building A is mapped to the shared feature space of this internal node by the wavefront potential function Φ, sharing the alignment parameters with school E.

[0059] Office building A obtains the aligned wave field ψ_aligned(A)=ρ(A)·Φ(ψ_A)+ρ(E)·Φ(ψ_E), where ρ(A)=0.6 (the office building's own data weight) and ρ(E)=0.4 (the school's alignment contribution weight).

[0060] Based on this aligned wave field, office building A can obtain a reasonable set of initial scheduling parameters on the first day of access, without having to start data collection and model training cycles from scratch.

[0061] Step S3: Chaotic hybrid variation wavefield scheduling.

[0062] Taking office building A as an example (assuming it ranks first in the priority queue in step S1), this step is executed every 1 minute.

[0063] First, the pulse source intensity is initialized using the Fuch chaotic mapping.

[0064] For the three pulse sources of office building A, the lighting source S_light∈[0,P_rated(A)] (P_rated(A) = the rated total power of office building A lighting 48kW), the shading source S_shade∈[-1,0] (the negative sign indicates that the shading generates pulses that are out of phase with the lighting demand), and the air conditioning source S_HVAC∈[0,P_HVAC_rated(A)], each generates a Fuch chaotic sequence value as the initial intensity: S_light^(0)=0+Fuch(0.317)·48=28.1kW (generating an initial lighting power of 28.1kW). The iteration formula of Fuch chaos is Fuch(z_{n+1})=cos(1 / z_n²). The seed value z_0 takes the millisecond part of the current microsecond-level timestamp of the system and maps it to the (0,1) interval to ensure that the initial distribution of each scheduling is different.

[0065] Then, the sine and cosine position update operators are used for iterative optimization.

[0066] Construct the wave field potential energy function Ω(S_light,S_shade,S_HVAC)=-[Σ_construct-Δ_destruct], where Σ_construct is the total amount of constructive interference (when natural lighting is enhanced, S_light decreases while S_shade increases, resulting in bidirectional energy saving), and Δ_destruct is the total amount of destructive interference (when the peak value of S_light coincides with the peak value of S_HVAC, the heating from lighting increases the air conditioning load).

[0067] In each iteration round n, a sine and cosine update is performed for each pulse source S_i: If the random number rand < 0.5, S_i^(n+1) = S_i^(n) + r1·sin(r2)·|r3·S_best-S_i^(n)|; Otherwise, S_i^(n+1) = S_i^(n) + r1·cos(r2)·|r3·S_best-S_i^(n)|. r1 = a·(1-n / 200), where a is the initial amplitude of 0.5 and N_max = 200 is the maximum number of iterations. Therefore, r1 decreases linearly from 0.5 to 0, and the search radius gradually shrinks.

[0068] r2 is randomly sampled in the range [0, 2π] in each round, r3 is randomly sampled in the range [0, 2], and S_best is the currently found optimal pulse source configuration.

[0069] Perform a stall detection: If the potential energy level improvement δΩ after 10 consecutive iterations of Δn=10 is less than δ_stag=10 -4 This triggers a hybrid mutation.

[0070] Choosing Cauchy mutation (far jump) with probability p_mix=0.6: S_i'=S_i+μ_C·Cauchy(0,1.0)·(S_max-S_min), where the Cauchy distribution with scale parameter σ=1.0 makes the variation length potentially 3~5 times (S_max-S_min) (because the probability density of the Cauchy distribution is still about 0.012 at ±5σ, which is a non-zero value); choosing Tent mutation (fine jump) with probability 0.4: S_i'=S_i+μ_T·(2·Tent(y)-1)·(S_max-S_min), μ_T=0.05·(S_max-S_min).

[0071] After mutation, Ω(S') is recalculated, and accepted only if Ω(S') < Ω(S_current).

[0072] After 200 iterations, the optimal pulse source configuration was output as {S_light*=35.2kW,S_shade*=-0.72,S_HVAC*=18.6kW}. This configuration indicates that under the current natural lighting conditions, the optimal lighting power for office building A is 35.2kW (73.3% of the total rated power of 48kW, due to sufficient sunlight compensation). The shading curtain opening is 72% to balance sunlight utilization and heat radiation blocking. The air conditioning cooling load is 18.6kW (the air conditioning load is reduced by about 12% compared to the fully open shading curtain state due to the reduction of solar heat radiation caused by shading).

[0073] See Figure 3 The complete process of step S3 above is as follows: Fuch chaotic initialization generates the initial intensity distribution of 3 pulse sources, enters the iterative loop, performs sine and cosine position updates in each round, calculates the current Ω and compares it with the previous round, if the improvement is lower than the stagnation threshold for 10 consecutive rounds, triggers hybrid mutation, after mutation, elites are retained and screened, and the optimal configuration is output after reaching the maximum number of iterations.

[0074] Step S4: Information entropy-driven adaptive erosion deposition.

[0075] See Figure 4 This step is executed on a 1-hour cycle. The residual wave field ψ_residual (actual energy consumption distribution - energy consumption distribution of the optimal scheduling scheme, discretized into 24×6=144 grids on a time-space grid, representing 24 time periods × 6 floors of lighting zones) after the execution of the scheduling scheme output by S3 is substituted into the information entropy formula H=-Σ_kp_k·log(p_k) for calculation. p_k=|ψ_residual(k)| / Σ_j|ψ_residual(j)| is the normalized residual energy consumption of the k-th grid.

[0076] In the ψ_residual distribution of a typical hour, the residual peak in the conference room area on the west side of the 3rd floor is approximately 1.8kW during the period from 11:00 to 12:00 (this conference room had a temporary meeting in the morning, and the scheduling plan did not fully cover it), while the residual trough in the corridor area on the 5th floor is approximately -1.2kW (this corridor is currently unoccupied, but the lighting maintains the minimum safe illuminance). The p_k values ​​of each grid show a significant bipolar distribution (a few peaks + a large number of flat areas), and the calculated H=0.73>H_start=0.65, thus triggering erosion-deposition.

[0077] Erosion phase: For the residual peak area in the west meeting room on the 3rd floor (ψ_residual=1.8kW), the scene lithological impedance coefficient ρ_rock=0.55 (office buildings belong to medium impedance scenarios), the base erosion rate ε_0=0.3, the calculated erosion rate ε=0.3·(1.8 / 0.35)·(1 / 0.55)=2.81, and the erosion peak shaving amount E_eroded=2.81·1.8=5.06kW·h (this calculation result means that the lighting power in this area is further reduced to shave the peak in the current hour period, and the duty cycle of the lighting circuit of the west meeting room on the 3rd floor is adjusted down from the current 85% to 68% through the DALI dimming interface).

[0078] Deposition phase: Energy consumption gradient G points to the 5th floor corridor (ψ_residual=-1.2kW, gradient ‖ G‖=0.42), information entropy gradient H points to the restroom area on the 4th floor (although the absolute value of the energy consumption offset in this area is small, it affects the global uniformity, H‖=0.31). Calculate the weighted deposition direction d_deposit with λ_G=0.7, λ_H=0.3, and the total deposition amount D_total=5.06kW·h.

[0079] The 5th floor corridor obtains a deposition amount δ_5=(1.2·0.42 / total weight)·5.06≈2.8kW·h (the corridor illuminance is raised from the safety lower limit to close to the comfort standard), and the 4th floor restroom obtains a deposition amount δ_4≈2.26kW·h. After the deposition is performed, H is recalculated. If H drops to 0.35<H_stop=0.40, the erosion-deposition of this cycle is stopped, indicating that the residual energy consumption distribution has reached full uniformity.

[0080] Step S5: Full-lifecycle compensation for pre-empted margin deferral.

[0081] This step is executed independently for each scenario on a 1-week cycle. Taking industrial plant C as an example, assume its lighting system has been in cumulative operation for 28,000 hours, and the LED lamp model is PhilipsMasterLEDtube1500mm24W840 (initial luminous flux Φ0=3700lm, characteristic decay time constant τ=50,000 hours, decay shape factor β=1.2, that is, a slight accelerated decay type).

[0082] Current wave field intensity ψ(t)=ψ_max·exp(-(28,000 / 50,000)^1.2)=ψ_max·exp(-0.504)=0.604·ψ_max.

[0083] The illuminance safety standard requirement for this factory building is ψ_min≥0.55·ψ_max (which is a proportional conversion of the standard GB50034-2013 "Standard for Lighting Design of Buildings" which stipulates that general lighting for industrial plants should not be less than 100 lux).

[0084] See Figure 5 Calculation maintenance can be postponed with margin: Solve for ψ(t+τ_margin+T_maint)=ψ_min, that is 0.604·ψ_max·exp(-((28,000+τ_hours+48) / 50,000)^1.2)=0.55·ψ_max, Where T_maint = 48 hours (including 2 people × 8 hours / day of lamp replacement work hours).

[0085] The inverse solution yields τ_margin ≈ 4,200 hours (approximately 175 days or 5.8 months). Currently, τ_margin > 0, providing ample safety margin for maintenance.

[0086] Further calculate the overall benefits of delaying maintenance.

[0087] Assuming the local time-of-use electricity price peak-valley difference is 0.35 yuan / kWh, delaying to the off-peak period can save C_energy_saving=175·24·48kW·0.35 / 2≈35,280 yuan (industrial plants operate 24 hours a day, and the off-peak electricity price is about half of the peak price).

[0088] C_risk = p_failure(τ_margin)·C_consequence, where p_failure uses an exponential growth model: p_failure = exp(τ_margin / τ·β)·C_base, and C_base = 10. -5 Based on the basic failure rate (referencing extrapolated reliability data from the TM-21 standard), p_failure ≈ 2.5 × 10⁻⁶. -4 C_consequence = 200,000 yuan (estimated loss due to industrial plant lighting interruption), C_risk ≈ 50 yuan. Benefit_delay = 35,280 - 50 = 35,230 yuan > 0, postponing maintenance is economically rational, and the system postpones this maintenance operation.

[0089] Step S6: Control of scheduling switching overhead.

[0090] After S3 outputs a new scheduling scheme but before execution, the system calculates the switching overhead.

[0091] Taking the S3 output scheme switching from the current configuration (S_light=38.0kW, S_shade=-0.45, S_HVAC=21.0kW) to the optimal configuration (S_light*=35.2kW, S_shade*=-0.72, S_HVAC*=18.6kW) as an example: C_LED = Σ_jP_j·|duty_j_new-duty_j_old|·t_ramp. Assume office building A has 200 dimmable LED lights (average rated power 240W), with an average duty cycle change of |0.733-0.792|=0.059, and a ramp time t_ramp=2.0 seconds (typical dimming transition time for the DALI-2 protocol). C_LED = 200 × 240W × 0.059 × 2.0s ≈ 5,664J ≈ 0.0016kWh.

[0092] C_shade = P_motor·N_motors·|angle_new-angle_old| / v_motor. Assume office building A is equipped with 8 sets of electric roller blinds (motor power 25W / unit), the shading angle change is |0.72-0.45|=0.27 (normalized value, approximately 24.3°), the motor speed v_motor corresponds to 15° / second, C_shade = 25W×8×0.27 / (15 / 90)≈32.4J.

[0093] C_HVAC=P_compressor·I_start_ratio·N_starts.

[0094] During this switchover, the air conditioning compressor was reduced from 21.0kW to 18.6kW. It is a continuously variable speed air conditioner (inverter air conditioner) with no additional start-stop impact. N_starts=0, C_HVAC=0.

[0095] C_switch=0.0016+0.000009+0≈0.0016kWh.

[0096] The expected energy saving benefit of this switchover is ΔE_saving = (38.0-35.2)kW × 1h × (1-C_switch / ΔE) = (38.0-35.2) × 1h ≈ 2.8kWh (direct energy saving from lighting) + approximately 0.5kWh energy saving from air conditioning linkage = a total of 3.3kWh. C_switch / ΔE_saving = 0.0016 / 3.3 = 0.00048 < θ_switch = 0.20. The switching cost is far less than 20% of the energy saving benefit, and the switchover is permitted.

[0097] Meanwhile, the system performs redundant monitoring removal: the traditional full polling of all 200 lamps and 24 sensors per second is changed to on-demand querying, and an active request is only sent to the edge gateway of a certain area when the S3 step needs real-time illuminance and personnel data of that area.

[0098] Data from similar sensors in adjacent areas (such as the illuminance meters on the east and west sides of the third layer) are first fused by the moving average at the edge nodes (within a 30-second window) before being uploaded; duplicate reports are automatically discarded. Testing showed that this optimization reduced the system's control network data traffic by approximately 65%–72% and the CPU utilization of the edge nodes by approximately 35%.

[0099] Example 2 Based on Example 1, this embodiment optimizes the configuration of the combined urgency index in step S1.

[0100] In terms of weight allocation, a differentiated weight allocation scheme is adopted for different types of building scenarios.

[0101] For scenarios with extremely high safety lighting requirements, such as hospitals (D) and schools (E), the urgency weight of equipment health w_H is increased to 0.5 (w_E=0.5) to enhance sensitivity to the risk of equipment degradation, as the consequences of lighting failure in operating rooms are far greater than in offices.

[0102] For scenarios with low safety illumination requirements, such as parking lot F, w_H is reduced to 0.2 (w_E=0.8), focusing almost entirely on energy consumption optimization.

[0103] In the calculation of U_E, the estimation of dψ / dt uses Theil-Sen slope estimation with 10 sampling points (compared to 5 sampling points in Example 1) instead of ordinary least squares regression. The Theil-Sen estimator takes the median value of the slope for each sampling point, which has natural robustness to occasional abnormal sensor readings (such as illuminance spikes caused by transient sunlight).

[0104] This improvement enables U_E to maintain computational stability in environments where sensor data may occasionally be noisy.

[0105] Example 3 This embodiment describes in detail the construction process of the recursive scenario access tree as described in claim 3 (see [link]). Figure 2 ).

[0106] The formation of the tree structure is illustrated using the access process of six typical scenarios of this invention as examples.

[0107] When a scenario is first accessed, there is only one leaf node (e.g., commercial complex B), and the tree structure consists of only single leaf nodes. When the second scenario (industrial plant C) is accessed, the waveform similarity σ(B,C) between the two scenarios is calculated. If σ>σ_merge (merging threshold 0.3), an internal node "Industrial and Commercial Class" is created, with B and C as its child nodes; if σ≤σ_merge, C is an independent leaf node located on a different branch from B at the root node.

[0108] When subsequent scenes are added, the attribution is determined recursively from top to bottom. At the internal node of the i-th layer, the average similarity between the new scene and all sub-scenes under that internal node is calculated, and the branch with the highest average similarity is selected to continue the recursion. The common characteristics of Hospital D and School E are "high-frequency use + fixed time period + high safe illumination", and they are clustered into the "public education and medical care" internal node.

[0109] The construction process of scene feature kernel e_i: The illumination standard (lux), typical usage time vector (24-dimensional 0-1 vector, each dimension indicating whether the hour is a high-frequency usage time), daily average population density (person / m²), and safety redundancy coefficient (0~1, 0 indicates no redundancy requirement, 1 indicates the highest redundancy requirement) of each scene are concatenated into a feature vector after Z-score normalization, and then encrypted and stored as feature kernel e_i by SHA-256 hash (only the first 16 bytes are taken as short identifier).

[0110] This encryption process ensures that while calculating the similarity between scenes, the specific operating parameters of a scene (such as the standard illuminance value of 300 lux for a hospital) are not exposed to the system modules of other scenes, which is suitable for the privacy protection needs of multi-tenant building management platforms.

[0111] Example 4 This embodiment provides a complete configuration of the preferred parameters in step S3, chaotic hybrid variation wavefield scheduling.

[0112] Sine and cosine position update operators: The decay curve of r1 is improved from linear decay to nonlinear decay based on cosine annealing, r1(n)=a_min+0.5·(a_max-a_min)·(1+cos(n·π / N_max)); Where a_max=0.5, a_min=0.001, N_max=200.

[0113] The advantage of cosine annealing is that it decays slowly in the early stage (n is close to 0), giving the global exploration a more ample time window; and it decays faster in the later stage (n is close to N_max), which speeds up the final convergence.

[0114] Random sampling of r2∈[0,2π] uses Sobol low-discrepancy sequence instead of uniform random sampling. The uniformity of sampling points in each interval of Sobol sequence is better than that of pure random sampling, avoiding the problem of "some phase angles being oversampled and some angles being missed" in random sampling.

[0115] Tent-Cauchy mixture mutation parameters: The μ_T of the Tent mutation is improved from a fixed value of 0.05 to an adaptive value, μ_T=0.05·(1-n_stag / N_max)·(1+H_current / H_initial), where n_stag is the number of iterations that the current stag has lasted, and H_current and H_initial are the current and initial population diversity (measured by the standard deviation of the Euclidean distance of the solutions in the population).

[0116] When n_stag increases (without breaking through for a long time) or H_current decreases (the population tends to be homogeneous), μ_T automatically increases to strengthen the local perturbation intensity. The Cauchy mutation σ is improved from a fixed value of 1.0 to σ=max(0.5,min(3.0,1.0+0.1·n_stag)). As the stagling time increases, the expected value of the jump step size gradually increases, continuously breaking through the "sphere of influence" of the local extremum.

[0117] Example 5 This embodiment provides a scenario-differentiated configuration scheme for information entropy dual thresholds and gradient weights in step S4.

[0118] For different building types, the settings for H_start and H_stop need to take into account the energy consumption distribution characteristics of the scene.

[0119] For industrial plant C (which operates continuously 24 hours a day with small load fluctuations and naturally uniform load), H_start is reduced to 0.50 and H_stop is reduced to 0.25 because its energy consumption distribution is naturally relatively uniform. An excessively high start threshold will cause erosion-deposition to hardly be triggered, thus losing the value of secondary optimization.

[0120] For commercial complex B (which has a drastic load shift between daytime and nighttime operations, resulting in natural unevenness), H_start is increased to 0.80 and H_stop is increased to 0.45 to allow for higher unevenness and avoid misinterpreting the inherent difference between operating and closed loads as an anomaly requiring "peak shaving and valley filling".

[0121] The weight configuration of λ_H is also based on scene features.

[0122] For multi-hotspot scenarios (such as a school with three high-energy-consuming areas used simultaneously: classrooms, laboratories, and a library), the information entropy gradient... H has additional value in distinguishing "which hotspot most needs priority peak shaving," in which case λ_H is increased to 0.4. For single-hotspot scenarios (such as a parking lot where only the entrance area has continuous high energy consumption, and the rest of the area has occasional low energy consumption), the energy consumption gradient... G is already clear enough in direction, so λ_H is reduced to 0.15 to reduce unnecessary calculations for direction correction.

[0123] Example 6 This embodiment provides a refined extension of the switching overhead model in step S6.

[0124] Based on C_LED=ΣP_j·|Δduty_j|·t_ramp in Example 1, an additional starting surge loss term for the LED driver switching power supply is added: C_LED_startup=N_switched·P_surge·t_surge, where N_switched is the number of lamps starting from zero duty cycle during this switch (i.e., lamps switching from off to on), P_surge=3.5·P_rated_j is the driver starting surge power multiple (refer to the measured surge current data of commercially available MeanWell HLG series LED drivers), and t_surge=0.15 seconds is the surge duration.

[0125] This refined modeling provides a more accurate estimate of switching overhead in scheduling schemes that involve a large number of "off, on" switches (such as turning on all the lights in the building in the morning), avoiding underestimating the energy cost of such large-scale cold starts.

[0126] Dynamic adjustment of θ_switch: During peak electricity price periods (such as 09:00~12:00, 17:00~21:00), θ_switch is reduced from the default 0.20 to 0.12. Switching energy consumption under peak electricity price has a higher economic cost, and the system more conservatively refuses to switch.

[0127] During off-peak electricity prices (such as 23:00~07:00), θ_switch is increased to 0.30. The economic cost of switching energy consumption is lower under off-peak electricity prices, and the system can more aggressively allow switching to approach the theoretically optimal scheduling scheme.

[0128] Example 7 This embodiment provides a multi-scenario building lighting full lifecycle energy consumption optimization and scheduling system that implements all the method steps of the present invention. See [link to documentation]. Figure 1 .

[0129] The system comprises six functional modules, which communicate asynchronously via RedisStreams message queues and shared memory: The urgency priority sorting module runs in 1-second cycles, reads the current ψ_actual and α from the real-time database of each scenario, calculates U(P) and maintains a priority queue, and outputs the queue to the message queue topic "priority_queue".

[0130] The recursive scene access tree alignment module automatically updates the tree structure when a scene access event is triggered or every Sunday morning. It uses Python's scikit-learn AgglomerativeClustering (ward linking criterion) to achieve automatic scene clustering and outputs the aligned wavefield ψ_aligned to the shared memory region "aligned_wavefields".

[0131] The chaotic hybrid mutation wave field scheduling module reads ψ_aligned from shared memory as the initial condition, executes all sub-steps of S3 in priority queue order with a 1-minute period, and outputs the optimal pulse source configuration to the message queue topic "optimal_pulses". The Fuch chaotic sequence generation of this module adopts a hardware entropy source, using hardware random numbers generated by the server CPU's RDRAND instruction as seeds to ensure that the chaotic sequence scheduled each time is not reproducible.

[0132] The information entropy-driven adaptive erosion deposition module subscribes to the topic "optimal_pulses", calculates H in 1-hour cycles, performs erosion-deposition when H > H_start, and outputs the corrected scheduling scheme to the message queue topic "refined_schedule".

[0133] The preemptive margin delay full lifecycle compensation module subscribes to the topic "refined_schedule", reads the cumulative runtime and ψ historical records of each scenario on a 1-week cycle, calculates τ_margin and Benefit_delay, and outputs maintenance decisions to the alarm system and WeChat message push interface.

[0134] The scheduling switching overhead control module intercepts each scheduling change message for topics "optimal_pulses" and "refined_schedule". Before execution, it calculates C_switch and compares it with ΔE_saving. If successful, it sends the message to the DALI / Modbus / KNX protocol gateway for execution; otherwise, it records the reason for rejection and marks it on the monitoring dashboard. The redundancy monitoring and removal submodule uses stream processing rules (eBPF filters, which directly filter duplicate sensor data packets in the Linux kernel space) on the edge gateway to achieve data deduplication and on-demand querying.

[0135] Although the illustrative specific embodiments of the present invention have been described above to enable those skilled in the art to understand the invention, the invention is not limited to the scope of the specific embodiments. For those skilled in the art, all inventions utilizing the concept of the present invention are protected as long as various variations are within the spirit and scope of the invention as defined and determined by the appended claims.

Claims

1. A multi-scenario building lighting life-cycle energy consumption optimization scheduling method, characterized in that, Includes the following steps: S1) Dynamic urgency priority sorting steps: Calculate the energy consumption urgency U_E, equipment health urgency U_H, and neighborhood propagation urgency U_P for each building scenario, and arrange the scheduling task queue in descending order according to the weighted combination U(P_i)=α(t)·U_E+β(t)·U_H+γ·U_P; Where α(t) and β(t) are weighting coefficients that are adaptively adjusted with the total system load L_total(t). The higher the load, the higher α(t) becomes, which increases the priority of energy consumption. The lower the load, the lower α(t) becomes, which increases the priority of health. U_P is the urgency component transmitted from adjacent scenarios that share equipment and facilities or have overlapping usage periods with this scenario; S2) Adaptive incremental scene tree alignment step: Using the scene wave field feature similarity matrix in the historical running data as input, a recursive scene access tree T is constructed through hierarchical clustering. The leaf nodes carry the scene feature kernel e_i, and the internal nodes represent the shared wave field feature basis space of the sub-scene groups. When a new scene is introduced, the matching degree of the wave field interferometer detector κ is calculated layer by layer from top to bottom along the tree structure. After the belonging path is determined, only the alignment parameters of the nodes on that path are updated. Adaptive pruning is performed on inactive subtrees that have not received new data for a long time to reduce the maintenance overhead of the tree structure. S3) Terrain-aware hybrid mutation wavefield scheduling steps: Initialize the pulse source intensity distribution of each subsystem of lighting, shading, and air conditioning with Fuch chaotic mapping; In each iteration, calculate the local roughness ρ_rough of the wavefield potential energy surface in the neighborhood of the current search point, and adaptively select the search strategy according to ρ_rough. When ρ_rough is lower than the flatness threshold, perform a large step sine and cosine surround search. When ρ_rough is higher than the ruggedness threshold, start the hybrid mutation escape. The mutation operator is selected from the parameterized mutation family {M_θ} according to the current search stage and stagnation count. After mutation, the new solution is accepted with the improved potential energy surface conditions. S4) Adaptive erosion deposition steps driven by multi-scale information entropy: Calculate the distribution information entropy H_fast and H_slow of the residual wavefield ψ_residual in two time windows, one fast and one slow, respectively; Erosion initiation is controlled by comparing H_fast with the initiation threshold H_start; erosion termination is controlled by comparing H_slow with the stopping threshold H_stop; the deposition direction d_deposit is determined by the energy consumption gradient. G, Information Entropy Gradient The weighted combination of H, the historical depositional direction inertia term d_inertia, and the other three factors determines the determination. S5) Full-lifecycle compensation steps for multi-factor preemption margin postponement: Calculate and maintain the deferable margin based on the inverse function of the wave field amplitude decay model ψ(t)=ψ_max·exp(-(t / τ)^β). τ_margin=τ·[ln(ψ_max / ψ_min)^(1 / β) ln(ψ_max / ψ(t))^(1 / β)] T_maint; Calculate the overall benefits of delaying maintenance: Benefit_delay=C_energy_saving C_risk C_opportunity; Where C_opportunity is the opportunity cost of missing a better scheduling window due to delayed maintenance. Maintenance is delayed only if Benefit_delay>0 and τ_margin>0; otherwise, it is executed immediately. S6) Predictive scheduling handover overhead control steps: Based on the statistical characteristics of historical scheduling handover records, learn the overhead parameters of various types of scheduling handover online; calculating a predicted handover overhead value C_switch_pred for the current policy change, and rejecting the handover when C_switch_pred exceeds the adaptive rejection ratio θ_switch(t) of the current expected energy saving gain ΔE_saving from the handover, wherein θ_switch(t) is dynamically adjusted according to the recent handover gain realization rate, and the rejection threshold is automatically tightened when the actual gains of consecutive multiple handovers are lower than expected.

2. The multi-scenario building lighting full-lifecycle energy consumption optimization scheduling method according to claim 1, characterized in that in step S1: the energy consumption urgency: U_E(scene_i)=|ψ_actual(i) ψ_target(i)| / ψ_target(i)·dψ / dt; the equipment health urgency: U_H(scene_i)=α(i) / θ_main; wherein α(i) is the wave field amplitude attenuation rate of the i-th scenario, and θ_main is a preset maintenance threshold; the neighborhood propagation urgency: U_P(scene_i)=Σ_{j∈N(i)}σ_share(i,j)·U(j) / |N(i)|; wherein N(i) is a set of adjacent scenarios that share equipment or have overlapping usage periods with scenario i, and σ_share(i,j)∈[0,1] is the shared coupling strength; The weight coefficient α(t)=α_0·(1+k_α·L_total(t)), β(t)=β_0·(1 k_β·L_total(t)); wherein L_total(t) is the current total system load rate, and k_α and k_β are sensitivity coefficients.

3. The multi-scenario building lighting full life cycle energy consumption optimization and scheduling method according to claim 1, characterized in that, The adaptive incremental scenario tree alignment described in step S2 includes: constructing a similarity matrix through the wave field waveform similarity σ(ψ_i,ψ_j), and constructing a recursive scenario access tree T from bottom to top by average linkage hierarchical clustering, wherein scenarios with higher σ are merged first; the wave field interference detector κ calculates the interference matching degree between the current scenario wave field ψ_cur and the representative wave field ψ_rep of each subspace at the current tree layer: μ=|ψ_cur ψ_rep| / (‖ψ_cur‖·‖ψ_rep‖); selecting the subspace with the highest μ as the downlink direction; removing subtree nodes that have no new wave field data injected for consecutive T_inactive cycles from the active alignment path, so as to prune the tree structure and reduce maintenance overhead.

4. The multi-scenario building lighting full life cycle energy consumption optimization and scheduling method according to claim 1, characterized in that, In step S3: The local roughness ρ_rough=(1 / |Ω|)·Σ_{x∈Ω}|Ω(x) Ω_mean| / Ω_range; wherein Ω is the set of potential energy surface values of M sampling points in the neighborhood of the current search point, and Ω_mean and Ω_range are respectively the mean value and the range thereof; when ρ_rough<θ_flat, executing a sine-cosine surrounding search, and performing oscillatory perturbation on the pulse source intensity in a direction randomly selected with equal probability by a sine or cosine function; when ρ_rough≥θ_rough, selecting an operator from the mutation family {M_θ} according to the current stagnation count c_stag: when c_stag<N_low, selecting a directional mutation with decreasing amplitude; when N_low≤c_stag<N_high, selecting a mixed mutation with medium amplitude; when c_stag≥N_high, selecting a large-amplitude long-jump mutation.

5. The multi-scenario building lighting full life cycle energy consumption optimization and scheduling method according to claim 1, characterized in that, In step S4: the fast-scale information entropy H_fast is calculated within a sliding time window W_fast, the slow-scale information entropy H_slow is calculated within a sliding time window W_slow, and W_slow>W_fast; the historical deposition direction inertia term: d_inertia(t)=(1 η)·d_deposit(t 1)+η·Δd; wherein η is the inertia attenuation coefficient, and Δd is the direction variation of the latest effective deposition. Deposition direction: d_deposit=λ_G· G+λ_H· H+λ_I·d_inertia; Where λ_G+λ_H+λ_I=1.

6. The multi-scenario building lighting full life cycle energy consumption optimization and scheduling method according to claim 1, characterized in that, In step S5: The maintenance can be postponed with a margin: τ_margin(t)=τ·[ln(ψ_max / ψ_min)^(1 / β) ln(ψ_max / ψ(t))^(1 / β)] T_maint; Where ψ_min is the minimum wave field intensity corresponding to the scene's safe illumination, and T_maint is the time required for maintenance operations; The overall benefits of the postponement: Benefit_delay=C_energy_saving C_risk C_opportunity; Where C_energy_saving is the operating electricity cost saved by delaying the operation to off-peak electricity price periods, C_risk = p_failure(τ_margin)·C_consequence is the product of the probability and consequences of sudden failures during the delay period, and C_opportunity = max(0, C_optimal) C_current represents the opportunity cost of missing a better scheduling window due to delayed maintenance.

7. The multi-scenario building lighting full life cycle energy consumption optimization and scheduling method according to claim 1, characterized in that, In step S6: The predicted switching overhead value C_switch_pred=Σ_kωk·C_k, where C_k includes the state transition energy consumption of each actuator; The adaptive rejection ratio θ_switch(t) = θ_0·(1+ε·f_under(t)); Where θ_0 is the basic rejection threshold, ε is the adjustment step size, and f_under(t)=N_under(t) / N_total(t) is the frequency at which the actual benefit of recent switching is lower than expected. The higher f_under(t) is, the more automatically θ_switch(t) increases, and the stricter the standard for rejecting policy switching. When C_switch_pred > θ_switch(t)·ΔE_saving, the current switch is rejected and the current scheduling configuration is maintained.

8. The multi-scenario building lighting full life cycle energy consumption optimization and scheduling method according to claim 1, characterized in that: The six steps operate in a heterogeneous, collaborative manner: S1 continuously updates the priority queue at the second level; S2 is executed and adaptively pruned when a scene is accessed; S3 is executed at the minute level in descending order of the priority output by S1; S4 runs asynchronously in both minute and hourly windows; S5 evaluates τ_margin at the weekly / monthly level; and S6 performs predictive go / no-go judgments before each scheduling strategy change. The six-step closed loop covers all dimensions of optimization, from priority sorting to cross-scene alignment to scheduling optimization to secondary correction to full lifecycle management to overhead control.

9. A multi-scenario building lighting full life cycle energy consumption optimization and scheduling system, characterized in that, include: The dynamic urgency priority sorting module is used to calculate the energy consumption urgency U_E, equipment health urgency U_H, and neighborhood propagation urgency U_P of each scenario, and to weight them with adaptive weights α(t) and β(t) and arrange the scheduling task queues in descending order. The adaptive incremental scene tree alignment module is used to construct a recursive scene access tree T through hierarchical clustering. When a new scene is accessed, the matching degree of the wave field interferometer detector κ is calculated layer by layer along the tree structure to determine the belonging path. Only the alignment parameters of the nodes on the path are updated, and adaptive pruning is performed on the inactive subtrees. The terrain-aware hybrid variation wavefield scheduling module is used to initialize the pulse source intensity distribution with Fuch chaotic mapping and adaptively select sine and cosine surround search or hybrid variation escape strategy according to the local roughness ρ_rough of the wavefield potential energy surface. The multi-scale information entropy-driven adaptive erosion deposition module is used to calculate the residual wavefield distribution information entropy H_fast and H_slow in fast and slow dual time windows respectively, and enhance the stability of the deposition direction with the historical deposition direction inertia term. The full-lifecycle compensation module for multi-factor preemption margin postponement is used to calculate the maintenance deferable margin τ_margin based on the inverse function of the wave field amplitude decay model. The decision on whether to postpone maintenance is made through a comprehensive benefit assessment of three factors including opportunity cost. The predictive scheduling switching overhead control module is used to learn the overhead parameters of various types of scheduling switching online, and to determine whether to allow policy changes based on the adaptive rejection ratio θ_switch(t).

10. The multi-scenario building lighting full life cycle energy consumption optimization and scheduling system according to claim 9, characterized in that, The terrain-aware hybrid variation wavefield scheduling module further includes: a local roughness estimation submodule, an adaptive strategy selection submodule, and a parameterized variation family management submodule; the multi-scale information entropy-driven adaptive erosion deposition module further includes: a fast and slow dual-window entropy calculation submodule, a historical deposition direction inertia preservation submodule, and a deposition direction three-factor weighted synthesis submodule.

Images