Liquid leakage detection method and system for liquid cooling system based on analysis of multiple load scenarios

Abstract

The application relates to the technical field of liquid cooling systems, and discloses a liquid cooling system leakage detection method and system based on analysis of multiple load scenarios, which comprises the following steps: obtaining the response templates of two execution modules arranged in series, extracting the control time intersection of the two as a cooperative time set and dividing the static and dynamic time periods; collecting the flow data of each time period according to a preset period, and obtaining frequency domain data through spectrum conversion; calculating the matching value and the matching comparison value of the response data and the corresponding template, and triggering a leakage early warning if the value exceeds a preset range. The static / dynamic time period data calculation logic can also be refined, the weight coefficient can be adjusted through the time period ratio, an algorithm error early warning can be realized through the addition of an overrun time period analysis, and the collection period can be dynamically adjusted according to the early warning frequency. The application can accurately distinguish between load fluctuations and leakage parameter changes, realize real-time early warning of leakage in the early stage of burst, adapt to multiple load scenarios, improve the detection reliability and adaptability, and meet the fine control requirements of the liquid cooling system.

Description

Technical Field

[0001] This application relates to the technical field of liquid cooling systems, and in particular to a method and system for detecting leakage in liquid cooling systems based on analysis of multiple load scenarios. Background Technology

[0002] With the rapid development of industries such as data centers and artificial intelligence, higher demands are being placed on the heat dissipation efficiency, stability, and energy consumption control of cooling systems. Traditional air-cooling technology, limited by its heat dissipation capacity, cannot meet the long-term stable operation requirements of equipment. Liquid cooling technology, with its advantages of high heat dissipation efficiency, low energy consumption, and low noise, has become the mainstream development direction in the future cooling and heat dissipation field.

[0003] The surge in cooling demand from data centers and the artificial intelligence industry is driving the development of liquid cooling systems towards higher integration and intelligence. The liquid cooling distribution unit (CDU), as the core control component of a liquid cooling system, undertakes key functions such as coolant flow regulation, temperature control, and pressure stabilization. Precise control of the CDU enables optimized allocation of coolant resources, significantly improving the heat dissipation performance of the liquid cooling system and reducing overall system energy consumption. It is a crucial element in ensuring the stable operation of the liquid cooling system.

[0004] Current traditional liquid cooling distribution units employ relatively outdated control methods, with control logic primarily designed around basic functions such as energy saving, consumption reduction, and stable temperature control. While they can detect abnormal conditions in cases of severe leakage, such as destructive pipe ruptures, they cannot provide real-time detection and early warning in the pre-rupture stages, such as pipe leaks, interface leaks, or internal equipment leaks. Summary of the Invention

[0005] In order to identify the abnormal state of the liquid cooling system in the early stage of bursting, this application provides a method and system for detecting liquid leakage in the liquid cooling system based on the analysis of multiple load scenarios.

[0006] Firstly, this application provides a method for detecting leakage in a liquid cooling system based on analysis of multiple load scenarios, employing the following technical solution:

[0007] A method for detecting leakage in a liquid cooling system based on analysis of multiple load scenarios includes the following steps:

[0008] The first execution module obtains the corresponding first response template, and the second execution module obtains the corresponding second response template. The first execution module and the second execution module are set in series. The first response template and the second response template are dynamic response data of the first execution module and the second execution module, respectively.

[0009] The intersection of the control times of the first execution module and the second execution module is obtained as the collaborative time set. Static time periods and dynamic time periods are extracted from the collaborative time set. In the static time period, both the first execution module and the second execution module are in a steady-state adjustment state. In the dynamic time period, both the first execution module and the second execution module are in a dynamic adjustment state, and the adjustment range of the dynamic adjustment state is greater than that of the steady-state adjustment state.

[0010] Based on a preset acquisition cycle, during static periods, the first set of preset flow meters is used to acquire the first set of metering data for the first execution module, and the second set of preset flow meters is used to acquire the second set of metering data for the second execution module; during dynamic periods, the third set of metering data for the first execution module is acquired, and the fourth set of metering data for the second execution module is acquired; a spectrum conversion algorithm is used to obtain the first frequency domain data, the second frequency domain data, the third frequency domain data, and the fourth frequency domain data.

[0011] The first response data is calculated based on the first frequency domain data and the third frequency domain data. The first matching value between the first response data and the first response template is calculated. The second response data is calculated based on the second frequency domain data and the fourth frequency domain data. The second matching value between the second response data and the second response template is calculated. The matching comparison value is calculated based on the first matching value and the second matching value. If the matching comparison value is outside the preset comparison value range, a leakage warning is issued.

[0012] By adopting the above technical solution, a first response template is obtained based on the first execution module, and a second response template is obtained based on the second execution module. The first execution module can be a water pump, and the second execution module can be a valve. Combining the characteristic of the two execution modules being set in series, the control time intersection is extracted to form a collaborative time set, which is then divided into static and dynamic time periods. The static time period is in a steady-state adjustment state, and the dynamic time period is in a dynamic adjustment state. Then, according to a preset acquisition cycle, the first set of flow meters and the second set of flow meters acquire the first, second, third, and fourth metering data sets in the corresponding time periods, respectively. The data is then processed by a spectrum conversion algorithm to obtain the first... The system uses first-frequency domain data, second-frequency domain data, third-frequency domain data, and fourth-frequency domain data to calculate first-response data and second-response data. It then obtains the first and second matching values ​​with the corresponding response templates and gets a matching comparison value. When the matching comparison value exceeds the preset comparison value range, a leakage warning is triggered. This effectively combines the operating characteristics of the liquid cooling system under different load scenarios, accurately identifies the differences in parameter changes caused by load fluctuations and leakage, and successfully achieves real-time detection and early warning of abnormal states in the pre-burst stage, such as pipeline leakage, interface leakage, and internal equipment leakage, meeting the refined control requirements of the liquid cooling system for pre-burst warning.

[0013] Optionally, the static time period may also include the following sub-steps:

[0014] Within a preset data collection period, multiple first flow data from the first group of flow meters and multiple second flow data from the second group of flow meters are acquired. The multiple first flow data and multiple second flow data are cross-arranged and combined to obtain a comprehensive flow data set. The flow fluctuation value is calculated based on the comprehensive flow data set.

[0015] If the flow fluctuation value is within the preset reference fluctuation range, the recorded collection period is used to extract and form a static period.

[0016] The step of calculating the first response data based on the first frequency domain data and the third frequency domain data also includes the following sub-steps:

[0017] Obtain the response frequency in the first response template, and calculate the steady-state energy value by weighting the energy values ​​corresponding to the response frequency in the first frequency domain data and the energy values ​​corresponding to the response frequency in the third frequency domain data. Obtain the first response data based on the response frequency and the corresponding steady-state energy value.

[0018] The step of calculating the first match value between the first response data and the first response template also includes the following sub-steps:

[0019] The first temporary value is obtained by weighting the response frequency as the weighting coefficient and the corresponding energy value as the calculated value; the difference between the first temporary value of the first response data and the first temporary value of the first response template is calculated as the first temporary difference.

[0020] The difference between the energy value of the first response data and the corresponding response frequency in the first response template is used as the weighting coefficient, and the response frequency is used as the calculated value. The weighted calculation is performed to obtain the second temporary value. The difference between the second temporary value of the first response data and the second temporary value of the first response template is calculated as the second temporary difference.

[0021] Calculate the average of the first temporary difference and the second temporary difference as the first matching value.

[0022] By adopting the above technical solution and optimizing the calculation logic of static time period filtering, first response data, and first matching value, the accuracy of traffic parameter analysis under different load scenarios is further improved, making the first response data more consistent with the actual operating conditions and the calculation of the first matching value more reflective.

[0023] Optionally, the step of calculating the steady-state energy value by weighted average may further include the following sub-steps:

[0024] In the most recent static and dynamic time periods, the ratio of the duration of the static time period to the duration of the dynamic time period is calculated as the static time period ratio.

[0025] The weighting coefficients of the energy values ​​of the first frequency domain data are adjusted based on the positive correlation of the static time period ratio, and the weighting coefficients of the energy values ​​of the third frequency domain data are adjusted based on the negative correlation.

[0026] By adopting the above technical solution and dynamically adjusting the weight coefficients according to the ratio of static time periods, the adaptability of steady-state energy value calculation is further improved, and the adaptability of the first response data and the first matching value is strengthened.

[0027] Optionally, the dynamic time period may also include the following sub-steps:

[0028] Within a preset data collection period, multiple first flow data from the first group of flow meters and multiple second flow data from the second group of flow meters are acquired. The multiple first flow data and multiple second flow data are cross-arranged and combined to obtain a comprehensive flow data set. The flow change value is calculated based on the comprehensive flow data set, and the flow change value is positive.

[0029] If the flow rate change value is within the preset reference change range, the recorded collection period is used to extract and form a dynamic period.

[0030] The step of calculating the second response data based on the second frequency domain data and the fourth frequency domain data also includes the following sub-steps:

[0031] Obtain the response frequency in the second response template, and take the response frequency that is higher than the preset set frequency as the dynamic frequency; calculate the dynamic energy value by weighting the energy value corresponding to the dynamic frequency in the second frequency domain data and the energy value corresponding to the dynamic frequency in the fourth frequency domain data respectively, and obtain the second response data based on the dynamic frequency and the corresponding dynamic energy value.

[0032] The step of calculating the second matching value between the second response data and the second response template also includes the following sub-steps:

[0033] The dynamic frequency is used as the weighting coefficient, and the corresponding energy value is used as the calculated value. The dynamic temporary value is obtained by weighted calculation. The difference between the dynamic temporary value of the second response data and the dynamic temporary value of the second response template is calculated as the second matching value.

[0034] By adopting the above technical solution, the filtering logic of dynamic time periods and the calculation logic of the second response data and the second matching value are optimized, and the key parameter characteristics under dynamic adjustment scenarios are obtained, making the second response data more consistent with the dynamic operation and the calculation of the second matching value more reflective.

[0035] Optionally, the step of calculating the dynamic energy value by weighted average may further include the following sub-steps:

[0036] In the most recent operating cycle, among multiple static time periods and multiple dynamic time periods, the ratio of the number of static time periods to the number of dynamic time periods is calculated as the dynamic time period ratio.

[0037] The weighting coefficients of the energy values ​​of the second frequency domain data are adjusted based on the negative correlation of the dynamic time period ratio, and the weighting coefficients of the energy values ​​of the fourth frequency domain data are adjusted based on the positive correlation.

[0038] By adopting the above technical solution and dynamically adjusting the weight coefficients according to the dynamic time period ratio, the adaptability of the calculation of the eastern energy value is further improved, and the adaptability of the second response data and the second matching value is strengthened.

[0039] Optionally, the method further includes the following steps:

[0040] If the flow rate change value is outside the preset reference change range within the preset collection period, the collection period is recorded and used to form the over-limit period.

[0041] Obtain the fifth measurement data group from the first execution module and the sixth measurement data group from the second execution module; use a spectrum conversion algorithm to obtain the fifth and sixth frequency domain data;

[0042] The third response data is calculated based on the fifth and sixth frequency domain data. The third matching value between the third response data and the preset third response template is calculated. If the third matching value is outside the preset matching value range, an algorithm error warning is issued.

[0043] By adopting the above technical solution, and by adding analysis of the over-limit period and processing of the fifth and sixth frequency domain data, combined with the calculation of the third response data and the third matching value, the algorithm error warning is realized, and the accuracy of the algorithm results is verified.

[0044] Optionally, the step of calculating the third response data based on the fifth and sixth frequency domain data further includes the following sub-steps:

[0045] Obtain the response frequency from the third response template, and take the response frequency that is lower than the preset set frequency as the stable frequency; calculate the stable energy value by weighting the energy value corresponding to the dynamic frequency in the fifth frequency domain data and the energy value corresponding to the stable frequency in the sixth frequency domain data, and obtain the third response data based on the stable frequency and the corresponding stable energy value.

[0046] The step of calculating the third match value between the third response data and the third response template also includes the following sub-steps:

[0047] The stable frequency is used as the weighting coefficient, and the corresponding energy value is used as the calculated value. The stable temporary value is obtained by weighted calculation. The difference between the stable temporary value of the third response data and the stable temporary value of the third response template is used as the third matching value.

[0048] By adopting the above technical solution, and by focusing on optimizing the calculation process of the third response data and the third matching value at a stable frequency, the accuracy of the algorithm error warning can be improved.

[0049] Optionally, the step of calculating the steady-state energy value by weighted average may further include the following sub-steps:

[0050] In the most recent operating cycle, the ratio of the total duration of the out-of-limit periods to the total duration of the operating cycle is the out-of-limit period ratio.

[0051] The weighting coefficients of the energy values ​​of the fifth frequency domain data are adjusted based on the negative correlation of the over-limit time period ratio, and the weighting coefficients of the energy values ​​of the sixth frequency domain data are adjusted based on the positive correlation.

[0052] By adopting the above technical solution, the weight coefficient is dynamically adjusted according to the ratio of the over-limit period, which optimizes the dynamic adaptability of the stable energy value calculation and helps to improve the accuracy of the third response data and the third matching value.

[0053] Optionally, the method further includes the following steps:

[0054] In response to algorithm error warnings, the frequency of these warnings is calculated, and the data acquisition cycle is adjusted based on the negative correlation between the warning frequency and the frequency of warnings.

[0055] By adopting the above technical solution, the data acquisition cycle is adjusted negatively according to the frequency of algorithm error warnings, and the data acquisition density is dynamically adapted. This improves the timeliness of data capture when the error risk increases and enhances the efficiency of algorithm error correction.

[0056] Secondly, this application provides a liquid cooling system leakage detection system based on multiple load scenario analysis, employing the following technical solution:

[0057] A liquid cooling system leakage detection system based on multiple load scenario analysis includes a processor, wherein the processor executes the steps of the liquid cooling system leakage detection method based on multiple load scenario analysis as described in any one of the above.

[0058] In summary, this application includes at least one of the following beneficial technical effects: by constructing a response template through serial execution modules, dividing static, dynamic and over-limit collaborative time periods, and combining flow data spectrum conversion, dynamic weight adjustment and matching value calculation, it accurately distinguishes between load fluctuations and changes in leakage parameters, realizes real-time early warning of leakage in the early stage of liquid cooling system burst, adapts to multiple load scenarios, reduces the risk of false alarms and missed alarms through algorithm error verification, and dynamically optimizes the acquisition cycle according to the early warning frequency, thereby improving the accuracy of detection results. Attached Figure Description

[0059] Figure 1 This is a step-by-step diagram of a liquid cooling system leakage detection method based on analysis of multiple load scenarios.

[0060] Figure 2 This is a diagram showing the sub-steps for calculating the first matching value between the first response data and the first response template. Detailed Implementation

[0061] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0062] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0063] This application discloses a method for detecting leakage in a liquid cooling system based on analysis of multiple load scenarios, referring to... Figure 1 and Figure 2 It includes the following steps:

[0064] Step 1: Selecting and cascading execution modules and constructing response templates. A variable frequency centrifugal water pump is selected as the first execution module, and an electric regulating ball valve as the second execution module. Both are connected in series via flanges to the liquid-cooled main circulation pipeline, with identical pipeline inner diameters. The pump outlet is connected to the valve inlet. Under standard leak-free conditions, covering full load, 50% load, and 30% load, the water pump is controlled to switch between 25Hz, 35Hz, and 50Hz, and the valve is adjusted in stages at 30%, 60%, and 90%. Data is continuously collected for 30 minutes for each condition, with a sampling interval of 100ms. The pump flow-pressure and valve differential pressure-flow response data are recorded. After smoothing and filtering, a first response template corresponding to the pump dynamic response dataset and a second response template corresponding to the valve dynamic response dataset are constructed. The templates include response frequencies from 0.1Hz to 10Hz and their corresponding energy values.

[0065] Step 2: Extract the coordinated time set and divide it into static / dynamic time periods. Set the pump control time to 8:00-24:00 daily, and the valve control time to 8:00-23:30 daily. The intersection of these two periods, 8:00-23:30, constitutes the coordinated time set. Using 1 minute as the minimum acquisition period, monitor equipment operating parameters in real time: If the pump frequency fluctuation is ≤±2Hz, the valve opening fluctuation is ≤±3%, and the flow rate fluctuation is within the 0-5% reference fluctuation range for 5 consecutive minutes, this period is marked as a static period. The flow rate fluctuation value is calculated as (maximum flow rate - minimum flow rate) / average flow rate. If the pump frequency fluctuation is >±2Hz or the valve opening fluctuation is >±3%, and the flow rate change is within the 5%-30% reference change range, this period is marked as a dynamic period. The flow rate change value is calculated as |current period average flow rate - previous period average flow rate|.

[0066] Step 3: Collect flow data and perform spectrum conversion. At least one electromagnetic flow meter is installed at both the pump outlet and inlet as the first group of flow meters, and at least one electromagnetic flow meter of the same model is installed at both the valve outlet and inlet. The average of the three data points is taken for each group. A preset acquisition period of 1 second is used: during the static period, 60 first flow data points are collected to form the first metering data group, and 60 second flow data points are collected to form the second metering data group; during the dynamic period, 60 third flow data points are collected to form the third metering data group, and 60 fourth flow data points are collected to form the fourth metering data group. The FFT algorithm is used to zero-padded each data group to an integer power of 2 before frequency domain conversion, extracting the energy values ​​in the 0.1Hz-10Hz frequency band to obtain the first to fourth frequency domain data.

[0067] Step 4: Calculate response data and matching value to trigger leakage warning. Calculate the first response data and first matching value: Extract the response frequencies of 0.1Hz, 0.5Hz, 1Hz, 3Hz, 5Hz, and 10Hz from the first response template. Calculate the static time period ratio = static duration / dynamic duration. Adjust the weights of the first frequency domain data according to the positive correlation with the static time period ratio, and adjust the weights of the third frequency domain data according to the negative correlation. For example, when the ratio is 2, the weights are 0.6 and 0.4 respectively. Calculate the steady-state energy value by weighted averaging to form the first response data. Calculate the first and second temporary differences by weighting twice, and take the average value as the first matching value. Calculate the second response data and the second matching value: Extract dynamic frequencies higher than 3Hz from the second response template, such as 3Hz, 5Hz, and 10Hz. Calculate the dynamic time period ratio = static quantity / dynamic quantity. Adjust the weights of the second frequency domain data based on the negative correlation of the dynamic time period ratio and the positive correlation of the fourth frequency domain data. For example, when the ratio is 2, the weights are 0.3 and 0.7 respectively. Calculate the dynamic energy value using the dynamic frequency as the weight, and take the difference from the template as the second matching value. Calculate the matching comparison value = 0.4 × first matching value + 0.6 × second matching value. If it exceeds the preset range of 0-0.15, trigger a local audible and visual alarm and a warning from the monitoring center.

[0068] The static time period also includes the following sub-steps:

[0069] The preset data acquisition period is 1 minute, consistent with the smallest unit of time segmentation. The first group of flow meters is installed at the outlet and inlet of the first execution module, acquiring 60 first flow data points at a 1-second acquisition cycle, denoted as Q11, Q12, ..., Q160. The second group of flow meters is installed at the outlet and inlet of the second execution module, simultaneously acquiring 60 second flow data points, denoted as Q21, Q22, ..., Q260. The average value of each group of data is taken to reduce measurement error.

[0070] Data is combined using a "timestamp-aligned cross-arrangement" rule. First and second flow rate data are extracted alternately in ascending order of collection timestamp, forming an arrangement structure of "Q11-Q21-Q12-Q22-...-Q160-Q260". This results in a comprehensive flow rate data set containing 120 data points, denoted as Qcomprehensive1 to Qcomprehensive120. This structure can simultaneously characterize the flow coordination characteristics of pumps and valves connected in series, avoiding the limitations of single-module data in reflecting the overall steady-state of the pipeline.

[0071] The flow fluctuation value is calculated by subtracting the mean from the standard deviation, and then de-normalized to ensure comparability across load scenarios. The formula is as follows: Flow fluctuation value V = (σ_Q_comprehensive / μ_Q_comprehensive) × 100%, where σ_Q_comprehensive is the standard deviation of the comprehensive flow data set, calculated using the built-in algorithm of the CDU controller, reflecting the data dispersion, and μ_Q_comprehensive is the arithmetic mean of the comprehensive flow data set, reflecting the central tendency of the data. For example, σ_Q_comprehensive of the comprehensive flow data set = 0.12m. 3 / h, μ_Q_synthesis = 6.0m 3 If / h, then V = (0.12 / 6.0) × 100% = 2.0%.

[0072] The preset reference fluctuation range is 0-5%, which is based on the measured steady-state fluctuation data of the 30%-100% load range under standard operating conditions with no leakage. If the flow fluctuation value V of a certain collection period falls within this range, such as 2.0% as mentioned above, then the period is determined to be a valid steady-state period, its start and end times are recorded, and it is included in the static period set; if V exceeds the range, such as 7.5%, then it is determined to be a non-steady-state period and is not included in the static period set.

[0073] To further refine the calculation of the first response data, the following steps are also included:

[0074] Extract the preset response frequency set F={f1,f2,……,f6}={0.1Hz,0.5Hz,1Hz,3Hz,5Hz,10Hz} from the first response template corresponding to the water pump no leakage dynamic response dataset. This set covers the main characteristic frequencies of the water pump steady-state operation, with low and medium frequencies as the main ones, and high-frequency interference signals are excluded.

[0075] The most recent consecutive static and dynamic time periods are selected as the basis for calculation. The static time period ratio R_static = T_static / T_dynamic, where T_static is the duration of the static time period (in minutes) and T_dynamic is the duration of the adjacent dynamic time period (in minutes). For example, if the most recent static time period is 4 minutes long and the adjacent dynamic time period is 2 minutes long, then R_static = 4 / 2 = 2; if both the static and dynamic time periods are 3 minutes long, then R_static = 1.

[0076] The weight coefficient ω1 of the first frequency domain data (obtained by FFT transformation of static time period traffic data) is positively correlated with R_static, and the weight coefficient ω3 of the third frequency domain data (obtained by FFT transformation of dynamic time period traffic data) is negatively correlated with R_static, and satisfies ω1+ω3=1. The specific adjustment formula is as follows: ω1=R_static / (R_static+1), ω3=1 / (R_static+1). Using the previous example where R_static=2, we calculate ω1=2 / (2+1)≈0.67, ω3=1 / (2+1)≈0.33; if R_static=3, then ω1=0.75, ω3=0.25, realizing dynamic adaptation of weights according to the proportion of static / dynamic time periods.

[0077] Extract the energy values ​​E11, E12, ..., E16 (unit: μV·s) corresponding to each response frequency from the first frequency domain data, and the energy values ​​E31, E32, ..., E36 corresponding to each response frequency from the third frequency domain data. Calculate the steady-state energy value for each response frequency using weighting coefficients: Esteady-statei = ω1 × E1i + ω3 × E3i, i = 1 to 6; Example: For f3 = 1Hz, E13 = 10.5μV·s, E33 = 9.8μV·s, ω1 = 0.67, ω3 = 0.33, then Esteady-state3 = 0.67 × 10.5 + 0.33 × 9.8 ≈ 10.27μV·s.

[0078] Using "response frequency - steady-state energy value" as key-value pairs, the first response data D1={(f1, E steady-state 1), (f2, E steady-state 2), ..., (f6, E steady-state 6)} is formed. This data comprehensively reflects the dynamic response characteristics of the water pump under static load scenarios.

[0079] The detailed calculation steps for the first matching value are as follows:

[0080] The first matching value is achieved through two-dimensional weighted difference calculation or curve graph similarity calculation, both of which are de-normalized to ensure the comparability of the matching results, as detailed below:

[0081] One approach: Weighted average of two temporary values:

[0082] The calculation of the first temporary value and the first temporary difference: The first temporary value uses the response frequency as the weighting coefficient and the energy value as the calculated value. The weighted average formula is: T1=[Σ(f_i×E_i)] / [Σf_i], i=1 to 6, where E_i is the steady-state energy value of the first response data (calculating the current data) or the standard energy value of the first response template (calculating the template data). Example: The first response data Σ(f_i×E_stable_i)=0.1×8.3+0.5×9.2+1×10.27+3×11.5+5×10.1+10×8.0=180.7; Σf_i=0.1+0.5+1+3+5+10=19.6, then T1=180.7 / 19.6≈9.22; the first response template T1'=9.15, which is pre-stored in the CDU controller, and the first temporary difference ΔT1=|9.22-9.15|=0.07.

[0083] The second temporary value and the second temporary difference are calculated as follows: The second temporary value uses the difference between the energy value of the first response data and the template as the weighting coefficient, and the response frequency as the calculated value. The weighted average formula is: T2=[Σ(|E_i-E_i'|×f_i)] / [Σf_i], i=1 to 6, where E_i' is the standard energy value of the corresponding response frequency in the first response template. Example: For f1=0.1Hz, |E_stable1-E1'|=|8.3-8.2|=0.1. After calculating the frequency difference in sequence, Σ(|E_stablei-E_i'|×f_i)=0.1×0.1+0.2×0.5+0.07×1+0.3×3+0.2×5+0.1×10=3.08, then T2=3.08 / 19.6≈0.157; T2'=0 for the first response template, the template's own energy value difference is 0, and the second temporary difference ΔT2=|0.157-0|=0.157.

[0084] The final calculation of the first matching value: First, calculate the average energy value of the first response template E_avg=ΣE_i' / 6, such as E_avg=(8.2+9.0+10.2+11.2+9.9+7.9) / 6≈9.4. Then, take the average of the two temporary differences and divide by E_avg (de-normalize): the first matching value M1=[(ΔT1+ΔT2) / 2] / E_avg=[(0.07+0.157) / 2] / 9.4≈0.0121.

[0085] In other implementations, the following approach can be used: the curve graph similarity method can be used instead of the first temporary difference calculation, and the specific process is as follows:

[0086] The response frequency and steady-state energy value of the first response data D1 are plotted as a characteristic curve L1, and the response frequency and standard energy value of the first response template are plotted as a standard curve L0.

[0087] The cosine similarity S between two curves is calculated using the following formula:

[0088] S=[Σ(E stable i×E_i')] / [√Σ(E stable i)²×√Σ(E_i')²];

[0089] Substituting the example data, we get S≈0.992;

[0090] The first temporary difference ΔT1' = 1 - S. The closer the similarity is to 1, the smaller the difference. This is consistent with the logic of the default scheme. Therefore, ΔT1' = 1 - 0.992 = 0.008.

[0091] The first matching value is calculated using the same formula: M1=[(ΔT1'+ΔT2) / 2] / E_avg=[(0.008+0.157) / 2] / 9.4≈0.0088, ensuring consistency with the output format of the default scheme for easy calculation of subsequent matching comparison values.

[0092] Through the above steps, static time period filtering combines the flow coordination characteristics of the serial dual execution modules, cross-arranges data groups to avoid the one-sidedness of data from a single module, and de-unitized flow fluctuation values ​​improve adaptability across load scenarios; dynamically adjusting the weight coefficient based on the static time period ratio makes the steady-state energy value more consistent with the distribution ratio of static / dynamic loads in actual operation, enhancing the adaptability of the first response data; the two-dimensional calculation or curve similarity of the first matching value takes into account both the weighting influence of response frequency and the degree of deviation of energy value, and provides a flexible precision adjustment method, effectively distinguishing between load fluctuations and parameter changes caused by leakage, and improving the accuracy of leakage early warning judgment.

[0093] The dynamic time period also includes the following sub-steps:

[0094] The preset data collection period is 1 minute, consistent with the smallest unit of time division. The first group of flow meters is installed at the outlet and inlet of the first execution module, acquiring 60 first flow data points at a 1-second acquisition cycle, denoted as Q31, Q32, ..., Q360. The second group of flow meters is installed at the outlet and inlet of the second execution module, simultaneously acquiring 60 second flow data points, denoted as Q41, Q42, ..., Q460. The average value of each group of data is taken to eliminate single-point measurement errors.

[0095] Data is combined using a "timestamp-aligned cross-arrangement" rule. First and second flow rate data are extracted alternately in ascending order of collection timestamp, forming an arrangement structure of "Q31-Q41-Q32-Q42-...-Q360-Q460". This results in a comprehensive flow rate data set containing 120 data points, denoted as Qcomprehensive2 to Qcomprehensive220. This structure can simultaneously characterize the coordinated flow rate changes of series-connected pumps and valves during regulation, avoiding the limitations of single-module data in reflecting the overall dynamic regulation state of the pipeline.

[0096] The flow change value is defined as the absolute value of the difference between the average comprehensive flow rate of the current collection period and the average comprehensive flow rate of the previous collection period, and must be positive. The calculation formula is as follows: Flow change value ΔQ = |μ_Qcurrent - μ_Qprevious| where μ_Qcurrent is the arithmetic mean of the comprehensive flow rate data set of the current collection period, and μ_Qprevious is the arithmetic mean of the comprehensive flow rate data set of the previous collection period, both in meters (m). 3 / h. For example: Current time period μ_Q_sum_current = 7.2m 3 / h, previous time interval μ_Q_sum = 6.0m 3 If the value is / h, then ΔQ = |7.2 - 6.0| = 1.2m 3 / h.

[0097] The preset reference variation range is 5%-30%, based on measured flow rate changes during pump frequency / valve opening adjustments under dynamic control conditions. This range covers dynamic control characteristics within a 30%-100% load range. The range is calculated using the flow rate change rate: Flow rate change rate = ΔQ / μ_Qsum × 100%. Using the previous example, the flow rate change rate = 1.2 / 6.0 × 100% = 20%, falling within the reference variation range. This period is considered a valid dynamic period, and its start and end times are recorded and included in the dynamic period set. If the flow rate change rate = 35%, exceeding the range, it is considered an invalid dynamic period and is not included.

[0098] The detailed calculation steps for the second response data are as follows:

[0099] Response frequency extraction and dynamic frequency filtering: A preset response frequency set F={f1, f2, ..., f6}={0.1Hz, 0.5Hz, 1Hz, 3Hz, 5Hz, 10Hz} is extracted from the second response template (valve leak-free dynamic response dataset). This set covers the main characteristic frequencies of valve dynamic adjustment. The preset setting frequency is 3Hz. Response frequencies higher than the preset frequency (3Hz, 5Hz, 10Hz) are filtered as dynamic frequencies, focusing on high-frequency characteristic signals in the dynamic adjustment process.

[0100] Dynamic time period ratio calculation: Select the most recent operating cycle (set as 24 hours) as the statistical range, and calculate the ratio of the number of static time periods to the number of dynamic time periods, defined as Rdynamic = Nstatic / Ndynamic, where Nstatic is the total number of valid static time periods in that operating cycle, and Ndynamic is the total number of valid dynamic time periods. For example: if Nstatic = 90 and Ndynamic = 60 in 24 hours, then Rdynamic = 90 / 60 = 1.5; if Nstatic = 60 and Ndynamic = 120, then Rdynamic = 0.5.

[0101] Dynamic adjustment of weight coefficients: The weight coefficient ω2 of the second frequency domain data (obtained by FFT transformation of static time period traffic data) is negatively correlated with R_dynamic, and the weight coefficient ω4 of the fourth frequency domain data (obtained by FFT transformation of dynamic time period traffic data) is positively correlated with R_dynamic, and satisfies ω2+ω4=1. The specific adjustment formula is as follows: ω2=1 / (R_dynamic+1), ω4=R_dynamic / (R_dynamic+1). Using the previous example where R_dynamic=1.5, we calculate ω2=1 / (1.5+1)=0.4, ω4=1.5 / (1.5+1)=0.6; if R_dynamic=0.5, then ω2=0.67, ω4=0.33, realizing dynamic adaptation of weights according to the distribution ratio of static / dynamic time periods.

[0102] Dynamic energy value calculation: Extract the energy values ​​E23, E25, and E210 corresponding to each dynamic frequency from the second frequency domain data (unit: μV·s), and the energy values ​​E43, E45, and E410 corresponding to each dynamic frequency from the fourth frequency domain data. Calculate the dynamic energy value for each dynamic frequency using weighting coefficients: Edynamicj = ω2 × E2j + ω4 × E4j, where j corresponds to dynamic frequencies of 3Hz, 5Hz, and 10Hz. Example: For a dynamic frequency f = 3Hz, E23 = 12.3 μV·s, E43 = 14.5 μV·s, ω2 = 0.4, ω4 = 0.6, then Edynamic3 = 0.4 × 12.3 + 0.6 × 14.5 = 13.62 μV·s.

[0103] Second response data construction: Using "dynamic frequency-dynamic energy value" as key-value pairs, the second response data D2={(3Hz, Edynamic 3), (5Hz, Edynamic 5), (10Hz, Edynamic 10)} is formed. This data accurately reflects the dynamic response characteristics of the valve under dynamic load scenarios.

[0104] Refined calculation steps for the second matching value:

[0105] Dynamic Temporary Value Calculation: Using dynamic frequency as the weighting coefficient and energy value as the calculated value, the dynamic temporary value is calculated using a weighted average formula as follows: Tdynamic = [Σ(f_j × E_j)] / [Σf_j], where j corresponds to dynamic frequencies of 3Hz, 5Hz, and 10Hz, and E_j is the dynamic energy value of the second response data (calculating the current data) or the standard dynamic energy value of the second response template (calculating the template data). Example: Σ(f_j × Edynamicj) of the second response data = 3 × 13.62 + 5 × 15.2 + 10 × 11.8 = 234.86; Σf_j = 3 + 5 + 10 = 18, then Tdynamic = 234.86 / 18 ≈ 13.05; Tdynamic' of the second response template = 12.8, pre-stored in the CDU controller, calibrated based on a leak-free dynamic operating condition.

[0106] The second matching value is finally calculated as follows: The second matching value is the absolute value of the difference between the second response data and the dynamic temporary value of the second response template. The formula is as follows: Second matching value M2 = |Tdynamic - Tdynamic'| Substituting the example data, we get: M2 = |13.05 - 12.8| = 0.25. This value directly reflects the response difference between the current dynamic scenario and the standard no-leakage scenario.

[0107] Through the above-mentioned detailed implementation, the dynamic time period screening combines the flow coordination and change characteristics of the series dual execution modules. The cross-arrangement of data groups avoids the limitation that data from a single module cannot capture the overall dynamics of the pipeline. The positive value of the flow change value and the determination of the change rate improve the accuracy of dynamic time period identification. Based on the dynamic time period ratio, the weight coefficient is dynamically adjusted so that the dynamic energy value is more in line with the distribution ratio of static / dynamic time periods in actual operation, which enhances the scenario adaptability of the second response data. The second matching value uses dynamic frequency as the weight, focuses on the core feature differences of dynamic adjustment, and has a simple and targeted calculation logic. It effectively distinguishes between dynamic load fluctuations and parameter changes caused by leakage, and provides accurate dynamic scenario basic data for subsequent matching comparison value calculation, ensuring that even small leaks in the early stage of bursting can still be accurately identified during dynamic adjustment.

[0108] This implementation method establishes an algorithm error verification mechanism by adding a process for identifying excessive time periods, dedicated data processing, and third-response data matching, ensuring the accuracy of the leakage detection algorithm results. The specific method includes the following steps:

[0109] Continuing the basic collection logic of dynamic time period filtering, the preset collection time period is still 1 minute, and the first and second groups are collected in 1 second cycles.

[0110] The first group of flow meters acquires 60 flow data points from the first actuator (pump), forming the fifth measurement data group, denoted as Q51-Q560. The second group of flow meters acquires 60 flow data points from the second actuator (valve), forming the sixth measurement data group, denoted as Q61-Q660. The average value of each data group is taken to eliminate the influence of measurement errors on subsequent analysis. The 60 first and 60 second flow data points acquired simultaneously are cross-arranged to form a comprehensive flow data group of 120 data points, arranged according to the following rule: Q51-Q61-Q52-Q62-…-Q560-Q660. The flow change value is calculated using the formula ΔQ=|μ_Qcomprehensive current - μ_Qcomprehensive previous|, where μ_Qcomprehensive current is the average comprehensive flow rate for the current period, and μ_Qcomprehensive previous is the average comprehensive flow rate for the previous period, converted to a flow change rate: Flow change rate = ΔQ / μ_Qcomprehensive previous × 100%. The preset reference change range is 5%-30%. If the flow rate change rate is <5% or >30%, such as the calculated change rate = 3% or 45%, then the collection period is determined to be an over-limit period, and its start and end times are recorded and included in the over-limit period set.

[0111] The CDU controller has a built-in dedicated storage partition for out-of-limit periods, recording information including: start and end times of the period, flow rate change value, flow rate change rate, and raw data from the first / second set of flow meters. This facilitates subsequent algorithm error tracing and analysis. Simultaneously, the system automatically skips the leakage warning judgment process for that period, avoiding misjudgments caused by abnormal flow data.

[0112] The steps for acquiring and converting the fifth / sixth measurement data sets to the frequency domain are as follows:

[0113] First, zero-padding is performed on the two sets of data, filling 60 data points to 64, which meets the data length requirement of the FFT algorithm.

[0114] The time-domain flow data is converted into frequency-domain data using the FFT algorithm, resulting in the fifth frequency-domain data corresponding to the fifth measurement data group and the sixth frequency-domain data corresponding to the sixth measurement data group.

[0115] The energy values ​​of the 0.1Hz-10Hz frequency band in the two sets of frequency domain data are extracted as the basis for the calculation of the third response data to ensure consistency with the frequency band of the previous frequency domain analysis.

[0116] Detailed calculation steps for third response data:

[0117] Third Response Template Construction: The third response template is a dynamic response dataset collected under standard leak-free operating conditions, simulating over-limit scenarios (such as sudden flow changes or abnormal equipment adjustments). The construction method is as follows:

[0118] In a leak-free pipeline environment, the pump frequency / valve opening is forcibly set to change abruptly through the CDU controller, such as the pump frequency suddenly increasing from 35Hz to 50Hz and the valve opening suddenly decreasing from 60% to 30%, to simulate the change of excessive flow.

[0119] The flow-response data of the water pump and valve in this scenario are collected. After FFT transformation, the response frequency and corresponding energy value of the 0.1Hz-10Hz frequency band are extracted to construct a third response template. The preset response frequency set in the template is consistent with the first and second response templates, F={0.1Hz, 0.5Hz, 1Hz, 3Hz, 5Hz, 10Hz}.

[0120] Stable frequency selection: Using the previously set frequency (3Hz), select response frequencies (0.1Hz, 0.5Hz, 1Hz) lower than the set frequency from the response frequency set of the third response template as stable frequencies, focusing on low-frequency characteristic signals that remain stable under over-limit scenarios, and avoiding high-frequency interference affecting error judgment.

[0121] Stable energy value calculation: Extract the energy values ​​E53, E55, and E510 corresponding to the dynamic frequencies (3Hz, 5Hz, 10Hz, and high-frequency characteristics defined by the preceding dynamic period) from the fifth frequency domain data, and the energy values ​​E61, E62, and E63 corresponding to the stable frequencies (0.1Hz, 0.5Hz, 1Hz) from the sixth frequency domain data. The stable energy value corresponding to each stable frequency is calculated using a weighted average algorithm. The weighting coefficients are preset to be equal, ω5=0.5, ω6=0.5, and the formula is as follows: Establek=ω5×E5j+ω6×E6k, where j corresponds to the dynamic frequency and k corresponds to the stable frequency. A one-to-one matching is used: 0.1Hz matches 3Hz, 0.5Hz matches 5Hz, and 1Hz matches 10Hz. For example, the stable frequency 0.1Hz corresponds to the dynamic frequency 3Hz, E53=11.2μV·s, E61=7.8μV·s, then Estable1=0.5×11.2+0.5×7.8=9.5μV·s; similarly, the stable energy values ​​corresponding to 0.5Hz and 1Hz are calculated as Estable2=10.3μV·s and Estable3=12.1μV·s.

[0122] Third response data construction: Using "stable frequency - stable energy value" as key-value pairs, the third response data D3={(0.1Hz, E-stable 1), (0.5Hz, E-stable 2), (1Hz, E-stable 3)} is formed. This data focuses on the stability characteristics under the over-limit scenario, providing a reliable basis for the algorithm error judgment.

[0123] The detailed calculation steps for the third matching value are as follows:

[0124] Stable Temporary Value Calculation: Using the stable frequency as the weighting coefficient and the energy value as the calculated value, the stable temporary value is calculated using a weighted average formula as follows: T_stable = [Σ(f_k × E_k)] / [Σf_k], where k corresponds to the stable frequency of 0.1Hz, 0.5Hz, and 1Hz, and E_k is the stable energy value of the third response data (calculating the current data) or the standard stable energy value of the third response template (calculating the template data). Example: Σ(f_k × E_stable_k) of the third response data = 0.1 × 9.5 + 0.5 × 10.3 + 1 × 12.1 = 18.2; Σf_k = 0.1 + 0.5 + 1 = 1.6, then T_stable = 18.2 / 1.6 ≈ 11.375; T_stable' of the third response template = 11.1, pre-stored in the CDU controller, calibrated based on a standard over-limit leak-free scenario.

[0125] The final calculation of the third matching value: The third matching value is the absolute value of the difference between the third response data and the stable temporary value of the third response template. The formula is as follows: Third matching value M3 = |T_stable - T_stable'| Substituting the example data, we get: M3 = |11.375 - 11.1| = 0.275.

[0126] Algorithm error warning trigger logic:

[0127] Matching value range setting: The default matching value range is 0-0.3, which is based on the measured data of algorithm error under the standard over-limit leak-free scenario, covering the normal error range of the 95% confidence interval.

[0128] Warning Triggering and Notification Format: If the third matching value M3 falls outside the set matching value range, such as when M3 is calculated to be 0.35, the CDU controller will trigger an algorithm error warning notification.

[0129] Local alert: Yellow indicator light flashes and buzzer sounds intermittently, distinguishing it from the red indicator light and continuous buzzer for leak warning;

[0130] Remote notification: Send an "algorithm error warning" message to the monitoring center, along with source tracing information such as the time period exceeding the limit, the third matching value, and the original traffic data;

[0131] System logging: Automatically stores warning-related data to the log, facilitating technicians to investigate the cause of errors, such as flow meter failure or frequency domain conversion parameter drift.

[0132] Through the above detailed implementation, the screening of excess time periods continues the dual-module traffic collaborative analysis logic, ensuring the accuracy of excess scenario identification. The third response data focuses on stable frequencies, combining dynamic and stable frequency energy values ​​for weighted fusion, enhancing the stability of characteristic signals under excess scenarios and avoiding errors and misjudgments caused by high-frequency interference. The third matching value uses stable frequencies as weights, with a simple and targeted calculation logic, accurately reflecting the degree of algorithm deviation under abnormal traffic scenarios. This mechanism enables real-time verification of the leakage detection algorithm, timely detection and warning of algorithm errors, and effectively reducing the risk of missed or false leaks due to algorithm deviation.

[0133] This implementation method, based on the analysis of excessive time periods and algorithm error early warning, adds dynamic weight adjustment of stable energy values ​​and adaptive optimization logic for the acquisition cycle, further improving the accuracy of algorithm error judgment and error correction efficiency. The specific method includes the following steps:

[0134] Refined steps for dynamic weight adjustment of stable energy values:

[0135] 1. Calculation of the ratio of over-limit periods:

[0136] Statistical range definition: The most recent operating cycle (set to 24 hours) is used as the statistical benchmark. This cycle is consistent with the statistical range of the dynamic time period ratio to ensure the consistency of load scenario analysis.

[0137] Core parameter statistics: Through the built-in statistics module of the CDU controller, all valid over-limit periods in the over-limit period set within 24 hours are extracted, and their total duration is recorded as T_over (unit: minutes); the total duration of the operation cycle T_total = 24 × 60 = 1440 minutes.

[0138] The ratio calculation formula is: Rover = Tover / Ttotal. This ratio reflects the proportion of over-limit scenarios in the overall operation, and its value ranges from [0, 1]. Example: If the cumulative over-limit time Tover = 36 minutes within 24 hours, then Rover = 36 / 1440 = 0.025, or 2.5%; if Tover = 72 minutes, then Rover = 0.05, or 5%.

[0139] 2. Dynamic adjustment rules for weighting coefficients:

[0140] The regulation logic is defined as follows: The weight coefficient ω5 of the fifth frequency domain data (derived from the conversion of water pump flow during the over-limit period) is negatively correlated with R_over. The higher the proportion of over-limit, the lower the reference value of the dynamic frequency energy of the water pump. The weight coefficient ω6 of the sixth frequency domain data (derived from the conversion of valve flow during the over-limit period) is positively correlated with R_over. The higher the proportion of over-limit, the higher the reference value of the stable frequency energy of the valve, and ω5+ω6=1 is satisfied.

[0141] Specific adjustment formula:

[0142] ω5 = 1 - Rsuper / (Rsuper + 0.1);

[0143] ω6 = Rsuper / (Rsuper + 0.1);

[0144] The formula design balances the smoothness of weight adjustment with scenario adaptability, avoiding weight fixation caused by extreme ratios. Example:

[0145] When Rsuper = 0.025 (2.5%), ω5 = 1 - 0.025 / (0.025 + 0.1) = 0.8, ω6 = 0.2;

[0146] When Rsuper = 0.05 (5%), ω5 = 1 - 0.05 / (0.05 + 0.1) ≈ 0.667, ω6 ≈ 0.333;

[0147] When R approaches 0, ω5 approaches 1 and ω6 approaches 0, degenerating into a calculation logic based primarily on the dynamic frequency energy of the water pump.

[0148] 3. Recalculate the stable energy value:

[0149] Following the one-to-one matching rule between stable and dynamic frequencies, 0.1Hz is matched with 3Hz, 0.5Hz with 5Hz, and 1Hz with 10Hz. The stable energy value is calculated based on the dynamic weight coefficient, and the formula is adjusted to: Establek = ω5 × E5j + ω6 × E6k, where j corresponds to the dynamic frequency and k corresponds to the stable frequency. For example, when Roverclock = 0.025, ω5 = 0.8 and ω6 = 0.2; the stable frequency of 0.1Hz corresponds to E53 = 11.2μV·s and E61 = 7.8μV·s, so Estable1 = 0.8 × 11.2 + 0.2 × 7.8 = 9.92μV·s; similarly, the Estable2 corresponding to 0.5Hz is calculated as 0.8 × 13.5 + 0.2 × 9.6 = 12.72μV·s, and the Estable3 corresponding to 1Hz is 0.8 × 15.1 + 0.2 × 11.3 = 14.34μV·s. Compared to the results of equal weight calculation, dynamic weights better reflect the actual load distribution in overload scenarios, improving the accuracy of stable energy value representation.

[0150] Detailed steps for dynamic adjustment of the acquisition cycle:

[0151] 1. Calculation of algorithm error warning frequency:

[0152] Statistical period setting: The statistical period is set to 1 hour to balance the timeliness of error trend capture with statistical reliability, and to avoid erroneous adjustments caused by accidental warnings in a short period of time.

[0153] Alarm frequency definition: Alarm frequency F_alarm = the total number of algorithm error warnings within 1 hour, in units of times / hour, which is counted and updated in real time by the CDU controller. Example: If 3 algorithm error warnings are triggered within 1 hour, then F_alarm = 3 times / hour; if 2 warnings are triggered within 2 hours, then F_alarm = 1 time / hour.

[0154] 2. Data Acquisition Period Adjustment Rules:

[0155] The adjustment of the data acquisition cycle is negatively correlated with the alert frequency. The more frequent the alerts, the higher the data acquisition density and the more timely the error correction. The preset initial acquisition cycle is 1 second, consistent with the static / dynamic time period. The specific adjustment levels are shown in Table 1 below:

[0156] Table 1. Data Acquisition Cycle Adjustment Levels

[0157]

[0158] Table 1 shows the data acquisition cycle adjustment levels, which are used to clarify the data acquisition cycle adjustment values ​​and applicable scenarios corresponding to different algorithm error warning frequencies. By dynamically adjusting the data acquisition cycle, the data acquisition density can be adapted to different error risk scenarios, thereby improving the efficiency of algorithm error correction.

[0159] Example: When the F alarm frequency is 4 times / hour, exceeding the threshold of 3 times, the CDU controller automatically adjusts the collection cycle from 1 second to 0.2 seconds during the over-limit period. If the F alarm frequency drops to 2 times in the following hour, it will be adjusted back to 0.5 seconds. The adjustment process takes effect in real time and only applies to the collection of the fifth and sixth metering data groups during the over-limit period. It does not affect the regular collection logic during static / dynamic periods, thus avoiding additional energy consumption.

[0160] 3. Adjusted data processing adaptation:

[0161] After the data collection period was adjusted, the sample sizes of the fifth and sixth measurement data groups changed synchronously:

[0162] With a collection period of 0.5 seconds, 120 data points are acquired within a 1-minute collection period, and these data points are then cross-arranged and combined to form a comprehensive traffic data set of 240 data points.

[0163] With a collection cycle of 0.2s, 300 data points are acquired within a 1-minute collection period, forming a comprehensive traffic data set of 600 data points.

[0164] The Spectrum Transformation (FFT) algorithm automatically adapts to the data length, ensuring that the data length is an integer power of 2 through zero padding. For example, 120 data points are padded to 128, maintaining the consistency and accuracy of frequency domain transformation.

[0165] By dynamically adjusting the weighting coefficients based on the proportion of out-of-limit periods, the calculation of stable energy values ​​no longer relies on fixed weights but adaptively optimizes based on the actual proportion of out-of-limit scenarios. This makes the third response data more closely reflect the out-of-limit characteristics under different operating loads, significantly improving the accuracy of the third matching value's error judgment. Furthermore, the acquisition cycle is dynamically adjusted to optimize the alert frequency for algorithm error warnings. When the error risk increases, the data acquisition density is increased to capture more refined flow change characteristics, providing richer basic data for algorithm error correction and enhancing error correction efficiency. The synergistic effect of these two mechanisms further improves the algorithm error verification and self-optimization mechanism, effectively reducing the risk of misjudgment due to fluctuations in out-of-limit scenarios or drift in algorithm parameters. This provides accurate leak warnings, algorithm error verification, and acquisition strategy optimization for liquid cooling system leak detection.

[0166] This application also discloses a liquid cooling system leakage detection system based on multiple load scenario analysis, including a processor, wherein the processor executes the steps of the liquid cooling system leakage detection method based on multiple load scenario analysis as described in any one of the above embodiments.

[0167] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for detecting leakage in a liquid cooling system based on analysis of multiple load scenarios, characterized in that, Includes the following steps: The first execution module obtains the corresponding first response template, and the second execution module obtains the corresponding second response template. The first execution module and the second execution module are set in series. The first response template and the second response template are dynamic response data of the first execution module and the second execution module, respectively. The intersection of the control times of the first execution module and the second execution module is obtained as the collaborative time set. Static time periods and dynamic time periods are extracted from the collaborative time set. In the static time period, both the first execution module and the second execution module are in a steady-state adjustment state. In the dynamic time period, both the first execution module and the second execution module are in a dynamic adjustment state, and the adjustment range of the dynamic adjustment state is greater than that of the steady-state adjustment state. The first execution module is a water pump, and the second execution module is a valve. Based on a preset acquisition cycle, during static periods, the first set of preset flow meters is used to acquire the first set of metering data for the first execution module, and the second set of preset flow meters is used to acquire the second set of metering data for the second execution module; during dynamic periods, the third set of metering data for the first execution module is acquired, and the fourth set of metering data for the second execution module is acquired; a spectrum conversion algorithm is used to obtain the first frequency domain data, the second frequency domain data, the third frequency domain data, and the fourth frequency domain data. The first response data is calculated based on the first frequency domain data and the third frequency domain data. The first matching value between the first response data and the first response template is calculated. The second response data is calculated based on the second frequency domain data and the fourth frequency domain data. The second matching value between the second response data and the second response template is calculated. The matching comparison value is calculated based on the first matching value and the second matching value. If the matching comparison value is outside the preset comparison value range, a leakage warning is issued. The data corresponding to the response template are pump flow rate-pressure and valve differential pressure-flow rate.

2. The method for detecting leakage in a liquid cooling system based on multiple load scenarios according to claim 1, characterized in that, The static period also includes the following sub-steps: Within a preset data collection period, multiple first flow data from the first group of flow meters and multiple second flow data from the second group of flow meters are acquired. The multiple first flow data and multiple second flow data are cross-arranged and combined to obtain a comprehensive flow data set. The flow fluctuation value is calculated based on the comprehensive flow data set. If the flow fluctuation value is within the preset reference fluctuation range, the recorded collection period is used to extract and form a static period. The step of calculating the first response data based on the first frequency domain data and the third frequency domain data also includes the following sub-steps: Obtain the response frequency in the first response template, and calculate the steady-state energy value by weighting the energy values ​​corresponding to the response frequency in the first frequency domain data and the energy values ​​corresponding to the response frequency in the third frequency domain data. Obtain the first response data based on the response frequency and the corresponding steady-state energy value. The step of calculating the first match value between the first response data and the first response template also includes the following sub-steps: The first temporary value is obtained by weighting the response frequency as the weighting coefficient and the corresponding energy value as the calculated value; the difference between the first temporary value of the first response data and the first temporary value of the first response template is calculated as the first temporary difference. The difference between the energy value of the first response data and the corresponding response frequency in the first response template is used as the weighting coefficient, and the response frequency is used as the calculated value. The weighted calculation is then performed to obtain the second temporary value. The difference between the second temporary value of the first response data and the second temporary value of the first response template is calculated as the second temporary difference. Calculate the average of the first temporary difference and the second temporary difference as the first matching value.

3. The method for detecting leakage in a liquid cooling system based on multiple load scenarios according to claim 2, characterized in that, The step of calculating the steady-state energy value using a weighted average also includes the following sub-steps: In the most recent static and dynamic time periods, the ratio of the duration of the static time period to the duration of the dynamic time period is calculated as the static time period ratio. The weighting coefficients of the energy values ​​of the first frequency domain data are adjusted based on the positive correlation of the static time period ratio, and the weighting coefficients of the energy values ​​of the third frequency domain data are adjusted based on the negative correlation.

4. The method for detecting leakage in a liquid cooling system based on multiple load scenarios according to claim 1, characterized in that, The dynamic time period also includes the following sub-steps: Within a preset data collection period, multiple first flow data from the first group of flow meters and multiple second flow data from the second group of flow meters are acquired. The multiple first flow data and multiple second flow data are cross-arranged and combined to obtain a comprehensive flow data set. The flow change value is calculated based on the comprehensive flow data set, and the flow change value is positive. If the flow rate change value is within the preset reference change range, the recorded collection period is used to extract and form a dynamic period. The step of calculating the second response data based on the second frequency domain data and the fourth frequency domain data also includes the following sub-steps: Obtain the response frequency in the second response template, and take the response frequency that is higher than the preset set frequency as the dynamic frequency; calculate the dynamic energy value by weighting the energy value corresponding to the dynamic frequency in the second frequency domain data and the energy value corresponding to the dynamic frequency in the fourth frequency domain data respectively, and obtain the second response data based on the dynamic frequency and the corresponding dynamic energy value. The step of calculating the second matching value between the second response data and the second response template also includes the following sub-steps: The dynamic frequency is used as the weighting coefficient, and the corresponding energy value is used as the calculated value. The dynamic temporary value is obtained by weighted calculation. The difference between the dynamic temporary value of the second response data and the dynamic temporary value of the second response template is calculated as the second matching value.

5. The method for detecting leakage in a liquid cooling system based on multiple load scenarios according to claim 4, characterized in that, The step of calculating the dynamic energy value using a weighted average also includes the following sub-steps: In the most recent operating cycle, among multiple static time periods and multiple dynamic time periods, the ratio of the number of static time periods to the number of dynamic time periods is calculated as the dynamic time period ratio. The weighting coefficients of the energy values ​​of the second frequency domain data are adjusted based on the negative correlation of the dynamic time period ratio, and the weighting coefficients of the energy values ​​of the fourth frequency domain data are adjusted based on the positive correlation.

6. The method for detecting leakage in a liquid cooling system based on multiple load scenarios according to claim 1, characterized in that, The method also includes the following steps: If the flow rate change value is outside the preset reference change range within the preset collection period, the collection period is recorded and used to form the over-limit period. Obtain the fifth measurement data group from the first execution module and the sixth measurement data group from the second execution module; use a spectrum conversion algorithm to obtain the fifth and sixth frequency domain data; The third response data is calculated based on the fifth and sixth frequency domain data. The third matching value between the third response data and the preset third response template is calculated. If the third matching value is outside the preset matching value range, an algorithm error warning is issued.

7. The method for detecting leakage in a liquid cooling system based on multiple load scenarios according to claim 6, characterized in that, The step of calculating the third response data based on the fifth and sixth frequency domain data also includes the following sub-steps: Obtain the response frequency from the third response template, and take the response frequency that is lower than the preset set frequency as the stable frequency; calculate the stable energy value by weighting the energy value corresponding to the dynamic frequency in the fifth frequency domain data and the energy value corresponding to the stable frequency in the sixth frequency domain data, and obtain the third response data based on the stable frequency and the corresponding stable energy value. The step of calculating the third match value between the third response data and the third response template also includes the following sub-steps: The stable frequency is used as the weighting coefficient, and the corresponding energy value is used as the calculated value. The stable temporary value is obtained by weighted calculation. The difference between the stable temporary value of the third response data and the stable temporary value of the third response template is used as the third matching value.

8. The method for detecting leakage in a liquid cooling system based on multiple load scenarios according to claim 7, characterized in that, The step of calculating the steady-state energy value using a weighted average also includes the following sub-steps: In the most recent operating cycle, the ratio of the total duration of the out-of-limit periods to the total duration of the operating cycle is calculated as the out-of-limit period ratio. The weighting coefficients of the energy values ​​of the fifth frequency domain data are adjusted based on the negative correlation of the over-limit time period ratio, and the weighting coefficients of the energy values ​​of the sixth frequency domain data are adjusted based on the positive correlation.

9. The method for detecting leakage in a liquid cooling system based on multiple load scenarios according to claim 6, characterized in that, The method also includes the following steps: In response to algorithm error warnings, the frequency of these warnings is calculated, and the data acquisition cycle is adjusted based on the negative correlation between the warning frequency and the frequency of warnings.

10. A liquid cooling system leakage detection system based on multiple load scenario analysis, characterized in that, The system includes a processor that performs the steps of the liquid cooling system leakage detection method based on multiple load scenario analysis as described in any one of claims 1-9.

Images